Volume 11 Issue 4 Article 3 12-31-2009 Behavioural aspects influencing the performance of Turkish fund managers Omar Masood Bruno S. Sergi Follow this and additional works at: https://www.ebrjournal.net/home Recommended Citation Masood, O., & S. Sergi, B. (2009). Behavioural aspects influencing the performance of Turkish fund managers. Economic and Business Review, 11(4). https://doi.org/10.15458/2335-4216.1272 This Original Article is brought to you for free and open access by Economic and Business Review. It has been accepted for inclusion in Economic and Business Review by an authorized editor of Economic and Business Review. 301 ECONOMIC AND BUSINESS REVIEW | VOL. 11 | No. 4 | 2009 | 301–320 BEHAVIOURAL ASPECTS INFLUENCING THE PERFORMANCE OF TURKISH FUND MANAGERS OMAR MASOOD* BRUNO S. SERGI** ABSTRACT: Using original survey data collected by the authors in 2005 we investigate the determinants of Turkish fund managers’ performance as measured by the number of clients that a fund manager has, the number of investment funds that the manager is responsible for and the size of the manager’s portfolio. All three measures of Turkish fund manager performance systematically vary with fund manager characteristics. Th is is con- sistent with Chevalier and Ellison’s (1999) fi nding for the USA that some managers are better than others. Further, the number of training courses attended by a manager and years of experience (in a particular organisation and/or as a fund manager) are found to positively infl uence all three measures of performance. Th is may suggest that senior man- agers and those with more training are given more responsibility than less experienced and less trained managers. Keywords: Turkish fund managers; Performance and ordered choice models UDC: 330.142:338.121(560) JEL: C25, C51, C52, G21 1. INTRODUCTION Th ere is a large and growing literature that links fund manager performance to the char- acteristics of fund managers. For example, Fama (1980), Lazear and Rosen (1981) and Hol- strom (1982) emphasised agency confl icts and career concerns. Smith and Goudzwaard (1970) and Chevalier and Ellison (1999) looked at the relevance of education. Golec (1996) examined a wide range of characteristics including tenure, MBA qualifi cation, performance, risk-taking and expenses. Other studies focus on the concept of herding borrowed from behavioural fi nance. Scharfstein and Stein (1990) focus on herding due to signal jamming between diff erent types of managers, Banerjee (1992), Bikhchandani * University of East London, Business School, Docklands Campus, University Way, E16 2RD, UK, Email: omar@uel.ac.uk ** Corresponding author: University of Messina, DESMaS “V. Pareto”, Via T. Cannizzaro 278, 98122 Messina, Italy, Email: bsergi@unime.it ECONOMIC AND BUSINESS REVIEW | VOL. 11 | No. 4 | 2009302 et al. (1992) and Welch (1992) on herding due to ineffi cient information transmission and King (1995) on herding due to free riding in information gathering. Trueman (1994) and Zwiebel (1995) suggest that herding among managers who are evaluated relative to their peers might be a result of reputational concerns. Mcnabb and Whitfi eld (2003) state that recent years have revealed extensive innova- tions in compensation systems and, in particular, a variety of attempts to link pay to a measure of performance. Such innovations have oft en been related to broader initiatives to improve the performance of organisations and especially eff orts to increase employee involvement in decision-making (Appelbaum and Batt, 1994; Walsh, 1993). Th e paper’s two major features are the use of survey methodology to obtain primary data and the application of ordered choice models for analysing this data. Th us, this research is unique and stands in contrast to other empirical studies on banking crises that are based principally on published annual data, such as Kaminski and Rhinehart (2000, 1998), Demirguc-Kunt and Detragiache (1998a) Eichengreen and Rose (1998) the IMF (1998) and Gavin and Haussman (1996). Most of the related empirical studies focus on industrialised countries with developed fi nancial systems, especially the USA. However, the link between performance and the characteristics of fund managers has now become a relevant concern in emerging markets due to the recent growth of fund management in these markets. Further, there is ongoing evidence that emerging market fi nancial sys- tems are more vulnerable to political interference, corruption and insider trading than those of developed countries. Conditions like these could conceivably have a signifi cant infl uence on fund manager characteristics and behaviour. Perhaps the lack of literature can be explained by the lack of data. Here we use data col- lected from questionnaire interviews with 110 diff erent fund managers and regulators from the four most signifi cant banks in Turkey. In this paper we take a fi rst step towards studying the link between fund manager per- formance and fund manager characteristics in the context of an emerging market, Tur- key. More specifi cally, we test the statistical signifi cance between three measures of fund manager performance and fund manager characteristics such as education, job experi- ence and the like. Our study is similar in spirit to Chevalier and Ellison (1999) and Golec (1996) but diff ers in one important way. Rather than use aggregated, observable data across some fund industry or sub-industry, our analysis is based on the statistical infor- mation gathered from personal interviews with 110 fund managers in four of Turkey’s largest banks. Our analysis includes characteristics such as age, highest level of educa- tion, number of years of experience and training. Henceforth, in this paper we explore the experience of fund managers based on their relative traits and how their performance and effi ciency is aff ected in terms of invest- ment decision-making and important implications. Th e paper aims to expose all fund managers to a series of questions that may help in analysing the associations between various inputs and their performance. Th e paper is as organised as follows. We continue O. MASOOD, B. S. SERGI | BEHAVIOURAL ASPECTS INFLUENCING THE PERFORMANCE OF ... 