ECONOMIC AND BUSINESS REVIEW VOLUME 21 j NUMBER 2 j 2019 j ISSN 1580 0466 Andrej Kuštrin A DSGE Model for the Slovenian Economy: Model estimates and Application Barbara Čater, Julijana Serafimova The Influence of Socio-Demographic Characteristics on Environmental Concern and Ecologically Conscious Consumer Behaviour among Macedonian Consumers Byunghoon Jin, John C Cary Are Middle Managers' Cost Decisions Sticky? Evidence from the Field Črt Lenarčič Inflation - the Harrod-Balassa-Samuelson effect in a DSGE Model Setting Eva Repovš, Mateja Drnovšek, Robert Kaše Change ready, resistant, or both? Exploring the concepts of individual change readiness and resistance to organizational change E/B/R E/B/R ECONOMIC AND BUSINESS REVIEW E / D / D Economic and Business Review is a refereed journal that aims to further the research and dis- seminate research results in the area of applied business studies. Submitted papers could be conceptual, interpretative or empirical studies. Literature reviews, conceptual frameworks and models that cover substantive insights are welcomed. Papers that employ either qualitative or quantitative research approaches, surveys, case studies, experiments or historical models can be considered for publication. It is expected that submitted articles contribute to increased understanding of the phenomenon studied and are efficiently written, methodologically clear, conceptually and empirically rigorous, readable and bias free. Authors need to highlight how the paper adds to domain knowledge or improves existing knowledge in the area of applied business research. The topics covered by the journal include (alphabetically): Accounting and Auditing, Bank and Financial Management, Business Informatics, Business Logistics, Economics, Entrepre-neurship and small business management, International Business, Management, Marketing, Money and Finance, Tourism. Today's Economic and Business Review is the successor to the Economic Review (1950-1990), the Slovenian Economic Review (1991-1999) and the Economic and Business Review for Central and South-Eastern Europe (1999-2010). EDITOR-IN-CHIEF Tjasa Redek, University of Ljubljana, School of Economics and Business associate editors Neven Borak, Union of Economists of Slovenia, Slovenia Guido Bortoluzzi, University of Trieste, DEAMS Department, Italy Barbara Cater, University of Ljubljana, School of Economics and Business Matej Cerne, University of Ljubljana, School of Economics and Business Marina Dabic, Nottingham Trent University, UK & University of Zagreb, Croatia Miro Gradisar, University of Ljubljana, School of Economics and Business Mateja Kos Koklic, University of Ljubljana, School of Economics and Business Darja Peljhan, University of Ljubljana, School of Economics and Business Roman Stollinger, The Vienna Institute for International Economic Studies, Austria Maja Vehovec, The Institute of Economics, Zagreb, Croatia Miroslav Verbic, University of Ljubljana, School of Economics and Business Katja Zajc Kejzar, University of Ljubljana, School of Economics and Business editorial board Mary Amity, Federal Reserve Bank of New York, United States Adamantios Diamantopoulos, Universität Wien, Austria Polona Domadenik, University of Ljubljana, Slovenia Jay Ebben, University of St. Thomas, United States Neil Garrod, University of Greenwich, United Kingdom Anja Geigenmüller, Technische Universität Bergakademie Freiberg, Germany Laszlo Halpern, Hungarian Academy of Sciences, Hungary Nevenka Hrovatin, University of Ljubljana, Slovenia Robert Kaše, University of Ljubljana, Slovenia Gayle Kerr, Queensland University of Technology, Australia Josef Konings, Katholieke Universiteit Leuven, Belgium Maja Makovec Brenčič, University of Ljubljana, Slovenia Igor Masten, University of Ljubljana, Slovenia Rasto Ovin, University of Maribor, Slovenia Daniel Ortqvist, Lulea University of Technology, Sweden Marko Pahor, University of Ljubljana, Slovenia Danijel Pucko, University of Ljubljana, Slovenia John Romalis, University of Chicago, United States Friederike Schroder-Pander, Vlerick Leuven Gent Management School, Belgium Christina Sichtmann, University of Vienna, Austria Sergeja Slapnicar, University of Ljubljana, Slovenia Beata Smarzynska Javorcik, Oxford University, United Kingdom Jan Svejnar, University of Michigan, United States Marjan Svetlicic, University of Ljubljana, Slovenia Miha Skerlavaj, University of Ljubljana, Slovenia Bobek Suklev, University „St. Cyril and Methodius", N. Macedonia Janez Sustersic, University of Primorska, Slovenia Fiti Taki, University „St. Cyril and Methodius", N. Macedonia Bob Travica, University of Manitoba, Canada Peter Trkman, University of Ljubljana, Slovenia Aljosa Valentincic, University of Ljubljana, Slovenia Irena Vida, University of Ljubljana, Slovenia Joakim Wincent, Umea University, Sweden Jelena Zoric, University of Ljubljana, Slovenia Vesna 2abkar, University of Ljubljana, Slovenia publisher: University of Ljubljana, School of Economics and Business, Kardeljeva ploščad 17, SI-1001 Ljubljana, Slovenia. The journal is co-financed by Slovenian Research Agency. the review's office: Economic and Business Review , Kardeljeva ploščad 17, SI-1001 Ljubljana, Slovenia tel: + 386 1 58 92 607, fax: + 386 1 58 92 698, email: ebr.editors@ef.uni-lj.si TEHNICAL EDITOR: Layout by Nina Kotar, Printed by Copis d.o.o., Ljubljana Economic and Business Review is indexed in: AJG, Cabell's Directory of Open Access Journals Publishing Opportunities, DOAJ, Ebsco, Econlit, IBSS and ProQuest URL: http://www.ebrjournal.net Tomaž Ulčakar ISSN 1580-0466 e-ISSN 2335-4216 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 141 CONTENTS 143 Andrej Kuštrin A DSGE Model for the Slovenian Economy: Model estimates and Application 213 Barbara Čater Julijana Serafimova The Influence of Socio-Demographic Characteristics on Environmental Concern and Ecologically Conscious Consumer Behaviour among Macedonian Consumers 243 Byunghoon Jin John C Cary Are Middle Managers' Cost Decisions Sticky? Evidence from the Field 275 Črt Lenarčič Inflation - the Harrod-Balassa-Samuelson effect in a DSGE Model Setting 309 Eva Repovš Mateja Drnovšek Robert Kaše Change ready, resistant, or both? Exploring the concepts of individual change readiness and resistance to organizational change E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 142 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 143 A DSGE MODEL FOR THE SLOVENIAN ECONOMY: MODEL ESTIMATES AND APPLICATION* Received: June 29, 2016 ANDREJ KUSTRIN1 Accepted: February 24, 2019 ABSTRACT: The paper presents the estimation of a dynamic stochastic general equilibrium (DSGE) model for the Slovenian economy and its applications. The model, which is built in the tradition of New Keynesian models, closely follows the structure of the model developed by Adolfson et al. (2007) and Masten (2010). We estimate the model using a Bayesian method on quarterly Slovenian macroeconomic data covering the period 1995-2014. Beyond evaluating the properties of the estimated model, we discuss the role of various shocks in explaining macroeconomic fluctuations in the Slovenian economy to illustrate the model's potential in structural business cycle analysis. Key words: DSGE models, Bayesian methods, business cycle JEL classification: C11, E32 DOI: 10.15458/ebr.87 INTORDUCTION New-Keynesian dynamic stochastic general equilibrium (DSGE) models have recently become a standard tool for macroeconomic analysis. The key feature of this class of models is that they are derived from the microeconomic foundations meaning that they assume optimizing agents which usually form rational expectations and maximize their objective functions subject to their respective constraints in the presence of imperfect competition and nominal rigidities.3 In recent years there have been many theoretical and empirical contributions developing and estimating DSGE models. The most influential papers in this area include Clarida et al. (1999, 2001), Beningo & Beningo (2003), Gali & Monacelli (2005), Christiano et al. (2005), Smets & Wouters (2003, 2007), Adolfson et al. (2007) and many others. * I am grateful for guidance and helpful comments from my supervisor Igor Masten. For helpful comments on earlier versions of this paper, I would also like to thank the editor and two anonymous referees. Last, I would like to acknowledge financial support from the Slovenian Research A gency. All remaining errors are my own responsibility. 1 University of Ljubljana, Faculty of Economics, Young Researcher, Ljubljana, Slovenia, e-mail: andrej.kustrin@ ef.uni-lj.si. 2 For further information on the New-Keynesian models, see Gali (2008) and Woodford (2003). 144 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 144 Although the literature on the estimation of DSGE models and the subsequent use of these models to study macroeconomic fluctuations in various countries has rapidly expanded in recent years, no attempt has as yet been made to estimate a New-Keynesian DSGE model for Slovenia, at least to the best of our knowledge.2'1 This paper therefore seeks to fill this gap by presenting an estimated DSGE model for the Slovenian economy. The model that we use was inspired in the work of Adolfson et al. (2007) and Masten (2010). Masten (2010) extended the baseline model of Adolfson et al. (2007) in two directions, namely by (i) adapting the model to the small open economy case within the euro area and (ii) enriching the fiscal block of the model. We use a Bayesian approach to estimate key model parameters on 15 time series for Slovenia: GDP, consumption, investment, exports, imports, government consumption, real effective exchange rate, real wage, employment, GDP deflator, CPI price index, short-run interest rate, and three foreign variables (that is output, inflation and interest rate), which refer to the first 12 euro area countries. With this paper we want to contribute to the large literature on estimated DSGE models by applying the Bayesian method to the estimation of the DSGE model for the Slovenian economy and therefore presenting evidence for an additional country on the estimates of the structural parameters, and by identifying the shocks responsible for the recent recession and the key sources of macroeconomic fluctuations in Slovenia. After the estimation, we first present our estimates of the structural parameters. We then perform several checks of the model's empirical performance. Specifically, we evaluate how well the model fits the data. To do so, we compare the actual data with the one-sided predicted values from the model. Next, we calculate statistics of the data generated by the estimated model and compare them with those based on the actual data. Finally, we look at the smoothed estimates of the shock innovation paths to check whether they look stationary. In the last part of the paper, we apply the estimated DSGE model to analyse the contribution of the structural shocks on business cycle fluctuations in the Slovenian economy. We proceed here in three steps. First, using traditional impulse response analysis, we look at the partial effects of the most important shocks included in the model on key 21 Despite their wide use, DSGE models have also certain drawbacks. The most problematic issues which are currently much discussed in the literature are mainly concerned with: (i) unrealistic assumptions (e.g. Ricardian equivalence, rational expectations hypothesis, infinitely-lived households, ...), (ii) unconvincing method of estimation (which is a combination of calibration and Bayesian estimation), (iii) questionable assumption about the structural parameters that are assumed to be invariant to policy changes, (iv) issue related with the use of revised or real-time data when estimating the model, and (v) poor performance during the recent crisis. For more detailed discussion of these issues, see Romer (2016), Blanchard (2016) and the other contributions (see Blanchard (2017) for an extensive list of references). Despite these shortcomings, we decided to use a DSGE framework as we believe that it is flexible enough to be used for our purposes, while other models are more limited in terms of their ability to fully address the research questions under study. E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 145 macroeconomic variables. Second, to assess how much of the volatility of the observed variables can be explained by the shocks included in the model, we also produce variance decomposition analysis. Finally, we compute historical decompositions of GDP growth and its main components in terms of various structural shocks of the model to examine the importance of respective shocks in explaining the observed macroeconomic dynamics over the sample period, with particular attention to the recent recessionary periods. Previewing the results, we find that investment-specific technology shocks mostly accounted for a significant portion of the drop in output growth from 2008 onwards. This result accords with a drop in foreign and domestic orders followed by a decline in investment (mostly at the beginning of the crisis) and the large amount of losses of the corporate sector that accumulated on balance sheets of the banks in the form of non-performing bad loans, further contributing to a contraction of lending activity, which in turn reduced investment and impeded economic activity. Furthermore, consumption preference and export mark-up shocks were another sources that contributed negatively to GDP growth, most likely reflecting the reduction in households' income (in combination with the precautionary saving) and the fall in exports due to the deterioration of external competitiveness, as wages increased faster than productivity before the crisis years, respectively. The results also suggest that fiscal shocks had a stimulating impact during the first stage of the crisis. However, starting from 2010 there was a turnaround in fiscal policy due to austerity measures adopted to consolidate public finances. The slowdown in GDP growth was also accompanied by permanent (unit-root) technology shocks that could be considered as associated with the lack of productivity-enhancing and other structural reforms in the run-up to the crisis. By contrast, the historical decomposition suggests that transitory (stationary) technology shocks were stimulative for GDP growth from 2013 onwards, which may be interpreted as resulting from a temporary greater tendency of the corporate sector to take restructuring measures in response to the crisis to enhance its technology and production efficiency. Finally, our results show that the recovery phase after 2013 is explained in our model mainly by consumption preference shocks, which could be explained by the increased consumer confidence, resolution of banking system problems, as well as by the improvement in the labour market situation. The rest of the paper is organized as follows. Section 2 presents the model. Section 3 presents the estimation methodology and discusses the calibration of the model, the choice of priors and presents the data used in the estimation. Section 4 contains the estimation results and evaluation, which are followed by an analysis of the impulse responses of the various structural shocks and their contribution to the developments in the Slovenian economy in Section 5. Section 6 concludes with a summary of the main findings. 2 THE MODEL As mentioned in the introduction, to describe the Slovenian economy we use a DSGE model presented in Adolfson et al. (2007) and Masten (2010), which is an extended ver- 146 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 146 sion of basic closed-economy new-Keynesian models, including the benchmark models of Christiano et al. (2005), Altig et al. (2011), and Smets & Wouters (2003, 2007). The model economy consists of households, domestic goods producing firms, importing consumption and importing investment firms, exporting firms, a government which conducts fiscal policy, and an exogenous foreign economy. As it is common in the DSGE literature, the model incorporates several real and nominal rigidities, such as habit persistence in consumption, variable capacity utilization of capital and investment adjustment costs, as well as the price and wage stickiness. The stochastic dynamics of the model is driven by sixteen exogenous structural shocks. The shocks considered are: permanent (unit-root) technology, transitory (stationary) technology, investment-specific technology, markup shocks (domestic, imported consumption, imported investment and export markup shocks), consumption preference and labour supply shocks, asymmetric technology, risk premium, foreign VAR shocks (foreign output, inflation and interest rate shocks) and fiscal shocks (rate of transfers to households and government spending shocks). One feature of the model worth noting is that it includes a stochastic unit-root technology shock, which implies a common trend in the real variables of the model. Consequently, the model can be estimated with raw data without any pre-filtering. In the following we summarize the main features of the model. To this end we follow quite closely the mode of presentation from Section 2 of Adolfson et al. (2014).3 2.1. Supply side of the economy 2.1.1. Domestic firms The domestic firms use labour together with capital to produce intermediate goods Yi, which are sold to the final good producer. The production function of the final good firm is of the Dixit-Stiglitz form: where \d}t is a stochastic process determining the time-varying markup in the domestic goods market. The final good producers operate in a perfectly competitive environment, taking the prices of the intermediate goods Pft and final goods Pd as given. The production function for each intermediate good firm i which operates under monopolistic competition is of the Cobb-Douglas type: where Hi t denotes homogeneous labour input hired by firm i, and Ki}t is the amount of capital services used by firm i, which can differ from capital stock since the model assumes (1) Yit = z— etK*tHl-a - ztt, (2) 3The detailed description of the model (including the first order conditions) is available in Adolfson et al. (2007). E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 147 a variable capital utilization rate. Furthermore, zt is a permanent (unit-root) technology shock, whereas et is a transitory (stationary) technology shock. The term ztt indicates fixed costs, which grow with the technology rate. Fixed costs are set in such a way that profits are zero in steady state. Cost minimization yields the following nominal marginal cost function for intermediate firm i: J1-a /1 \ a ii —-, (3) where R is the gross nominal rental rate per unit of capital, Rt is the gross nominal interest rate, and Wt is the nominal wage rate per unit of aggregate, homogeneous labour Hi t. Besides solving the cost minimization problem, intermediate good firms have to decide on price for their output. The model assumes the Calvo type staggered-price setting. This means that at each period, each firm faces a random probability (1 — ^d) that it can reoptimize its price. The reoptimized price is denoted Pt ,opt. With probability a firm is not allowed to set its prices optimally, and its price is then set according to the following indexation rule (Smets & Wouters, 2003): P+1 = wr Pd, where _ Pf / Pf-1 is the (gross) inflation rate and Kd is an indexation parameter. The optimization problem of a firm setting a new price in period t is the following: max Et ^ (PUT U+ Pt s=0 s s (nfnd+i...nf+s-i>Kd Pf'^+s —MC!t+s Wt+s + zt+st) (4) where (3£d)s ut+s denotes the stochastic discount factor, which is used to make profits conditional upon utility. / is the discount factor, and vt denotes the marginal utility of households' nominal income in period t + s, which is exogenous to the intermediate firms. 2.1.2. Importing and exporting firms The importing sector consists of two types of firms: firms which import consumption goods and firms which import investment goods. There is a continuum of importing firms, indexed by i € (0,1). These firms buy a homogeneous good in the world market at price Pf and transform it into a differentiated consumption Cmt or investment good I™. In addition, there is also a continuum i € (0,1) of exporting firms that buy a homogeneous good on the domestic market and transform it into a differentiated exported good which is sold on the foreign market. The marginal cost of importing and exporting firms are Pf and Pt, respectively. The aggregate import consumption, import investment and export good is a composite of a continuum of i differentiated imported consumption, imported investment and exported goods, each supplied by a different firm, which follows the CES function: Cm _ cmy di im di \ \ 1 1 u 148 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 148 Xt (Xi,t)xx di L./0 (5) where 1 < < œ for j = {mc, mi, x} is the time-varying flexible-price mark-up in the import consumption (mc), import investment (mi) and export (x) sector. The model assumes monopolistic competition among importers and exporters and Calvo-type staggered price setting. The price setting problems are completely analogous to that of the domestic firms in Equation (4). From the optimization problems four specific Phillips curves, determining inflation in the domestic, import consumption, import investment and export sectors, can be derived. i 2.2. Demand side of the economy 2.2.1. Households In the model economy there is also a continuum of households j e (0,1), which attain utility from consumption and leisure. The households decide on their current level of consumption and their domestic and foreign bond holdings. They also choose the level of capital services provided to the firms, their level of investment and their capital utilization rate. The households can increase their capital stock by investing in additional physical capital, taking one period to come in action, or by directly increasing the utilization rate of the capital at hand. The jth household's utility function is: Eßt t=0 Zt ln (Cjt - bCjt-i ) - ChAL ^ 1+ °L (6) where Cj,t and hj,t denotes levels of real consumption and labour supply of household j, respectively. AL is a constant representing the weight that the worker attaches to disutility of work. The model also allows for habit formation in consumption by including bCj ,t-1. ZC and Zh are preference shocks, consumption preference shock and labour supply shock, respectively. The aggregate consumption Ct is a CES index of domestic Cf and imported Cm consumption goods: Ct = [(1 - u>cfn° (C(flc-1)hc + (c™)^-^ j^-1), (?) where uc is the share of imported consumption goods in total consumption, and nc is the elasticity of substitution between domestic and imported consumption goods. The corresponding consumer price index is given by: PC [(1 - ^^ Pf- + ^ (pr)l-n^ (8) The model also assumes that households can purchase investment goods in order to increase their capital stock. The law of motion of capital is given by: Kt+i = (1 - S) Kt + YtF (It, It-i) + A t, (9) E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 149 where Kt is a physical capital stock, S is the depreciation rate of capital stock, F (It, It-1) is a function that transforms investment into capital. Following Christiano et al. (2005), F (It, It-1) is of the following form: F (It,It-i)= 1 - S(IuI-i) It (10) where S determines the investment adjustment costs through the estimated parameter S". Yt denotes the investment-specific technology shock and At represents either newly boughi capital if it is positive or sold capital if it is negative. The investment (It) is a bundle be- tween domestic and imported investment goods (Id and Itm, respectively): It (1 - u*)1/ni (If) .(Vi-^/Vi + U i/m (ID Xvi-i)/vi ni/(ni-i) (11) where wi denotes the share of imported investment goods in total investment, and ni is elasticity of substitution between domestic and imported investment goods. It is worth noting that domestically produced consumption and investment goods have the same price Pf. The aggregate investment price index is therefore given by: PI (1 - Ui) (Pf )i-ni + u* (PD'*) i-ni i/(i-ni ) (12) Furthermore, the model assumes that each household is a monopolistic supplier of differentiated labour service, which implies that they can determine their own wage. Each household sells its labour hj,t to a firm which transforms it into a homogeneous input good Ht according to the following production function: Ht (hj't) dj Xv (13) where Xw is the wage markup. The demand function for each differentiated labour service is given by: h j't W, j't Wt Ht (14) Following Erceg et al. (2000) and Christiano et al. (2005), the households are subject to the Calvo wage rigidities, which means that in every period each household faces a random probability 1 - gw that it can change its nominal wage. If a household is allowed to re-optimize its wage, it will set its wage to Wtopt taking into account the probability that the wage will not be re-optimized in the future. The households that cannot re-optimize set their wages according to the following indexation rule: Wj j't+i (nC)Kw Vz't+iWj't, (15) where kw is an indexation parameter, nf is the inflation rate measured by the consumer price index, and ¡i,Ztt = zt/zt-i is the growth rate of the unit-root technology shock. The A i 0 A 1-A 150 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 150 household j that can re-optimize solves the following optimization problem: h A (h,,t+s)1 + 'L zt+sAL l+aL max (ßtw)s<| +,t+s (1 - t« + r»a) ft -.ft+s-i)" \ > (16) X (mz,t+l ■ ■ ■ Mz,t+s) WOfhjt+s where ty is a labour income tax and tfr is a time-varying rate of social transfers to households defined in more detail in Subsection 2.4. 2.3. Monetary policy The monetary policy is modelled in a highly simplified way. It is assumed that the domestic interest rate (Rt) depends on the exogenously given foreign interest rate (R£) adjusted for the risk premium on foreign bonds ($ ^at, ):4 Rt = (at,t-i) - l) B*t + n with tc, ty and tk being the tax rates on private consumption, labour income and capital income, respectively, which are assumed to be fixed. In the above expression, nt are total profits, which are equal to the sum of profits earned by domestic, importing and exporting firms, nd, n™ and n®, respectively: nt = nd + nD + (22) where: nd = Ptd (Cdd + id + Gt) + Ptd (cx + IX) - MCd (Cd + id + Gt + C'x + I'x) - MCfzrf nm = pmcc™ + P™,ii™ - P* (C™ + ID (24) and: nx = PX (CX + IX) - Pd (CX + IX). (25) these simplifications, the model in such a structure fits the data, including the short-term nominal interest rates, reasonably well. It is also worth noting that similar approach neglecting existence of diverse monetary policies and flexible exchange rates prior to the EMU-start was used in the literature (see, for example, Adolfson et al. (2007), Almeida (2009), Smets & Wouters (2003), Marcellino & Rychalovska (2014) among others). The authors estimated their models under implicit assumption that, even before the establishment ofthe currency area there was a common monetary policy in the European Union. Finally, as a robustness check we re-estimated our model using the data for the period 2004Q3 onwards, when Slovenia entered the ERM 2. Our analysis reveals that in this case parameter estimates do not substantially vary from the estimates reported in Tables 3 and 4 in the main text. But what is more important, we find that our main results reported in the paper (e.g., those of the historical decompositions) persist. We choose not to report this robustness check in the paper to save space, but it is available upon request from the author. 152 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 152 Furthermore, the government expenditures are given by: GEXt = TRt + PtdGt + (Rt-i - 1) Bt, (26) where TRt denotes transfers to households, Gt is government consumption of goods and services and (Rt-1 — 1) Bt stands for public debt interest payments. We assume that transfers to households are indexed to wages Wt and hours worked Ht with an exogenously given rate of transfers r^ according to the following expression (D'Auria et al., 2009): TRt = rtr Wt Ht. (27) For the rate of transfers to households it is simply assumed that follow an AR(1) process (in deviations from its steady state): T^ = pTtr T- i + £Ttr f (28) Finally, government consumption follows the log-linear rule of the following form:6 gt = Pg gt-i — nct — $y yt — $bbt — $d def t + £g,f (29) In this equation, gt is the percentage deviation of real government consumption (station-arized with the unit-root technology level, zt) from its steady state level, nt is the CPI inflation, yt reflects the output gap, bt is the public debt and def t denotes the government deficit which is expressed as a difference from its steady state, that is, deft = deft — def. £g,t defines the exogenous shock aimed at capturing discretionary changes in government consumption. , $y, $b and $d denote the feedback coefficients towards inflation, output gap, public debt and government deficit deviations, respectively. pg reflects the degree of government consumption smoothing. 2.5. Market equilibrium In equilibrium all markets clear. The market clearing condition for the domestic goods market is given by: Cd + Id + Gt + Cf + If < z¡-aetK?H¡-a - zh - a (ut) Kt, (30) where Cf and If are the foreign demand for export goods which follow CES aggregates with elasticity r¡f. Furthermore, the net foreign assets' market clears when domestic investment in foreign bonds (denoted by B*) equals the net position of exporting/importing firms: B* = px (Cf + If) - p* (Ctm + im) + R-1* (at, 4>t) B¡. (31) 6Our specification for the fiscal rule is similar to those used by Erceg & Lindé (2013), with the only exception that they do not include the inflation rate. E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 153 2.6. Foreign economy Since the domestic economy is a small open economy, we assume that the foreign economy is exogenous. In particular, foreign output (y;), foreign inflation (n^) and foreign interest rate (R;) are exogenously modelled as an identified VAR model with two lags:7 m; = fiX-1 + $2X(-2 + Sx* £x«tU (32) where X; = R; j , £x*,t ~ N (0,Ix*), Sx* is a diagonal matrix with standard deviations and $-1Sx*£x*,t ~ N (0, Sx*). 2.7. Structural shocks In total, the dynamics of the model is driven by 16 exogenous shock processes that are assumed to be characterized in log-linearized form by the univariate representation: it = Peit-1 + £t,t, £t,t ~ N (0, af ) , (33) where it = ^z,t,et,^t,QXt, Yt,t,z*t ,rlrfor j = {d,mc,mi,x}. £g,t is assumed to be a white noise process (that is, p£g = 0). There are also three foreign shocks (that is, foreign output, foreign inflation and foreign interest rate shock) provided by the exogenous (pre-estimated) foreign VAR model. 3 MODEL SOLUTION AND ESTIMATION In this section, we present how the DSGE model is solved and estimated. 