ORIGINAL SCIENTIFIC PAPER RECEIVED: JANUARY 2015 REVISED: MAY 2015 ACCEPTED: JULY 2015 DOI: 10.1515/ngoe-2015-0018 UDK: 336.71:519.8 JEL: C58 NG NASE GOSPODARSTVO OUR ECONOMY Vol. . 61 No. 5 2015 pp . 12-22 Model for Determining the Stability of Retail Deposits with Higher Outflow Rates Aleksandra Murks Bašič Zavarovalnica Triglav d.d., Ljubljana, Slovenia sandra666.basic@gmail.com Abstract Retail deposits are treated as one of the cheapest and most stable funding sources for banks, especially for those with high volumes of retail deposits. A bank defines three main categories of retail deposits that are subject to different outflow rates for the purpose of liquidity coverage requirements in reporting and compliance. The outflow rates for the first two main groups are 5% and 10% respectively, but for the third main group the bank calculates its own outflow rates. We analyzed the latter in this paper. Each bank should assign retail deposits to one of the three categories based on the number and type of predetermined risk factors. Risk factors are divided into two groups according to the degree of risk. The paper first describes the legislative framework, followed by the method of calculating higher outflow rates for retail deposits according to the historical movements and the expected volatility assessment in the situation of stress conditions. At the end of the paper, we briefly provide the future treatment of retail deposits with higher outflow rates. Keywords: retail deposits, stability, transactional account, LCR, deposit guarantee scheme, depositors, higher outflow rates, Basel III 1 Introduction On 6 December 2013, the European Banking Authority (EBA) published "Guidelines on retail deposits subject to higher outflows for the purposes of liquidity reporting under the Regulation (EU) no. 575/2013 on prudential requirements for credit institutions and investment firms and amending Regulation (EU) No 648/2012 (CRR 2013)." For the purpose of the liquidity coverage ratio (LCR) of CRR Part 6, the guidelines (EBA/GL/2013/01) defined detailed rules on retail deposits with higher outflows in accordance with the third paragraph of Article 421 of CRR (Guidelines). For the purposes of the LCR calculation, institutions report the amount of retail deposits covered by a deposit guarantee scheme (DGS), which are either part of an established relationship making withdrawal highly unlikely or held in a transactional account. In principle, these retail deposits are considered the most stable and will be subject to the lowest outflow rates of 5%. All other retail deposits not included in the previous items and that do not fulfill the conditions of retail deposits with a higher outflow rate are subject to the 10% outflow rates. A retail deposit should be considered part of an established relationship when the depositor meets at least one of the following criteria: (a) has an active contractual relationship with the institution of a minimum duration of 24 months, 12 Aleksandra Murks Bašič: Model for Determining the Stability of Retail Deposits with Higher Outflow Rates (b) has a borrowing relationship with the institution for mortgage loans or other long-term loans, or (c) has a minimum of two active products, other than loans, with the institution. A retail deposit should be considered as being held in a transactional account when salaries, income, or transactions are regularly credited and debited against that account. Guidelines cover the methodology for the identification of retail deposits that are subject to higher outflows. Retail deposits are grouped in three main categories. Each category is formed based on the number and riskiness of the risk factors they meet. The guidelines do not prescribe the outflow rates for the three categories, but stipulate that institutions are to report retail deposit amounts allocated to each of the three categories together with their own estimates of expected outflows under stress conditions. This paper demonstrates how a commercial bank can develop a simple internal model for calculating the stability of the retail deposits with higher outflow rates that are subject to LCR reporting. The bank's own econometric model is based on the historical data and expected stability/volatility for the specific retail deposits. Thus, we have developed our model based on two main hypotheses: how the historical data impact the stability of considered retail deposits and what level of stability we can expect in the subsequent 22 working days. This paper is divided into three main parts. The first part considers the regulatory issues that are the basis for the empirical part, which is the second part, including the interpretation of the results of the internal model. The last part describes the future challenges in the field of measuring the stability of the retail deposits with higher outflow rates according to the Delegated Act. 2 Results of the EBA Discussion Paper and the Basis for the Guidelines In February 2013, the EBA published its discussion paper (DP) on retail deposits subject to higher outflows for the purposes of LCR reporting. The scope of the DP covered all retail deposits as per Article 409 of CRR, but emphasis was put on the process of identifying retail deposits that carry a higher outflow risk. The result of the DP was presented in the form of 18 questions that should be answered by any potential respondent. EBA received 25 responses, of which 21 were published on the