303 with a brief literature review in the next section. Section 3 describes the data, the meth- odology and discusses the principal empirical fi ndings. Th e last section concludes with a short summary. 2. LITERATURE REVIEW Avery and Chevalier (1998) state that the probability of termination decreases steeply with a performance when managers have negative excess returns, but it is fairly insen- sitive to diff erences at positive excess return levels. As a result, young managers may have an incentive to avoid unsystematic risk when selecting their portfolios. Modigliani and Pogue (1975), Starks (1987), Grinblatt and Titman (1989) and Admati and Peiderer (1997) consider the incentive eff ects of explicit performance contracts between a mutual fund company (or manager) and mutual fund investors. Starks (1987) and Grinblatt and Titman (1989) show that mutual fund fee schedules which are nonlinear in fund per- formance may distort the fund’s risk incentive. Smith and Goudzwaard (1970) analysed the relevance of education for investment man- agement and found that education does not have a clear eff ect on the performance of graduates in their jobs as fund managers. However, using cross sectional data Chevalier and Ellison (1999) fi nd strong evidence between age and education as explanatory varia- bles for fund performance, measured as risk-adjusted excess returns, even aft er adjusting for behavioural diff erences and selection biases. From pension schemes, mutual funds, banks and other fi nancial institutions portfolio decisions rest with the fund managers. Th ere has been a growing concern that these managers adopt investment strategies that are too similar. One possible explanation of this phenomenon may be found in the in- centives schemes related to performance (Masood & Tunaru, 2006). Another explana- tion is based on herding, a concept from behavioural fi nance. For the latter, existing literature focuses on herding due to either signal jamming between diff erent types of managers (Scharfstein and Stein, 1990), ineffi cient information transmission (Banerjee, 1992, Bikhchandani et al. 1992; Welch, 1992) or free riding in information gathering (King, 1995). Fama (1980) and Lazear and Rosen (1981) show that a manager’s investment decision can be infl uenced by career concerns. Holstrom (1982) confi rms their conclusion but argues that it is only one of a number of other factors that infl uence the investment decision process. Following this line of reasoning, Scharfstein and Stein (1990), Zwiebel (1995), Morris (1997), and Avery and Chevalier (1999) argue that the career concern factor leads to herd behaviour in the fund manager community. Chevalier and Ellison (1997) em- phasise that career issues of mutual fund managers play a signifi cant role in their deci- sions about risk. Golec (1996) fi nds that the portfolio return is aff ected by the manager’s tenure, age, and MBA status. Th e subsequent academic literature (following Modigliani and Pogue [1975]) has noted that a number of ways remain in which investment decisions may be aff ected both by ECONOMIC AND BUSINESS REVIEW | VOL. 11 | No. 4 | 2009304 the explicit compensation schemes of fund companies, and by implicit incentives which derive from a desire to attract new customers. Chevalier and Ellison (1998) argue that a manager being terminated is aff ected by the manager’s actions, past performance, that aspects of the relationship might cause behaviour to vary systematically across manag- ers, and they then examine these predictions by looking at how behaviour actually dif- fers between younger and older managers. Starks (1987) studied the impact of performance incentive fees on portfolio investment management decisions and fi nds that a symmetric compensation contract1 is better than a bonus contract2 and yields better results for the investor. In their study of the relation- ship between managers’ compensation and the relative performance of the funds they manage, Brown, Harlow and Starks (1996) fi nd empirical evidence suggesting that mid- year “loser” managers3 tend to increase the volatility of the funds they manage in the second part of the assessment year. Yet Lemmon, Schallheim and Zender (2000) found that fi nancial contracts play an important role in providing incentives and the perform- ance of a fund. All previous research used information about fund managers from the outside, without specifi c questioning of the managers under analysis. Here, we attempt to break this bar- rier and reveal the inside story. 3. EMPIRICAL MODELLING Th e primary data were collected by a questionnaire (reported in the Appendix) given to fund managers of banks in Turkey to determine three measures of fund manager performance. We had face-to-face interviews with 110 senior Turkish fund managers in the four major commercial banks in Turkey. Th e fi rst dependent variable is the number of clients a fund manager has, denoted Clients (question 7 from the survey) while the second is the number of investment funds that the manager is responsible for, NoFunds, (question 8). Th e third, PortSize (question 23) is the size of the manager’s portfolio.4 Th e respondents were also asked if they thought that any factors other than those addressed by our questionnaire were relevant determinants of their performance. An ordered probit model is applied to the survey data as one of these variables, PortSize, is ordinal. As we assign it with three ranked categories i.e., values 0, 1, 2, we applied or- 1 With a symmetric contract, the manager receives a percentage of the market value of the assets and a bonus if the portfolio return exceeds the return on the designated benchmark or incurs a penalty in the opposite case. 2 With a bonus performance incentive fee the manager receives a percentage of the market value of the assets and a bonus if the portfolio return was higher than the return on some benchmark index; no penalties are imposed 3 A “loser” manager is defi ned as a manager who is underperforming as regards the designated benchmark. 4 Clients, NoFunds and PortSize are arguably accurate and objective measures of fund manager performance because they are terms that are relatively easy to determine and unlikely to be misreported. Th is is partly why investigating such measures of performance may be of particular interest. O. MASOOD, B. S. SERGI | BEHAVIOURAL ASPECTS INFLUENCING THE PERFORMANCE OF ... 