7The foreign VAR model is estimated for the first 12 Euro area countries over the period 1995Q1-2014Q4 and includes the following variables: output (GDP at market prices, chain linked volumes (2005), million units of national currency); GDP deflator (GDP at market prices, price index (implicit deflator), 2005=100, national currency); interest rate (12-month money market interest rate in percent). To make the observed data consistent with the model's concepts, we adjusted the data before entering the VAR model. Specifically, we used HP-detrended log of GDP (we set the smoothing parameter to 1600, which is typically used with quarterly data), the demeaned first difference of the log of GDP deflator and the demeaned interest rate which is divided by 400. All data series are seasonally adjusted and adjusted by working days. The lag order of the VAR model was chosen using the Hannan-Quinn information criterion, which suggests an optimal lag order of two periods (Lütkepohl & Krätzig, 2004). We also removed variables with lowest t-ratios until all remaining variables had t-ratios greater than 2, which is often used in applied work. The estimated foreign VAR model is, therefore, given by: nt =0.028y_1 + 0.12177— + 0.279tT—2 + en* ,t yt = 1.667y_i - 0.698y—2 + £y* ,t Rt =1.190R—1 + 0.321yt_1 - 0.306y—2 - 0.27lRt*_2 + eR*,t. 154 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 154 3.1. Model solution The model presented in the previous section consists of a set of optimality conditions and laws of motion of the shock processes. Since the model comprises the unit-root technology shock that induces a stochastic trend in the levels of the real variables, the first step prior to model solution is rendering the model stationary. To this end all real variables are divided with the trend level of technology zt. The resulting stationary variables are then denoted by lower-case letters, that is, xt = Xt for a generic variable xt. We then proceed with the log-linearisation8 to the model's equations of the transformed model around the deterministic steady state9, where the variables are expressed as logarithmic deviations from their steady state values, that is, xt = xt-x & lnxt — ln x, where x denotes the steady state value of a generic variable xt. Once the model has been stationarized and log-linearized, it can be written in the following compact form: Et {«0rt-1 + + «2rt+1 + ft^t+i + ft} = 0, (34) where rt is a vector of endogenous variables, is a vector of exogenous variables, and a0, a1, a2, ft and ft are coefficient matrices. It is assumed that evolves according to: = ptft-i + £t £t ~ N (0, S). (35) We use Dynare 4.4.310 to solve the model. The solution of the model takes the form:11 rt = Ar-1 + BVf (36) 3.2. Data and measurement equations For estimation purposes the solved model can be written in the following state-space form (Hamilton, 1994): 6+1 = Fit + ut+i (37) and: Yt = A'xt + H 'it + ut- (38) 8However, it is important to notice that dynamics in the log-linearized model is only approximation of the true non-linear dynamics. Therefore, studying the log-linearized models is only valid for small deviations from the model's steady state. For a complete list of the log-linearized equations of the model, see Appendix A. 9We compute the non-stochastic steady state of the model following the procedure described in Adolfson et al. (2007). It is important to note that the steady state also depends on estimated parameters. For this reason, when estimating the model, it is of great importance to take into account parameter dependence by using model-local variables. For further discussion, see Pfeifer (2014a), Remark 4 (Parameter dependence and the use of model-local variables). 10Dynare is a software package for solving and estimating DSGE models. For more information regarding Dynare refer to the official Dynare web page http://www.dynare.org and see Mancini Griffoli (2011), as well as Adjemin, Bastani, Karamé, Juillard, Maih, Mihoubi, Perendia, Pfeifer, Ratto & Villemot (2014). 11 Dynare uses solution algorithms proposed by Klein (2000) and Sims (2002). For a detailed look at what exactly is going on behind the scenes of Dynare's computations, the interested reader is referred to Villemot (2011). E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 155 The first equation is called the state equation, whereas the second is called the observation (measurement) equation. The symbols appearing in (37) and (38) have the following meaning: Yt is an (n x 1) vector of observed variables at time t, £t is an (r x 1) vector of unobserved variables at time t (also referred to as state vector) and xt is a (k x 1) vector with exogenous or predetermined variables (e.g. a constant). Furthermore, F, A' and H' are matrices of dimension (r x r), (n x k) and (n x r), respectively. The (r x 1) vector ut and the (n x 1) vector ut are uncorrelated, normally distributed white noise vectors, therefore: Q for t = t 0 W < M \ Q f0r T = 1 E (ut Vt ) = \0 otherwise E (wtdT) = R for t = t 0 otherwise, where Q and R are (r x r) and (n x n) matrices, respectively. The disturbances ut and ut are assumed to be uncorrelated at all lags: E (vt,JT) = 0 for all t and t. (39) In what follows, we describe how the raw data were converted to the form used in estimation. In addition, we present the exact measurement equations that are employed to relate the observed data to the model state variables. The estimates are based on quarterly Slovenian macroeconomic data covering the period 1995Q1-2014Q4. We employ the following 14 variables as observables:12 the GDP deflator (Pf), the real wage (Wt/Pf), consumption (Ct), investment (It), government consumption (Gt), the real exchange rate (xt), the short-run interest rate (Rt), employment13 (Et), GDP (Yt), exports (Xt), imports (Mt), the CPI price index (Ptc), foreign output (for the first 12 euro area countries) (Yt*), the foreign GDP deflator (for the first 12 euro area countries) (Pt*) and the foreign interest rate (12-month money market interest rate of the euro area) (R¿). Regarding the foreign variables, GDP for the first 12 euro area countries is used for foreign output, and the GDP deflator for the first 12 euro area countries is used for foreign inflation, while the foreign interest rate refers to the 12-month money market interest rate of the euro area. Data come from four different sources. Data on the employment and gross wages are taken from the Statistical Office of the Republic of Slovenia. The sources for domestic interest rate are the Bank of Slovenia and the Institute of Macroeconomic Analysis and Development of the Republic of Slovenia. The rest of the data are taken from Eurostat. Since the model comprises a stochastic unit root technology shock that induces a common stochastic trend in the real variables of the model, we use first differences to make these variables stationary. When estimating the model, the following variables are matched in growth rates 12 A detailed description of the data used in the estimation together with their sources is provided in Appendix B. Additionally, the data are plotted in Appendix D. 13We assume that the employment variable (Et) is related to the hours worked variable (Ht) by an auxiliary equation (expressed as a percentage deviation from the steady state): AE 13 E A E , (1 - ge)(1 - fce) (H E A Et = — AEt + 1 + (1+ 3) U {Ht - Et The Calvo parameter, £e, representing the fraction of firms that in any period is able to adjust employment to its desired total labour input, is estimated. 156 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 156 measured as quarter-to-quarter log-differences: GDP, consumption, investment, exports, imports, government consumption, real wage, GDP deflator, CPI price index, foreign output and foreign GDP deflator. The rest of the variables are used in levels: domestic and foreign interest rate, employment and real exchange rate. The real wage is calculated as the nominal gross wage per employee deflated by the GDP deflator. All interest rates are divided by 4 to express them in quarterly rates consistent with the variables in the model. The stationary variables, xt and Et, are measured as follows: we take the logarithm of real exchange rate and remove a linear trend, so that it is expressed in percentage deviations from the trend, consistently with the model concepts, that is Xfata = xt-x, while the employment is measured as deviation around the mean, that is Edata = ^eE . Furthermore, in order to align the data with the model-based definitions, some additional transformations are made. First, since the model assumes that all real variables are growing at the same rate as output, we match the sample growth rates of private consumption, investment, government consumption, exports, imports and real wage with the sample growth rate of real GDP by removing the remaining growth rate differentials. Second, the model assumes that in steady-state, the interest rates (that is, domestic and foreign interest rate) as well as different measures of inflation (that is, domestic, CPI and foreign inflation) are identical, that is R = R* and nd = nc = n*, respectively. This assumption is clearly rejected by the data. To circumvent this issue, we demean all these time series before the model estimation and add the sample mean of domestic interest rate to the foreign interest rate and the sample mean of domestic inflation to the CPI and foreign inflation, so that the data match the model assumptions. All variables (except the nominal interest rates) are seasonally adjusted and adjusted by working days. The vector of observable variables, Yt, is then given by: Yt = [AlnPdAata Aln (Wt/Pf)data AlnCfata Alnífata xdtata ... Rdata Edata A ln Ytdata A ln Xdata A ln Mfata ... (40) A ln Gdtata A ln P¡,data A ln Y**,data A ln Pt*'data R*'data]', E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 157 where A is the first difference operator. The corresponding measurement equation that matches observed data with model's variables is: A ln Pd'data A ln (Wt/Pfj A ln Cdata A ln !data Adata xt Rdata EE data A ln Ytdata A ln Xdata A ln Mdata A ln Gdata .data data .data A ln Pt A ln Yt A ln P*'data R *,data t ■ nd -1) ■ ln Hz ln Hz ln Hz 0 4R (R - 1) 0 ln Hz ln Hz ln Hz ln Hz (nd - 1) ln Hz nd -1) 4R (R - 1) + Ay, Awt + Hz,t Act + Hz,t Ait + Hz,t Xt 4RRt Et Ayt + Hz,t Act + Hz,t Arn t + Hz,t Agt + Hz,t nct An Hz.t 4RRÎ + me nd,t me w,t me c,t £ i,t me x,t me R,t me E,t me y,t me x.t me m.t me g,t me nc,t me y*,t me n*,t £me , (41) where elf denotes the measurement error for the respective variable. The standard deviation of specific measurement error is calibrated at 10% of the standard deviation of the corresponding observed domestic variables, while the measurement errors for the foreign variables are set to 0, as in Adolfson et al. (2007). d t 3.3. Estimation methodology Structural parameters of the model are either calibrated or estimated. The values for the parameters that are calibrated (and thus kept fixed throughout the estimation) are chosen in accordance with the practice in the literature calibrating small open-economy models. Their values are discussed in Subsection 3.4. All remaining parameters are estimated with a Bayesian estimation method, which has become a standard econometric technique for estimating DSGE models.14 In the following, we briefly describe the main features of the method. The key idea of the Bayesian estimation method15 is that it combines the prior belief of the parameters with empirical data to form the posterior distributions of the parameters. The posterior distributions are obtained by using the Bayes theorem: p(Yt |0) p (6) t| 6Y = A-, (42) V 7 p (Yt) 14All estimates are performed using Dynare version 4.4.3 in Matlab R2012b. 15A detailed explanation of the estimation method can be found in An & Schorfheide (2007), Adolfson et al. (2007), Canova (2007), Fernandez-Villaverde (2010) and Smets & Wouters (2003, 2007) among many others. The reader is also referred to Dynare Manual for additional explanation of the estimation method. p 158 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 158 where 0 is a vector of the parameters to estimate, p is the density of the parameters conditional on data (the posterior), p (Yt\0^ is the density of the data conditional on the parameters (the likelihood), p (0) is the unconditional density of the parameters (the prior) and p is the marginal density of the data.16 Given that the marginal density of the data is a constant term or equal for any parameter, equation (42) can be rewritten as: p(0\Yt) « p(?t\0^p (0) =k(0\Yt) , (43) where K ^Yt) is the posterior kernel. Taking logs of (43), we get: lnK (0ft) = lnp(Yt\0) + lnp (0) = ln L (?t\0) + lnp (0). (44) Before the estimation can begin, we need to specify the priors for the parameters to be estimated and evaluate the likelihood function of the observed data. The choice of priors is discussed in Subsection 3.5. The likelihood function of the observed data is evaluated by generating forecasts from the state-space system, (37) and (38), with the use of the Kalman filter. Conceptually, the Kalman filter consists of calculating the sequence and j E^ 11 j , where £t+111 denotes the optimal forecast of £t+1 based on observation of yt = [y(, Y— i, y/_2, • • •, x't, x't_i,x't_2, • • •, x'i) and denotes the mean squared error of this forecast. The algorithm works forward in time and is conducted as follows:17 For t = 1, the algorithm needs to be provided with initial values for a one-step ahead forecast of time t states, £tit-1, and respective forecast error variance-covariance matrix, Based on this a one-step ahead forecast of time t data, Ytit-1 and respec- tive variance-covariance matrix, E'Yit_1 are computed. The algorithm then updates the forecasts of time t states, £t|t, and a respective variance-covariance matrix, E^. The final step is to compute a one-step ahead forecast of time t + 1 states, £t+'|t, and respective variance-covariance matrix, E^^. These steps are iterated for t = 2, 3,4, • • • ,T. The log-likelihood function (based on the data up to time t) can be written as follows (Hamilton, 1994): T E t=i lnL (?t\xt, yt-i,F, A', H', Q, R T - E t=i n 1 ~ 2 log 2n + 2 log \£)t-i\ +2 EL (Yt - Yt—y (4-1)" (Yt - Yt\t-i) (45) 16It is defined as: p(Yt) = p(e,Yt) de, where p denotes the joint density of the parameters and the data. 17The presentation here follows Hamilton (1994). E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 159 Finally, the posterior distribution is obtained in two steps: first, by maximizing the log posterior density with respect to 0, the posterior mode 0m and an approximate covari-ance matrix, based on the inverse Hessian matrix evaluated at the posterior mode, Y,gm = H [om , is obtained and second, the posterior distribution is simulated by using the Monte-Carlo Markov-Chain (MCMC) sampling method, specifically the Metropolis-Hastings (MH) algorithm. The idea behind this algorithm is the following (Mancini Grif-foli, 2013): first, the algorithm chooses a starting point (posterior mode), then it draws a candidate value 0* from an arbitrary candidate (or jumping) distribution: j(e*\e-i) ~ n(e-ucZem): (46) where 0i-1 is the last accepted draw, Y,gm denotes the inverse of the Hessian computed at the posterior mode, and c is the scale factor, which is chosen to ensure an appropriate acceptance rate. In the next step, the algorithm computes the acceptance ratio: a = mm e >\Yt) K (ei-i\Yt (47) The algorithm then accepts or discards the proposal 0* according to the following rule: ei { e* { ei- with probability a otherwise If the parameter value is accepted, the mean of the distribution is updated with the new draw 0j. These algorithm steps are repeated many times to simulate the posterior distribution. 1 i 3.4. Calibrated parameters In this section, we present the calibrated parameters of the model.18 Their values are taken mainly from Adolfson et al. (2007) unless otherwise stated. The discount factor, 3, is fixed to 0.993, implying a steady-state interest rate of 11%,19 which matches the average interest rate in the sample period. The share of capital in production, a, is calibrated to 0.30. The depreciation rate of capital, 5, is set to 0.013. We calibrate the capital utilization cost parameter, oa, to 106. The elasticity of substitution between domestic and foreign goods, nc, is calibrated to 5. Labour disutility constant, AL, is calibrated to 7.5. As in Christiano, Eichenbaum and Evans (2005), we set the labour supply elasticity, aL, to 1, and the wage mark-up, Aw, to 1.05. The steady state mark-ups are calibrated at: 1.222 in the domestic goods market (\d), 1.633 in the imported consumption goods market (Am,c) and 1.275 in the imported investment goods market (Am,i). The steady state foreign terms of trade, , is calibrated to 1. The rest of the parameters, as well as the steady state relationships, are It is important to note that a time period is taken to be a quarter. k a 19This follows from the first order condition of the households' bond holdings, R = ^z ~ k)^ 160 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 160 calibrated using the averages of Slovenian data for the period 1995Q1-2014Q4. The shares of imports in consumption and investment, wc and wj, are set to 0.67 and 0.40, respectively. The steady state rate of transfers to households, ttr, is calibrated to 0.50. The ratios of government expenditures (^^), taxes (y), government consumption (g), and debt services (y) in GDP are 0.37, 0.36, 0.19, and 0.02, respectively. Further, the share of government consumption, social transfers and debt services in total government expenditures, , — 1 ' or' gex' gex and grx, are set to 0.51, 0.44 and 0.05, respectively. The target value of debt-to-GDP ratio, by, is assumed to be 240% in the steady state, which is consistent with the reference value of public debt established by the Maastricht Treaty, which equals 60% of yearly output. The steady state quarterly gross inflation rate, nd, is equal to 1.01. Finally, the average effective tax rates on consumption, labour income and capital income, tc, ty and tk, amount to 0.17, 0.48 and 0.22, respectively. An overview of the calibrated parameters is found in Table 1, while Table 2 provides an overview of the steady state relationships. Table 1: Calibrated parameters of the model Parameter Description Calibrated value ß Households' discount factor 0.993 a Capital share in production 0.30 nc Substitution elasticity between Cf and Cm 5 Capital utilization cost parameter 106 al Labour disutility constant 0.3776 cl Labour supply elasticity 1 s Depreciation rate of physical capital 0.013 ^w Wage mark-up 1.05 \d Mark-up in the domestic goods market 1.168 \m,c Mark-up in the imported consumption goods market 1.619 Mark-up in the imported investment goods market 1.226 Wi Share of imports in investment 0.40 Uc Share of imports in consumption 0.67 Tc Consumption tax rate 0.114 Ty Labour income tax rate 0.48 Tk Capital tax rate 0.22 Ttr Rate of transfers to households 0.50 3.5. Prior distributions of the estimated parameters Before the Bayesian estimation method starts, the prior distributions of the estimated parameters need to be specified. As the name suggests, prior distribution describes the available information about the parameters prior to observing the data used in the estimation. The observed data is then used to update the prior, through the Bayes theorem, to the posterior distribution of the model's parameters. In specifying the prior distributions we mainly E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 161 Table 2: Steady state relationships Parameter Description Value n Steady state quarterly gross inflation rate 1.01 ^yX Share of government expenditures in GDP 0.37 1 Share of taxes in GDP 0.36 y y Share of government consumption in GDP 0.19 yj*X Share of government consumption in government expenditures 0.51 r Share of debt services in GDP 0.02 y Share of debt services in government expenditures 0.05 yex ° r — Share of social transfers in government expenditures 0.44 yex b*y Target value of debt-to-GDP ratio 2.4 rely on choices from Adolfson et al. (2007). Throughout the analysis we use four main distributions: beta distribution, inverse gamma distribution, normal distribution and gamma distribution. For the parameters bounded between 0 and 1 we choose beta distribution. Parameters belonging to this group are nominal stickiness parameters indexation parameters k, the habit persistence b and the persistence parameters of the shock processes p. We set the mean of prior distributions for the price stickiness parameters to 0.5 with standard deviation 0.2, while the mean for the indexation parameters is set to 0.4 with standard deviation 0.1. However, there are three exceptions. For the Calvo parameter for domestic firms we set the prior mean to 0.85 with a standard deviation of 0.1, while for the Calvo parameter for exporting firms we choose a prior mean equal to 0.75 with a standard deviation of 0.1. For the wage indexation parameter we impose a prior mean of 0.5 with a standard deviation of 0.2. The prior on habit persistence has a mean of 0.65 and a standard deviation of 0.2. With the exception of the shocks to the unit-root technology, stationary technology and government consumption, we set the prior means of the persistence parameters for the structural shocks equal to 0.5 with a standard deviation of 0.2. For the unit-root technology, stationary technology and government consumption shocks we choose a mean of 0.6 and a standard deviation of 0.2. We use inverse gamma distribution to describe our priors about the parameters that are assumed to be positive. These parameters are the standard deviations of shocks and the substitution elasticities between goods, n We set the prior mean of the substitution elasticity between domestic and foreign investment goods, r^, equal to 0.8, while the prior mean of the substitution elasticity among goods in the foreign economy, nf, is set to 1.5. Continuing with the standard deviations of shocks20, we set the standard deviation of the stationary technology shock, ae, to 0.007, and the standard deviation of the unit-root technology shock, , is assumed to be 0.002, which is the value used by Altig et al. (2011). 20 In order to decrease the degree of non-linearity when estimating the model, the mark-up shocks in the Phillips curves, as well as the investment-specific technology shock, the labour supply shock and the consumption preference shock enter into the equations in an additive way. 162 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 162 The size of the risk premium shock, ag, and the prior on the risk premium parameter related to net foreign assets, <, are set to 0.0005 and 0.045, respectively. Based on the residuals from a first-order autoregression of the series obtained when substracting the HP-trend in domestic output from the HP-trend in foreign output, we set the size of the asymmetric technology shock, a, to 0.003. The consumption preference, labour supply and investment-specific technology shocks, a^c, a^h and aY, respectively, are assumed to have the prior mean of 0.002, which is similar to Adolfson et al. (2007). Since we have little information about the properties of these shocks, we choose very loose priors with infinite variances. Regarding the foreign shocks, there are three standard deviations of shocks which need to be specified, namely the standard deviation of the foreign output shock, foreign inflation shock and foreign interest rate shock. We fix their values at the standard deviations of residuals obtained from a pre-estimated foreign VAR model. The standard deviation for the foreign output shock, ay*, is, therefore, set to 0.004, the foreign inflation shock, an*, is assumed to have a standard deviation of 0.002, while the standard deviation for the foreign interest rate shock, aR*, is set to 0.003. Finally, turning to the parameters of the fiscal rule, the prior on the persistence parameter (pg) follows a beta distribution with a mean of 0.6 and a standard deviation of 0.2. The priors on the feedback coefficients are assumed to be gamma distributed. We set their values as follows: the prior mean of the feedback coefficient on inflation (a, is 0.03. Regarding the parameters in the fiscal policy rule, we find that the feedback coefficient of government consumption to inflation, , is estimated at 0.22, the estimated feedback coefficient of output gap, $y, is 0.08, while the estimated feedback coefficients of public debt and government deficit, and are equal to 0.06 and 0.05, respectively. It is worth noting that the latter two parameters are driven by a prior. This can be explained by the fact that we do not use the data on public debt and government deficit in the estimation. The persistence parameter in the fiscal rule, pg, is estimated to be 0.50, which indicates a moderate degree of persistence in government consumption. Finally, we consider the parameters associated with the persistence and volatility of shock processes (see Table 4). We find that the autoregressive parameters are estimated to lie between 0.22 for the consumption preference shock and 0.96 for the unit-root technology shock. In general, the level of persistence of stochastic processes is not very high, indicating that the model contains a sufficiently persistent endogenous propagation mechanism. Turning to the estimated standard deviations of shocks, we find that the most volatile are the imported investment mark-up shocks and the investment-specific technology shock, with standard deviations of 0.3345 and 0.0309, respectively, while the least volatile is the unit-root technology shock with a standard deviation equal to 0.0013. Table 3: Prior and posterior distribution of structural parameters Prior distribution Posterior distribution Description Parameter Type Mean Std. Dev./Df Mode Std. Dev. Mean 5 0 b 95 % Calvo wages tw Beta 0.500 0.200 0.5775 0.0881 0.5568 0.4173 0.6918 Calvo domestic prices id Beta 0.850 0.100 0.9044 0.0206 0.9018 0.8639 0.9365 Calvo import consumption prices "C771, C Beta 0.500 0.200 0.7569 0.0957 0.7051 0.5352 0.8673 Calvo import investment prices ^m,i Beta 0.500 0.200 0.6293 0.1099 0.5195 0.2910 0.7509 Calvo export prices Beta 0.750 0.100 0.8954 0.0439 0.8655 0.7689 0.9702 Calvo employment te Beta 0.500 0.200 0.8112 0.0319 0.8232 0.7762 0.8757 Indexation wages t%iD Beta 0.500 0.200 0.5927 0.1770 0.6016 0.3491 0.8291 Indexation domestic prices Kd Beta 0.400 0.100 0.2013 0.0643 0.2181 0.1064 0.3172 Indexation import consumption prices Beta 0.400 0.100 0.3379 0.1110 0.3510 0.1951 0.5056 Indexation import investment prices Beta 0.400 0.100 0.3250 0.0957 0.3349 0.1814 0.4848 Indexation export prices Kx Beta 0.400 0.100 0.3293 0.0996 0.3309 0.1786 0.4764 Investment adjustment cost s* Normal 7.694 1.500 8.6319 1.2547 8.6526 6.5778 10.6581 Habit formation b Beta 0.650 0.200 0.9442 0.0214 0.9413 0.9081 0.9761 Substitution elasticity investment m Inv. Gamma 0.800 inf 0.2860 0.0552 0.2925 0.2001 0.3847 Substitution elasticity foreign Vf Inv. Gamma 1.500 2 1.1934 0.4019 1.3696 0.6160 2.1375 Technology growth ßz Beta 1.006 0.005 1.0061 0.0005 1.0061 1.0054 1.0068 Risk premium 4>a Beta 0.045 0.02 0.0234 0.0103 0.0296 0.0115 0.0460 Policy parameters Policy rule: lagged gov. consumption Pa Beta 0.600 0.200 0.5149 0.1585 0.4987 0.3314 0.6619 Policy rule: inflation 4>ir Gamma 0.25 0.15 0.1900 0.0990 0.2234 0.0623 0.3765 Policy rule: output gap 4>y Gamma 0.25 0.15 0.0649 0.0491 0.0834 0.0148 0.1451 Policy rule: public debt 4>b Gamma 0.05 0.01 0.0508 0.0160 0.0553 0.0381 0.0716 Policy rule: gov. deficit 4>def Gamma 0.05 0.01 0.0495 0.0101 0.0514 0.0339 0.0674 Table 4: Prior and posterior distribution of shock processes Prior distribution Posterior distribution Description Parameter Type Mean Std. Dev./Df Mode Std. Dev. Mean 5 % 95 % Persistence parameters Unit-root technology shock iVs Beta 0.600 0.200 0.9707 0.0167 0.9577 0.9272 0.9899 Stationary technology shock Pf- Beta 0.600 0.200 0.8697 0.1096 0.7887 0.6071 0.9613 Investment-specific technology shock pr Beta 0.500 0.200 0.2872 0.1381 0.2775 0.0823 0.4677 Asymmetric technology shock Pz* Beta 0.500 0.200 0.9872 0.0091 0.9822 0.9678 0.9971 Consumption preference shock P Beta 0.500 0.200 0.9329 0.0153 0.9317 0.9076 0.9577 Domestic mark-up shock p^d Beta 0.500 0.200 0.4599 0.0988 0.4742 0.3051 0.6394 Import consumption mark-up shock P^m.c Beta 0.500 0.200 0.4727 0.2155 0.5425 0.2412 0.8520 Import investment mark-up shock P\n,i Beta 0.500 0.200 0.1670 0.1219 0.3121 0.0500 0.5801 Export mark-up shock P^x Beta 0.500 0.200 0.2893 0.1458 0.3566 0.0859 0.6103 Rate of transfers shock PrtT Beta 0.500 0.200 0.6550 0.1105 0.5941 0.3635 0.8286 Standard deviations Unit-root technology shock Inv. Gamma 0.002 inf 0.0011 0.0002 0.0013 0.0008 0.0017 Stationary technology shock o-e Inv. Gamma 0.007 inf 0.0126 0.0056 0.0210 0.0071 0.0380 Investment-specific technology shock ay Inv. Gamma 0.002 inf 0.0304 0.0040 0.0309 0.0246 0.0367 Asymmetric technology shock a-,* Inv. Gamma 0.003 inf 0.0037 0.0005 0.0038 0.0030 0.0047 Consumption preference shock Inv. Gamma 0.002 inf 0.0044 0.0006 0.0044 0.0035 0.0054 Labour supply shock Inv. Gamma 0.002 inf 0.0010 0.0005 0.0015 0.0006 0.0023 Risk premium shock a4> Inv. Gamma 0.0005 inf 0.0030 0.0003 0.0032 0.0026 0.0036 Domestic mark-up shock Inv. Gamma 0.003 inf 0.0024 0.0003 0.0024 0.0018 0.0030 Import consumption mark-up shock CAm,c Inv. Gamma 0.003 inf 0.0026 0.0004 0.0028 0.0019 0.0037 Import investment mark-up shock a>-m,i Inv. Gamma 0.003 inf 0.2533 0.0756 0.3345 0.1470 0.5169 Table continued on next page Description Parameter Prior distribution Posterior distribution Type Mean Std. Dev./Df Mode Std. Dev. Mean 5 0 b 95 % Export mark-up shock Inv. Gamma 0.003 inf 0.0112 0.0043 0.0113 0.0044 0.0184 Government consumption shock aa Inv. Gamma 0.002 inf 0.0049 0.0006 0.0050 0.0041 0.0058 Rate of transfers shock z,t + a%H t + agCjh + «lo^f l a0 \ «i «2 «3 «4 a.5 a6 «7 a8 «9 \aw/ Euler equation for consumption: i bw £,w \ vlXw - bw (1 + ßtl) bw ßtw bw tw bw ߣw bw tw Kw bw ߣw Kw (1 - Xw ) - (1 - Xw ) O-L - (1 - Xw ) (1 - Xw) (1 - t y + T b) [(1 - ßtw )(1 - tw )] . = 0, (57) Et -bßct+i + (M + b2ß) ß - b^zct-i + b^z (ßz,t - ßßz,t+i) + (nz - bß) (nz - b) ibz,t + (Mz - bß) ((Hz - b) Ytt'd - (Hz - b) Uz et - bßct+i (58) First order condition w.r.t. it: E t {Pk',t + "Tt - hIS'^H - it-i) - ß (it+i - it^ + ßz,t - ßßz,t+i] } = 0 = 0. (59) x n = t x b b w E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY 185 First order condition w.r.t. bt+1 : Et -4>z,t + n^z - t k ß Hz n t k +ß- Hz n z,t+1 - ßz,t+1 - nt+1 + Rt ( z,t+1 - Hz,t+1 - nt+1 (60) First order condition w.r.t. kt+1 : Et ? - ? ß (! - S)Y Y Wz,t + ßz,t+1 - Wz,t+1--Pk',t+1 + Pk',t Hz Hz - ß (1 - 5)fi k 't+1 (61) Law of motion for capital: kt+i =(1 - s) — kt - (1 - s) — Hz,t + Hz Hz 1 - (1 - s) — Hz Yt + 1 - (1 - S) — Hz «t- (62) Capacity utilization rate: ^ T k 1 ^k Ut = kt - kt = — rt. (63) Aggregate resource constraint: (1 - uc) (r'T - {ct + VcYC'd) +(1 - (y^T - (it + ni7i'd) + 9 „ ~Yt y +y [yt - nf Yt ' + ¿t \d [êt + a(kt - Hz,t) + (1 - a) Ht] - (1 - tk) rkk-(kt - %) - y Hz (64) Equilibrium law of motion for net foreign assets: ^ *----- x , * , * , / m I -m\ ^T at = - y mct - nf y Yt + y yt + y zt + (c + « ) Yt m T /1 \ ( cd)-(1 — nc) s-mc.d . ^ 1 -cm [-nc (1 - Wc) (7c'd) Yt + Ctj , -m [ M \ ( i d )-(1-Vi) „ mi,d . Y 1 . R ~ +«m [-ni (1 - Wi) {7id) 7t +«tj + nHz at 1 . (65) CPI inflation: H z a a 186 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 186 n = [(i - *c) h d'c) 1—nd + [K) (Yrn1-Vc] n?'c- (66) Investment price inflation: ni = [(1 - -i) (7d'i)1-ni] nd + [(<*) ^r^] ? (67) Gross domestic product: Vt = [êt + a (kt - jjz^ + (1 - a) Êt] . (68) Real effective exchange rate: / c,mc\ — (1—ric) "mC'd -—-x xt = --c (y ' ) y Yt - Yt - mct. (69) Employment equation: AÊt EtAÊt+1 + (1 -(1- 3e) (Êt - Êt) . (70) t 1 + ( t t+1 (1 + () V t 7 y ' Domestic interest rate: Rt = RI + 4>t - 4>aàt- (71) Government budget constraint: bbt + ttt = (Rt-i + bt-i - ndt - jz't) + trtrt + ggt. (72) ndjz V / Government expenditures: - tr r g ^ r R g r b— 1 r n gext =-trt +--gt +---——— Rt-i + t 1 »t i / D 1 \ "t—1 i gex gex gexn^,z (R — 1) gex gexn^z (73) r ßz,t gex Transfers to households: trt = if + wt + Ht. (74) Fiscal policy rule for government consumption: gt = Pggt-1 - n - Vt - fobt - $ddef t + £g,t- (75) E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY 187 Tax on consumption: ta = u (y cmc)-(i-nc ) Ymc,d+¿t. (76) Taxes and contributions on wages: tbt = êt + Ht. (77) Public debt interest payments: Rb nßz -Rt-1 + (R - 1) b t (R - 1) K (R - 1) b - bt-1 -nßz n^z -nt - nßz ßz,t- (78) Interest on the amount of the capital services: tdt = — (ft + kt - ßz,t) Hz V ) (79) Interest on the amount of foreign bond holdings: ae = tt = R1 nßz -at-1. (80) Profit of domestic firms: nd fXd - i nt = y OV ) yt - jt (y+a ißz,t+Ht - ¿t) + êt - ¿t (81) Profit of importing firms: It = < c"' l y Ht = jc-^ - Cm^mc,d - j.) et |i^Ymi -d - - (1 - Uc) ( 1 ï c \ Yc,mCYmc,d J 1—nc + Ymc, dcm{ Ymc,d + Ymi,d - yd)tt - (1 - Ui) (-r-1—) rh + Ymi-dim\ YT'' i Yi,—iY—i,d ¡ I 1 + —¡T - +(-m—-) r yn—,c - 1j yn—,i - 1 j Yt (53) c c Y Profit of exporting firms: 188 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 188 n4= --- X t = -y mct ■ Total tax revenue: Deficit: t cta T y tb T k tt = T^îat + —îb + - de+td+n+f Debt-to-GDP ratio: ttt (n+td+n+f def t = gexgext — îît. (83) (84) (85) Deficit-to-GDP ratio: Relative prices: îy,t = ît — yt. J— def t def „ def yt =---yt. yy * mc,d * mc,d , ^ m,c * d it ' = ît-i' + nt' — n (86) (87) (88) /v mi'd /v mi'd . ^ m,i * d Yt = Yt-i + nt — nt (89) Yt' = Yt' i + n — n (90) ^ f ----- X , S--X'* Yt = mct + Yt (91) Yc/ = uc (jmc'c)(1-nc) Ymc'd (92) ¿,i'd , , mi'i\(1 ni) ¿mi'd Yt = "i [Y ) Yt Exogenous shock processes: it = Piit-1 + £ç,t, £Ç't ~ N (0, af ) (93) (94) X E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 189 where it = {XtcXt, Yt,4>t,Z*t ,r};rfor j = {d,mc,mi,x}. 