EBA website (EBA, 2013b). Before presenting a summary of the key points arising from the consultation, here are the most significant, from a technical point of view, questions (EBA, 2013a): • Availability of data: the introduction of liquidity requirements to some extent is based on actual behavior observed during a stressed situation that may be considered realistic. Thus, the institutions should use local historical data and a forward-looking approach (the latter within the circumstances of a combined idiosyncratic and market-wide stress scenarios). • Factors affecting the stability of retail deposits: the value of deposits, products that are rate-driven or have preferential conditions, maturing fixed term or notice period deposits, high risk distribution channels including Internet-only access and brokered deposits, the currency and location of deposits, non-resident deposits, depositors who are sophisticated or high net worth individuals, product-linked deposits and any other characteristics that might indicate a retail deposit with a higher outflow rate. • Factors divided into two groups: high risk and very high risk. • Mix of characteristics according to their riskiness and three categories: category 1 with an outflow rate of 15%, category 2 with an outflow rate of 20%, and category 3 with an outflow rate of 25%. A summary of the key points arising from the DP is provided below (EBA, 2013b): • Most respondents considered that the proposed risk factors concerned the characteristics of the deposits rather those of the depositors. • Some respondents proposed additional risk factors; these were basically the financial position of the bank, the financial stability of the country, the customer as a single product user, and concentration risk. • Some respondents considered that some risk factors may be closely correlated; such a situation was mentioned regarding sophisticated and rate-driven deposits, which are by default deemed to be associated. • Some respondents did not agree that deposits with virtual (Internet only) banks must be subjected to a more stringent regime than those linked to Internet accounts with traditional banks, which have a network of physical branches. • Many respondents considered that the non-resident condition itself could be potentially discriminative. • Some respondents rejected the idea that rate-driven deposits must be subjected to higher outflow rates. • Some respondents saw the imposition of higher outflows on maturing term deposits or fixed-notice accounts as counterintuitive and should be seen as providing stable funding. • Some respondents argued that the imposition of specific higher outflow rate places EU banks at a disadvantage compared to non-EU banks. • Many respondents stated that the DGS amount in each relevant jurisdiction should be the threshold for determining whether a retail deposit should be subjected to a higher outflow rate. For a split between high and very 13 NAŠE GOSPODARSTVO / OUR ECONOMY Vol. 61 No. 5 / October 2015 high-risk retail deposits, a threshold of 1 million € was proposed instead of 500,000 €. • Some respondents raised concerns about the technical difficulties inherent in the data collection process. • Some respondents stated that significant investment in IT infrastructure would be necessary to implement the identified process and new staff would be needed to analyze databases. In August 2013, the EBA published its Consultation Paper (CP) considering the draft guidelines as a further step of the previously published DP. As a change to the approach proposed in the DP, the draft guidelines will not prescribe the associated outflow rates for the three categories. Instead, the draft stipulates that credit institutions shall report retail deposit amounts allocated to each of the three categories together with their own estimates of expected outflows under stress conditions. 3 Data Preparation, Collection, and Analysis The bank is in the process of data preparation, collection, and analysis, which includes detailed examination of the criteria and methodology for the definition of retail deposits with higher outflow rate, fully taking into account the guidelines. The latter continue to include a three-tiered "bucket" approach to allocate retail deposits subject to higher outflows for the purposes of liquidity reporting. At the aggregated level of reporting liquidity outflows within CRR, three main groups of retail deposits are considered (CRR, 2013 and ZBS, 2013): • retail deposits covered by a DGS and which are either part of an established relationship making withdrawal highly unlikely or held in a transactional account, including accounts to which salaries are regularly credited. Such retail deposits are subject to an outflow rate of 5%. • retail deposits that do not meet criteria from the previous point or are not identified as deposits with a higher outflow rate. These are subject to an outflow rate of 10%. • retail deposits based on the number and risk level of the risk factors are grouped into three categories set out in the guidelines. Such retail deposits have three different outflow rates depending on the credit institution's own estimates of expected outflows under stress conditions. Our paper emphasizes the retail deposits that are subject to a higher outflow rates. First, the criteria for their definition need to be clarified. Second, the methodology for the identification of three categories needs to be determined. Third, the time series that represent the core for the econometric modeling needs to be identified in order to calculate the outflow rates for all three categories. Above all, we need to emphasize what represents the starting point for the development of the model presented in this paper. The respective bank has developed a number of simple econometric models that have been inspired primarily by the following papers: Stesevic (2008), Perusko and Zenzerovic (2011), Takemura and Kozu (2009), OENB (2008), and von Feilitzen (2011). The latest has also been a fundamental issue within the development process of the internal model, which has been described in the article. 