305 dered choice estimation techniques to model this ordinal dependent variable. Here lower values indicate a smaller size. We also used ordered logit models which yielded similar results. Th e ordered dependent variable model assumes the following latent variable form (see Greene 2003, pp 736-740):5 i K k ikki uXY += ∑ =1 * β (3.1) where, ikX are the explanatory variables, iu is a stochastic error term and * iY is the unobserved dependent variable that is related to the observed dependent variable, iY , (assuming three categories) as follows: * 2 2 * 1 1 * if 3 if 2 if 1 ii ii ii YY YY YY <= ≤<= ≤= λ λλ λ (3.2) where 1λ and 2λ are unknown parameters (limit points) to be estimated with the coef- fi cients (the β s). Th e probit form of this model assumes that the error, iu , is distributed as a standard normal random variable.6 Th ere are three forms of this model. Th e logit form assumes the error has a logistic distribution, while the Gompit model specifi es the extreme value distribution for the error term. Th e probit form assumes that the error, εi, is distributed as a standard normal random variable, hence we employed this form for our approach. Th e remaining two variables (Clients and NoFunds) are based upon interval/ratio data so that the appropriate estimation method is Ordinary Least Squares (OLS) – we employ White’s heteroscedasticity consistent standard errors to calculate t-ratios. For all three proxies of a fund manager’s performance, we took into consideration 13 explanatory factors as follows. Where a fund manager is male this is denoted Male (ques- tion 1 from the survey),7 where they are married this is denoted Married (question 2)8 and where they are single this is denoted Single (question 2).9 Th e manager’s years of 5 Our interest is primarily confi ned to the general direction of the correlation between the dependent and independent variables. Th erefore, we use the sign of βk to provide guidance on whether the estimated signs of coeffi cients concur with our a priori expectations. Th is is instead of looking at the marginal eff ects which indicate the direction of change of the dependent variable (for each value of the dependent variable) to a change in Xik. 6 Greene (2003) suggests that probit and logit (the error has a logistic distribution) models yield results that are very similar in practice. 7 Th is is a dichotomous variable that takes the value of 1 if the manager is male and zero if they are female. 8 Th is variable takes the value of 1 if the manager is married and zero otherwise. 9 Th is variable takes the value of 1 if the manager is single and zero otherwise. We allowed three categories for marital status being, married, single and divorced, hence we can only include two of them to avoid col- linearity problems. if if if ECONOMIC AND BUSINESS REVIEW | VOL. 11 | No. 4 | 2009306 experience in the organisation, YOE Organisation (question 3) and their years of experi- ence as a fund manager, YOE Manager (question 4). Where the manager has a Master’s degree this is denoted as Master’s degree (question 5a),10 a business degree, Business de- gree (question 5b),11 a degree from a Turkish institution, Turkish degree (question 5c),12 a degree from a UK institution, UK degree (question 5c),13 or a degree from a US institu- tion, US degree (question 5c).14 Th e number of training courses a manager has attended is included under Training (question 6), their age under Age (question 21), and the return on the investment under Return.15 We provide the regression results for each of the three dependent variables (Clients, No- Funds and PortSize) in the next section of this paper. For each of the regressions we re- port a general model (including all the variables specifi ed) and a parsimonious specifi ca- tion (or a small number of parsimonious models) obtained using the general-to-specifi c methodology.16 3.1 Regression Results for the Number of Clients Th e results of the OLS regressions for Clients are reported in Table 1. For the two mod- els that are reported there is no evidence of misspecifi cation, except for non-normally distributed residuals for both models and heteroscedasticity in the general specifi cation, Model 1.17 Since we use t-ratios based upon White’s heteroscedasticity consistent coeffi - cient standard errors, the results are considered robust to heteroscedasticity. Estimation of the model excluding the outlying observation (fund manager 23) removed the evident non-normality and yielded qualitatively similar results to those reported in Table 1 (see Table 1b). Hence, we believe that the non-normality evident in Table 1 does not sub- stantially aff ect our inference and we therefore present the results reported in Table 1 as valid. 10 Th is variable takes the value of 1 if the manager has an MA, MSc or MBA and is zero otherwise. 11 Th is variable takes the value of 1 if the manager’s degree is in the area of business and is zero otherwise. 12 Th is variable takes the value of 1 if the manager’s degree is from Turkey and is zero otherwise. 13 Th is variable takes the value of 1 if the manager’s degree is from the UK and is zero otherwise. 14 Th is variable takes the value of 1 if the manager’s degree is from the USA and is zero otherwise. Th ere were four options for the country from which a degree was obtained, Turkey, the UK, the USA and other, so only three variables could be included (for Turkey, the UK and the USA) to avoid collinearity problems. 15 Th is is an ordinal variable that is measured in percentages. 16 In this method, for the models that are considered valid for inference, we fi rst delete all variables with t-ra- tios below one (or, exceptionally, 0.5 if the t-ratios are very small for a large number of variables) and apply an F-test (or likelihood ratio, LR, test) relative to the general model. If the restrictions cannot be rejected we then delete all variables with t-ratios below 1.5 and then all explanatory factors with t-ratios below 1.96 (applying F/LR tests relative to the general model). If any F/LR test for joint restrictions is rejected, we experiment to fi nd the variable(s) that cause this rejection and retain them in the model. 