190 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 190 B DATA SOURCES AND DESCRIPTION Table B.1: List of variables used in the estimation and their sources Symbol Description Country Source Yt GDP. Gross domestic product in millions of euro, chain-linked vol- SI Eurostat umes, reference year 2005, SA Ct Private consumption. Household and NPISH final consumption ex- SI Eurostat penditure in millions of euro, chain-linked volumes, reference year 2005, SA It Investment. Gross fixed capital formation in millions of euro, chain- SI Eurostat linked volumes, reference year 2005, SA Gt Government consumption. Final consumption expenditure of gen- SI Eurostat eral government in millions of euro, chain-linked volumes, refer- ence year 2005, SA Xt Exports. Exports of goods and services in millions of euro, chain- SI Eurostat linked volumes, reference year 2005, SA Mt Imports. Imports of goods and services in millions of euro, chain- SI Eurostat linked volumes, reference year 2005, SA Wt Gross wages and salaries. Gross wages and salaries (income struc- SI SORS ture of GDP), current prices, millions of euro, SA Et Employment. Employment (domestic concept), persons (in 1000), SA SI SORS jjd Pt GDP deflator. Price index, reference year 2005, SA SI Eurostat PC CPI index. Consumer price index, current month/average of the SI Eurostat/ECB year 2005, not SA xt Real exchange rate. Real effective exchange rate, consumer price SI Eurostat index deflator, reference year 2005, 28 trading partners Rt Domestic interest rate. Monetary interest rate on new loans to non- SI BS/IMAD financial corporations in domestic currency in percent Y* Foreign GDP. Gross domestic product in millions of euro, chain- EA12 Eurostat linked volumes, reference year 2005, SA P* Foreign GDP deflator. Price index, reference year 2005, SA EA12 Eurostat R* Foreign interest rate. 12-month money market interest rate in per- EA12 Eurostat cent Notes: SA: seasonally adjusted; SORS: Statistical Office of the Republic of Slovenia; IMAD: Institute of Macroeconomic Analysis and Development of the Republic of Slovenia; BS: Bank of Slovenia E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 191 C PRIOR AND POSTERIOR DISTRIBUTIONS Figure C.1a: Prior and posterior distributions of the structural parameters, friction parameters Notes: Prior (black) vs. posterior (red) distributions for the estimated structural parameters. The gray dashed vertical line is the posterior mode obtained from the posterior kernel maximization. Estimates obtained from Bayesian estimation of the DSGE model using Slovenian macroeconomic data from 1995Q1-2014Q4. E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 192 Figure C.1b: Prior and posterior distributions of the structural parameters, friction parameters (cont.) .004 1. 005 1.006 1. 007 1.008 Notes: Prior (black) vs. posterior (red) distributions for the estimated structural parameters. The gray dashed vertical line is the posterior mode obtained from the posterior kernel maximization. Estimates obtained from Bayesian estimation of the DSGE model using Slovenian macroeconomic data from 1995Q1-2014Q4. E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 193 Notes: Prior (black) vs. posterior (red) distributions for the estimated structural parameters. The gray dashed vertical line is the posterior mode obtained from the posterior kernel maximization. Estimates obtained from Bayesian estimation of the DSGE model using Slovenian macroeconomic data from 1995Q1-2014Q4. 194 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 194 Figure C.1d: Prior and posterior distributions of the structural parameters, shock processes parameters (cont.) Notes: Prior (black) vs. posterior (red) distributions for the estimated structural parameters. The gray dashed vertical line is the posterior mode obtained from the posterior kernel maximization. Estimates obtained from Bayesian estimation of the DSGE model using Slovenian macroeconomic data from 1995Q1-2014Q4. E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 195 Notes: Prior (black) vs. posterior (red) distributions for the estimated structural parameters. The gray dashed vertical line is the posterior mode obtained from the posterior kernel maximization. Estimates obtained from Bayesian estimation of the DSGE model using Slovenian macroeconomic data from 1995Q1-2014Q4. 196 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 196 D DATA AND ONE-SIDED PREDICTED VALUES FROM THE MODEL Figure D.1: Data (thick black) and one-sided Kalman-filteredpredictions (thin red) Domestic inflation 051-•-•-•-•-•-— 0.1 0 Consumption 0.5 0 Investment 1995 1998 2001 2004 2007 2010 2013 Output 1995 1998 2001 2004 2007 2010 2013 Export 1995 1998 2001 2004 2007 2010 2013 Import 0 0 0 1995 1998 2001 2004 2007 2010 2013 Government consumption 1995 1998 2001 2004 2007 2010 2013 CPI inflation 1995 1998 2001 2004 2007 2010 2013 Real exchange rate 0 ^rw^y^ 0 0 1995 1998 2001 2004 2007 2010 2013 Real wage 1995 1998 2001 2004 2007 2010 2013 Employment 1995 1998 2001 2004 2007 2010 2013 Foreign output 0 0 1995 1998 2001 2004 2007 2010 2013 Foreign inflation 1995 1998 2001 2004 2007 2010 2013 Foreign interest rate 1995 1998 2001 2004 2007 2010 2013 Domestic interest rate 01 y^/w^^ 0'—1—1—1—1—1— 0.1 0 19 0.2 0 19 —— Data -Model (one-sided Kalman filtered prédictions)! Notes: The plot shows deviations from steady state/trend. E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 197 E SMOOTHED SHOCKS Figure E.1a: Smoothed (two-sided Kalman filtered) estimates of the structural shocks (deviations from steady state) Unit-root technology 0.01 0.005 0 -0.005 -0.01 0.015 0.01 0.005 0 -0.005 1995 1998 2001 2004 2007 2010 2013 Asymmetric technology Stationary technology 0.1 I-■-■-■-■-- 0.05 0 -0.05 -0.1 I-■-■-■-■-■-- 1995 1998 2001 2004 2007 2010 2013 Consumption preference 0.01 0.005 0 -0.005 -0.01 Investment-specific technology 0.015 0.01 0.005 0 -0.005 Domestic markup 0.05 0 -0.05 -0.1 1995 1998 2001 2004 2007 2010 2013 x 10-3 Labour supply 1995 1998 2001 2004 2007 2010 2013 Imported consumption markup 0.015 0.01 0.005 0 -0.005 - Higest posterior density interval (HPDI) Notes: The plot shows deviations from steady state. -0.0151-1-1-'-1-1-— 1995 1998 2001 2004 2007 2010 2013 -0.0151-1-1-'-1-1-— 1995 1998 2001 2004 2007 2010 2013 Risk premium -0.01 1-1-1-'-1-1-— 1995 1998 2001 2004 2007 2010 2013 Mean 198 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 198 Figure E.1b: Smoothed (two-sided Kalman filtered) estimates of the structural shocks (deviations from steady state) (cont.) Imported investment markup 0.01 0.005 0 -0.005 -0.01 1995 1 998 2001 2004 2007 2010 2013 Foreign inflation 1995 1 998 2001 2004 2007 2010 2013 Foreign interest rate ,SM Mi „jU fpf/i...... U> 0.01 0.005 0 -0.005 -0.01 -0.015 1995 1998 2001 2004 2007 2010 2013 Government spending 1995 1998 2001 2004 2007 2010 2013 0.03 0.02 0.01 0 -0.01 -0.02 1995 1998 2001 2004 2007 2010 2013 - Higest posterior density interval (HPDI) Export markup Foreign output 0.05 x 10 Mean Notes: The plot shows deviations from steady state. E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 199 F IMPULSE RESPONSE FUNCTIONS Figure F. 1: Impulse responses to a unit-root technology shock Notes: The solid line shows the average impulse responses results over the MCMC parameter draws; the dashed lines at the 5% and 95% posterior intervals. The impulse horizon is measured in quarters. 200 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Figure F.2: Impulse responses to a stationary technology shock Notes: The solid line shows the average impulse responses results over the MCMC parameter draws; the dashed lines at the 5% and 95% posterior intervals. The impulse horizon is measured in quarters. E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 201 Figure F.3: Impulse responses to an investment-specific technology shock Notes: The solid line shows the average impulse responses results over the MCMC parameter draws; the dashed lines at the 5% and 95% posterior intervals. The impulse horizon is measured in quarters. 202 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Figure F.4: Impulse responses to a consumption preference shock Notes: The solid line shows the average impulse responses results over the MCMC parameter draws; the dashed lines at the 5% and 95% posterior intervals. The impulse horizon is measured in quarters. E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 203 Figure F.5: Impulse responses to a labour supply shock Notes: The solid line shows the average impulse responses results over the MCMC parameter draws; the dashed lines at the 5% and 95% posterior intervals. The impulse horizon is measured in quarters. 204 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Figure F.6: Impulse responses to a domestic mark-up shock Notes: The solid line shows the average impulse responses results over the MCMC parameter draws; the dashed lines at the 5% and 95% posterior intervals. The impulse horizon is measured in quarters. E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 205 Figure F.7: Impulse responses to an imported consumption mark-up shock Notes: The solid line shows the average impulse responses results over the MCMC parameter draws; the dashed lines at the 5% and 95% posterior intervals. The impulse horizon is measured in quarters. 206 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Figure F.8: Impulse responses to an imported investment mark-up shock Notes: The solid line shows the average impulse responses results over the MCMC parameter draws; the dashed lines at the 5% and 95% posterior intervals. The impulse horizon is measured in quarters. E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 207 Figure F.9: Impulse responses to an export mark-up shock Notes: The solid line shows the average impulse responses results over the MCMC parameter draws; the dashed lines at the 5% and 95% posterior intervals. The impulse horizon is measured in quarters. 208 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 G HISTORICAL DECOMPOSITIONS Figure G.1: Historical decomposition of consumption growth in terms of structural shocks Notes: The smoothed observed time series is plotted excluding its mean. E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 209 Figure G.2: Historical decomposition of investment growth in terms of structural shocks Notes: The smoothed observed time series is plotted excluding its mean. 210 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Figure G.3: Historical decomposition of import growth in terms of structural shocks Notes: The smoothed observed time series is plotted excluding its mean. E/B/R A. KUSTRIN | A DSGE MODEL FOR THE SLOVENIAN ECONOMY ... 211 Figure G.4: Historical decomposition of export growth in terms of structural shocks Notes: The smoothed observed time series is plotted excluding its mean. 212 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 213-242 213 The influence of socio-demographic characteristics on environmental concern and ecologically conscious consumer behaviour among macedonian consumers BARBARA CATER1 Received: April 10, 2017 JULIJANA SERAFIMOVA2 Accepted: February 25, 2019 ABSTRACT: Western Balkan countries face a decisive moment in the development of their economies, societies and the environment. According to the European Environment Agency, household consumption patterns in these countries have changed rapidly in the recent years and are of key interest due to the fact that unsustainable patterns of consumption are an important cause of environmental problems. The main purpose of this paper is to add to the body of knowledge on environmental consumer profiling, especially in the context of posttransition economies. We present the results of a survey on 323 Macedonian consumers, relating their attitudes and consumption patterns to socio-demographic characteristics. Key words: environmental concern, ecologically conscious consumer behaviour, socio-demographic characteristics, the Republic of North Macedonia JEL classification: M31, Q01 DOI: 10.15458/ebr.84 1 INTORDUCTION Over the last decades, substantial efforts have been put into policies aimed at production processes to cope with the depletion of natural resources, climate change, air pollution and waste generation. However, more recently the focus has shifted to the consumption perspective, as high levels of consumption endanger the quality of the environment and the processes of sustainable development (Liobikene & Bernatoniene, 2017). Unsustainable consumption puts a threefold of environmental burdens to the environment: via the natural resource depletion, pollution and biodiversity reduction. Consumption is directly related to global climate change, identified as the major environmental issue of modern life. Hence, one of the main responsibilities for environmental degradation lies with the consumers and their consumption choices (Berglund & Matti, 2006). Therefore, in order to reduce the environmental consequences of consumption, it is essential to stimulate 1 Corresponding author, University of Ljubljana, School of Economics and Business, Ljubljana, Slovenia, e-mail: Barbara.cater@ef.uni-lj.si 2 MOD, North Macedonia 214 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 the consumption of environmentally friendly products (Liobikiene, Grinceviciene, & Bernatoniene, 2017). Understanding consumer behaviour is important for any marketer and it is especially critical for environmental products. There is a general belief among researchers and environmental activists that by buying environmentally friendly products consumers can contribute significantly to improve the quality of the environment (Abdul-Muhmim, 2007). Groening, Sarkis and Zhu (2018) point out that the need to understand green purchasing behaviour is especially relevant owing to environmental, scientific, and communication developments, such as the internet and social media, and increases environmental awareness and concerns in consumers. Green consumers are those who associate the act of purchasing or consuming products with the possibility of acting in line with preservation of the environment (Hailes, 2007). In a similar vein, Roberts (1996) defines ecologically conscious consumers as individuals who try to consume only products that produce the least or do not cause any impact on the environment. When profiling green consumers, companies can use standard bases for customer segmentation. On the one hand, many companies focus primarily on socio-demographics when segmenting the market for green products, due to the fact that these segmentation measures are easily available and simple to implement (Park, Choi, & Kim, 2012; Patel, Modi, & Paul, 2017). Furthermore, socio-demographic variables are often used to improve the accessibility of segments for subsequent profiling and targeting strategies (Park et al., 2012). However, a review of literature indicates that several studies on socio-demographic profiling of green consumers report mixed results, therefore limiting the value of the use of socio-demographic variables for consumer segmentation and profiling (Diamantopoulos et al., 2003; Fisher, Bashyal, & Bachman, 2012). Further studies are therefore needed to determine whether these characteristics play a significant role in green consumer profiling, especially in markets where marketing research is not very developed. The reason why the present study focuses on socio-demographics is that in transition and post-transition markets, which are less developed in terms of marketing research, it is easier for companies to use simple variables for consumer profiling. However, it is important to establish how relevant they are in profiling green consumers and this is where this study aims to make a contribution. The main purpose of this paper is to add to the body of knowledge on environmental concern and ecologically conscious consumer behaviour, especially in the context of transition and post-transition economies. Past studies on the attitudes of consumers toward the environment and ecologically conscious consumer behaviour have been conducted mostly in developed or developing countries (for an overview see Patel, Modi, & Paul, 2017), with less focus on transition and post-transition countries. However, according to the European Environment Agency (EEA Report No 1/2010, 2010), household consumption patterns in the Western Balkan countries have changed rapidly and are of key interest due to the fact that unsustainable patterns of consumption are an important cause of environmental problems. Therefore, it is important to advance our knowledge about E/B/R B. CATER, J. SERAFIMOVA | THE INFLUENCE OF SOCIO-DEMOGRAPHIC 215 environmental attitudes and consumer behaviour in these markets. Of Western Balkan countries this study focuses on the Republic of North Macedonia, which has the worst air quality in Europe (Migrio, 2018). The problem intensifies every winter as a consequence of industrial emissions, smoke from wood-burning stoves and exhaust fumes from old cars (Georgievski, 2018), of which the last two pertain to consumers and could be better managed by having a deeper insight in consumer environmental concern and behaviour. The contribution of this study is therefore not only academic, but it gives implications for every day practice of policy makers and domestic and international marketers that are present or plan to enter this market. The main goal of this research is to analyse consumers' environmental concern and ecologically conscious consumer behaviour and to discover if significant differences exist based on socio-demographic profiles that would enable companies to use them in profiling green consumers. This study should therefore provide answers to the following core research questions: (1) What is the awareness of the importance of environmental issues in the examined context? (2) What is the presence of ecologically conscious consumer behaviour in the market? (3) How are environmental concern and ecologically conscious consumer behaviour related to socio-demographic characteristics? The paper is structured as follows. First, we define environmental concern and ecologically conscious consumer behaviour. This is followed by the section on demographic characteristics and their influence on environmental concern and ecologically conscious consumer behaviour. In the next section we present research design and research results. This is followed by a discussion of implications for theory and practice, limitations and opportunities for future research. 2 ENVIRONMENTAL CONCERN AND ECOLOGICALLY CONSCIOUS CONSUMER BEHAVIOUR 2.1 Environmental concern There are some variations in the definition of environmental concern across the literature, but most researchers use the term to refer to attitudes about environmental issues or perceptions that such issues are important (Cruz, 2017). Liu, Vedlitz, and Shi (2014) stress that identifying and understanding the determinant factors of consumers' environmental concern is one of the major necessary conditions to make sound policies and promote consumers' engagement in pro-environmental behaviour. As evidenced, almost all Europeans say that environmental protection is important to them personally and over 75% believe that environmental problems have a direct effect on their lives (Special Eurobarometer 416, 2014). By recognizing the severity of environmental problems, people in general have become more environmentally aware (Han, Hsu, & Lee, 216 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 2009) and their sensitivity and consciousness toward environmental issues should have an effect on their buying behaviour (Brochado, Teiga, & Oliveira-Brochado, 2017). Despite traditional beliefs that environmental concern is limited to the wealthy nations, research shows that consumer environmental concern is not dependent on national wealth (Dunlap & York, 2008). People in poor and developing countries have shown as much concern about environmental issues as those in developed countries, which is confirmed in North Macedonia as well (Angelovska, Sotiroska, & Angelovska, 2012). 2.2 Ecologically conscious consumer behaviour Kuchinka et al. (2018) point out that in general consumer behaviour is primarily motivated by benefits and costs, and can bring instant personal gain or gratification benefit, while environmentally conscious behaviour is attempting to achieve a future outcome with benefits for the entire society. If consumers care about the environment, they will most likely consider the consequences of their purchasing decisions (Brochado et al., 2017). There has been a lot of research attention devoted to the study of consumers' environmentally friendly behaviour because it is extremely beneficial for companies to understand what factors influence consumers' behaviour (Fisher et al., 2012). The growing importance of protecting the environment has changed the way people see the market, and consumers now believe that their purchasing behaviour will find a better match in products (Akehurst, Afonso, & Gon^alves, 2012). As already pointed out in the introduction, green (named also pro-environmental or ecologically conscious) consumers associate the act of purchasing or consuming products with the possibility of acting in line with preservation of the environment (Hailes, 2007). In this study, the focus is on the pro-environmental purchase behaviour (e.g., eco-labelled products, reusable packaging, lower emission cars, and low-energy appliances) and not on the pro-environmental consumption (e.g., household waste separation, noise control, use of recycling points and water saving) (Sánchez, López-Mosquera, & Lera-López, 2016). Researchers have studied several factors leading to ecologically conscious consumer behaviour. Groening, Sarkis, and Zhu (2018) provide a comprehensive overview of green marketing and green consumerism theoretical relationships. They draw upon existing models and include topics featuring factors affecting relationships between attitudes and behaviours (e.g., situational, sociological and psychological factors) and barriers to environmental action. Based on the prior consumer decision making literature, Groening et al. (2018) propose six theory groupings: values and knowledge, beliefs, attitudes, intentions, motivations, and social confirmation. Values and knowledge are the foundation for beliefs, which in turn form attitudes that predict behaviour (as in Theory of Reasoned Action by Fishbein & Ajzen, 2011). However, contradictory results were found regarding the relationship between attitude and behaviour, leading to conclusion that the fact that E/B/R B. CATER, J. SERAFIMOVA | THE INFLUENCE OF SOCIO-DEMOGRAPHIC 217 consumers exhibit a positive attitude towards green products does not necessarily indicate they will engage in green purchase behaviour (Kuchinka et al., 2018). Groening et al. (2018) also present theory groupings that could explain why attitudes do not directly result in green purchase behaviour, including intentions, motivations, facilitators or instantiaters, and social confirmation. 3 SOCIO-DEMOGRAPHIC CHARACTERISTICS AND THEIR INFLUENCE ON ENVIRONMENTAL CONCERN AND ECOLOGICALLY CONSCIOUS CONSUMER BEHAVIOUR The latest green marketing consumer-level literature has among others illustrated the focus on identifying the profile of the environmentally conscious consumers (e.g., Akehurst et al., 2012; Brochado et al., 2017; Sánchez et al., 2016; Pinto et al., 2014), including the socio-demographic characteristics of environmentally conscious consumers, such as age, gender, education, income and so on. The inconsistency of the results in a variety of studies (for an overview see Diamantopoulos et al., 2003; Fisher et al., 2012; Verain et al., 2012) has perhaps shown how complicated it is to accurately identify the demographic profile of an environmentally conscious consumer. Even though these results provide insufficient data for profiling environmentally conscious consumers, they can be a useful tool to marketers in describing market segments (D'Souza et al., 2007). In the following sections we present the socio-demographic characteristics that have been most often related to environmental concern and environmentally conscious consumer behaviour (Diamantopoulos et al., 2003; Fisher et al., 2012) and we propose hypotheses about the Macedonian consumers. Groening et al. (2018) provide a large-scale review of more than 20 consumer-level theories used in the field of green marketing. This study builds on role theory (Biddle, 1986) to explain the differences in consumers' environmental concern and ecologically conscious consumer behaviour. Biddle (1986) proposes that individuals hold social positions in society which reflect their roles and create expectations for their own behaviours and others' expectations of behaviour. Role theory can be used both to explain and predict social behaviour of individuals based on situations and identities. According to role theory, different groups of people playing different roles exhibit different patterned behaviours. Gender role theory argues that women and men behave according to roles related with their genders. Han, Hsu and Lee (2009) provide a review of studies that found differences in gender roles analysed in environmental studies. These studies show that women are more nurturing, which is associated with their greater concern for the environment and willingness-to-pay more for green products (Han et al., 2009). Role theory has also been utilised to explain the differences in pro-environmental behaviours among sustainable and apathetic consumers (Park & Ha, 2012). In line with role theory this study proposes that there are differences in attitudes and behaviour of consumers based on the roles they play in the society (for example, based on gender, educational level, income level and similar). Argumentations for the differences are provided in the next sections. 218 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 This study therefore focuses on socio-demographic characteristics and with those related social roles in explaining environmental concern and ecologically conscious consumer behaviour. Due to the low explanatory power of socio-demographic characteristics to predict ecologically conscious consumer behaviour (e.g., Roberts, 1996; Diamantopolous et al., 2003; Brochado et al., 2017), in the last step the analysis will be complemented by adding environmental concern as an additional predictor of ecologically conscious consumer behaviour. Various studies report that consumers with higher environmental concern are more likely to evaluate the environmental consequences of their purchase behaviour and that environmental concern positively influences ecologically conscious consumer behaviour (Mainieri et al., 2007; Nath et al., 2013; Brochado et al., 2017). 3.1 Gender Gender has been one of the most often used variables when profiling green consumers. One important, well-established finding is that females are more environmentally sensitive about general environmental issues than males and more likely to express concern about the social and environmental impacts of their consumption (Koos, 2011; Zelezny, Chua, & Aldrich, 2000; Park et al., 2012). They consider the environmental issues in the purchase decisions to a larger extent and are more willing to engage in ecologically conscious consumption than men (Brochado et al., 2017; Liobikiene et al., 2017; Sánchez et al., 2016; Diamantopoulos et al., 2003; Luchs & Mooradian, 2012). Furthermore, women show more willingness to buy and pay a premium price for environmentally benign products (Laroche, Bergeron, & Barbaro-Forleo, 2001). On the other hand, Mostafa (2007) found that men possess a deeper knowledge of environmental issues, express higher levels of environmental concern and have more positive attitudes towards green purchase, while Chen at al. (2011) and Rice (2006) found no significant relationship of gender with environmental variables. Based on the results of the study of purchase differences of environmentally labelled products in 18 European countries, women are more likely to consider the environmental issues when they do their shopping (Koos, 2011). Similarly, Zelezny et al. (2000) evaluated 13 studies on environmentally responsible consumption and state that in nine of them women appeared to have a higher level of pro-environmental attitudes and behaviours, three reported no significant differences between sexes, but only one has shown that males were more environmentally concerned than females. Based on the above, we can conclude that gender is an important socio-demographic predictor of environmental concern and ecologically conscious consumer behaviour; women appear to be more concerned about the environment and are more likely to act in accordance to those concerns when making a purchase decision. Therefore, it is hypothesised that: E/B/R B. CATER, J. SERAFIMOVA | THE INFLUENCE OF SOCIO-DEMOGRAPHIC 219 H1a: Females are more concerned about the environment than males. H1b: Females demonstrate more ecologically conscious consumer behaviour than males. 3.2 Age Age is another demographic variable that has been widely examined in past studies. Findings about the age of consumers can provide a useful base in market segmentation, however, the results in relation to this demographic variable have been inconsistent. Most studies reveal that younger individuals are likely to be more sensitive and concerned about environmental issues (Chen & Peng, 2012; Diamantopoulos et al., 2003). On the other hand, Liu et al. (2014) found a positive relationship between age and environmental concern. When researching consumer behaviour, the results are somewhat different. Roberts (1996) found that age is significantly related to ecologically conscious consumer behaviour, concluding that middle aged consumers are more prone to ecologically conscious consumption activities. Likewise, Anic, Jelenc and Sebetic (2015) and Mohr and Schlich (2016) examining sustainable food consumption detected that middle aged respondents show the highest level of environmentally conscious consumption behaviour. Also, Brochado et al. (2017) found that older consumers (compared to the youngest group) are more prone to ecologically conscious consumer behaviour. These results might be due to the fact that younger individuals are mostly students without jobs who have a lower buying power and who cannot afford environmentally friendly products or more expensive alternatives (Jain & Kaur, 2006). On the other hand, some researchers have found that the relationship between age and ecologically conscious consumption is significant and negative (Zimmer, Stafford, & Stafford, 1994). In relation to these mixed findings, Chan (1996) in his two-country study, found that the respondents' age has a significant influence on the environmentally sustainable purchases in Canada (i.e., younger respondents more frequently purchase recyclable products), while no association between these two variables was found for respondents in Hong Kong. Due to the contradicting results related to the relationship between the age of consumers and their environmental concern and environmentally conscious consumer behaviour, we posit exploratory hypotheses, only assuming that differences exist, but not predicting the direction of these differences. H2a: Younger and older consumers differ in terms of environmental concern. H2b: Younger and older consumers differ in terms of ecologically conscious consumer behaviour. 3.3 Educational level A consumer's level of education is in many studies considered as a socio-demographic factor that affects environmental practices of the consumer. In terms of education, most empirical studies have shown that more educated people are more sensitive and 220 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 aware of environmental issues (Zsoka et al., 2013; Zhao, Wu, & Wang, 2014). They show higher preferences for environmental protection and willingness to pay leading to environmentally conscious consumer behaviour (Diamantopoulos et al., 2003; do Pa^o, Raposo, & Filho, 2009; Zhao et al., 2014). For illustration, Koos (2011) in his study on sustainable consumption across Europe states that buying environmentally-labelled products increases with education. Because higher educated people in general are better informed and could understand environmental issues better, they express higher concern about the quality of the environment and have strong desire to protect it. Consequently, they are more willing to practice ecologically conscious consumer behaviour (Torgler & Garcia-Valinas, 2007; Zhao et al., 2014). Based on these findings, it is hypothesised that: H3a: Less educated people are less environmentally concerned than people with higher educational levels. H3b: Less educated people exhibit less ecologically conscious consumer behaviour than people with higher educational levels. 