17 We tested for autocorrelation, non-linear functional form, non-normally distributed residuals, hetero- scedasticity, and parameter non-constancy. O. MASOOD, B. S. SERGI | BEHAVIOURAL ASPECTS INFLUENCING THE PERFORMANCE OF ... 307 Although only two variables in Model 1 are statistically signifi cant at the 5% level (YOE Manager and Training) the removal of insignifi cant variables yields Model 2 in which four variables are signifi cant (YOE Organisation, YOE Manager, UK degree and Training).18 Th is is our favoured model for inference given that it features no insignifi - cance variables, the restrictions placed on Model 1 to obtain Model 2 cannot be rejected and it features a better fi t compared to Model 1 – it explains about 70% of the variation in Clients according to 2 R .19 All the variables retained in Model 2 have positive coeffi cients, except UK degree, which is broadly consistent with the expectations. Th e coeffi cients on the variables indicate that for each extra year of experience in the organisation (or as a manager) the fund manager gains, on average, 0.164 (0.374) clients. Holding a degree from a UK institution reduces the number of clients by, on average, 0.753, while each additional training course attended by a fund manager increases the number of clients by 0.155, on average.20 3.2 Regression Results for the Number of Funds Table 2 reports the OLS regression results for the number of funds variable (NoFunds). Th ere is evidence of autocorrelation, non-normality and heteroscedasticity for both reported models, Model 3 and Model 4. Since we could reorder the data to remove the autocorrelation and the statistics would remain unchanged we do not consider the presence of autocorrelation as adversely aff ecting the results.21 As before, the use of White’s coeffi cient standard errors addresses the problem of heteroscedasticity. Esti- mating the same models with the removal of the outlying observation (fund manager 23) from the sample (see Table 2b) removed the evident non-normality and yielded qualitatively and quantitatively similar results. Hence, non-normality is not regarded as adversely aff ecting our inferences. Th ere is also some evidence of parameter non- constancy at the 5% level across the sample for Model 4 but not Model 3. Th is suggests that, for this model, the coeffi cients of the variables for the fi rst 55 fund managers are diff erent from the last 55 managers. However, in the models estimated without observation 23 neither model has unstable coeffi cients at the 1% level (although Model 4 exhibits non-constant parameters at the 5% level). Th us, to the extent that there are departures from coeffi cient equality they are arguably not serious. Nevertheless, we note that there may be some heterogeneity across the sample and interpret the coef- fi cients as averaged eff ects for the whole sample that provide generalisations for the 18 Variables that are insignifi cant in the general model may become signifi cant through model reduction due to, for example, increased effi ciency and reduced collinearity. Hence our focus on the results of the parsimo- nious model is identifying the variables of statistical signifi cance. 19 Model 2’s regression standard error indicates that the model incorrectly predicts the number of clients that each fund manager has by, on average, 1.5 clients. Th is compares to the standard deviation of the data on Clients of 2.7 clients. 20 To place these numbers in perspective, the number of clients that a fund manager had, in our sample, ranged from 3 to 13 with a mean value of 7.036. 21 Th e use of cross-sectional data here contrasts with the use of time-series data where the order of the obser- vations matters and reordering the data is not appropriate. ECONOMIC AND BUSINESS REVIEW | VOL. 11 | No. 4 | 2009308 population.22 We therefore present our results from Table 2 as valid for inference given their similarity to those reported in Table 2b. Two variables in the general model, Model 3, are statistically signifi cant at the 5% lev- el (YOE Manager and Training).23 Following the general-to-specifi c model reduction method we identifi ed the parsimonious specifi cation, Model 4. Model 4 contains four statistically signifi cant variables (at the 5% level), YOE Organisation, YOE Manager, Business degree and Training. We favour Model 4 over Model 3 because it has a supe- rior fi t and the zero coeffi cient restrictions cannot be rejected.24 In Model 4 we see that three variables have the anticipated positive sign (YOE Organisation, YOE Manager and Training) while Business degree has an unexpected negative sign. However, the impact of holding a business degree is small: it reduces the number of funds by, on average, 1.191 (this is relative to an average number of funds of approximately 11.5). Th us, because such a negative eff ect is diffi cult to rationalise and it is numerically small we interpret this result as suggesting that holding a business degree has little infl uence on the number of funds in a manager’s portfolio. Th e coeffi cients on the other signifi cant variables indicate that for each extra year of experience in the organisation (as a manager) the fund man- ager increases the number of funds by, on average, 0.258 (0.383), while each additional training course attended by a fund manager increases the number of funds by 0.141, on average.25 3.3 Regression Results for Portfolio Size Th e ordered probit regression results for portfolio size (PortSize) are reported in Table 3.26 Model 5 is the general model and Model 6 is the associated parsimonious model obtained aft er applying the general-to-specifi c method. Five variables are signifi cant in the general model while seven are signifi cant in the favoured parsimonious speci- fi cation, namely, Married, YOE Manager, Master’s degree, Business degree, Training, Age and Return. Five variables have a positive coeffi cient suggesting that being mar- ried, having more years of experience as a manager, holding a Master’s degree, attend- ing more training courses and being older will increase the fund manager’s portfolio 22 Coeffi cient inequality does not represent structural breaks in cross-sectional data as it does in time-series data. It simply suggests that sub-groups of the sample may have diff erent coeffi cients from each other. Given that we have split the sample arbitrarily in half and have not ordered the sample in any particular way, it is diffi cult to identify any particular feature that distinguishes each sub-group in a way that could explain the diff erences in coeffi cients. 