3.4 Income level Consumers with higher income have less economic problems and can turn to other concerns; at the same time they have higher willingness and ability to pay for goods (Franzen & Vogl, 2013). Results from previous research show that consumers with higher income are more interested in protecting the environment (Royne, Levy, & Martinez, 2011) and prefer life style based on environmentally friendly consumption (Anic et al., 2015). A positive relationship between respondents' income and their environmental concern is also confirmed in the studies by Zimmer, Stafford and Stafford (1994) and Roberts (1996). On the other hand, Park et al. (2012) report a non-linear relationship between these two variables. In their study, consumers in the lowest and in the highest income group were found to be the most environmentally concerned. In relation to ecologically conscious consumer behavior, the results from previous research are somehow mixed but still mostly indicate that income has positive and meaningful influence on purchase decision (do Pa^o et al., 2009; Hines, Herald, & Audrey, 1987; Anic et al., 2015; Welsch & Kuhling, 2009). This notion is mainly based on the fact that pro-environmental products are usually priced higher than conventional ones, and people with higher income may be more likely to buy these products because they can bear the associated marginal increase in their cost (Zhao et al., 2014). On the other hand, some researchers have found that people with a lower level of income are more prone to ecologically conscious consumer behaviour (Roberts, 1996) or even that the income level does not affect their green consumption decisions significantly (Straughan & Roberts, 1999; Ci-Sheng, Xiao-Xia, & Meng, 2016). Therefore, due to contradicting results related to the relationship between income of consumers and their environmental concern and environmentally conscious consumer behaviour, we posit exploratory hypotheses, only assuming that differences exist, but not predicting the direction of these differences. E/B/R B. CATER, J. SERAFIMOVA | THE INFLUENCE OF SOCIO-DEMOGRAPHIC 221 H4a: There are differences in the concern about the environment based on the income level. H4b: There are differences in the ecologically conscious consumer behaviour based on the income level. 3.5 Marital status There have been some attempts to link environmental attitude and behaviour to marital status (Diamantopoulos et al., 2003; Fisher et al., 2012; Chen et al., 2011). The argument behind these relationships is that spouses can act as a social referent in influencing environmental attitude and behaviour (Neuman, 1986). Not many studies found support for the influence of marital status on environmental concern (e.g. Research 2000 in Diamantopoulos et al., 2003). On the other hand, few studies indicate that married people are more likely to participate in green activities (Diamantopoulos et al., 2003; Fisher et al., 2012). Although this is a rarely tested variable in environmental research, we build on argumentation developed by Neuman (1986) and for transitional context expect positive relationships between these variables. H5a: Single people are less concerned about the environment. H5b: Single people exhibit less ecologically conscious consumer behaviour. 3.6 Number of children Research shows that the presence of children in the household positively affects environmental concern and environmentally conscious behaviour (Laroche et al., 2001; Loureirro, McCluskey, & Mittlehammer, 2002). The reason would be that due to discussions on ecology at school children have certain expectations regarding environmentally friendly behaviour of their parents (Schlossberg, 1992). On the other hand, Diamantopoulos et al. (2003) did not find significant relationships between the number of children and environmental consciousness measures (knowledge, attitudes and behaviour), while Fisher et al. (2012) found that only one part of behaviour (usage of recyclable bags) is related to the number of children in the household. In line with role theory and findings of Laroche et al. (2001) and Loureirro et al. (2002) we expect a positive relationship between the number of children and environmental concern and behaviour. H6a: The more children a consumer has, the stronger the concern about the environment. H6b: The more children a consumer has, the greater the participation in ecologically conscious consumer behaviour. 222 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 4 RESEARCH DESIGN 4.1 Questionnaire design Existing scales were used to measure constructs under study. To measure environmental concern we used statements from the Socially Responsible Consumption Behaviour scale (Antil, 1984), while for ecologically conscious consumer behaviour we used statements from the Ecologically Conscious Consumer Behaviour scale (Roberts, 1996). Respondents were presented with statements and they were asked to evaluate them on a five point Likert scale (1 = I entirely disagree, 5 = I entirely agree). The last set of questions was related to demographic characteristics of the respondents. Gender, age, educational level, income, marital status and number of children under 15 years were included. The questionnaire applied for collecting the primary data was translated twice, from English into Macedonian and vice versa, to ensure that all difficulties due to language differences would be minimized and that the meanings of the statements were properly transferred. Then, the questionnaire was tested on a small sample of 15 respondents of different age, gender and educational level. The questionnaire testing was made in order to identify possible problems related to the questionnaire's clarity, bias and possible ambiguity. The participants were asked for their opinion regarding the wording, sequencing and timing as well. No difficulties in understanding the statements were indicated and it was not suggested that the time needed for answering the questions was too long. 4.2 Data collection and sample characteristics The research population is defined as persons over the age of 18 years living in Skopje, the capital of the Republic of North Macedonia. Printed questionnaires were administered to teachers in four primary schools in different areas in Skopje and their students later forwarded them to their parents or grandparents. In addition, questionnaires were distributed to students at a private university and to additional known citizens with different demographic characteristics. Altogether, we distributed 399 questionnaires and 368 were returned (response rate of 81%), while the number of fully filled questionnaires bearing the status of "completed" was 323, on which the final analysis was done. Sample characteristics were compared to the latest attainable official statistical data for the inhabitants of Skopje and the population of North Macedonia acquired from the State Statistical Office of the Republic of North Macedonia. The inspection indicated that despite some deviations the sample was close enough to the population to continue the analysis. Some of the respondents' socio-demographic characteristics used in further analysis are presented in Table 1. Regarding the gender structure, 46.7% of respondents were male and 53.3% female. The average age was 39.6 years (standard deviation 13.4). Regarding the level of education, a substantial number (48.9%) of the respondents completed at least a bachelor degree. The majority reported to have an average monthly household income E/B/R B. CATER, J. SERAFIMOVA | THE INFLUENCE OF SOCIO-DEMOGRAPHIC 223 (62.5%). Additionally, the majority were married or living with a partner (71.8%), while the rest were single, separated, divorced or widowed. The average number of children under the age of 15 years was 1.0 (standard deviation 0.9), where one third of the sample had no children. Table 1: Some demographic characteristics of the respondents Demographic characteristics Frequency Relative frequency in % Age 00 - 20 44 13.6 21 - 30 33 10.2 31 - 40 98 30.3 41 - 50 96 29.7 51 - 60 25 7.7 61 - 70 20 6.2 71 + 7 2.2 Total 323 100.0 Level of education Elementary school 11 3.4 Vocational school 117 36.2 Secondary (high) school 37 11.5 Bachelor degree 139 43.0 Master's degree 12 3.7 PhD 7 2.2 Total 323 100.0 Household average monthly income Below average/ in lower half of below average 12 3.7 Below average/ in upper half of below average 15 4.6 Average 202 62.5 Above average/ in lower half of above average 42 13.0 Above average/ in upper half of above average 37 11.5 I do not know 15 4.6 Total 323 100.0 Marital status Single 73 22.6 Married 229 70.9 Living together without being married 3 0.9 Divorced 8 2.5 Separated 3 0.9 224 ECONOMIC AND BUSINESS REVIEW | VOl. 21 | No. 2 | 2019 Demographic characteristics Frequency Relative frequency in % Widowed 7 2.2 Total 323 100.0 Number of children 0 109 33.7 1 104 32.2 2 105 32.5 3 3 0.9 4 1 0.3 5 1 0.3 Total 323 100.0 4.3 Data analysis We used univariate statistical techniques (frequencies, means and standard deviations) to present sample characteristics and results for the statements measuring environmental concern and ecologically conscious consumer behaviour. The reliability of measurement for the individual constructs (Table 2) was evaluated before the hypotheses test. We tested the hypotheses using independent samples t-test, one-way ANOVA and correlation analysis. In the end, multiple regression analysis was carried out to test the effect of all variables at the same time. Further results validation was performed using clustering and discrimination analysis. The value of reliability coefficient (Cronbach's a) for the ecologically conscious consumer behaviour scale consisting of eleven items is 0.859, which shows good internal consistency of the scale. Cronbach's alpha coefficient for environmental concern (0.610) is below the recommended 0.7 threshold, but since the value of over 0.60 for Cronbach alpha can be still considered acceptable (Kline, 2000, p. 13), we can use both constructs in further analyses. Both constructs are also sufficiently different from each other (correlation coefficient is 0.509, p < 0.01). Table 2: Statistics for environmental concern and ecologically conscious consumer behaviour Summary statistics Environmental measures Number of items Mean Standard deviation Possible range Cronbach's a Environmental concern 6 24.05 3.23 6 - 30 0.610 Ecologically conscious consumer behaviour 11 38.90 7.06 11 - 55 0.859 E/B/R B. CATER, J. SERAFIMOVA | THE INFLUENCE OF SOCIO-DEMOGRAPHIC 225 5 FINDINGS 5.1 Descriptive statistics for environmental concern and ecologically conscious consumer behaviour Descriptive statistics for statements measuring the focal constructs are presented in Tables 3 and 4. Consumer environmental concern was measured with six items. As presented in Table 3, all items have a mean value above the neutral/undecided response option in the range between 3.77 and 4.27, which means that on average, Macedonian consumers are environmentally concerned. The highest average agreement was expressed with the statement that pollution affects their life. Table 3: Descriptive statistics for consumer environmental concern Scale item M SD You feel that pollution affects your life personally. 4.27 0.77 You think all the worried comments made about air and water pollution are all justified. 4.11 0.90 You become incensed when you think about the harm being done to the plant and animal life by pollution. 4.11 0.85 You have often thought that if we could just get by with a little less there would be more left for future generations. 4.00 1.01 Natural resources must be preserved even if people must do without some products. 3.81 0.94 Pollution is presently one of the most critical problems facing this nation. 3.77 1.04 Descriptive statistics for individual scale items of ecologically conscious consumer behaviour are presented in Table 4. All items have a mean value above the neutral/ undecided response option in the range between 3.18 and 4.02. The overall conclusion is that on average the respondents seem to engage in ecologically conscious consumer behaviour, yet the average scores are lower than at environmental concern. The easier behaviour (When you have a choice between two equal products, you always purchase the one less harmful to other people and the environment; M = 4.02, SD = 0.93) is more practiced than the more demanding forms (for example, buying only products that can be recycled and avoiding or not buying products that have excessive packaging). 226 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Table 4: Descriptive statistics for ecologically conscious consumer behaviour Scale item M SD When you have a choice between two equal products, you always purchase the one less harmful to other people and environment. 4.02 0.93 If you understand the potential damage to the environment that some products can cause, you do not purchase those products. 3.78 0.90 When you purchase products, you always make a conscious effort to buy those products that are low in pollutants. 3.74 0.99 You do not buy a product if the company that sells it is ecologically irresponsible. 3.69 1.10 When there is a choice, you always choose the product that contributes to the least amount of pollution. 3.66 0.98 Whenever possible you buy products packaged in reusable containers. 3.54 1.06 You have switched products for ecological reasons. 3.46 1.03 You have convinced some members of your family and friends not to buy some products that are harmful to the environment. 3.35 1.04 You normally make a conscious effort to limit the use of products that are made of or use scarce resources. 3.27 0.84 You try only to buy products that can be recycled. 3.21 1.05 You do not buy products that have excessive packaging. 3.18 0.99 5.2 Testing individual influences of socio-demographics on environmental concern and ecologically conscious consumer behaviour With the first set of hypotheses we tested the effect of gender on environmental concern and ecologically conscious consumer behaviour. Based on an extensive literature review we proposed that women demonstrate more ecologically conscious consumer behaviour than men. The results (Table 5) are in line with the proposed hypotheses. Women are on average more environmentally concerned and report more sustainable consumer behaviour than men. Therefore, H1a and H1b are supported. Table 5: Impact of gender on environmental concern and ecologically conscious consumer behaviour Gender Female Male t-value (1-tailed sig.) Environmental measures M (SD) M (SD) Environmental concern 24.57 (3.07) 23.45 (3.30) 3.16 (0.001) Ecologically conscious consumer behaviour 39.75 (6.43) 37.94 (6.43) 2.28 (0.011) E/B/R B. CATER, J. SERAFIMOVA | THE INFLUENCE OF SOCIO-DEMOGRAPHIC 227 With the second set of hypotheses we tested the effect of age on consumers' attitudes and behaviour. The results of the correlation analysis indicate that there is a significant positive relationship between age and environmental concern (r = 0.229, p < 0.01), as well as age and ecologically conscious consumer behaviour (r = 0.303, p < 0.01). In order to test the differences among age groups we used one-way ANOVA. We used three age groups (30 years and less, 31 to 50 years old, and 51 years and above) to differentiate consumers. The analysis of variance shows that the effect of age for both environmental concepts is significant (F = 16.341, P = 0.000 for environmental concern; F = 28.215, P = 0.000 for ecologically conscious consumer behaviour). The Bonferroni post hoc test indicates that the average for environmental concern is significantly lower in the youngest age group (M = 22.38, SD = 3.16), compared to the other two age groups (for 31 to 50 years old M = 24.40, SD = 2.99, and for 51 years and above M = 25.24, SD = 3.30). The results are similar to the ones about ecologically conscious consumer behaviour. The youngest age group (M = 34.18, SD = 7.82) scored significantly lower than the other two age groups (for 31 to 50 years old M = 39.97, SD = 6.12, and for 51 years and above M = 41.92, SD = 5.93). We can therefore support H2a and H2b that differences exist between younger and older consumers regarding environmental concern and ecologically conscious consumer behaviour. With the third set of hypotheses we tested the influence of educational level on the consumers' environmental concern and ecologically conscious consumer behaviour. The educational level of respondents as an independent variable originally presented with six groups (1 - elementary, 2 - vocational, 3 - secondary, 4 - bachelor degree, 5 - master and 6 - PhD) was regrouped in two groups (respondents with lower education comprising groups 1 to 3 and respondents with higher education comprising groups 4 to 6). Although the results indicate that the respondents with lower education exhibit lower environmental concern and ecologically conscious consumer behaviour, the differences between the two groups are not statistically significant (Table 6). Therefore, at a = 0.05 we cannot conclude that in this research context less educated people exhibit lower environmental concern and less ecologically conscious consumer behaviour than people with higher educational levels. We also conducted a more detailed analysis (one-way ANOVA), comparing environmental concern and ecologically conscious consumer behaviour among all six educational groups. The results indicate there are no statistically significant differences among different educational groups (F = 0.911, P = 0.474 for environmental concern; F = 1.167, P = 0.325 for ecologically conscious consumer behaviour). Thus, hypotheses H3a and H3b are not supported. 228 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Table 6: Impact of educational level on environmental concern and ecologically conscious consumer behaviour Educational level Lower Higher t-value (1-tailed sig.) Environmental measures M (SD) M (SD) Environmental concern 23.78 (3.05) 24.32 (3.39) -1.51 (0.065) Ecologically conscious consumer behaviour 38.36 (7.28) 39.46 (6.81) -1.40 (0.081) Next, we tested the effect of household income on environmental variables. We regrouped the original five categories of household income into three (below average, average and above average) to ensure sufficiently large groups for analysis. The results indicate that significant differences exist between these three groups for environmental concern (F = 6.635, P = 0.002) but not for ecologically conscious consumer behaviour (F = 1.720, P = 0.181). There are statistically significant differences in environmental concern between consumers with below average household income (M = 25.81, SD = 3.24) and those with above average household income (M = 23.27, SD = 3.41), indicating that those coming from less wealthy households are more concerned about the environment. H4a is therefore supported, while H4b is not. The results for the influence of marital status on environmental variables (Table 7) indicate that on average single people are less environmentally concerned and practice less ecologically conscious consumer behaviour. Therefore, H5a and H5b are supported. Table 7: Impact of marital status on environmental concern and ecologically conscious consumer behaviour Marital status Single Married t-value (1-tailed sig.) Environmental measures M (SD) M (SD) Environmental concern 22.96 (3.27) 24.48 (3.11) -3.89 (0.000) Ecologically conscious consumer behaviour 34.76 (7.24) 40.53 (6.30) -7.09 (0.000) The last set of hypotheses tested the relationship between the number of children (under the age of 15) and environmental variables. The results of the correlation analysis indicate that there is a significant positive relationship between the number of children and environmental concern (r = 0.172, P < 0.01) and the number of children and ecologically conscious consumer behaviour (r = 0.235, P < 0.01). H6a and H6b are thus supported. E/B/R B. CATER, J. SERAFIMOVA | THE INFLUENCE OF SOCIO-DEMOGRAPHIC 229 5.3 Testing the joint influence of socio-demographics on environmental concern and ecologically conscious consumer behaviour In the next section we present the results of multiple regression analyses that were carried out to test the joint explanatory value of socio-demographics for environmental attitudes and behaviour. We performed two regression analyses, where environmental concern and ecologically conscious consumer behaviour were separately used as dependent variables and the earlier discussed socio-demographic characteristics as the independent variables. Age and number of children were measured on ratio scales, so they were directly entered in the regression analysis. Gender, marital status, educational level and income had to be transformed into dummy variables. In the case of the first three each was represented by a single dummy variable, while income was measured with two dummy variables (the details are explained below in Table 9 and Table 10). The nspection of correlations among the predictors did not indicate collinearity concerns (the highest correlation coefficient was 0.481), which was also confirmed by multicollinearity checks with assessment of tolerance (values in the range 0.643 - 0.948) and variance inflation factor (values in the range 1.055 - 1.555). Both regressions are significant and independent variables account for 13.2% of variance in environmental concern and 18.4% in ecologically conscious consumer behaviour (Table 8). Table 8: Regression results Summary statistics Environmental measures Multiple R Adj. R2 F value Significance Environmental concern 0.388 0.132 7.962 0.000 Ecologically conscious consumer behaviour 0.449 0.184 11.313 0.000 Table 9: Regression coefficients for environmental concern Summary statistics Independent variables ß t Significance Gender 0.200 3.708 0.000 Age 0.156 2.559 0.011 Educational level 0.048 0.880 0.380 Income below average 0.139 2.606 0.010 Income above average -0.085 -1.537 0.125 Marital status 0.099 1.520 0.130 Number of children 0.102 1.767 0.078 Codes for dummy variables: Gender (1 = female, 0 = male), Education level (1 = bachelor and higher, 0 = secondary or lower), Income below average (1 = below average, 0 = otherwise), Income above average (1 = above average, 0 = otherwise), Marital status (1 = married, 0 = single). 230 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Table 10: Regression coefficients for ecologically conscious consumer behaviour Summary statistics Independent variables ß t Significance Gender 0.178 3.414 0.001 Age 0.178 3.012 0.003 Educational level 0.008 0.150 0.881 Income below average 0.018 0.339 0.735 Income above average -0.034 -0.641 0.522 Marital status 0.251 3.984 0.000 Number of children 0.112 2.006 0.046 Codes for dummy variables: Gender (1 = female, 0 = male), Education level (1 = bachelor and higher, 0 = secondary or lower), Income below average (1 = below average, 0 = otherwise), Income above average (1 = above average, 0 = otherwise), Marital status (1 = married, 0 = single). Environmental concern (Table 9) is predicted by gender, age and income below average, with gender having the strongest influence. As already indicated in hypothesis testing, women and those consumers that reported to have below average income tend to be more concerned about the environment. Environmental concern on average also increases with age. On the other hand, ecologically conscious consumer behaviour (Table 10) is predicted by gender, age, marital status and number of children. The main difference to the previous analysis is that while in the regression analysis marital status and number of children do not seem to significantly influence environmental concern, they still have a positive effect on ecologically conscious consumer behaviour. When environmental concern is included as a predictor in the regression analysis of ecologically conscious consumer behaviour, this substantially increases the percentage of explained variance (adjusted R2 is 0.336 compared to R2 of 0.184 without environmental concern), as expected. In this case ecologically conscious consumer behaviour is explained by environmental concern (ß = 0.417, P = 0.000), marital status (ß = 0.211, P = 0.000), age (ß = 0.117, P = 0.031) and gender (ß = 0.099, P = 0.042). To validate the results we additionally performed a cluster analysis on attitudinal and behavioural variables (the seventeen variables measuring environmental concern and ecologically conscious consumer behaviour). The TwoStep cluster analysis revealed a two cluster solution (with cluster quality rated as fair) where variables related to behaviour carry a heavier importance at predicting cluster membership than those related to attitudes. The largest cluster (55.8% of sample elements) consisted of consumers that rank consistently lower in environmental concern and ecologically conscious consumer behaviour than the smaller group (44.2% of sample elements). The results for the summated scales of ecologically conscious consumer behaviour (M1 = 34.63, SD = 6.00; M2 = 44.32; SD = 3.93) and environmental concern (M1 = 22.16; SD = 2.68; M2 = 26.43; SD = 2.10) also E/B/R B. CATER, J. SERAFIMOVA | THE INFLUENCE OF SOCIO-DEMOGRAPHIC 231 revealed greater variability in the less ecological group. In the discriminant analysis that we performed with the previously mentioned socio-demographic variables, the percentage of variance explained was similar to our previous analyses (16%). The correlation between the discriminant scores and the levels of the dependent variable was weak to moderate (0.371) and Wilks' lambda (0.862) was statistically significant (P = 0.000). The analysis revealed that the two groups differ significantly in marital status, age, number of children, gender and education, while the difference in income is not statistically significant. In line with the results of the previous analysis, consumers in the more ecological group are to a larger extent married, older, female, with higher education and have on average more children. 6 DISCUSSION AND CONCLUSIONS The main goal of this research was to analyse consumers' environmental concern and ecologically conscious consumer behaviour and discover if significant differences exist based on socio-demographic characteristics that would enable companies and policy makers to use these variables in profiling green consumers. In regards to the recognition of the importance of environmental issues among consumers, it can be said that Macedonian consumers seem to be quite concerned about the general issues related to environmental protection. Although people seem to be highly concerned about the state of the environment due to high pollution the country experiences, this has not yet translated into their buying decisions. 6.1 Theoretical implications The broad theoretical underpinning of this research is role theory (Biddle, 1986) that can be used both to explain and predict social behaviour of individuals based on situations and identities. In line with role theory this study proposes that there are differences in attitudes and behaviour of consumers based on the roles they play in the society (for example, based on gender, educational level, income level and similar). Testing these relationships in the examined context can give better insights to companies and policy makers with more prominent roles. Although the results of previous studies are quite mixed and ambiguous (Verain et al., 2012), the majority of the proposed hypotheses were supported in our research. Women are on average more environmentally concerned and report to engage more in ecologically conscious consumer behaviour than men, which is in line with the findings of several authors (e.g., Brochado et al., 2017; Diamantopoulos et al., 2003; Koos, 2011; Luchs & Mooradian, 2012; ). We can conclude that gender is a socio-demographic variable that seems to work across cultures and level of market development and can be used in posttransition contexts, as well as for profiling green consumers. 232 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Age is also an important predictor of environmental variables in the examined context. The results indicate that age is positively related to both environmental concern and ecologically conscious consumer behaviour. Further analyses revealed that the youngest age group (30 and below) is less environmentally concerned and less engaged in ecologically conscious consumer behaviour than the other two age groups (31 to 50 years and 51 years and above). Mixed results exist on these relationships in the literature and our research adds to the group of authors that found that older consumers are more environmentally concerned (Liu et al., 2014) and more engaged in ecologically conscious consumer behaviour (e.g. Anic et al., 2015; Brocado et al., 2017; Mohr & Schlich, 2016). Furthermore, our research did not find statistically significant differences in environmental concern and ecologically conscious consumer behaviour regarding educational level, which is in contradiction to previous research. Most empirical studies have shown that higher educated people tend to perceive environmental issues better and are more sensitive and aware of environmental issues (e.g. Zhao et al., 2014; Zsoka et al., 2013) and that highly educated people are more prone to ecologically conscious consumption in developed (Diamantopoulos et al., 2003; do Pa^o et al., 2009) and developing countries (Zhao et al., 2014; Zsoka et al., 2013). A closer inspection of the results reveals that differences among the groups exist and are statistically significant at P = 0.065 and P = 0.081, respectively, but not at our threshold (a = 0.05). Therefore, at a less stringent threshold (a = 0.10) both hypotheses regarding education would be supported. However, the results of clustering and discriminant analysis reveal that when ecologically conscious consumer behaviour and environmental concern are jointly analysed, the level of education discriminates between the more and less ecological groups. Regarding income, the results indicate that significant differences exist in environmental concern between consumers with below average household income and those with above average household income, indicating that those coming from less wealthy households are more concerned about the environment. This is in contradiction with most previous studies, except partially with Park et al. (2012) who also found people from less wealthy households to be more environmentally concerned compared to the group with average income. No differences regarding income exist for ecologically conscious consumer behaviour, which is in line with mixed findings in the published literature, especially with Ci-Sheng et al. (2016) and Straughan and Roberts (1999) who also found that income level does not affect green consumption decisions significantly. The explanation for these findings could be in line with the discussion offered by Roberts (1996) that pollution and environmental degradation may have reached the point where consumers from all (also the lower) socioeconomic strata are becoming involved. Skopje is one of the most polluted European cities and it is possible that consumers from poorer households live in more polluted areas and are consequently more concerned about the environmental problems. In the last section, we tested the influence of spouses and children on environmental concern and ecologically conscious consumer behaviour. Regarding the marital status (married were those living together with a significant other in a household), our results E/B/R B. CATER, J. SERAFIMOVA | THE INFLUENCE OF SOCIO-DEMOGRAPHIC 233 support that on average married people are more environmentally concerned and report to exhibit more ecologically conscious consumer behaviour. This study therefore adds to the scarce empirical evidence of the influence of marital status on environmental concern (e.g. Research 2000 in Diamantopoulos et al., 2003) and ecologically conscious consumer behaviour (Diamantopoulos et al., 2003; Fisher et al., 2012). The relationship of the number of children in the household is closely related to environmental variables. The results indicate that the number of children is positively related to environmental concern and ecologically conscious consumer behaviour, which supports the results of previous studies on environmental concern and environmentally friendly behaviour (Laroche et al., 2001; Loureirro et al., 2002). We can conclude that in this context, possibly due to discussions on ecology at school, children influence environmentally friendly behaviour of their parents. The other explanation could be in line with role theory that parents play the role of responsible adults and try to lead by example. When testing the joint influence of socio-demographics on environmental concern and ecologically conscious consumer behaviour, there are some differences compared to hypotheses testing. Environmental concern is predicted by gender, age and income below average, with gender having the strongest influence, which is in line with the findings using role theory (Han et al., 2009). Marital status and number of children that were significantly related to environmental concern when tested individually do not have a statistically significant effect on environmental concern. When age was not in the equation, marital status had a statistically significant effect on environmental concern, while the effect of the number of children became significant only after also marital status was excluded from the equation. Despite multicollinearity not being an evident issue in this dataset, a close inspection of the correlations reveals that correlations between the independent variables (marital status, age and number of children below 15 years) are higher than correlations between the respective independent variables and environmental concern), which is a possible explanation why not all of the above mentioned regression coefficients are statistically significant when examined jointly. Ecologically conscious consumer behaviour is predicted by gender, age, marital status and number of children, which is in line with our previous analyses. The results indicate that in the examined context, socio-demographic variables have substantially larger explanatory power for environmental concern and ecologically conscious consumer behaviour than in more developed economies. For example, in the study on U.S. consumers, conducted by Roberts (1996), socio-demographic variables explained 6% of variance in ecologically conscious consumer behaviour, while for the UK, with slightly different scales, Diamantopoulos et al. (2003) had less than 6% of variance in environmental measures explained (5.7% for environmental attitudes and 3.9% for purchasing behaviour). More recently, Brochado et el. (2017) explained 12.9% of variance in ecologically conscious consumer behaviour with socio-demographic variables, compared to 13.2% for environmental concern and 18.4% for ecologically conscious consumer behaviour in our study. The percentage of variance that remains unexplained indicates there might be other influences, such as psychographic characteristics or the 234 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 impact of other situational factors on consumers' purchase decisions rather than socio-demographics. When we included environmental concern as a predictor in the regression analysis of ecologically conscious consumer behaviour, this, as expected, considerably increased the percentage of the explained variance (adjusted R2 is 0.336 compared to 0.184 without environmental concern). However, in transition or post-transition markets where companies do not spend a lot of money on marketing research, this R2 indicates that socio-demographic variables do offer a relevant, although not ideal, base for profiling green consumers. 