23 Th is is exactly the same as was found for the general specifi cation for Clients, Model 1. 24 Model 4 explains about 55.3% of the variation in the dependent variable, while the regression standard error indicates that the model incorrectly predicts the number of funds that each fund manager has by, on average, 2. Th is compares to the standard deviation of the data on the Number of Funds of about 2.926. 25 To place these numbers in perspective, the number of funds that a fund manager had ranged from 5 to 19 with a mean value of 11.491. 26 Th is variable has a minimum value of zero, a maximum of two and a mean value of 0.955. Th e standard deviation is 0.596. O. MASOOD, B. S. SERGI | BEHAVIOURAL ASPECTS INFLUENCING THE PERFORMANCE OF ... 309 size.27 Th ese coeffi cients all seem plausible. Indeed, the fi nding that holding a Master’s degree has signifi cant positive infl uences on fund manager performance is consistent with Chevalier and Ellison (1999). Th ey found, for the USA, that fund managers with higher average SAT scores at their undergraduate institutions achieved higher returns. In contrast to Chevalier and Ellison (1999) who fi nd that older fund managers typically secure lower returns, our results indicate that age has a signifi cant positive eff ect on port- folio size.28 Th is may be due to the diff erent measures of performance used (Chevalier and Ellison, 1999, model returns). Older managers may be trusted with more responsi- bility and hold larger portfolios while career concerns may explain why older managers do not achieve as high returns as younger managers who may, for example, feel the need to work harder. Th e coeffi cients on the variables Business degree and Return exhibit an unexpected negative sign in our favoured model. Th e coeffi cient on Return is very small (–0.078), as are the marginal eff ects (see Table 3b), suggesting that this eff ect is numeri- cally minor. Hence, given the diffi culty in rationalising this negative sign we interpret this result as suggesting that return has little impact on portfolio size. Similarly, and as argued for the number of funds model, we interpret the negative sign for the business degree variable as suggesting that holding a business degree has little infl uence on the portfolio size. 3.4 Comparison of Inferences of the Performance Regressions All three measures of performance have very similar determinants (and non-deter- minants). Clients is determined by YOE Organisation, YOE Manager, UK degree and Training while the signifi cant explanatory factors of NoFunds are the same as for Cli- ents, except UK degree is replaced by Business degree. Th e determinants of PortSize are Married, YOE manager, Master’s degree, Business degree, Training, Age and Return. Th e signs of the coeffi cients on the determinants that are common to favoured models with diff erent dependent variables are always the same. Th is may be expected given the simple correlations among these performance proxies: Clients, NoFund and PortSize have high positive correlations (all exceed 0.8).29 Notably the number of training courses attended and years of experience in a particular organisation and/or as a fund manager have a positive and signifi cant eff ect on all of our measures of performance. Th is sug- gests that senior managers and those with more training are given more responsibility than less experienced and less trained managers. Further, our fi nding that Turkish fund managers have systematically diff erent performances is consistent with Chevalier and Ellison’s (1999) fi ndings that, for the USA, some managers are better than others. 27 Marginal eff ects are reported in Table 3b. However, it is diffi cult to comment on these in a way that is of in- terest to us here. We confi ne ourselves to interpreting the coeffi cients as indicating the sign, but not marginal eff ect, of the explanatory factors on the dependent variable. 28 It should be noted that this was a “fragile” fi nding for Chevalier and Ellison (1999) because age was signifi - cant in some of their regressions but not others. 29 Th e correlation between Clients and NoFund (and PortSize) is 0.887 (and 0.886) and between NoFund and PortSize it is 0.833. ECONOMIC AND BUSINESS REVIEW | VOL. 11 | No. 4 | 2009310 4. CONCLUSION Using data from a survey of 110 Turkish fund manages we have estimated models for three diff erent measures of fund manager performance (number of clients, number of funds and portfolio size). Th e number of clients is positively correlated with years of ex- perience in the organisation, years of experience as a manager and the number of train- ing courses attended and negatively associated with holding a UK degree. Th e number of funds is also positively determined by years of experience in the organisation, years of experience as a manager and the number of training courses attended but is negatively related to holding a business degree. Th e determinants of portfolio size are being mar- ried, years of experience as a manager, holding a master’s degree, holding a business degree, the number of training courses attended, the manager’s age and the return on investment. All of these variables’ coeffi cients have a plausible positive sign, except for holding a business degree and return which exhibit unexpected negative signs. However, we note that the eff ects for business degree and return are numerically small and inter- pret them as having little eff ect on our measures of fund manager performance. All three measures of performance are positively determined by the number of training courses attended and years of experience in a particular organisation and/or as a fund manager. Th is suggests that senior managers and those with more training are given more responsibility than less experienced and less trained managers. Further, all three measures of performance systematically vary with fund manager characteristics. Th is is consistent with Chevalier and Ellison’s (1999) fi nding for the USA that some managers are better than others. O. MASOOD, B. S. SERGI | BEHAVIOURAL ASPECTS INFLUENCING THE PERFORMANCE OF ... 311 TABLE 1: Number of Clients Regressions (OLS) Model 1 Model 2 Variables Coeffi cients t-ratio Coeffi cients t-ratio Intercept 0.734 0.476 0.720 1.070 Male 0.468 0.831 Married 0.345 0.930 Single –0.221 –0.514 YOE Organisation 0.206 1.279 0.164 2.106 YOE Manager 0.386 4.498 0.374 4.629 Master’s degree 0.717 1.041 Business degree –0.435 –1.386 Turkish degree –0.047 –0.112 UK degree –0.857 –1.658 –0.753 –2.597 US degree –0.077 –0.157 Training 0.143 2.671 0.155 3.099 Age –0.021 –0.307 Return –0.012 –0.347 Fit (Test) Statistic Probability (Test) Statistic Probability 2 R 0.675 0.695 S 1.555 1.507 SBC 4.184 3.825 F(R2=0) 18.428 0.000 63.139 0.000 F(1→2) NA 0.282 0.978 Misspecifi cation Test Statistic Probability Test Statistic Probability FA 0.334 0.565 1.529 0.219 FFF 0.815 0.369 0.028 0.868 χ2N 34.618 0.000 28.619 0.000 FH 1.979 0.019 2.074 0.053 FCH 0.269 0.996 0.561 0.730 FPF 0.424 0.998 0.521 0.990 Th e dependent variable is Number of Clients, the number of observations in the sample is 110 and White’s heteroscedasticity adjusted t-ratios are reported. 2 R is the coeffi cient of determination adjusted for degrees of freedom, s denotes the unbiased estimate of the regression standard error, F(R2=0) gives the F-test for the signifi cance of the overall explanatory power of the model and F(1→2) is an F-test for the deletion of variables from Model 1 to obtain the parsimonious specifi cation. Th e reported misspecifi cation tests (Misspecifi cation) are F-versions of Breusch-Godfrey’s test for fi rst-order autocorrela- tion (FA), Ramsey’s Rest test for non-linear functional-form (FFF) and Chow’s fi rst and second tests for parameter non-constancy (FCH and FPF, respectively). Th e chi-squared distributed Jarque-Bera test for non-normally distributed residuals (χ2N) is also reported. Th e Chow and Predictive Failure tests split the sample between observations 55 and 56. All statistics are produced using EViews 5.0. ECONOMIC AND BUSINESS REVIEW | VOL. 11 | No. 4 | 2009312 TABLE 1b: Number of Clients Regressions (OLS) Model 1 Model 2 Variables Coeffi cients t-ratio Coeffi cients t-ratio Intercept 0.013 0.009 0.376 0.633 Male 0.412 0.745 Married 0.349 0.980 Single –0.146 –0.353 YOE Organisation 0.198 1.249 0.188 2.465 YOE Manager 0.401 4.613 0.377 4.543 Master’s degree 0.882 1.318 Business degree –0.508 –1.616 Turkish degree –0.018 –0.042 UK degree –0.838 –1.618 –0.760 –2.604 US degree –0.002 –0.004 Training 0.143 2.718 0.156 3.163 Age 0.001 0.002 Return –0.023 –0.675 Fit (Test) Statistic Probability (Test) Statistic Probability 2 R 0.732 0.745 S 1.417 1.381 SBC 4.000 3.652 F(R2=0) 23.634 0.000 79.823 0.000 F(1→2) NA 0.425 0.919 Misspecifi cation Test Statistic Probability Test Statistic Probability FA 0.721 0.398 2.051 0.155 FFF 0.065 0.799 0.599 0.441 χ2N 3.270 0.195 2.407 0.300 FH 1.253 0.239 0.650 0.713 FCH 0.348 0.985 0.507 0.771 FPF 0.606 0.958 0.677 0.919 Th e dependent variable is Number of Clients, the number of observations in the sample is 109 (the outlying 23rd observation has been omitted) and White’s heteroscedasticity adjusted t-ratios are reported. 2 R is the coeffi cient of determination adjusted for degrees of freedom, s denotes the unbiased estimate of the regression standard error, F(R2=0) gives the F-test for the signifi cance of the overall explanatory power of the model and F(1→2) is an F-test for the deletion of variables from Model 1 to obtain the parsimoni- ous specifi cation. Th e reported misspecifi cation tests (Misspecifi cation) are F-versions of Breusch-Godfrey’s test for fi rst-order autocorrelation (FA), Ramsey’s Rest test for non- linear functional-form (FFF) and Chow’s fi rst and second tests for parameter non-con- stancy (FCH and FPF, respectively). Th e chi-squared distributed Jarque-Bera test for non- normally distributed residuals (χ2N) is also reported. Th e Chow and Predictive Failure tests split the sample between observations 55 and 56. All statistics are produced using EViews 5.0. O. MASOOD, B. S. SERGI | BEHAVIOURAL ASPECTS INFLUENCING THE PERFORMANCE OF ... 313 TABLE 2: Number of Funds Regressions (OLS) Model 3 Model 4 Variables Coeffi cients t-ratio Coeffi cients t-ratio Intercept 5.748 2.526 4.322 3.818 Male 0.011 0.021 Married –0.318 –0.624 Single 0.121 0.220 YOE Organisation 0.383 1.863 0.258 1.938 YOE Manager 0.330 2.294 0.383 3.003 Master’s degree –0.569 –0.610 Business degree –1.356 –2.302 –1.191 –2.051 Turkish degree 0.002 0.004 UK degree –0.831 –1.116 US degree –0.152 –0.209 Training 0.138 2.195 0.141 2.365 Age –0.057 –0.769 Return 0.015 0.315 Fit (Test) Statistic Probability (Test) Statistic Probability 2 R 0.533 0.553 S 2.000 1.956 SBC 4.686 4.347 F(R2=0) 10.565 0.000 34.758 0.000 F(3→4) NA 0.489 0.879 Misspecifi cation Test Statistic Probability Test Statistic Probability FA 5.775 0.018 5.818 0.018 FFF 0.001 0.971 0.098 0.755 χ2N 18.232 0.000 19.333 0.000 FH 2.171 0.009 3.926 0.001 FCH 1.676 0.077 3.414 0.007 FPF 0.813 0.765 0.867 0.698 Th e dependent variable is Number of Funds, the number of observations in the sample is 110 and White’s heteroscedasticity adjusted t-ratios are reported. 