6.2 Implications for managers and policy makers Even though in general consumers want to take a part in ecologically conscious behaviour and there are varieties of available options to do so, the environmental impacts from consumption are continuously increasing. Therefore, it is essential that researchers shed more light on consumer behaviour. In that line, this research gives its own impact investigating attitudes toward the environment and ecologically conscious consumer behaviour in the context of a post-transition and heavily polluted country, where this type of research is quite scarce. Companies can use the results presented in this research in several ways. First, the research offers information about the level of environmental concern and ecologically conscious consumer behaviour in the examined market. This information can be used to assess market readiness for green products and initiatives. Second, the results of testing individual and joint influences on environmental variables can be used in profiling green consumers. Due to not very developed market in terms of marketing research, it is easier for companies to use socio-demographic variables for segmentation of green consumers. This research suggests which variables could be used. This study also offers some implications for policy makers. It is evident from the results that the general public needs more education to raise environmental awareness and motivation for ecologically conscious consumer behaviour. This is especially the case for younger consumers who scored lower on environmental variables compared to older consumers. The implication for policy makers is to incorporate more environmental content in the curriculum to properly educate the youngest population in the country, even though it might take years to see the effect of the educational system on their higher awareness of environmental issues. Thus the country could be on the right way to create a more environmentally responsible society of active, environmentally conscious consumers and citizens. In the short term, policy makers should offer more financial stimulation for replacing old wood-burning stoves and old cars with greener ones in order to reduce air pollution. In this context ecologically conscious behaviour is not significantly affected by income, but environmental awareness is. The results show that consumers from households with below average income are more environmentally aware than others, but they do not have the budget to transform their environmental attitudes to behaviour. E/B/R B. CATER, J. SERAFIMOVA | THE INFLUENCE OF SOCIO-DEMOGRAPHIC 235 Additionally, by accepting and implementing the concept of sustainable development, the government develops strategies to promote more ecologically conscious consumer behaviour. Regarding their effectiveness, it is important to understand and evaluate consumer behaviour in order to develop ways which can help to influence consumer behaviour in the desired direction. Thus, the results from the current study concerning the relation between socio-demographic, attitudinal and behavioural factors might be used by all relevant players involved in implementing the strategies for promoting more ecologically conscious consumption in the society. It seems a lot of additional efforts are needed to bring consumers' behaviour into accordance with the sustainable development policy on the national and international levels. 6.3 Limitations and opportunities for future research As with any research, the present study has its own limitations. One of the limitations is the use of non-probability sampling, which limits its generalization; although, due to a careful selection of respondents, the sample does resemble the population in several characteristics. Nevertheless, the results give insights into the situation on the Macedonian market regarding the current issues of ecologically conscious consumption. In order to achieve a more representative sample, the use of probability sampling is one of the options suggested for further research. Additionally, the respondents gave self-reported responses that might not be entirely accurate because they tended to show their perception of their own behaviour, rather than their actual behaviour. The data was collected outside of the actual buying situation, which might give an inaccurate picture of real decision-making processes. Thus, we suggest that further data collection needs to be performed in real purchase situations in order to examine the relevant product categories more effectively. The current study can be seen as the beginning of a journey into further research of ecologically conscious consumer behaviour in transition and post-transition contexts. Since the issue with all of its relevant factors has not yet been comprehensively studied in these contexts, there is a great opportunity for further research in the field by examining additional factors that may impact ecologically conscious consumer behaviour. Besides socio-demographic characteristics several psychographic characteristics could be included (e.g., values, attitudes and lifestyles), which would also increase explanatory power. Groening et al. (2018) offer future theoretical directions for green marketing research, especially in the area of behavioural intentions, which can also be tested in the context of transition and post-transition economies. One highly interesting topic for further research could also be the influence of eco-labels on consumer decision making. Another research with great potential could be examining young people's knowledge of sustainability issues in general, which could help find ways to implement appropriate educational strategies in order to motivate, enable, and empower future consumers to engage in more ecologically conscious consumer behaviour and sustainable development processes. 236 ECONOMIC AND BUSINESS REVIEW j VOl. 21 j No. 2 j 2G19 7 REFERENCES Abdul-Muhmin, A. G. (2007). Explaining consumers' willingness to be environmentally friendly. International Journal of Consumer Studies, 31, 237-247. Akehurst, G., Afonso, C., & Gonçalves, H. M. (2012). Re-examining green purchase behaviour and the green consumer profile: new evidences. Management Decision, 50(5), 972-988. Angelovska, J., Sotiroska, B. S., & Angelovska, N. (2012). The impact of environmental concern and awareness on consumer behaviour. Journal of International Environmental Application & Science, 7(2), 406-416. Anic, I. D., Jelenc, L., & Šebetic, N. (2015). Istraživanje demografskih obilježja i ponašanja kupaca ekoloških prehrambenih proizvoda u Karlovačkoj županiji. Ekonomska Misao i Praksa, 2(24), 367-388. Antil, J. H. (1984). Socially responsible consumers: Profile and implications for public policy. Journal of Macromarketing, 4(2), 18-39. Berglund, C., & Matti, S. (2006). Citizen and consumer: The dual role of individuals in environmental policy. Environmental Politics, 15(4), 550-571. Biddle, B. J. (1986). Recent development in role theory. Annual Review of Sociology, 12, 67-92. Brochado, A., Teiga, N., & Oliveira-Brochado, F. (2017). The ecological conscious consumer behaviour: are the activists different?. International Journal of Consumer Studies, 41, 138-146. Chan, T. S. (1996). Concerns for environmental issues and consumer purchase preferences: A two-country study. Journal of International Consumer Marketing, 9(1), 43-55. Chen, A., & Peng, N. (2012). Green hotel knowledge and tourists' staying behavior. Annals of Tourism Research, 39(4), 2211-2216. Chen, X., Peterson, M., Hull, V., Lu, C., Lee, G., Hong, D., & Liu, J. (2011). Effects of attitudinal and sociodemographic factors on pro-environmental behaviour in urban China. Environmental Conservation, 38(1), 45-52. E/B/R B. CATER, J. SERAFIMOVA | THE INFLUENCE OF SOCIO-DEMOGRAPHIC 237 Ci-Sheng, W., Xiao-Xia, Z., & Meng, S. (2016). Sustainable consumer behaviour in China: an empirical analysis from the Midwest regions. Journal of Cleaner Production, 134, 147165. Migrio, G. (2018, July 3). Cities with the worst air quality in Europe. Retrieved from https:// www.worldatlas.com/articles/cities-with-the-worst-air-quality-in-europe.html Cruz, S. M. (2017). The relationships of political ideology and party affiliation with environmental concern: A meta-analysis. Journal of Environmental Psychology, 53, 81-91. D'Souza, C., Taghian, M., Lamb, P., & Pretiatko, R. (2007). Green decisions: Demographics and consumer understanding of environmental labels. International Journal of Consumer Studies, 31(4), 371-376. Diamantopoulos, A., Schlegelmilch, B. B., Sincovics, R. R., & Bohlen, G. M. (2003). Can socio-demographics still play a role in profiling green consumers? A review of the evidence and empirical investigation. Journal of Business Research, 56(6), 465-480. do Pa^o, A. M. F., Raposo, M. L. B., & Filho, W. L. (2009). Identifying the green consumer: A segmentation study. Journal of Targeting, Measurement and Analysis for Marketing, 17(1), 17-25. Dunlap, R., & York, R. (2008). The globalization of environmental concern and the limits of the postmaterialist values explanation: Evidence from four multinational surveys. The Sociological Quarterly, 49(3), 529-563. EEA Report No 1/2010: Environmental trends and perspectives in the Western Balkans: future production and consumption patterns. Retrieved from http://www.eea.europa.eu/ publications/western-balkans Fishbein, M., & Ajzen, I. (2011). Predicting and Changing Behavior: The Reasoned Action Approach. New York, NY: Taylor & Francis. Fisher, C., Bashyal S., & Bachman, B. (2012). Demographic impacts on environmentally friendly purchase behaviors. Journal of Targeting, Measurement and Analysis for Marketing. 20(3/4), 172-184. Franzen, A., & Vogl. D. (2013). Two decades of measuring environmental attitudes: A comparative analysis of 33 countries. Global Environmental Change, 23(5) (2013), 10011008. 238 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Georgievski, B. (2018, January 10). Skopje: Welcome to Europe's most polluted city. Deutsche Welle. Retrieved from https://www.dw.com/en/skopje-welcome-to-europes-most-polluted-city/g-42083092 Groening, C., & Sarkis, J., & Zhu, Q. (2018). Green marketing consumer-level theory review: A compendium of applied theories and further research directions. Journal of Cleaner Production, 172, 1848-1866. Hailes, J. (2007). The New Green Consumer Guide. London: Simon and Schuster. Han, H., Hsu, L.-T., & Lee, J.-S. (2009). Empirical investigation of the roles of attitudes toward green behaviors, overall image, gender, and age in hotel customers' eco-friendly decision-making process. International Journal of Hospitality Management, 28(4), 519528. Hines, J. M., Herald, R. H., & Audrey, N. T. (1987). Analysis and synthesis of research on responsible environmental behavior: A meta-analysis. Journal of Environmental Education, 18(2), 1-8. Jain, S. K., & Kaur, G. (2004). Green marketing: An attitudinal and behavioural analysis of Indian consumer. Global Business Review, 5(2), 187-205. Kline, P. (2000). The handbook of psychological testing (2nd ed.). London: Routledge. Koos, S. (2011). Varieties of environmental labeling, market structures, and sustainable consumption across Europe: A comparative analysis of organizational and market supply determinants of environmental labeled goods. Journal of Consumer Policy, 34(1), 127-151. Kuchinka, D., Balazs, S., Gavriletea, M., & Djokic, B.-B. (2018). Consumer attitudes toward sustainable development and risk to brand loyalty. Sustainability, 10(4), 997. Laroche, M., Bergeron, J., & Barbaro-Forleo, G. (2001). Targeting consumers who are willing to pay more for environmentally friendly products. Journal of Consumer Marketing 18(6), 503-520. Liobikiene, G. & Bernatoniene, J. (2017). Why determinants of green purchase cannot be treated equally? The case of green cosmetics: Literature review. Journal of Cleaner Production. 162, 109-120. E/B/R B. CATER, J. SERAFIMOVA | THE INFLUENCE OF SOCIO-DEMOGRAPHIC 239 Liobikiene, G., Grinceviciene, S. & Bernatoniene, J. (2017). Environmentally friendly behaviour and green purchase in Austria and Lithuania. Journal of Cleaner Production. 142, 3789-3797. Liu, X., Vedlitz, A., & Shi, L. (2014). Examining the determinants of public environmental concern: Evidence from national public surveys. Environmental Science and Policy, 39, 77-94. Loureiro, M. L., McCluskey, J. J., & Mittelhammer, R. C. (2002). Will consumers pay a premium for eco-labeled apples?. Journal of Consumer Affairs, 36(2), 203-219. Luchs, M. G., & Mooradian, T. A. (2012). Sex, personality, and sustainable consumer behaviour: Elucidating the gender effect. Journal of Consumer Policy, 35, 127-144. Mainieri, T., Barnett, E. G., Valdero, T. R., Unipan, J. B., & Oskamp, S. (1997). Green buying: The influence of environmental concern on consumer behavior. Journal of Social Psychology, 137(2), 189-204. Mohr, M. & Schlich, M. (2016). Socio-demographic basic factors of German customers as predictors for sustainable consumerism regarding foodstuffs and meat products. International Journal of Consumer Studies, 40(2), 158-167. Mostafa, M. M. (2007). Gender differences in Egyptian consumers' green purchase behaviour: the effects of environmental knowledge, concern and attitude. International Journal of Consumer Studies, 31(3), 220-229. Nath, V., Kumar, R., Agrawal, R., Gautam, A., & Sharma, V. (2013). Consumer adoption of green products: Modeling the enablers. Global Business Review, 14(3), 453-470. Neuman, K. (1986). Personal values and commitment to energy conservation. Environment and Behavior, 18(6), 53-74. Park, J., & Ha, S. (2012) Understanding pro-environmental behavior: A comparison of sustainable consumers and apathetic consumers. International Journal of Retail & Distribution Management, 40(5), 388-403. Park, S., Choi, S., & Kim, E. (2012). The Relationships between socio-demographic variables and concerns about environmental sustainability. Corporate Social Responsibility and Environmental Management, 19, 343-354. 240 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Patel, J. D., Modi, A., & Paul, J. (2017). Pro-environmental behavior and socio-demographic factors in an emerging market. Asian Journal of Business Ethics, 6(2), 189-214. Pinto, D. C., Herter, M. M., Rossi, P., & Borges, A. (2014). Going green for self or for others? Gender and identity salience effects on sustainable consumption. International Journal of Consumer Studies, 38(5), 540-549. Roberts, J. A. (1996). Green consumers in the 1990s: Profile and implications for advertising. Journal of Business Research, 36(3), 217-231. Royne, M. B., Levy, M., & Martinez, J. (2011).The public health implications of consumers' environmental concern and their willingness to pay for an eco-friendly product. Journal of Consumer Affairs, 45(2), 329-343. Sánchez, M., López-Mosquera, N., & Lera-López, F. (2016). Improving pro-environmental behaviours in Spain. The role of attitudes and socio-demographic and political factors. Journal of Environmental Policy & Planning, 18(1), 47-66. Schlossberg, H. (1992). Kids teach parents how to change their buying habits. Marketing News, 26(8). Special Eurobarometer 416: Attitudes of European citizens towards the environment (2014). Retrieved from http://ec.europa.eu/public_opinion/archives/ebs/ebs_416_en.pdf Straughan, R. D., & Roberts, J. A. (1999). Environmental segmentation alternatives: A look at green consumer behaviour in the new millennium. Journal of Consumer Marketing, 16(6), 558-575. Torgler, B., & Garcia-Valinas, M. A. (2007). The determinants of individual attitudes towards preventing environmental damage. Ecological Economics, 63(2-3), 536-552. Verain, M. C. D., Bartels, J., Dagevos, H., Sijtsema, S. J., Onwezen, M. C., & Antonides, G. (2012). Segments of sustainable food consumers: A literature review. International Journal of Consumer Studies, 36(2), 123-132. Welsch, H., & Kühling, J. (2009). Determinants of pro-environmental consumption. The role of reference groups and routine behavior. Ecological Economics, 69(1), 166-176. Zelezny, L. C., Chua, P., & Aldrich, C. (2000). Elaborating on sex differences in environmentalism. Journal of Social Issues, 56(3), 443-457. E/B/R B. CATER, J. SERAFIMOVA | THE INFLUENCE OF SOCIO-DEMOGRAPHIC 241 Zhao, H., Wu, Y., Wang, Y. & Zhu, X. (2014). What affects green consumer behavior in China? A case study from Qingdao. Journal of Cleaner Production 63, 143-151. Zimmer, M. R., Stafford, T. F., & Stafford, M. R. (1994). Green issues: Dimensions of environmental concern. Journal of Business Research, 30(1), 63-74. Zsoka, A., Szerenyi, Z. M., Szechy, A., & Kocsis, T. (2013). Greening due to environmental education? Environmental knowledge, attitudes, consumer behavior and everyday pro-environmental activities of Hungarian high school and university students. Journal of Cleaner Production, 48, 126-138. 242 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 243-274 243 are middle managers' cost decisions sticky? evidence from the field BYUNGHOON JIN1 Received: February 9, 2018 JOHN C CARY2 Accepted: May 21, 2018 ABSTRACT: Anderson, Banker, and Janakiraman (2003) show that costs are "sticky" (i.e., costs change relatively less when sales decrease than when sales increase) because managers are reluctant to cut resources when sales decrease. We predict that cost behavior at the middle management level is sticky also when the magnitude of sales increase is sufficiently large, considering that middle managers have more limited ability in adding resources and are more risk averse. Using a survey instrument and interviews, we find evidence that middle managers' cost decisions are sticky at both ends. Our findings are supported by empirical evidence based on segment-level data. Key words: cost stickiness; asymmetric cost behavior; middle manager; resource allocation; budget constraints; risk aversion; business segments JEL classification: M10, M41, D24 DOI: 10.15458/ebr.85 1 INTRODUCTION Anderson, Banker, and Janakiraman (2003, hereafter ABJ) and subsequent studies in the management accounting literature document that costs decrease relatively less when sales decrease than they increase when sales increase by an equivalent amount; i.e., costs are "sticky". While the literature explains such asymmetric cost behavior as a result of asymmetric cost decisions by managers, most studies in cost stickiness literature either examine cost behavior at the corporate level or focus on CEOs as decision makers. In this study, we focus on middle managers who have significant influence on the corporate strategy through day-to-day operational decisions and also have characteristics distinct from those of CEOs or other top managers. Unlike prior studies that rely heavily on archival data to examine cost stickiness, we take a behavioral approach and more directly ask middle managers in practice about their cost decisions, using a survey instrument and interviews in addition to a regression analysis. We find that middle managers' cost decisions are sticky not only when sales decrease but also when the magnitude of sales increase is sufficiently large. 1 Marist College, School of Management, Poughkeepsie, USA, e-mail: byunghoon.jin@marist.edu 2 Corresponding author, Marist College, School of Management, Poughkeepsie, USA, e-mail: john.cary@marist. edu 244 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Cost decisions at the middle management level are important and thus worth examining because of two reasons. First, middle managers are more involved in the day-to-day operations of a company than top managers and are also likely to be the ultimate decision makers for the business unit and thus can have significant influence on the firm's overall costs (Kanter, 1982). Second, at the same time, middle managers' cost decisions are likely to be different from those of top managers because middle managers are likely to (1) have more limited ability in adding resources due to limited annual budgets and corporate-level policies or strategies to follow, which are typically set by top managers (Williamson, 1975; Mueller, 2003), and (2) be more risk averse because of their compensation structure, which is focused relatively more on fixed salary and less on incentives such as cash bonus and equity-based compensation. To examine cost behavior at the middle management level, we conducted both a survey and field interviews, directly asking middle managers in practice to describe their decisions related to various types of costs, including overall SG&A costs, under various situations regarding the change in sales revenue. The analysis results based on the detailed interviews and 152 survey responses indicate that middle managers' cost decisions are sticky when sales decrease (or, to be more accurate, when the magnitude of sales decrease is sufficiently large), consistent with the findings in the previous empirical studies, and also when the magnitude of sales increase is sufficiently large. To complement our behavioral findings, we also conducted an empirical analysis using segment level data. The regression results based on 26,050 segment/year observations support our prediction and behavioral findings. Our study contributes to the accounting and management literature in several ways. First, using a survey instrument and field interviews, we provide direct evidence that managers' resource capacity decisions are sticky, which supports the explanations in the previous studies based on empirical models and archival data (e.g., ABJ). Second, more importantly, we provide an additional insight that at least at the middle management level costs are sticky not only when sales decrease but also when a firm experiences a sufficiently large increase in sales revenue. The rest of the paper is organized as follows. In Section 2, we review the prior literature on cost stickiness and middle managers and provide our research hypothesis. In Section 3, we describe the design and procedures of the survey instrument and interviews. Section 4 presents our data and summary statistics. In Section 5, the results of the quantitative and qualitative analyses are presented, followed by the conclusion in Section 6. E/B/R B. JIN, J. C CARY | ARE MIDDLE MANAGER'S COST DECISIONS STICKY? EVIDENCE 245 2 LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT 2.1 Cost stickiness The asymmetric cost behavior, called "cost stickiness," was first documented by ABJ. Using archival data spanning 20 years (from 1979 to 1998), ABJ showed that costs decrease less when sales fall than they increase when sales rise by an equivalent amount. ABJ argued that the fundamental reason for cost stickiness is that changing the levels of committed resources is costly. Adjustment costs include severance pay when employees are laid off, recruiting and training costs when new employees are hired, as well as organizational costs such as loss of morale among the remaining employees when colleagues are terminated. Because of the adjustment costs, managers will choose to retain unutilized resources to some extent when sales decline and there is uncertainty about the permanence of a decline in demand. In contrast, when demand increases beyond the available resource capacity, managers do not have as much discretion in adding resources because not doing so would result in losing not only current sales but also future sales because of disappointed customers. As a result of the asymmetry in resource capacity decisions, costs become sticky, i.e., costs decrease relatively less when sales fall than they increase when sales increase by an equivalent amount. Consistent with this explanation, previous studies have shown that the degree of cost stickiness is related to macroeconomic factors and firm-specific factors which constrain resource adjustment. For instance, ABJ find that the cost stickiness is weaker when sales revenue also declined in the preceding period, stronger during periods of macroeconomic growth, and positively associated with the asset intensity and the employee intensity. Balakrishnan, Petersen, and Soderstrom (2004) find that the degree of cost stickiness is influenced by capacity utilization. Banker, Byzalov, and Chen (2013) focus on crosscountry differences and find that the degree of cost stickiness is increasing in the strictness of employment protection legislation, consistent with ABJ's adjustment cost theory. While the literature explains the asymmetric cost behavior using asymmetric cost decisions of managers, behavioral factors affecting the cost decisions have been largely ignored in the prior literature. A few exceptions are Dierynck, Landsman, and Renders (2012), Kama and Weiss (2013), Chen, Lu, and Sougiannis (2012), and Banker, Jin, and Mehta (2018), all of whom focused, either explicitly or implicitly, on CEOs as the ultimate decision makers. Dierynck, Landsman, and Renders (2012), and Kama and Weiss (2013) find that incentives to avoid losses and earnings decreases or to meet financial analysts' earnings forecasts managers expedite downward adjustments of slack resources when sales fall, lessening cost stickiness. Chen, Lu, and Sougiannis (2012) find that managers' incentives to grow the firm beyond its optimal size or to maintain unutilized resources with the purpose of increasing personal utility from status, power, compensation, and prestige (i.e., empire building incentives) induce greater cost stickiness. Banker, Jin, and Mehta (2018) focus on managerial decision horizon and show that short-term cash bonus provides managers with incentives to cut more slack resources and thus induce less cost stickiness while long- 246 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 term incentives, such as stock option and restricted stock award, extend the managerial decision horizon and thus induce more cost stickiness. 2.2 Middle managers While prior studies in the cost stickiness literature generally regard a firm's cost behavior as a result of the asymmetry in the cost decisions either at the corporate level or by top management, many cost-related decisions, including employment, asset acquisition, and overall SG&A spending decisions, are made by middle managers, such as department managers and regional managers, especially in decentralized firms. Middle managers and their business decisions are important mainly because middle managers have significant influence on strategic decision making process of the company. Middle managers are more involved in the day-to-day operations of a company than top managers and are often said to have their fingers on the "pulse of operation" (Kanter, 1982). Because of their deep involvement into the day-to-day operations, middle managers have the opportunity to report valuable information and suggestions from the inside of a company (Likert, 1961), which makes them play a critical role in the corporate level decision making process. By using bottom-up management processes, they communicate information and propose issues for top management (Floyd & Wooldridge, 1994; Dutton & Ashford, 1993; Dutton et al., 1997).3 The significant influence of middle managers on corporate decisions, including investment in resource capacity decisions, suggests that firm-level cost behavior is also heavily affected by middle management decisions. What makes middle managers and their cost decisions even more important and thus worth examining is that middle managers have characteristics distinct from those of top managers. First, middle managers are likely to have more constraints in the decision making process than top managers. The primary responsibility of a middle manager is to implement a strategy, set by the top management, in an effective and efficient manner (Floyd & Wooldridge, 1997; Huy, 2002; Delmestri & Walgenbach, 2005). During the implementation process, however, middle managers tend to have limited ability in adding resources, including human resources and long-term assets. Such a limit is typically set by top managers only. Managerial discretion arises, at least partly, from the authority to allocate the funds of the company to pursue their own interests (Mueller, 2003). This suggests that if middle managers are given too much power on resource allocation and pursue their own interests, for example, performance of the department, fewer resources or funds will be left for top managers who have their own interests, for example, companylevel performance (Fama & Jensen, 1983; Eisenhardt, 1989). Thus, top managers tend to restrict middle managers' ability and monitor their behaviors in order to prevent middle managers from wasting the resources of the company and thereby limiting the top management's ability to utilize such resources (Williamson, 1975; Mueller, 2003). 