2 R is the coeffi cient of determination adjusted for degrees of freedom, s denotes the unbiased estimate of the regression standard error, F(R2=0) gives the F-test for the signifi cance of the overall ex- planatory power of the model and F(3→4) is an F-test for the deletion of variables from Model 3 to obtain Model 4. Th e reported misspecifi cation tests (Misspecifi cation) are F-versions of Breusch-Godfrey’s test for fi rst-order autocorrelation (FA), Ramsey’s Rest test for non-linear functional-form (FFF) and Chow’s fi rst and second tests for parameter non-constancy (FCH and FPF, respectively). Th e chi-squared distributed Jarque-Bera test for non-normally distributed residuals (χ2N) is also reported. Th e Chow and Predictive Failure tests split the sample between observations 55 and 56. All statistics are produced using EViews 5.0. ECONOMIC AND BUSINESS REVIEW | VOL. 11 | No. 4 | 2009314 TABLE 2b: Number of Funds Regressions (OLS) Model 3 Model 4 Variables Coeffi cients t-ratio Coeffi cients t-ratio Intercept 4.868 2.201 3.808 3.624 Male –0.057 –0.111 Married –0.313 –0.638 Single 0.212 0.391 YOE Organisation 0.374 1.810 0.298 2.279 YOE Manager 0.348 2.388 0.395 3.060 Master’s degree –0.368 –0.394 Business degree –1.445 –2.650 –1.370 –2.581 Turkish degree 0.038 0.065 UK degree –0.807 –1.087 US degree –0.060 –0.083 Training 0.137 2.278 0.140 2.413 Age –0.030 –0.428 Return 0.002 0.042 Fit (Test) Statistic Probability (Test) Statistic Probability 2 R 0.602 0.619 S 1.842 1.802 SBC 4.525 4.184 F(R2=0) 13.574 0.000 44.930 0.000 F(3→4) NA 0.501 0.871 Misspecifi cation Test Statistic Probability Test Statistic Probability FA 6.164 0.015 5.393 0.022 FFF 0.304 0.583 0.308 0.580 χ2N 0.698 0.705 0.873 0.646 FH 1.595 0.078 2.437 0.024 FCH 1.642 0.086 2.601 0.030 FPF 1.030 0.465 1.102 0.478 Th e dependent variable is Number of Funds, the number of observations in the sample is 109 (the outlying 23rd observation has been omitted) and White’s heteroscedasticity adjusted t-ratios are reported. 2 R is the coeffi cient of determination adjusted for degrees of freedom, s denotes the unbiased estimate of the regression standard error, F(R2=0) gives the F-test for the signifi cance of the overall explanatory power of the model and F(3→4) is an F-test for the deletion of variables from Model 3 to obtain Model 4. Th e re- ported misspecifi cation tests (Misspecifi cation) are F-versions of Breusch-Godfrey’s test for fi rst-order autocorrelation (FA), Ramsey’s Rest test for non-linear functional-form (FFF) and Chow’s fi rst and second tests for parameter non-constancy (FCH and FPF, re- spectively). Th e chi-squared distributed Jarque-Bera test for non-normally distributed residuals (χ2N) is also reported. Th e Chow and Predictive Failure tests split the sample between observations 55 and 56. All statistics are produced using EViews 5.0. O. MASOOD, B. S. SERGI | BEHAVIOURAL ASPECTS INFLUENCING THE PERFORMANCE OF ... 315 TABLE 3: Portfolio Size (Ordered Probit Regressions) Model 5 Model 6 Variables Coeffi cients t-ratio Coeffi cients t-ratio Male 0.912 1.625 Married 1.609 2.847 1.416 3.338 Single –0.122 –0.275 YOE Organisation 0.030 0.203 YOE Manager 0.391 3.245 0.399 3.955 Master’s degree 1.983 2.655 1.613 2.188 Business degree –1.313 –2.461 –1.229 –2.855 Turkish degree 0.325 0.560 UK degree –0.084 –0.137 USA degree 0.554 0.791 Training 0.101 1.806 0.122 2.135 Age 0.077 1.196 0.079 1.992 Return –0.084 –2.245 –0.078 –2.103 Limit Points Coeffi cients t-ratio Coeffi cients t-ratio λ1 5.633 3.309 4.420 2.931 λ2 10.682 5.014 9.201 4.780 Fit (Test) Statistic Probability (Test) Statistic Probability Pseudo R2 0.578 0.564 SBC 1.394 1.164 LR statistic 113.633 0.000 110.778 0.000 LR(5→6) NA 2.855 0.827 Th e dependent variable is portfolio size which takes on values 1, 2 and 3, so there are two limit points, λi, i=1,2 – the intercept is not separately identifi ed from the limit points. Th e number of observations in the sample is 110. Th e z-statistics (in parentheses) are based upon Huber-White standard errors which are robust to certain misspecifi cations of the underlying distribution of the dependant variable (see E-Views 5.0 User Guide p. 651). Th e reported fi t measures are the Pseudo R2 [R2 = 1 – (lnL / lnL0), where lnL and lnL0 are the maximised values of the model’s likelihood function including all variables and only incorporating an intercept, respectively – see Greene, 2003, pp. 683-684] and Schwarz’s information criterion, SBC. Also included are chi-squared tests for the model’s explanatory power, LR Statistic, and the deletion of variables from Model 5 to obtain the restricted Model 6, LR(5→6) – probability values are given in parentheses. Th e probit model assumes that the cumulative distribution function of the error term is standard normal: Φ(λj – ΣkβkXik) = (2π)–½exp[–½(λj – ΣkβkXik)2], j=1,2. All Probit regressions were estimated using E-Views 5.0. ECONOMIC AND BUSINESS REVIEW | VOL. 11 | No. 4 | 2009316 TABLE 3b: Marginal Eff ects of Model 6 (Portfolio Size) Model 6 Pr(Y=1) Pr(Y=2) Pr(Y=3) Married –0.058 0.041 0.017 YOE Manager –0.016 0.012 0.005 Master’s degree –0.066 0.047 0.019 Business degree 0.050 –0.035 –0.015 Training –0.005 0.004 0.001 Age –0.003 0.002 0.001 Return 0.003 –0.002 –0.001 Marginal eff ects are reported for each value of the dependent variable, denoted Y, for all variables included in Model 6 using the ordered probit specifi cation. Th ey are calculated using the means of the explanatory variables, Xk. RECEIVED: MARCH 2009 REFERENCES Asmussen, S., and Edwards, D. (1983), “Collapsibility and Response Variables in Contingency Tables,” Biometrika 70, 566–578. Avery, C., and Chevalier, J. (1999), “Herding over the Career,” Economics Letters 63, 327–333. Bouchet, M. H., Clark, E. and Groslambert, B., (2003), Country Risk Assessment, London: John Wiley & Sons. Brown, K.C., Harlow, W.V., and Starks, L.T. (1996), “Of Tournaments and Temptations: An Analysis of Mana- gerial Incentives in the Mutual Fund Industry,” Journal of Finance 51, 85–110. Banerjee, A. (1992), “A Simple Model of Herd Behavior,” Quarterly Journal of Economics 107, 797–817. Bikhchandani, S., Hirshleifer, D., and Welch, I., (1992), “A Th eory of Fads, Fashion, Custom, and Cultural Change as Informational Cascades,” Journal of Political Economy 100, 992–1026. Brown, K.C., Goetzmann, W.N., and Park, J. (2001), “Careers and Survival: Competition and Risk in the Hedge Fund and CTA Industry,” Journal of Finance 56, 1869–1886. Carpenter, J.N. (2000), “Does Option Compensation Increase Risk Appetite?” Journal of Finance 55, 2311– 2331. Chevalier, J., and Ellison, G. (1997), “Risk Taking By Mutual Funds as a Response to Incentives,” Journal of Political Economy 114, 389–432. Chevalier, J., and Ellison, G. (1999), “Are Some Mutual Fund Managers Better Th an Others? Cross-Sectional Patterns in Behavior and Performance,” Journal of Finance, 54, 875–899. Christoff ersen, S.E.K. (2001), “Why Do Money Fund Managers Voluntarily Wave Th eir Fees?” Journal of Finance 56, 1117–1140. O. MASOOD, B. S. SERGI | BEHAVIOURAL ASPECTS INFLUENCING THE PERFORMANCE OF ... 317 Edwards, D. (1995), Introduction to Graphical Modelling, Springer: New York. Elton, E.J., Gruber, M.J., and Blake, C.R. (2003), “Incentive Fees and Mutual Funds,” Journal of Finance 58, 779–804. E-Views 5.0 User Guide (2004), Quantitative Micro Soft ware. Fama, E.F. (1980), “Agency Problems and the Th eory of the Firm,” Journal of Political Economy 88, 288–307. Frydenberg, M. (1990), “Marginalisation and Collapsibility in Graphical Interaction Models,” Annals of Sta- tistics 18, 790–805. Goodman, L.A. (1973), “Th e Analysis of Multidimensional Contingency Tables When Some Variables are Posterior to Others: A Modifi ed Path Analysis Approach,” Biometrika 60, 179–192. Golec, J. H. (1996), “Th e Eff ects of Mutual Funds Managers’ Characteristics on Th eir Portfolio Performance, Risk and Fees,” Financial Services Review 5, 133–148. Greene W H (2003) Econometric Analysis, 5th edition Englewood Cliff s: Prentice-Hall. Grinblatt, M. and Titman, S. (1992), “Th e Persistence of Mutual Fund Performance,” Journal of Finance 47, 1977–1984. Hand, D.J., McConway, K.J., and Stanghellini, E. (1996), “Graphical Models of Applicants for Credit,” IMA Journal of Mathematics Applied in Business and Industry 8, 143–155. Holstrom, B. (1982a), Managerial Incentive Problems-A Dynamic Perspective -Essays in Economics and Management in Honour of Lars Wahlbeck, Helsinki: Swedish School of Economics. Holmstrom, B. (1982b), “Moral Hazard in Teams,” Bell Journal of Economics 13, 324–340. Kreiner, S. (1987), “Analysis of Multidimensional Contingency Tables by Exact Conditional Tests: Techniques and Strategies,” Scandinavian Journal of Statistics 14, 97–112. Lauritzen, S.L. (1996), Graphical Models. Oxford: Oxford University Press. Modigliani, F., and Pogue, G.A. (1975), “Alternative Investment Performance Fee Arrangements and Implica- tions for SEC Regulatory Policy,” Bell Journal of Economics 6, 127–160. Lemmon, M.L., Schallheim, J.S., and Zender, J.F. (2000), “Do Incentives Matter? Managerial Contracts for Dual-Purpose Funds,” Journal of Political Economy 108, 273–299. Mookherjee, D. (1984), “Optimal Incentive Schemes with Many Agents,” Review of Economic Studies 51, 433–446. Morris, S. (1997), An Instrumental Th eory of Political Correctness, University of Pennsylvania. Scharfstein, D.S., and Stein, J.C. (1990), “Herd Behavior and Investment,” American Economic Review 80, 465–489. Smith, K.V., and Goudzwaard, M.B. (1970), “Survey of Investment Management: Teaching versus Practice,” Journal of Finance 25, 329–347. Stanghellini, E., McConway, K.J., and Hand, D.J. (1999), “A Discrete Variable Chain Graph for Applicants for Credit,” Applied Statistics 48, 239–251. ECONOMIC AND BUSINESS REVIEW | VOL. 11 | No. 4 | 2009318 Tannenbaum. R. (1950), “Managerial Decision-Making,” Journal of Business 23, 22–39. Tunaru, R., and Watson, A. (2000), “An Exploratory Statistical Analysis of Financial Measures,” IMA Journal of Mathematics Applied in Business and Industry 11, 101–110. Tunaru, R. (2001), “Models of Association versus Causal Models for Contingency Tables,” Th e Statistician 50, 257–269. Welch, I., (1992), “Sequential Sales, Learning, and Cascades,” Journal of Finance 47, 695–732. Zwiebel, J. (1995), “Corporate Conservatism and Relative Compensation,” Journal of Political Economy 103, 1–25. O. MASOOD, B. S. SERGI | BEHAVIOURAL ASPECTS INFLUENCING THE PERFORMANCE OF ... 319 Appendix: Questionnaire for Turkish Investment (Fund) Managers Name: Position held: Q1. Sex: Male/ Female Q2. Marital Status: Single Married Divorced Q3. How many years of experience do you have within the organisation? Q4. How many years of experience do you have as a fund manager? Q5. Specify your educational qualifi cation in terms of: (a) Level of study: BA MA/MBA PhD Other (please specify) (b) Subject of study: A business subject A non-business subject (c) Country of study: UK USA Turkey Other (please specify) Q6. How many training courses have you attended as a fund manager? Q7. How many clients do you have? Q8. How many investment funds are you responsible for? Q9. To what extent do you feel performance pressure as a fund manager? To a: Very high degree High degree Moderate degree Low degree Very low degree Q10. Are you satisfi ed with the incentives provided to fund managers? Very satisfi ed, satisfi ed, neither satisfi ed or unsatisfi ed, unsatisfi ed, very unsatisfi ed Q11. What is your level of satisfaction with the quality of risk management techniques applied? Very satisfi ed, satisfi ed, neither satisfi ed or unsatisfi ed, unsatisfi ed, very unsatisfi ed Q12. How accurate are the data available to you on a scale from zero to four (inclusive), with zero being highly inaccurate and four being highly accurate? 0 1 2 3 4 ECONOMIC AND BUSINESS REVIEW | VOL. 11 | No. 4 | 2009320 Q13. How much do you rely on data to make your decisions? Totally, To a large extent, To a moderate extent, To a limited extent Not at all Q14. To what extent are you concerned with the volatility of today’s fi nancial markets? Totally concerned, Highly concerned, Moderately concerned, A little concerned, Unconcerned Q15. To what extent are your investment decisions based on your personal judgement? Totally, To a large extent, To a moderate extent, To a limited extent Not at all Q16. How oft en do you use mathematical projections and statistical models for invest- ment decisions? Very oft en, oft en, sometimes, seldom, never Q17. How effi cient satisfi ed are you with these projections and models: Very satisfi ed, satisfi ed, neither satisfi ed or unsatisfi ed, unsatisfi ed, very unsatisfi ed Q18. What importance do you give to fi nancial statements of diff erent companies when making investment decisions? Very important important, neither important nor unimportant, unimportant , very unimportant Q19. What importance do you give to non-fi nancial data when making investment deci- sions? Very important, important, neither important nor unimportant, unimportant, very unimportant Q20. How much do you rely on credit rating agencies? Totally, a lot, moderately, a little, not at all Q21 What is your age? Q22. What is the amount (band) of performance-related pay? Q23. What is the size of your portfolio? Q24. What is the return on the investment (capital employed)?