3 For more insights into middle managers' involvement in the strategy or decision making process, see Burgelman (1983), Floyd & Wooldridge (1992a, 1992b, 1997, 1999), Huy (2001, 2002), Kanter (1988), Westley (1990), and Wooldridge & Floyd (1990). E/B/R B. JIN, J. C CARY | ARE MIDDLE MANAGER'S COST DECISIONS STICKY? EVIDENCE 247 Another distinctive characteristic of middle managers is that they are generally more risk averse than top managers because their future is narrowly dependent on their current tasks (Eisenhardt, 1989; Gomez-Mejia & Balkin, 1992; Shimizu, 2012). According to the agency theory, principals use various forms of non-salary components in the compensation package, such as cash bonus or long-term equity incentives, to provide risk-averse agents with incentives to take risk (Jensen & Meckling, 1976; DeFusco, Johnson, & Zorn, 1990; Murphy, 1999; Rajgopal & Shevlin, 2002). However, the portion of non-salary incentives is substantially smaller for middle managers compared to CEOs and other top managers (Belcher & Atchison, 1987), suggesting that managerial decisions of middle managers are likely to be more risk averse than those of top managers. 2.3 Research hypothesis The distinctive characteristics of middle managers suggest that cost behavior at the middle management level may look different from that at the company or top management level. In specific, the cost stickiness theory assumes that companies' or top managers' ability to add resources are relatively less limited than their ability to cut slack resources, and as a result the relation between sales change and cost change is kinked at the point where sales change equals zero, as illustrated in Figure 1A. On the other hand, the middle managers' ability to change the level of cost or investment is limited for both adding and cutting as discussed above. In addition, middle managers, who are relatively more risk averse than top managers, are less likely to increase cost or investment substantially when the company or the business unit experiences a huge increase in revenue, concerning the permanence of the increase in demand. Based on this intuition, we formulate our main hypothesis as follows: Hypothesis: Middle managers' decisions to change the level of cost or investment are "sticky" when the magnitude of sales change is sufficiently large. In other words, we predict that at the middle management level, costs change relatively less not only when sales decrease (or when the magnitude of sales decrease is large), but also when the magnitude of sales increase is sufficiently large. This suggests that the relation between sales change and cost change at the middle management level is expected to be kinked at two different points as illustrated in Figure 1B. The main objectives of this study include (1) examining how costs behave at the middle management level (and especially if the cost behavior is consistent with our prediction) and (2) providing an explanation for the observed behavior based on qualitative information obtained through the survey and the interviews. 248 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Figure 1: Graphical illustrations of SG&A cost behavior Figure 1A Asymmetric SG&A cost behavior (Anderson et al., 2003) % Change in Sales Revenue Figure IB Middle management's SG&A cost behavior with two kinks % Change in Sales Revenue Note: Figure 1A, drawn based on the theory of Anderson, Banker, and Janakiraman (2003), illustrates the asymmetric SG&A cost behavior, "cost stickiness". The relation can be described as SG&A costs changing relatively less when sales decrease than when sales increase by an equivalent amount. The line is kinked at % change in sales revenue = 0. The y-intercept is not necessarily zero. Figure 1B illustrates the behavior of SG&A costs at the middle management level. The non-linear costs-sales relation can be described as SG&A cost changing relatively less when the change in sales revenue is sufficiently large in magnitude. The flatter parts at both ends are not necessarily parallel to each other. 3 METHODOLOGY 3.1 Surveys and interviews To examine the characteristics of middle management cost decisions and also to complement prior studies in the cost stickiness literature, we use a combination of a survey instrument and field interviews in this study. The prior literature on cost stickiness relies heavily on archival firm-level data. The main advantage of using archival data is E/B/R B. JIN, J. C CARY | ARE MIDDLE MANAGER'S COST DECISIONS STICKY? EVIDENCE 249 that it enables researchers to perform relatively objective analyses based on historical real data. As discussed by Graham, Harvey, and Rajgopal (2005), however, studies based on archival analyses can also suffer from several weaknesses related to model/variable specification. In most cases, a regression analysis cannot be entirely free from model/ variable misspecification or measurement error. Sometimes it is also difficult to develop a good economic proxy. Another weakness of archival studies is the inability to ask qualitative questions. In contrast, surveys and interviews provide an opportunity to ask managers very specific and qualitative questions about the motivation behind managerial decisions without relying on potentially misspecified regression models (Graham, Harvey, & Rajgopal, 2005). On the other hand, potential caveats related to surveys and interviews include subjective or biased inputs from survey respondents or interviewees. In this study, we mainly use a combination of a survey instrument and field interviews for the purpose of complementing those archival studies in the prior literature. Specifically, surveys and interviews enable us to examine the characteristics of middle managers' resource capacity decisions without worrying about any model specification issues which have been previously addressed in the literature (e.g., Balakrishnan, Labro, & Soderstrom, 2014; Banker & Byzalov, 2014). In addition, surveys and field interviews provide us with an opportunity to identify factors affecting managerial resource capacity decisions, which are not easily identifiable using archival data. Considering the potential caveats associated with surveys and interviews, we also conduct an empirical analysis based on archival data as an additional analysis to back up our main findings from the surveys and interviews.4 3.2 Research design We developed a survey instrument based on a review of the cost stickiness literature. In specific, we designed the main survey questions to ask how a manager's decisions to adjust overall SG&A expenditure, as well as the capacity level of individual resources, including human resources, long-term assets, raw materials and merchandises, vary under hypothetical scenarios regarding sales change. In addition, qualitative questions were asked to identify limitations in the resource capacity decisions and other affecting factors. The survey contained 25 questions including: 13 questions about respondents and their companies and 12 quantitative and qualitative questions addressing their cost decisions. The interviews were designed to obtain more detailed qualitative information about decision behavior at the middle management level, as well as impact factors and limitations in the decision making process. The potential interviewees were contacted using our personal network, a basic introduction was provided through a telephone/email briefing and then the 25 survey questions were sent. The main telephone interviews asking about detailed decision-making mechanisms were conducted about a week after the survey questionnaires were sent. 4 See Section 5.3 for the detailed model and sample data for the empirical analysis. 250 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 4 DATA We used the Cint service to recruit 175 U.S.-based respondents who were identified as middle managers.5, 6 After manually identifying 23 responses with an error (e.g., using dollar amounts instead of percentages) and spam responses, 152 valid responses remained for quantitative and qualitative analyses. Table 1 presents self-reported summary information about demographic characteristics of the sample companies and respondents. The survey gathered information frequently used in empirical research for subsample analyses to consider potential conditioning effects. Table 1: Summary statistics Panel A - Demographic characteristics of sample companies (n = 152) Avg. sales revenue for past 5 years Percent Years of operation Percent < $200,000 4.6 0-5 years 3.9 $200,000 - $500,000 7.9 5-10 years 25.0 $500,000 - $1,000,000 17.8 10-20 years 28.9 $1,000,000 - $1,500,000 21.1 20-30 years 21.7 $1,500,000 - $2,000,000 15.8 > 30 years 20.4 > $2,000,000 32.9 SG&A as % of sales revenue Industry 0-5% 3.9 Construction 17.8 5-10% 21.7 Manufacturing 15.1 10-20% 27.6 Transportation and Utilities 5.3 20-30% 23.0 Wholesalers and Retailers 7.9 30-50% 16.4 Financial Services 12.5 > 50% 7.2 Business Services 17.8 Consumer Services 13.8 Number of employees Public Administration and Other 9.9 < 10 2.6 11-50 18.4 51-100 17.8 101-500 28.9 > 500 32.2 5 Cint is a market research company which has access to a large number of preregistered members who vary in demographics and other social characteristics (e.g., occupation or title). Once a client selects a target respondent group, Cint sends the client's survey until it collects a predetermined number of responses. Our survey was sent to 459 middle managers in the U.S. and completed by 175 of them (i.e., the response rate was 38.3%). 6 In the survey, a qualifying question asking respondents to self-identify their job title was also included. E/B/R B. JIN, J. C CARY | ARE MIDDLE MANAGER'S COST DECISIONS STICKY? EVIDENCE 251 Panel B - Demographic characteristics of sample managers (n = 152) Primary responsibility Hiring Purchasing Production Sales & Marketing Accounting & Finance Administration General management Experience at current position 0-3 years 3-5 years 5-10 years 10-15 years > 15 years Percent 5.9 7.2 15.1 11.2 11.8 19.1 29.6 15.1 21.7 40.1 17.8 5.3 Gender Male Female Mean 25th percentile 50th percentile (median) 75th percentile Total annual compensation Mean 25th percentile 50th percentile (median) 75th percentile Percent 64.5 35.5 Year 39 32 36 42 $ thousand 82.6 60.0 80.0 100.0 Experience in current industry Composition of compensation package (as % of total comp.) Avg. Percent 0-3 years 5.3 Fixed salary 70.9 3-5 years 14.5 Short-term cash bonus 11.7 5-10 years 32.9 Long-term incentives 7.0 10-15 years 23.7 Pension 5.3 > 15 years 23.7 Perks and other 5.1 Note: Table 1 presents demographic characteristics of sample companies (Panel A) and managers (Panel B). Revealing the dollar amount of total annual compensation was optional. 151 out of 152 respondents chose to answer this question. For the mean calculation, all amounts greater than $150,000 were treated equal to $150,000. Considering only six out of 151 valid responses were $150,000, the effect of potential understatement is expected to be minimal. Panel A of Table 1 presents descriptive statistics of the sample companies. Our sample companies range from small to large in terms of average sales revenue and number of employees. In specific, 30.3% of the sample firms were relatively small with less than $1 million of average sales revenue, while 32.9% were relatively large firms earning more than $2 million of sales revenue per year. Also, 32.2% of the firms had more than 500 employees. For more than half of the companies, SG&A costs were between 10% and 30% of sales revenue, comparable to the statistics reported in the previous archival studies (e.g., ABJ). Most of the companies (96.1%) have operated for more than five years. The industry 252 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 distribution indicates that the sample firms are from a wide range of industries, which reduces the concern with sample clustering. Panel B reports demographic information ofthe sample managers (i.e., survey respondents). While various roles are played by sample managers, the largest group consists of general managers (29.6%), who are expected to have the most influence over SG&A spending for the business unit. Most of the respondents have experience of 3 years or longer either at their current position or in the current industry. The mean age was 39 and about two thirds of the sample managers were male. On average, total annual compensation was $82.6 thousand, which consists of 70.9% of fixed annual salary, 18.7% of short-term or long-term incentives, and 10.4% of other types. The large portion of fixed salary suggests that the compensation structure of middle managers is very different from that of top executives who typically receive significant portions of total compensation as incentives.7 5 RESULTS 5.1 Quantitative analysis 5.1.1 SG&A cost decisions of middle managers To gauge the degree to which middle managers are willing to change the overall SG&A spending for a given sales change, we asked the following hypothetical question: Hypothetical question: Assume sales have been increasing for the past five years. How much change in SG&A costs would you make under the following situations?8 1. when sales growth this year is 0%? 2. when sales increase by 5%? 10%? 15%? 3. when sales decrease by 5%? 10%? 15%? The two extreme situations, 15% increase and 15% decrease, are still considered within the normal range of annual sales change, which also means that the responses for these scenarios are considered a normal operational decision. The assumption of past sales 7 Banker, Jin, and Mehta (2018) report that on average, a CEO of a S&P 1500 company receives 68.2% of the total compensation in the form of incentives. 8 The survey asked respondents' decisions regarding SG&A costs, as well as other cost items. The responses for SG&A cost, the main cost item in the cost stickiness literature, are separated from others for reporting purposes. See Table 3 for responses for the rest of the cost items. E/B/R B. JIN, J. C CARY | ARE MIDDLE MANAGER'S COST DECISIONS STICKY? EVIDENCE 253 increase was given considering that managers' positive expectation for future sales is the main assumption in the cost stickiness theory (ABJ).9 Table 2: Survey responses to the question: "How much change in SG&A costs would you make under the following situations?" Change in SG&A costs (%) Hypothetical situation Mean Comparison with prior range One-tailed p-value 25th percentile 50th percentile (median) 75th percentile When sales growth this year is 0% 4.53% 0.00% 5.00% 5.00% When sales increase by ... 5% 6.40% +1.87%*** 0.01 0.75% 5.00% 6.00% 10% 7.03% +0.63% 0.31 1.00% 5.00% 7.25% 15% 7.28% +0.26% 0.39 2.00% 5.00% 10.00% When sales decrease by . 5% 2.78% -1.75%*** < 0.01 0.00% 0.00% 5.00% 10% 2.85% +0.07% 0.54 0.00% 0.00% 5.00% 15% 2.06% -0.79%* 0.07 0.00% 0.00% 5.00% Note: Table 2 summarizes the survey response to the question "How much change in SG&A costs would you make" under various scenarios regarding sales change. Respondents are given the assumption that sales have been increasing for the past five years. "Comparison with prior range" column presents the mean comparison between ranges regarding sales change. For ranges of sales increase, it is tested whether the mean SG&A cost change for the range is statistically larger than that for the previous sales increase range. (E.g., for the situation of +10% sales change, it is tested whether the mean response is statistically greater than the mean response for the +5% sales change.) For ranges of sales decrease, it is tested whether the mean SG&A cost change for the range is statistically smaller than that for the previous sales decrease range. (E.g., for the situation of -10% sales change, it is tested whether the mean response is statistically smaller than the mean response for the -5% sales change.) *, **, and *** denote significance at levels of 0.1, 0.05, and 0.01, respectively. Table 2 presents the summary of the responses. Empirical studies in the cost stickiness literature generally use the zero sales change as the point where the slope of the sales-costs relation changes, meaning the cost decisions at zero sales growth may serve as a benchmark when examining whether the cost behavior is sticky. On average, the respondents indicate that they are willing to increase overall SG&A costs by 4.53% even when sales revenue does not grow at all in the current period. A potential explanation for this positive cost change is that the managers are optimistic and believe the sales will rise in the future. Considering 9 Prior literature also finds that costs are "anti-sticky" (i.e., costs change relatively more when sales decrease than when sales increase) when managers are pessimistic about future sales revenue (Banker et al., 2014). 254 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 the respondents are middle managers, another explanation is that there is a corporate-level strategy or policy to follow regarding the minimum level of SG&A spending. Next, the responses for the scenarios of sales increase indicate that middle managers tend to increase overall SG&A spending as expected sales growth increases, as intuitively expected. More interestingly, the increase in SG&A cost change is mitigated as sales growth increases, suggesting that middle managers increase SG&A spending relatively less when the magnitude of sales increase is large compared to when the magnitude of sales increase is small. In particular, the mean response was to add 1.87% (= 6.40% - 4.53%) extra SG&A spending when sales growth changes from 0% to +5%. However, the extra increase in SG&A spending drops to 0.63% (= 7.03% - 6.40%) when sales growth changes from +5% to +10% and further drops to 0.25% (= 7.28% - 7.03%) when sales growth changes from + 10% to +15%. The difference in means was statistically significant only for 0% vs. +5% and insignificant at the conventional level of significance for +5% vs. +10% and +10% vs. +15%. This is consistent with our expectation based on the characteristics of middle management including limited ability in adding resources and risk aversion. Last, the responses for the scenarios of sales decrease indicate that middle managers tend to reduce the increase in overall SG&A spending as sales decrease, again, as intuitively expected. Similarly to the case of sales increase, the degree of the SG&A cost change is relatively smaller when the sales decrease is large compared to when the sales decrease is small. In particular, the extra cut in the SG&A spending was 1.75% (= 4.53% - 2.78%, p-value < 0.01) when sales growth changes from 0% to -5%. However, the cut in the SG&A cost is substantially mitigated when sales growth drops further. In particular, the difference in mean cost changes between -10% and -15% sales growth scenarios is statistically insignificant. The additional SG&A cut when sales growth further drops from -10% to -15% was 0.79% (= 2.85% - 2.06%, p-value = 0.07), which is insignificant at the conventional level of significance (p-value < 0.05) and much smaller in magnitude compared to 1.75%, the SG&A cut for the sales growth range between 0% and -5%. The relatively smaller decrease in SG&A costs for a large sales decrease is consistent with the empirical findings in the prior cost stickiness literature (e.g., ABJ). It is also consistent with our expectation based on (1) limited ability of middle managers and (2) risk aversion by middle managers. E/B/R B. JIN, J. C CARY | ARE MIDDLE MANAGER'S COST DECISIONS STICKY? EVIDENCE 255 Figure 2: Sticky SG&A cost decision of middle managers Note: Figure 2 presents the mean and median survey responses to the question "How much change in SG&A costs would you make?" given the sales growth in this year is 0%, +5%, +10%, +15%, -5%, -10%, and -15%. The respondents are given the assumption that sales have been increasing for the past five years. Figure 2 graphically summarizes the non-linear SG&A cost decisions of middle managers observed from the survey responses. For the line representing the mean responses, the slope is relatively steeper when the sales change is relatively small in magnitude (from -5% to +5%) and relatively flatter when the sales change is relatively large in magnitude (-5% or lower and +5% or higher). Similarly, the median response of 0% of SG&A cost change for -5% sales change does not decrease further when the magnitude of sales decrease gets larger and the median response of 5% for zero sales growth does not rise when the expected sales growth increases. Overall, the non-linear cost behavior of middle managers shown in Figure 2 is consistent with our expectation. The shape of the two plots in Figure 2 also suggests that while the empirical models in the prior cost stickiness literature generally use zero sales growth as the point where the slope changes, the change in managerial behavior may not be triggered by a mere sales decrease. Figure 2 suggests that it is rather a "sufficient large" sales decrease. More generally, the cost behavior at the middle management level can be described as costs changing relatively less when the magnitude of sales change (i.e., sales increase or decrease) is sufficiently large.10 10 The criteria for being "sufficiently large" are not necessarily the same for sales increase and for sales decrease. 256 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 5.1.2 Other cost and investment decisions of middle managers While the prior literature on cost stickiness focuses on SG&A costs, where managers are supposed to have the most discretion, we also examine middle managers' decisions regarding other cost and investment items. Similarly to the main questions about SG&A cost decisions, we asked the following question for (1) human resources (i.e., hiring and firing), (2) investment in fixed assets (e.g., machine and equipment), and (3) investment in intangible assets (e.g., patent and software): Hypothetical question: Assume sales have been increasing for the past five years. How much change in cost or investment would you make under the following situations? 1. when sales growth this year is 0%? 2. when sales increase by 5%? 10%? 15%? 3. when sales decrease by 5%? 10%? 15%? For these cost and investment decisions on which managers are supposed to have relatively smaller discretion compared to that on SG&A cost decisions, we excluded responses of the managers who self-reported that they have weak or no discretion on the corresponding decision. The survey responses summarized in Table 3 and Figure 3 show a pattern very similar to that of SG&A cost decisions shown in Table 2 and Figure 2. In specific, the mean and median responses show that the change in the cost or investment is less sensitive to the change in sales revenue when the magnitude of sales change is relatively large. This suggests that first, similarly to the case of SG&A costs, the magnitude of employee layoffs or cut in asset investments by middle managers is relatively small when the magnitude of sales decline is sufficiently large, consistent with the cost stickiness theory and our prediction. Second, also similarly to the case of SG&A costs, middle managers do not want to substantially increase the number of employees or investments in assets when experiencing a sales boom, which is consistent with our hypothesis. E/B/R B. JIN, J. C CARY | ARE MIDDLE MANAGER'S COST DECISIONS STICKY? EVIDENCE 257 Table 3: Survey responses to the question: "How much change in cost or investment would you make under the following situations?" Change in number of Change in fixed asset Change in intangible employees (%) investment (%) asset investment (%) Hypothetical situation Mean Median Mean Median Mean Median When sales growth this year is 0% 6.02% 2.00% 6.24% 4.00% 4.98% 5.00% When sales increase by ... 5% 6.48% 5.00% 5.58% 4.00% 6.40% 5.00% 10% 9.15% 5.00% 7.03% 5.00% 7.03% 5.00% 15% 8.92% 5.00% 7.28% 5.00% 7.28% 5.00% When sales decrease by ... 5% 3.58% 0.00% 2.78% 0.00% 2.78% 0.00% 10% 2.75% 0.00% 2.85% 0.00% 2.85% 0.00% 15% 2.52% 1.00% 2.06% 0.00% 2.06% 0.00% Note: Table 3 summarizes survey responses to the question asking the intended level of change in number of employees, fixed asset investment, and intangible asset investment. The responses of managers who self-reported that they have weak or no discretion on the corresponding cost or investment item are excluded. The number of responses is 130 for employment, 117 for fixed asset investment, and 126 for intangible asset investment. Respondents are given the assumption that sales have been increasing for the past five years. Figure 3: Employment and asset investment decisions of middle managers Figure 3A Employment decision of middle managers (n = 130) 51 §20 -15 -10 -5 0 5 10 15 20 B ^ js % Change in Sales Revenue U —•—Mean — -Median 258 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Note: Figure 3 presents the mean and median survey responses to the question "How much change in cost or investment would you make?" given the sales growth in this year is 0%, +5%, +10%, +15%, -5%, -10%, and -15%. The respondents are given the assumption that sales have been increasing for the past five years. Figures 3A, 3B, and 3C are for the number of employees, fixed asset investment, and intangible asset investment, respectively. 5.1.3 Subsample analysis of the impact of compensation structure One of our explanations for the reverse Z-shaped cost behavior at the middle management level is that middle managers are likely to be more risk averse than top managers, due to their compensation structure which includes a relatively small portion of incentives. To test the validity of this explanation, we conducted a subsample analysis. Using the median value of total incentives as a percentage of total annual compensation (20.0%), we constructed two subsamples and repeated the main analysis described above for each of the two subsamples.11 11 Total incentive is defined as the sum of short-term cash bonus and long-term incentives. E/B/R B. JIN, J. C CARY | ARE MIDDLE MANAGER'S COST DECISIONS STICKY? EVIDENCE 259 Table 4: Subsample analysis of the impact of the compensation structure Change in SG&A costs (%) Managers with small incentives (< 20% of total compensation) (n = 101) Managers with large incentives (> 20% of total compensation) (n = 51) Hypothetical situation Mean Comparison with prior range 50 th percentile (median) Mean Comparison with prior range 50 th percentile (median) When sales growth this year is 0% 4.94% 5.00% 3.73% 2.00% When sales increase by ... 5% 7.06% +2.12%** 5.00% 5.10% +1.37%** 3.00% 10% 7.72% +0.66% 5.00% 5.65% +0.55% 5.00% 15% 7.37% -0.36% 5.00% 7.12% +1.47%** 5.00% When sales decrease by . 5% 2.45% -2.49%*** 0.00% 3.45% -0.27% 2.00% 10% 2.71% +0.27% 0.00% 3.12% -0.33% 2.00% 15% 2.07% -0.64% 0.00% 2.04% -1.0 Note: Table 4 presents the results of the subsample analysis performed to examine the impact of the compensation structure on cost decisions. Using the median value of total incentives (= cash bonus + long-term incentives) as a percentage of total compensation, two subsamples have been constructed. "Comparison with prior range" column presents the mean comparison between ranges regarding sales change. For ranges of sales increase, it is tested whether the mean SG&A cost change for the range is statistically larger than that for the previous sales increase range. (E.g., for the situation of+10% sales change, it is tested whether the mean response is statistically greater than the mean response for the +5% sales change.) For ranges of sales decrease, it is tested whether the mean SG&A cost change for the range is statistically smaller than that for the previous sales decrease range. (E.g., for the situation of -10% sales change, it is tested whether the mean response is statistically smaller than the mean response for the -5% sales change.) *, **, and *** denote significance at levels of 0.1, 0.05, and 0.01, respectively. Figure 4: Subsample analysis of the impact of compensation structure 260 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Note: Figure 4 presents the mean and median survey responses to the question "How much change in SG&A costs would you make?" for two subsamples constructed based on the compensation structure. The respondents are given the assumption that sales have been increasing for the past five years. Figure 4A summarizes the responses of the managers who receive equal to or less than 20% of total compensation as incentives. Figure 4B summarizes the responses of the managers who receive more than 20% of total compensation as incentives. Table 4 and Figure 4 present the results of the subsample analysis. For the middle managers who receive relatively small incentives (equal to or less than 20% of total compensation), the responses remain very similar to those for the main sample (i.e., change in SG&A costs is relatively small when the magnitude of sales change is large). On the other hand, the responses of the middle managers who receive relatively large incentives (greater than 20% of total compensation) show that the "sticky" cost behavior at the higher end is less significant. In specific, Table 4 shows that the increase in the mean response when sales growth increases from 10% to 15% is statistically significant (one-tailed p-value = 0.025), suggesting that the increase in SG&A spending is not mitigated even when sales growth reaches 15%. The median also rises at least until the sales growth reaches 10%, unlike the case for the main sample or the subsample of middle managers with small incentives where the median does not increase at all in the range of increasing sales. The difference in the cost behavior between the two subsamples can be more easily identified in Figure 4. Overall, the result of the subsample analysis suggests that middle managers who receive compensation relatively more in the form of incentives are less likely to slow down in adding resources when experiencing a sales boom, which supports our expectation that incentive compensation mitigates the risk-averse behavior of managers. E/B/R B. JIN, J. C CARY | ARE MIDDLE MANAGER'S COST DECISIONS STICKY? EVIDENCE 261 5.2 Qualitative analysis 5.2.1 Survey To obtain a better understanding of the cost behavior at middle management level, we also asked qualitative questions in the survey in addition to the quantitative questions discussed above. First, we asked which factors affected their cost decisions in the quantitative section. From the prior literature on cost stickiness, we obtained potential factors as follows: • Economy • Company's past performance • Long-term relation between company and employees • Morale of employees • Short-term cash bonus • Long-term incentives • Expenses related to hiring/firing process (e.g., training fees, severance pay) • Expenses related to machine/equipment (e.g., installation fees, transportation fees) The question has been asked using a 5-point Likert scale from 1 to 5 (1=No impact, 2=Minor impact, 3=Neutral, 4=Moderate impact, 5=Major impact). In addition, we also asked if there were any other factors which affected their decision-making process. Figure 5: Factors affecting middle managers' cost decisions Figure 5A Factors with Major or Moderate impact 0% 10% 20% 30% 40% 50% 60% 70% 80% 262 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Note: Figure 5 summarizes the survey responses regarding factors affecting cost decisions at the middle management level. For each factor obtained from the prior literature, respondents were asked to indicate the significance of the impact using a 5-point Likert scale (1=No impact, 2=Minor impact, 3=Neutral, 4=Moderate impact, 5=Major impact). Figure 5 summarizes the responses regarding the impact of each factor. Figure 5A shows that all the potential factors were identified to have at least a moderate impact by 50% or more respondents. A relatively small number of respondents indicated short-term cash bonus (50.0%) or long-term incentives (61.2%) as a factor with a major or moderate impact, consistent with the fact that only 11.7% and 7.0% of total compensation are received in the form of short-term cash bonus and long-term incentives, respectively. Figure 5B shows that the most respondents (32.2%) selected the economy as a factor with a major impact on their cost decisions, which supports the argument in the prior literature that the economic condition affects managers' belief about permanence of the current sales decline, ultimately affecting their cost decisions (ABJ; Banker et al., 2014). Again, a relatively small number of respondents (21.1%) chose short-term cash bonus as a factor with a major impact on their cost decisions. The respondents also indicated that their cost and investment decisions are affected by several factors in addition to those provided from the survey. Based on their nature, we classified those additional factors as follows: • Factors restricting middle managers' cost or investment decisions - Annual budget or availability of cash - Minimum acceptable rate of return - Availability of qualified labor force - Long-standing contracts with suppliers - Corporate level strategy E/B/R B. JIN, J. C CARY | ARE MIDDLE MANAGER'S COST DECISIONS STICKY? EVIDENCE 263 • Other additional factors - General trend in business or market - Behavior or strategy of main competitor(s) - Needs from customers or clients Consistent with our prediction, many respondents indicated that there are factors which limit their cost or investment decisions. First, annual budget and availability of cash directly limit the middle managers' ability to add resources. Also, minimum acceptable rate of return, which is often demanded by top managers, forces middle managers to limit their expenses to maintain a high return. In addition, middle managers' employment-related decisions are also affected by availability of qualified labor force for the current period. These factors are likely to set the upper limit in increasing costs, consistent with the relatively small increase in costs when the sales increase is large as shown in Table 2 and Figure 2. On the other hand, long-standing contracts with suppliers are likely to set a contractual minimum (i.e., the lower limit) for raw material or merchandise purchase per year, resulting in limited ability in cutting resources, consistent with the relatively small change in costs when the magnitude of sales decrease is large. Many respondents also indicated their decisions are significantly affected by corporate- or top management-level strategy such as globalization or increasing market share, which can set either an upward limit or a downward limit, depending on its nature. Respondents also reported additional factors which do not necessarily restrict their decisions. Those factors include (1) general trends in the market or industry, (2) strategy or behavior of their major competitors, and (3) needs from their clients or customers. These responses confirm the widely-accepted fact that management decisions are heavily influenced by Porter's (1979) five forces (i.e., industry rivalry, bargain powers of buyers/ suppliers, threats of new entrants/substitutes). Last, the survey directly asked the participants if there was any personal or corporate policy or strategy to follow regarding the maximum and minimum levels of cost or investment. The results summarized in Figure 6 show that a significant number of respondents have a certain policy to follow when making cost or investment decisions. In specific, 37.1% of valid responses indicated the existence of a personal or corporate policy regarding the maximum level of cost or investment. Specific examples include an increase in SG&A expenses by a maximum of 5% from the prior period's expenses, a maximum number of line workers limited due to factory or equipment capacity, maximum SG&A spending limited to the annual budget, etc. Regarding the minimum level of cost or investment, 42.3% of valid responses indicated the existence of a restricting policy. Examples include an increasing number of temporary workers by 1% every year, not cutting SG&A spending regardless of performance, spending all the budget given for the period, etc. Interestingly, the annual budget seems to serve as both the upper limit and the lower limit for cost and investment decisions. 264 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Overall, the result of the qualitative analysis suggests that middle managers are likely to face the upper limit and/or the lower limit when making a cost or investment decision, which explains the reverse Z-shape of cost-sales relation identified from the quantitative analysis. Figure 6: Existence of policy, strategy, or norm regarding the minimum or maximum level of annual investment Figure 6A Existence of policy or strategy for Maximum level of cost or investment Figure 6B Existence of policy or strategy for Minimum level of cost or investment Note: Figure 6 summarizes the survey responses to the question asking if there is any policy, strategy, or norm regarding the minimum or maximum level of annual investment. Many of the respondents who answered "Yes" to the question also provided a description of the policy or strategy. The examples of policies for the maximum level include (1) the increase in SG&A cost limited to a certain percentage of prior SG&A costs and (2) the maximum number of line workers limited due to the factory capacity. The examples for the minimum level include (1) not cutting SG&A cost regardless of the current performance and (2) spending all the budget given for the period. 5.2.2 Interview To obtain an even deeper understanding of the decision-making mechanism at the middle management level, we conducted interviews with two middle managers currently in E/B/R B. JIN, J. C CARY | ARE MIDDLE MANAGER'S COST DECISIONS STICKY? EVIDENCE 265 practice, who were selected and approached using our personal network; Manager A is a director of client services at a company which provides seismic data to the oil and gas industry; Manager B is a production manager at a manufacturer of custom molded plastic parts. As a part of the briefing, our survey questionnaires were provided to each of the interviewees and the actual interviews were conducted a few days later through telephone. Similarly to the survey respondents, both of the interviewees indicated that their decisions to add or cut resources are affected by top management and/or other factors, although the degree varies. Manager A, who self-reported that he has "a great deal of discretion" in terms of spending and resource allocation, stated: "If I think a $500 resource is needed for an operation or a project, I simply spend the capital and continue. However, if the resource needed approaches the $10,000 mark, I send it to upper management for confirmation before executing the order ... My discretion range to give raises (to the employees) is 3-5%, without consulting or push-back from top management. If I want to consider an employee for a 10% raise, then this requires approval at the executive level and from upper management." Similarly, Manager B, who exercises a "moderate level of discretion" in terms of spending and human resource allocation, stated: "(SG&A spending) is rarely my complete decision but rather the committee's that I work and consult with. I need to go through upper management for most of the major decisions." These statements suggest that their managerial decisions to increase spending are limited by top management, although the degree varies, which is consistent with the survey responses in general. Regarding the factors affecting their resource allocation decisions, Manager A stated: "We are in a "sales driven" business and have to maintain an operation that can react and bring a deal to fruition within a quick delivery window, closing out the few competitors we do have. There are about ten other companies we compete with domestically, so this makes it easy for customers to work with us, as they know who has the services in this field." This implies that competitors and customers are limiting his discretion in cutting resources to a certain degree, as many survey respondents also indicated. On the other hand, Manager B stressed the significant influence of company-level strategy: 266 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 "Prior to 2009, the company was in a growth stage as was the industry (and thus my discretion in cutting resources was limited) ... On the contrary, subsequent to the 2009 economic upheaval, the industry, and my company as well, have yet to truly recover from the recession (meaning my discretion in increasing costs is somewhat limited.)" To summarize, the interview statements are consistent with our intuition and observation from the survey. Although the real world decision making processes, identified during the interviews, are much more complicated and dynamic compared to the simplified plots we have drawn from the survey results, the interviews confirmed at least that middle managers' discretion in spending decisions is limited both upward and downward and the limiting factors include top managers and their strategies. 5.3 Empirical analysis Middle managers include heads of business segments, such as division managers and regional managers, who can be reasonably considered to have the most significant influence on the segment level cost decisions. As such, we also conducted an empirical analysis using segment level data obtained from Compustat, which covers all publicly traded companies in the U.S., to complement our findings from the survey and field interviews. Our sample period spans fiscal years 2008-2015 and the number of segment/year observations was 26,050.12 Cost behavior at the middle management (or segment) level was examined using the following regression model: ASG&At = P0 + ^ AREVt + DECt x AREVt + DECt x AREVt x SUCCESSIVE_DECt + B„ DEC, x AREV, x ASSETINT + ^ LARGE INC x AREVt ' 4 t t t ' 5 — t t + Industry/Year Fixed Effects (1) where ASG&A is natural logarithm of current SG&A costs over prior SG&A costs and AREV is natural logarithm of current sales revenue over prior sales revenue. Both ASG&A and AREV are winsorized at the 1% level. DEC is a dummy variable which takes the value of 1 if sales revenue of the firm decreases in the current period, and 0 otherwise. Similar to ABJ, a negative ^2 would indicate that costs decrease relatively less when sales decrease. We also include interaction terms containing a dummy variable for successive sales decrease (SUCCESSIVE_DEC = 1 if sales decrease for two consecutive years) and asset intensity (ASSETINT = log (total assets / sales revenue)), considering the factors affecting the degree of cost stickiness. We use dummy variables based on the two-digit Standard Industry Classification (SIC) codes and year dummies to control for the industry and year fixed effects, respectively. The main variable of interest is the interaction term containing 12 Our sample period spans 8 years (2008-2015), since our data source, Compustat's Current Segments database, provides information for the past 8 years. E/B/R B. JIN, J. C CARY | ARE MIDDLE MANAGER'S COST DECISIONS STICKY? EVIDENCE 267 LARGE_INC, a dummy variable for a large sales increase, which is defined using different values of sales increase. (See Note for Table 5 for detailed variable definitions.) A negative would indicate that SG&A costs become sticky when the magnitude of sales increase reaches a given level of sales increase. Table 5: Regression analysis ofSG&A cost behavior at the segment level (1) (2) (3) (4) (5) VARIABLES ASG&At ASG&At ASG&At ASG&At ASG&At AREVt 0.403*** 0.341*** 0.424*** 0.487*** 0.537*** (68.93) (9.67) (14.07) (18.68) (23.30) DEC txAREVt -0.093*** -0.029 -0.115*** -0.182*** -0.238*** (-6.59) (-0.74) (-3.37) (-5.96) (-8.50) DEC txAREVtxSUCCESSIVE_DEC t 0.093*** 0.094*** 0.093*** 0.093*** 0.092*** (6.26) (6.29) (6.25) (6.21) (6.18) DEC txAREVtxASSETINTt -0.046*** -0.047*** -0.046*** -0.046*** -0.045*** (-15.31) (-15.40) (-15.18) (-14.94) (-14.68) LARGE_INC15 lxAREVl 0.061* (1.79) LARGE_INC20 xAREV -0.021 (-0.72) LARGE_INC25 txAREVt -0.083*** (-3.30) LARGE_INC30 xAREV -0.133*** (-6.01) Constant 0.033*** 0.035*** 0.032*** 0.029*** 0.025*** (13.78) (13.57) (12.21) (10.65) (9.08) Industry/Year Fixed Effects Included Included Included Included Included Observations 26,050 26,050 26,050 26,050 26,050 Adjusted R-squared 0.217 0.217 0.217 0.218 0.218 Note: Table 5 presents the results of the multivariate regression analysis based on 26,050 segment/year observations. *, **, and *** denote significance at levels of 0.1, 0.05, and 0.01, respectively. T-statistics are in parentheses. SG&At = Selling, general, and administrative costs in year t (in million $); ASG&At = Log (SG&At/ SG&A ); REVt = Sales revenue in year t (in million $); AREVt = Log (REVt / REVJ; DECt = 1 if REVt < REV tl, = 0 otherwise; SUCCESSIVE_DECt = 1 if REV tl < REV = 0 otherwise; TAt = Total assets (in million $); ASSETINTt = Log (TAt/ REV); LARGE_INC15t = 1 if AREVt > 0.15, = 0 otherwise; LARGE_INC20t = 1 if AREVt > 0.20, = 0 otherwise; LARGE_INC25 t = 1 if AREVt > 0.25, = 0 otherwise; LARGE_INC30t = 1 if AREVt > 0.30, = 0 otherwise. 268 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 The regression results are presented in Table 5. Consistent with the prior literature, the coefficient on DECxAREV is significant and negative in general, indicating that cost becomes sticky when sales decrease. The coefficient on SUCCESSIVE_DEC interaction term is significant and positive in general, suggesting a lower degree of SG&A cost stickiness at the lower end when sales decline for two consecutive years. The significant and negative coefficients on ASSETINT interaction term indicate that SG&A costs are stickier at the lower end for firms that require relatively more assets to support their sales. Most interestingly, the coefficients on the interaction term for a large sales increase show that cost becomes sticky when the magnitude of sales increase is "sufficiently" large. In specific, the coefficients are not significantly negative when the sufficiently large sales increase is defined as AREV of 0.15 or higher (Column (2)) or 0.2 or higher (Column (3)), suggesting that a sales increase up to about 20% does not trigger the cost stickiness at the higher end. The coefficient becomes significantly negative when the sufficiently large sales increase is defined as AREV of 0.25 or higher (Column (4)), suggesting that approximately 25% change in sales revenue is sufficiently large to induce sticky cost behavior at the higher end. Considering that a significant portion of the sample (20.9%) has AREV of 0.25 or higher (untabulated), the conditions that trigger sticky cost behavior at the higher end (e.g., 25% sales increase) are still considered normal rather than extreme. The negative coefficient becomes even more significant and larger in magnitude when AREV of 0.3 is used to define the dummy variable (Column (5)), as intuitively expected. Overall, the regression results based on segment level data suggest that cost behavior at the segment level is sticky not only when sales decrease but also when the magnitude of sales increase is large, consistent with our findings from the survey instrument and the interviews. 6 DISCUSSION WITH CONCLUSIONS 6.1 Theoretical contributions Decisions at the middle management level are different from those at the top management or corporate level because middle managers are likely to have limited ability in both adding and cutting resources and also because the salary-focused compensation structure for middle managers are likely to induce more risk-averse behavior. In this study, we examine cost behavior at the middle management level using two different approaches. First, we take a behavioral approach and conduct a survey and field interviews. The analysis results based on the detailed interviews and 152 survey responses indicate that middle managers' cost decisions are sticky (i.e., change relatively less) when the magnitude of sales change is sufficiently large at both increasing and decreasing ends. Our findings contribute to the prior literature on cost stickiness by suggesting the existence of stickiness at the higher end (i.e., when the sales increase is large) at least at the middle management level and also by confirming the empirical findings in the literature using behavior approaches. E/B/R B. JIN, J. C CARY | ARE MIDDLE MANAGER'S COST DECISIONS STICKY? EVIDENCE 269 Second, we use archival data to empirically confirm our findings from the survey and the interviews. Using a regression analysis based on 26,050 segment-level observations for publicly traded companies in the U.S., we show that cost decisions at the segment level are sticky at both low and high ends, consistent with our findings from the survey and the interviews. Using segment level data also contributes to the prior literature which relies heavily on company level data and examines the cost asymmetry at the low end only (i.e., firms facing a sales decline). 6.2 Practical implications Middle managers' cost decisions, which are sticky not only when sales decrease but also when the magnitude of sales increase is large, have practical implications for both top managers and investors. For top managers, the sticky cost behavior at the high end suggests that the cost decisions of middle managers are restricted by annual budgets and corporate-level strategies or policies, as evidenced by the survey results and the interviews. This further suggests that a company may face an undesirable situation of losing an opportunity to grow because investments or expenditures at the middle management level are restricted for internal reasons. For investors and analysts, the sticky cost behavior at the high end suggests that analysts' earnings forecasts are likely to be biased when the magnitude of sales increase is large. Banker, Jin, and Mehta (2018) argue that if analysts fail to fully consider the cost stickiness (at the low end), costs of firms facing sales decline will be under-forecasted, and, by extension, earnings of those firms will be over-forecasted. In contrast, the cost stickiness at the high end that is documented in this study suggests that costs will be over-forecasted and thus earnings will be under-forecasted for firms facing a large increase in sales. 6.3 Limitations with future research directions As this study mainly uses a survey instrument and interviews, it is subject to potential caveats associated with behavioral studies, such as biased inputs from the survey/interview respondents and/or samples not representative of the whole population. To mitigate this concern, we also conduct an empirical analysis using archival data for publicly traded companies in the U.S. Another limitation in our study is that while we show that middle managers' cost decisions are sticky when the magnitude of sales increase is sufficiently large, whether the corporate-level cost behavior is also sticky at the higher end remains untested. This suggests that examining the corporate-level cost behavior at the high end will be an interesting venue for future research. Also, our survey and interview responses suggest that Porter's (1979) five forces have significant influence on cost and investment decisions, consistent with the common belief 270 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 in the management literature. This suggests that it will be interesting and thus worth exploring to examine how the five forces affect non-linearity in cost decisions individually and collectively. REFERENCES Anderson, M. C., Banker, R. D., & Janakiraman, S. N. (2003). Are selling, general, and administrative costs "sticky"?. Journal of Accounting Research, 41(1), 47-63. Balakrishnan, R., Labro, E., & Soderstrom, N. S. (2014). Cost structure and sticky costs. Journal of Management Accounting Research, 26(2), 91-116. Balakrishnan, R., Petersen, M. J., & Soderstrom, N. S. (2004). Does capacity utilization affect the "stickiness" of cost?. Journal of Accounting, Auditing & Finance, 19(3), 283-300. Banker, R. D., & Byzalov, D. (2014). Asymmetric cost behavior. Journal of Management Accounting Research, 26(2), 43-79. Banker, R. D., Byzalov, D., & Chen, L. T. (2013). Employment protection legislation, adjustment costs and cross-country differences in cost behavior. Journal of Accounting and Economics, 55(1), 111-127. Banker, R. D., Byzalov, D., Ciftci, M., & Mashruwala, R. (2014). The moderating effect of prior sales changes on asymmetric cost behavior. Journal of Management Accounting Research, 26(2), 221-242. Banker, R.D., Jin, B., & Mehta, M. (2018) Short-term Bonus for CEOs and Cost Stickiness (previously titled CEO Compensation and Cost Structure), Working Paper, Temple University [online] http://www.fox.temple.edu/cms/wp-content/uploads/2013/08 / Byunghoon Jin.pdf Belcher, D. W., & Atchison, T. J. (1987). Compensation Administration. Prentice Hall. Burgelman, R. A. (1983). A process model of internal corporate venturing in the diversified major firm. Administrative Science Quarterly, 223-244. Chen, C. X., Lu, H., & Sougiannis, T. (2012). The agency problem, corporate governance, and the asymmetrical behavior of selling, general, and administrative costs. Contemporary Accounting Research, 29(1), 252-282. E/B/R B. JIN, J. C CARY | ARE MIDDLE MANAGER'S COST DECISIONS STICKY? EVIDENCE 271 DeFusco, R. A., Johnson, R. R., & Zorn, T. S. (1990). The effect of executive stock option plans on stockholders and bondholders. The Journal of Finance, 45(2), 617-627. Delmestri, G., & Walgenbach, P. (2005). Mastering techniques or brokering knowledge? Middle managers in Germany, Great Britain and Italy. Organization Studies, 26(2), 197220. Dierynck, B., Landsman, W. R., & Renders, A. (2012). Do managerial incentives drive cost behavior? Evidence about the role of the zero earnings benchmark for labor cost behavior in private Belgian firms. The Accounting Review, 87(4), 1219-1246. Dutton, J. E., & Ashford, S. J. (1993). Selling issues to top management. Academy of Management Review, 18(3), 397-428. Dutton, J. E., Ashford, S. J., O'neill, R. M., Hayes, E., & Wierba, E. E. (1997). Reading the wind: How middle managers assess the context for selling issues to top managers. Strategic Management Journal, 407-423. Eisenhardt, K. M. (1989). Agency theory: An assessment and review. Academy of Management Review, 14(1), 57-74. Fama, E. F., & Jensen, M. C. (1983). Separation of ownership and control. The Journal of Law and Economics, 26(2), 301-325. Floyd, S. W., & Wooldridge, B. (1992a). Middle management involvement in strategy and its association with strategic type: A research note. Strategic Management Journal, 13(S1), 153-167. Floyd, S. W., & Wooldridge, B. (1992b). Managing strategic consensus: the foundation of effective implementation. The Executive, 6(4), 27-39. Floyd, S. W., & Wooldridge, B. (1994). Dinosaurs or dynamos? Recognizing middle management's strategic role. The Academy of Management Executive, 8(4), 47-57. Floyd, S. W., & Wooldridge, B. (1997). Middle management's strategic influence and organizational performance. Journal of Management Studies, 34(3), 465-485. Floyd, S. W., & Wooldridge, B. (1999). Knowledge creation and social networks in corporate entrepreneurship: The renewal of organizational capability. Entrepreneurship Theory and Practice, 23(3), 123-144. 272 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Gomez-Mejia, L. R., & Balkin, D. B. (1992). Determinants of faculty pay: An agency theory perspective. Academy of Management journal, 35(5), 921-955. Graham, J. R., Harvey, C. R., & Rajgopal, S. (2005). The economic implications of corporate financial reporting. Journal of Accounting and Economics, 40(1-3), 3-73. Huy, Q. N. (2001). In praise of middle managers. Harvard Business Review, 79(8), 72-9. Huy, Q. N. (2002). Emotional balancing of organizational continuity and radical change: The contribution of middle managers. Administrative Science Quarterly, 47(1), 31-69. Jensen, M. C., & Meckling, W. H. (1976). Theory of the firm: Managerial behavior, agency costs and ownership structure. Journal of Financial Economics, 3(4), 305-360. Kama, I., & Weiss, D. (2013). Do earnings targets and managerial incentives affect sticky costs?. Journal of Accounting Research, 51(1), 201-224. Kanter, R. M. (1982). The middle manager as innovator. Harvard Business Review, 60(4), 95-105. Kanter, R.M. (1988). When a thousand flowers bloom: structural, collective, and social conditions for innovation in organization. Research in Organizational Behavior, Vol. 10, 169-211. Likert, R. (1961). New Patterns of Management, McGraw-Hill, NY. Mueller, D. (2003). The Corporation: Growth, Diversification and Mergers. Routledge. Murphy, K. J. (1999). Executive Compensation. Handbook of Labor Economics, O. Ashenfelter and D. Card, eds. Porter, M. E. (1979). How competitive forces shape strategy. Harvard Business Review, 21-38. Rajgopal, S., & Shevlin, T. (2002). Empirical evidence on the relation between stock option compensation and risk taking. Journal of Accounting and Economics, 33(2), 145-171. Shimizu, K. (2012). Risks of corporate entrepreneurship: Autonomy and agency issues. Organization Science, 23(1), 194-206. E/B/R B. JIN, J. C CARY | ARE MIDDLE MANAGER'S COST DECISIONS STICKY? EVIDENCE 273 Westley, F. R. (1990). Middle managers and strategy: Microdynamics of inclusion. Strategic Management Journal, 11(5), 337-351. Williamson, O.E. (1975). Markets and Hierarchies: Analysis and Antitrust Implications. Free Press, NY. Wooldridge, B., & Floyd, S. W. (1990). The strategy process, middle management involvement, and organizational performance. Strategic Management Journal, 11(3), 231241. 274 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 275-308 275 inflation - the harrod-balassa-samuelson effect in a dsge model setting Received: February 26, 2018 ČRT LENARČIČ1 Accepted: September 15, 2018 ABSTRACT: This paper sets up a two-country two-sector dynamic stochastic general equilibrium model that introduces sector specific productivity shocks with quality improvement mechanism of goods. It provides a model-based theoretical background for the Harrod-Balassa-Samuelson phenomenon that describes the relationship between productivity and price inflation within different sectors in a particular economy. Both, the calibrated and the estimated model are able to show that the Harrod-Balassa-Samuelson effect is confirmed by inducing tradable sector productivity shocks as they drive the non-tradable sector price inflation higher than the tradable sector price inflation. By doing this, we overcome the problem that the tradable productivity increase in a typical open economy specification reduces the relative price of domestic tradable goods relative to the foreign ones. Key words: Harrod-Balassa-Samuelson effect, DSGE model, inflation, productivity, quality improvement JEL classification: C32, E31, E32 DOI: 10.15458/ebr.86 1 INTRODUCTION The relationship between productivity and price inflation is described by the theory of the Harrod-Balassa-Samuelson phenomenon (henceforth HBS). Harrod (1933), Balassa (1964) and Samuelson (1964) independently developed and formulated the HBS productivity approach in order to explain the purchasing power parity2. The HBS effect represents a tendency for countries that experience higher tradable-sector productivity growth compared to non-tradable sector productivity growth to have higher overall price levels (Obstfeld and Rogoff, 1996). In more detail, the basic idea behind it is that the growth in the productivity of a tradable sector influences the growth of wages in the tradable and later on in the non-tradable sector. Wage growth in the tradable sector consequently affects the growth of prices in the non-tradable sector. Depending on the nominal exchange rate regime of a particular economy, it affects the real exchange rate as well. However, Betts and Kehoe (2008) studied the relationship between the real exchange rate and the relative price of non-tradable to tradable goods. Their conclusion is that the relation between the 1 Bank of Slovenia, Ljubljana, Slovenia, e-mail: crt.lenarcic@bsi.si 2 Baumol and Bowen (1967) developed a similar model that only describes the relationship between productivity and wages, and presents an important part of the HBS hypothesis, as discussed by Wagner and Hlouskova (2004). 276 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 two variables is stronger in an intense trade environment. Therefore, the basic assumption is that the relationship between the relative growth in the productivities of the tradable to non-tradable sector and the relative price of non-tradable to tradable goods is relatively straightforward if we include sectoral data for European countries, for example. In addition to the close trade environment, the sole Euro area integration process suppresses the ability of economies to adjust through the nominal exchange rate channel, which could consequently put more pressure on non-tradable price inflation. The HBS hypothesis can be tested on different entities, which in general represent countries, regions, or in many cases, sectors. In our case, we divide these entities into a tradable sector and a non-tradable sector; we use a similar principle as the De Gregorio, Giovannini, and Wolf's (1994) methodology by using the ratio of exports to total production to define both sectors. In order to do that we include and combine the NACE Revision 2 10-sector breakdown statistical classification time series data of economic activities, which provides data on labour productivity and price levels across the two sectors, and the ratio of exports to total production data calculated from the input-output tables, which are available at the World Input-Output Database (WIOD)3. By obtaining the relevant tradable and non-tradable data for further analysis and adding other observable macroeconomic data, we estimate the constructed DSGE model. The problem of permanent tradable productivity increase in a typical dynamic open economy specification is reducing the relative price of domestic tradable goods relative to the foreign ones. This implies worsening the terms of trade for the domestic economy and consequently, its real exchange does not increase. These dynamics are not consistent with empirical evidence found for the new European Union member states. The main contribution of the paper is to overcome the typical dynamic open economy setting by constructing and estimating a two-country two-sector DSGE model with the quality improvement extension, proposed by Masten (2008), in a smaller calibrated version of a dynamic model. The basic assumption is therefore the separation of the economy into a tradable and a non-tradable sector. The tradable sector is open and allows domestic goods to be exported and foreign goods to be imported, whereas the non-tradable sector is closed to foreign markets (a similar structure was used by Masten, 2008; Rabanal, 2009; Micaleff and Cyrus, 2013). The assumption is that the tradable and non-tradable sectors are exposed to different productivity shocks; this means that non-stationary real variables can grow at a different pace, thus providing a case for the HBS effect. In specifying technology, we allow a quality improvement mechanism, which is needed to replicate the appreciation of prices, without resorting to the unrealistic assumption of perfect competition in the tradable sector (Masten, 2008). We find evidence for the HBS effect, based on an augmented technology process that considers a quality improvement mechanism, which affects marginal costs by requiring 3 In defining the tradable and the non-tradable sector we differ from the standard approach used in the literature by excluding the not distinctively tradable or non-tradable sectors from the analysis. E/B/R Č. LENARČIČ | INFLATION - THE HARROD-BALASSA-SAMUELSON EFFECT ... 277 the use of more advanced inputs in the production process. The quality improvement of goods overcomes the typical open economy theoretical specification that reduces the relative prices of domestic tradable goods relative to foreign prices, and consequently worsens the terms of trade for the domestic economy. By introducing a sector-specific domestic tradable technology shock, the modelled economy responds by increasing price differential of non-tradable relative to tradable prices and the overall domestic inflation. Doing this we are able to theoretically explain why the economies with higher economic and productivity growth during the catching-up phase experienced higher inflation. In Section 2, a review of the HBS related literature is presented and discussed. Section 3 provides a theoretical framework for the DSGE model. In section 4, the classification and definition of economic activities into a tradable and a non-tradable sector is presented, obtaining sectoral price indices and time series of sectoral labour productivity growths. The calibrated model is presented in Section 5, while the estimation results of the DSGE model are given and discussed in Section 6. Section 7 presents the conclusions. 2 LITERATURE REVIEW Despite treating the HBS theory as an old idea, in which the sectoral productivity differential is seen as the driver for price inflation in the non-tradable sector (Harrod, 1933; Balassa, 1964; and Samuelson, 1964), the empirical testing of the HBS effect only became more popular in recent years as econometric methods advanced and new (or additional) time series data became available. This availability was largely due to the establishment of the EU and later on its enlargement process, together with advances and convergence of methodologies in collecting data by the national statistical offices. At the same time, addressing the HBS issue became relevant from the economic policy perspective trying to identify the different sources of (structural) inflation. Betts and Kehoe (2008) show that a close trade environment lowers the significance of the nominal exchange rate adjustment. This was (and can still be) especially important for the future EU and euro area countries, which are obliged to satisfy the Maastricht criterion of low and stable inflation, as well as for other emerging economies in trying to stabilise their overall inflation. In their comprehensive survey, Tica and Družic (2006) gather empirical evidence regarding the HBS effect. They point out that most of the empirical work supports the HBS effect. Especially strong evidence comes from the work based on cross-section empirical studies, similar to Balassa's (1964) work. A large number of papers focus on studying the magnitude of the HBS effect in accession countries in the EU. Čihak and Holub (2001) for instance study the presence of the HBS effect in the Czech Republic vis-à-vis EU countries, while allowing for differences in structures of relative prices. Jazbec (2002) considers Slovenia as the HBS case of an accession country, while Dedu and Dumitrescu (2010) test the HBS effect using Romanian data. Papers by Cipriani (2000), Coricelli, and Jazbec (2004), Halpern and Wyplosz (2001), Arratibel, Rodriguez-Palenzuela, and Thimann (2002), Breuss (2003), Wagner and Hlouskova (2004), Mihaljek and Klau (2008) consider 278 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 a larger accession country panel. Some of the work focuses also on emerging economies. Jabeen, Malik, and Haider (2011) test the HBS hypothesis on Pakistani data, while Guo and Hall (2010) test the HBS effect on Chinese regional data. These empirical strands of the HBS effect related literature opened up new questions regarding data issues and were related mostly to availability in reliability of sectoral data. As databases, especially in Europe, became more complete, new available data enabled studying the HBS effect between individual tradable and non-tradable sectors of a particular economy. Since it is difficult to clearly divide tradable and non-tradable commodities in the real world, some of the early papers tried to identify the tradability/ non-tradability of commodities. Officer (1976) proposed that manufacturing and/or industry combine a tradable sector, while the services represent the non-tradable sector. De Gregorio, Giovannini, and Wolf (1994) used a ratio of exports to total production of each sector to define both sectors. In empirical studies, mostly total factor productivity (TFP) or average productivity of labour are used. Marston (1987), De Gregorio, Giovannini, and Wolf (1994), De Gregorio and Wolf (1994), Chinn and Johnston, (1997), Halikias, Swagel, and Allan (1999), Kakkar (2002), and Lojshova (2003) use total factor productivity as a productivity proxy, while due to the lack of data on TFP, many others, i.e., Coricelli and Jazbec (2004) and Zumer (2002), use average productivity of labour. Comparing total factor productivity and average productivity of labour, the argument against the use of average productivity of labour is that it is not completely clear if average labour productivity should be regarded as a reliable indicator for representing a sustainable productivity growth, which has a long-term effect on the economy (De Gregorio and Wolf, 1994). However, according to Canzoneri, Cumby, and Diba (1999), the argument against TFP is that TFP is a result of a possibly unreliable data collection of sectoral capital stocks comparing to data collection of sectoral employment and sectoral gross value added, especially in the case of the shorter-term series. Sargent and Rodriguez (2000) also conclude that if the intent of the research is to examine trends in the economy over a period of less than a decade or so, labour productivity would be a better measure than total factor productivity. According to Kovacs (2002), another setback of using TFP is that during the catching-up phase the capital accumulation intensifies faster in the transition/accession countries than in the developed countries, due to the lower starting point in fundamentals of transition/accession countries. Therefore, the HBS effect might be overestimated. Listing some of the arguments against using TFP, we rather include the average labour productivity as a productivity proxy in the model. Comparing to the vast HBS literature in the 2000s in the accession process of the countries to the EU and the monetary union, less theoretical work was done with regards to the HBS effect in more structural and more complex models. Rogoff (1992) was the first to implement a general equilibrium framework, introducing the demand side of the economy within the HBS theory. This opened the possibility to further investigate the effects of relative productivities of production factors and the effects of the demand side of the economy on price levels. For instance, Mihaljek and Klau (2002) conclud that the E/B/R Č. LENARČIČ | INFLATION - THE HARROD-BALASSA-SAMUELSON EFFECT ... 279 HBS effect could have important policy implications for the EU accession countries in order to satisfy the Maastricht inflation criterion. To further investigate Mihaljek's point, Masten (2008) constructs a two-sector DSGE model to see whether the HBS effect could represent an issue in satisfying the Maastricht inflation criterion. Further on, Natalucci and Ravenna (2002) compare the magnitude of the HBS effect within different exchange rate regimes in the general equilibrium model, while Restout (2009) allows for varying mark-ups in its general equilibrium framework. However, Asea and Mendoza (1994) conclude that the proof of the HBS theory within the framework of general equilibrium cannot reliably asses the relationship between output per capita and domestic relative prices. In other words, conclusions regarding the HBS theory from cross-country analyses can only be conditionally accepted since it is difficult to account for cross-country trend deviations from purchasing power parity (PPP). Even more, Bergin, Glick, and Taylor (2004) show that the relationship between output per capita and domestic relative prices had historically oscillated too much for the HBS theory to be proved by cross-section empirical studies. In order to test the HBS theory their suggestion is that it should be tested with a sector-specific analysis. Following the general equilibrium strand of the HBS related literature, Rabanal (2009) offers three explanations for studying sectoral inflation dynamics in Spain in a DSGE model structure. The first explanation relates to the role of productivity growth differentials, which directly brings the possibility to study the HBS effect. Altissimo et al. (2005) introduced a seminal paper on productivity growth differentials in a DSGE model setting. The second explanation adds the role of the demand-side effects in shaping the inflation dynamics (López-Salido et al., 2005). The third explanation suggests that, due to different product and labor market structures, there is heterogeneity of inflation dynamics processes in each country of the union (Angeloni and Ehrmann, 2007; Andrés et al., 2003). Rabanal (2009) concludes that even when economies are hit by symmetric external shocks, such as for example oil prices, world demand, or nominal exchange rate, the response of sectoral inflation will be different across countries. The Rabanal's model was adopted by Micaleff and Cyrus (2013) as well. They analyse the relative importance of the three main determinants of inflation differentials in Malta. Based on these considerations, a structured theoretical framework is presented in the following section. 3 MODEL In this section, we present the theoretical framework for the two-country two-sector DGSE model. The DSGE framework follows the Rabanal (2009) model, but the main contribution of the theoretical model is its extension for sectoral wage rigidites, thus making the model more realistic. Additionally, we introduce an augmented technology process with quality improvement (Masten, 2008). In order to investigate the HBS effect phenomenon, different sectoral productivity shocks have to be introduced providing assymetricity between sectors. The monetary union is made of two economies; a domestic and a foreign country with the common monetary policy rule. They are indexed on intervals [0,5] and [5,1], respectively, where s denotes the size of the domestic country with 280 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 respect to the two-country universe. In our case, we relate to Slovenia and the rest of the euro area. The following section only gives a structural domestic economy description since the foreign economy block is analogous to the domestic economy, which is in our case Slovenia. 3.1 Households The assumption is that the representative household maximizes its utility function, given by E0T.?= 0ß* = [ln(Ct(0 - hC^ii)) - (1) where Ct (i) and Lt (i) present consumption and quantity of work effort of a particular household. The parameter 0<3<1 is the discount factor of household. We assume that households value the current consumption more than the future one. The parameter O<0<^ is the inverse of the elasticity of work effort with respect to the real wage (Frisch elasticity parameter). We assume consumption habits as well, which is represented by the parameter 0TN presents the share of the tradable goods in the aggregate consumption basket. The parameter vm>1 presents the elasticity of substitution between tradable and non-tradable goods. Since the demand for tradable goods is not dependent only on domestic goods but foreign as well, the index of the tradable consumption good is written analogously to the equation (3) with which the aggregate consumption index is defined 4 We scale the variables in the model with Ztc = Ci""™) so that the variables enter the model detrended, for example, ct = ct/z$. The scaling variable Ztc ensures a constant steady-state level of utility and is determined by productivity dynamics (Masten, 2008). E/B/R Č. LENARČIČ | INFLATION - THE HARROD-BALASSA-SAMUELSON EFFECT ... 281 where wHF represents the share of domestic tradable goods in the tradable consumption basket. The parameter vHF>1 is therefore the elasticity of substitution between domestic tradable goods and tradable goods produced abroad. The indexes of individual goods are defined by the following equations and represent a continuum of differenced goods of the same type and The parameter v>1 denotes the elasticity of substitution within one type of differentiated goods: ctH, cf and ctN. The same principle can be applied to price indexes. The aggregate price index P is then given by pt = Kw(pni-v™ + ci - «wof)1"1'™]1-™. As above, the price index for tradable goods is given by (7) :>r _ KfW^ + O-toHFXPD1-^1']1-^ (8) Households have a set of contingent riskless euro area bonds BtEA at their disposal that pay one unit of currency in every possible state of nature in t+1. The assumption is that households can trade these bonds that pay a gross interest rate of RtEA. Since households are ex ante identical, they face the same budget constraint in each period: where W represents the real wage, while represents other income sources of households. As shown in Chari, Kehoe, and McGrattan (2002), the real exchange rate is given by 282 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 where the variables and represent the marginal utilities of domestic and foreign consumption, respectively. Labour market is, in comparison to the Rabanal (2009) model, differentiated, thus provides a more realistic model assumption. Further on, the aggregation of work effort of both sectors (i.e., tradable and non-tradable) holds Against this backdrop, each household working in the tradable or the non-tradable sector sets its own wage (Erceg et al., 2000; Christiano et al., 2005). Firms aggregate the differentiated supply of labour by transforming it into a homogenous input of labour Lj, where j=N,T, in accordance with the Dixit-Stiglitz (1977) aggregator The parameter vLj is defined as the wage elasticity within different varieties of labour services in a particular sector, where j=N,T. Based on that the labour demand function for a particular is given by Combining equations (12) and (13) we get the aggregate wage, which is obtained from differentiated labour Wt} = \£wt1(i)1 vUdif -■*LJ (14) In order to introduce wage frictions in the model, we apply the Calvo (1983) principle. Each household has monopolistic power over the setting of its wage Wj (i), where j=N,T. Yet not all the households can set their optimal wage at any point of time, but only a fraction of households (1-aj where the Calvo parameter is defined on the interval 0*. Each firm in the tradable sector follows the Cobb-Douglas production function, where work effort is the only production factor and The variable AJ is a sector-specific productivity process that is characterised by quality improvement of higher-quality goods in the tradable sector index %=(ZJ)BZ with quality improvement parameter 9Z>0 (Masten, 2008), so that \nATt = \nZ{ -ln/£ . (20) The variable xt represents a quality improvement of goods index that influences wages and marginal costs via positive productivity shocks. Masten (2008) finds that the problem of permanent tradable productivity improvement in a typical open economy specification reduces the relative price of domestic tradable goods relative to the foreign ones, thus worsens the terms of trade. Consequently, the real exchange does not increase and is not consistent with empirical evidence based on the new European Union member states. On the other hand, introducing quality improvement of higher-quality goods may require the use of more advanced inputs in the production process and will consequently increase the marginal costs and product prices. Sallekaris and Vijselaar (2004) introduce a similar mechanism, as they adjust capital with a simple quality correction.5 The variable ZJ represents a tradable sector productivity shock, which is country-specific We assume that productivity shocks of both sectors can be different and that their growth rates could be different. We let the tradable productivity process ZJ to be affected by two different productivity innovations eJJ, which are country and sector specific, and e(Z, which represents a euro-area wide innovation. For the labour supply it holds LJ=LJ,H+LJH•*. 5 The idea of adjusting prices with quality improvements goes back into the 90s, as the study of Gordon (1990) tried to empirically document these biases. Later research focused on constructing quality-adjusted price indexes (Hulten, 1992; Greenwood et al., 1997; Cummins and Violante, 2002), production based estimates (Bahk and Gort, 1993), and capital model (Hobijn, 2000). E/B/R Č. LENARČIČ | INFLATION - THE HARROD-BALASSA-SAMUELSON EFFECT ... 285 Tradable sector firms producing domestic goods for the domestic market maximize their profits according to subject to where the expression A( t+k=pk A fc/A represents the stochastic discount factor, and yt±k (h) is the tradable goods demand of a firm in time t+k. YtH is the aggregate domestic-made tradable goods demand. Similarly, we can write the maximization profit function for tradable sector firms producing domestic goods for the foreign market subject to where the expression A A t/ A represents the stochastic discount factor, and y"+kd W is ^e tradable goods demand of a firm in time t+k. YtH-" is the aggregate domestic tradable goods demand from abroad. Real marginal costs in the tradable sector for both types of firms are defined as MCtT. Marginal costs are defined as the real wage normalized for augmented productivity 286 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 Both types of tradable sector firms maximize their profit with respect to prices p" (h) and ptH- (f) and demands y"+t{h)and yt+kd(f)> respectively. The tradable price dynamics of domestic produced goods for the domestic market is where PtH-°p< is the optimal price and II ^ = P(LP^-2- The tradable price dynamics of domestic goods for the foreign market is if* ^ -h, (ptH-i( n^-^y^T+(i - «H.wn1 ,(28) where PtH-°p< is the optimal price and n^ = P^[/Pt t—2' 3.2.2 Non-tradable sector Analogously to the tradable sector, each non-tradable sector firm follows the Cobb-Douglas production function, where work effort is the only production factor The variable AtN is a sector-specific productivity process that is characterised by quality improvement index Xt = (,zt)Bz so that In this respect, we assume that the sector-specific productivity process AtN is affected by quality improvement of goods X in the tradable sector, while the variable Zf represents a non-tradable sector productivity shock, which is again country-specific where we let the non-tradable productivity process Zf to be affected by a sector-specific innovation, efZ,N. Non-tradable sector firms maximize their profits E/B/R Č. LENARČIČ | INFLATION - THE HARROD-BALASSA-SAMUELSON EFFECT ... 287 subject to where the expression A( t+k=lik A fc/Af represents the stochastic discount factor, and (n) is the non-tradable goods demand of a firm in time t+k. YtN is the aggregate non-tradable goods demand. Real marginal costs in the non-tradable sector are defined as MCtN. From the cost-optimization perspective, the marginal costs are defined as the real wage normalized for productivity A non-tradable sector firm maximizes its profit with respect to price ptN (n) and demand yt+k(n)- The non-tradable price dynamics should therefore be p*,= (35) where P^' is the optimal price and llf j = P" 1/Pf-2-3.3 Monetary policy Monetary policy is modelled as a Taylor rule (Taylor, 1993) and is the same for both economies (36) where e(mp represents the monetary policy shock, while the interest rate RtEA responds to inflation and output gaps. The total output of the euro area is defined by YtEA=(Yt)s (Y*)1-s, while the overall inflation in the euro area is defined by nEA=(nt)s (n*)1-s, where s is the size of the domestic country. The parameter Qr is the weight parameter for the responsiveness of the past interest rate, while yn and yv are Taylor type paramaters for the response of the interest rate accordingly to both gaps. 3.4 Market clearing The clearing conditions are 288 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 and where variables GtT and GtN represent exogenous government spending shocks. Combining equations (37) and (38), the real GDP is What is left to do is to define the government sectoral spending process In Gf = pG:i In Gl_1 + cf'', where i=N,T. (40) 4 TRADABILITY OF SECTORS AND DATA As the theoretical model is divided into tradable and non-tradable sectors, some attention is needed for the specification and the sectoral definition of the data. The dataset consists of quarterly Slovene and euro area sectoral data, which is available from the Eurostat6 website. The time series data spans from 1998Q4 to 2018Q1 and includes sectoral gross value added data and sectoral price indexes data. 4.1 Tradability of sectors To begin with, the tradability of the sectors has to be defined. Officer (1976) proposes the following sector division. Manufacturing and other industry activities represent the tradable sector, while the services represent the non-tradable sector. De Gregorio et al. (1994) use a ratio of exports to total production to define both sectors. Their division threshold is set to 10 percent, stating that the sector is defined as tradable if the ratio of exports exceeds the 10 percent threshold, and the sector is defined as non-tradable if the ratio of exports does not exceed the 10 percent threshold. Following the De Gregorio et al. (1994) sector division, we take a step further by strictly distinguishing between the tradable and the non-tradable sector. This means that we exclude those activities from the analysis that oscillate around the 10 percent threshold too much. We provide a more detailed specification below. 6 Available at the European Commmission's statistical database site http://epp.eurostat.ec.europa.eu/portal/ page/portal/eurostat/home/. E/B/R Č. LENARČIČ | INFLATION - THE HARROD-BALASSA-SAMUELSON EFFECT ... 289 First, data on the share of exports in total value added have to be extracted from the input-output tables available at the World Input-Output Database (WIOD). We use a standard ISIC/NACE Revision 2 aggregation category, which is used for reporting data from the System of National Accounts (SNA) for a wide range of countries. We present a 10-sector breakdown in Table 1. Table 1: NACE Revision 2 10-sector classification of economic activities NACE Revision 2 Sector description Ratio of exports (in %) Tradability A B, C, D, E F Agriculture, forestry and fishing Manufacturing, mining and quarrying and other industry Construction 18.32* 45.99 2.20 T N G, H, I Wholesale and retail trade, transportation and storage, accommodation and food services 17.25 T J Information and communication 10.42 K Financial and insurance activities 12.63 L Real estate activities 0.56 N M, N O, P, Q Professional, scientific, technical, administrative and support services Public administration, defence, education, human health and social work services 16.39** 0.95 N R, S, T, U Other services 6.27 N Source: European Commission, author's calculations. *Note: Countries, such as Belgium, the Netherlands and Luxembourg, stand out with their ratio-of-export figures, thus driving up the average of ratio of exports in the agriculture sector. **Note: Countries, such as Ireland, the Netherlands and Luxembourg, stand out with their ratio-of-export figures, thus driving up the average of ratio of exports in the professional services sector. As mentioned above, to divide the 10 sectors into tradable and non-tradable sectors, we use a similar approach as De Gregorio et al. (1994). However, in the present paper we put emphasis only on strictly tradable and non-tradable sectors, meaning that the sectors which are not distinctively tradable or non-tradable are exluded from the analysis. A sector is then treated as tradable if its ratio of exports exceeds the 10 percent threshold for at least 75 percent of time using the WIOD data in the 2000-2011 period. The same principle is applied for the definition of a non-tradable sector. A sector is treated as non-tradable if its ratio of exports is under the 10 percent threshold for at least 75 percent of time using the WIOD data in the 2000-2011 period. Applying stricter conditions regarding the division of sector means that NACE Rev. 2 sectors, such as agriculture, forestry and fishing (A), information and communication (J), financial and insurance activities (K), professional, scientific, technical, administration and support services (M and N), are excluded from the analysis. These excluded sectors account for around 20 percent in total value added. Based on this threshold the manufacturing, mining, quarrying and other industries (B, C, 290 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 D and E), wholesale, retail, transportation, storage, accommodation and food services (G, H and I) are treated as tradable sectors, while construction (F), real estate activities (L), public administration, defence, education, human health, social work services (O, P and Q), and other services (R, S, T and U) are treated as non-tradable sectors. 4.2 Sectoral inflation and productivity Based on quarterly data available from the Eurostat website and consideration of the classification of economic activities into a tradable and a non-tradable sector (as defined in Table 1), supported by time-varying sectoral gross value added weights expressed in millions of euros in 2015, growth rate in prices for the tradable and the non-tradable sector are obtained. We use the same principle that was applied to divide economic activities into the tradable and non-tradable sectors to divide sectoral growth rate of value added for both sectors, based on the aggregation done for sectoral inflation. This way we get growth rates for the output on a quarterly frequency basis for a separate sector, i.e. tradable and non-tradable. 4.3 Data entering the model After defining and obtaining the sectoral data, we can provide a full description of the dataset entering the model in Table 2. There are 9 observable variables at a quarterly frequency in the period of 1998Q4-2018Q1, thus providing 78 observations. Tradable sector figures stand out the most and have the highest variability. Intuitively, this means that the tradable sector is more responsive to changes in different phases of business cycles. Additionally, Slovene data in comparison to the euro area data varies more, thus providing a case that small open economies are more vulnerable to macroeconomic imbalances. Table 2: Descriptive statistics (in p.p. deviations from the steady state) Variable description Data transformation Country Minimum Maximum Standard deviation Weighted tradable sector inflation demeaned log-differences SI -2.59 2.21 0.92 Weighted tradable sector inflation demeaned log-differences EA -1.08 1.31 0.39 Weighted tradable sector gross value added demeaned log-differences SI -10.17 3.02 1.64 Weighted tradable sector gross value added demeaned log-differences EA -6.36 1.23 1.07 Weighted non-tradable sector inflation demeaned log-differences SI -1.22 1.84 0.69 Weighted non-tradable sector inflation demeaned log-differences EA -0.76 0.80 0.30 E/B/R Č. LENARČIČ | INFLATION - THE HARROD-BALASSA-SAMUELSON EFFECT ... 291 Variable description Data transformation Country Minimum Maximum Standard deviation Weighted non-tradable sector gross value added demeaned log-differences SI -3.20 5.13 1.51 Weighted non-tradable sector gross value added demeaned log-differences Interest rate given by EA -0.73 1.02 0.41 3-month Euribor log(1+r(-400), demeaned log-differences EA -0.55 0.78 0.42 Source: Eurostat, author's calculations. 5 CALIBRATION OF THE MODEL We set the values of the calibrated parameters accordingly to known empirical facts from the existing literature and characteristics of the modelled economies, in our case Slovenia and the euro area. The discount factor ft is set to 0.99, following Smets and Wouters' (2003) paper. The degree of habit formation parameter h for Slovenia is set to 0.80 (as in Kilponen et al., 2015), while for the euro area it is set to 0.60 (as in Smets and Wouters, 2003), thus making Slovenia's consumption slower to respond and more persistent. The Slovene economy size parameter s is set to 0.01.7 The Frisch elasticity or the inverse of the elasticity of work effort for both economies has a typical parameter value of 2 (Smets and Wouters, 2003; Rabanal, 2009; Rabanal, 2012; Micallef and Cyrus, 2013). The elasticities of substitution between tradable and non-tradable goods for both, domestic (vm) and foreign (vm>), economies, take the value of 0.44, following the values set by Stockman and Tesar (1995). The elasticities of substitution between domestic produced and foreign produced goods for both, domestic (vHF) and foreign (vHF) economies, take the value of 1.5, following Chari, Kehoe, and McGrattan (2002). Furthermore, the shares of important economic variables are calibrated as well. The share of government spending relative to GDP in Slovenia is set to 0.17 and for the euro area it is set to 0.20, while the average share of tradable goods in the consumption basket is set to 0.58 in Slovenia and 0.61 in the euro area. The Calvo wage parameters for both areass and both sectors are set to 0.81, while the price stickiness is set to 0.75, following the values set for Slovenia in Clancy, Jacquinot, and Lozej (2014) and Kilponen et al. (2015). The wage indexation parameters are set to 0.75, according to Rabanal (2012). The quality improvement parameters 9Z and 9Z* for both economies are set to 0.25. The Taylor rule values inflation and output gap response parameters y= 1.5 and y= 0.1 take usual values when modelling the euro area monetary policy close to Four^ans and Vranceanu's (2004) estimation of the euro area parameters. 7 In comparison to the euro area the size of the Slovene economy is even smaller. The reason behind a slightly bigger economy size parameter is that very small numbers of the parameters could represent numerical difficulties for the model. These are shown in a very slow convergence after shocking the model or even in the inability of computing the responses of the shocks. However, 0.01 economy size parameter does not significantly influences the universum of both economies, which would be the case for small open economies. 292 E/B/R ECONOMIC AND BUSINESS REVIEW | VOL. 21 | No. 2 | 2019 | 143-212 200 The calibrated model is able to produce the HBS type of productivity shock. The following figure shows the impulse responses of the main macroeconomic variables to a 1 p.p. domestic tradable sector productivity shock, based on the calibrated model. The productivity shock increases the production of both sectors, tradable and non-tradable. As the quality improvement mechanism takes place, firms are compelled to raise wages since more sophisticated labour force is needed with the productivity picking up. The pick-up in wages increases inflation and consumption in both sectors. What is noteworthy is that inflation in the non-tradable sector increases more than in the tradable sector, thus providing a case for the HBS effect. Figure 1: Impulse responses of the main variables to a 1 p.p. domes tic tradable sec tor productivityshock (deviationsfromsteadystate, inp.p.) 6 ESTIMATION OF THE MODEL AND COMPARISON WITH THE CALIBRATED MODEL With the obtained dataset and the calibration parameters set, the two-country two-sector DSGE model is ready to be estimated. Doing that, we use the Bayesian inference methodology. We set the prior distribution of the estimated parameters, given in Table 3. The prior and the posterior distribution of the estimated parameters and the shocks is presented in Table 3, while the figures with comparisons between the prior and the posterior distribution of the parameters are presented in Appendix A, and in Appendix B the dynamics of the exogenous shocks is presented. The Metropolis-Hastings MCMC algorithm is used with 300,000 steps and two sequential chains with the acceptance rate per chain of around 30%. E/B/R Č. LENARČIČ | INFLATION - THE HARROD-BALASSA-SAMUELSON EFFECT ... 293 We estimate the quality improvement parameters 9Z and 9Z * for both economies. The priors of both parameters were set to 0.25, while the estimates of both parameters took the values of 0.1676 and 0.2127, respectively. The estimated values of both quality improvement parameters are below the calibrated value of the parameter for the domestic economy in Masten (2008). Since Slovenia was catching up the average of the euro area and experienced higher growth and inflation, the estimate of the quality improvement mechanism had to be stronger during this period. With respect to the other estimated parameters, the shock persistence parameters seem to suggest that the productivity persistence parameters show less persistence than the demand shocks entering both the non-tradable and the tradable sector. The parameter Qr of the monetary policy rule is estimated as well and takes the value of 0.6250, suggesting a relatively high persistence of the past interest rate. In comparison to the calibrated model, the Calvo price and wage rigidity parameters (as) are estimated to be higher, meaning that the prices and wages respond slower to exogenous shocks. The values of the Calvo parameters are similar comparing the foreign or domestic economy. Table 3: Prior and posterior distribution of the estimated parameters and shocks Parameter Calibration model values Prior mode Posterior mode 90% HPD interval Prior distribution Prior distribution ez 0.250 0.250 0.1676 0.1061 0.2268 inv. gamma 0.100 0.250 0.250 0.2127 0.1153 0.3393 inv. gamma 0.100 aH 0.750 0.750 0.6259 0.5828 0.6676 beta 0.150 aF 0.750 0.750 0.8955 0.8524 0.9355 beta 0.150 aH* 0.750 0.750 0.8742 0.7620 0.9975 beta 0.150 "P.' 0.750 0.750 0.9200 0.8963 0.9412 beta 0.150 aN 0.750 0.750 0.8519 0.8250 0.8746 beta 0.150 0.750 0.750 0.9550 0.9415 0.9686 beta 0.150 awT 0.810 0.810 0.9010 0.8395 0.9659 beta 0.070 aW,T,' 0.750 0.750 0.8249 0.7279 0.9198 beta 0.070 aW,N 0.810 0.810 0.8889 0.8392 0.9367 beta 0.070 aW,N,' 0.750 0.750 0.7920 0.6909 0.8920 beta 0.070 V TN 0.440 0.500 0.5471 0.1888 0.8864 gamma 0.200 VHF 1.500 1.500 1.1671 0.5602 1.7190 gamma 0.500