3.1 Criteria for Definition of Retail Deposits with Higher Outflow Rates Criteria are presented as a list of factors that form the basis for the bank's calculations of retail deposits with higher outflow rates. High value deposits are particularly sensitive in a combined idiosyncratic and market-wide stress scenario and may therefore be subject to higher and faster outflows. In addition, high value deposits contribute to the concentration of the deposit base, and over-reliance on such deposits can compromise the stability of the deposit base. Depositors influenced by higher yield, preferential conditions, or negotiated rates can be more responsive to competitors and other attractive offers. Consequently, these deposits may prove to be less stable. Maturing fixed-term retail deposits or deposits with fixed-notice periods may be less stable during stress periods because they are more likely to be funds that depositors do not need for day-to-day transactions. The clients of Internet-only banks have access to their bank only through the Internet, and the absence of direct contact with staff can have a negative effect on confidence and stability under stress conditions. Moreover, exchange rate volatility can also affect the stability of both foreign and local currency denominated deposits. The ability of retail non-resident depositors to transfer deposits may impact the stability of such deposits. The following summarized list of risk factors or harmonized criteria will be used for our empirical analysis (EBA, 2013c): 1.) High value deposits: over 100,000 € to 500,000 € (C1) 2.) Very high value deposits: over 500,000 € (C2) 3.) Deposits that are rate-driven or have preferential conditions: exceeding the average rate for similar retail products offered by peers, return is derived from the return on a market index or set of indices, or return is derived from any market variable other than a floating interest rate (C3) 14 Aleksandra Murks Bašič: Model for Determining the Stability of Retail Deposits with Higher Outflow Rates 4.) Maturing fixed-term or notice period deposits: fixed-term deposit with an expiry date maturing within the 30-day period or deposit with fixed notice period shorter than 30 days (C4) 5.) High-risk distribution channels: including Internet-only access and brokered deposits (C5) 6.) The currency of deposits: deposits denominated in foreign currencies (C6) 7.) Non-resident deposits: the statistical or tax definition at the depositor level (C7) 8.) Product-linked deposits: banking product to which the deposit is linked terminates during the 30-day period and the client can then disburse the savings (C8) 3.2 Method for Determining the Retail Deposits with Higher Outflow Rates Risk factors or criteria described in the previous chapter are classified into two groups according to the level of risk, which impacts the stability of retail deposits (EBA, 2013c): a) Group 1 (high risk - VR), which includes the following risk factors: • VR_C1: deposits over 100,000 € to 500,000 € • VR_C3: interest-rate sensitivity • VR_C5: Internet bank • VR_C6: deposits in foreign currency • VR_C8: product-linked deposits b) Group 2 (very high risk - ZR), which includes following risk factors: • ZR_C2: deposits over 500,000 € • ZR_C4: fixed-term deposits with residual maturity up to 30 days or notice period deposits shorter than 30 days • ZR_C7: non-resident deposits Using the scoring system, the bank assessed the retail deposits from points (a) and (b) and assigned retail deposits to one of the three following tiered buckets defined based on the number of risk factors attributed to the underlying deposit (Nova KBM d.d., 2014): • CATEGORY 1 (KAT01): retail deposits with two factors from Group 1 or written for the econometric modeling use a = VR_C1 + VR_C3 + VR_C5 + VR_C6 +VR_C8 = 2 and b = ZR_C2 + ZR_C4 + ZR_C7 = 0 • CATEGORY 2 (KAT02): retail deposits with three factors from Group 1 or with one factor from Group 1 and one factor from Group 2 a = 3 and b = 0 or a = 1 and b = 1 • CATEGORY 3 (KAT03): retail deposits with two factors from Group 2 or two factors from Group 1 and one factor from Group 2 or with any other mix of factors a = 2 and b = 0 or a = 2 and b = 1 or a > 2 and b > 1 3.3 Time Series Time series are data that are collected over a certain period of time (e.g., unemployment rate, salaries, rents, inflation, Euribor). These data can be collected daily, weekly, monthly, quarterly, or annually. The main purpose of the time series is to observe the evolution of economic phenomena over time and to establish the general findings of this movement. The latter enables the prediction of further development and the acceptance of appropriate measures. Included are daily data from 1 September 2013 to 31 March 2014. A longer time series, from a historical point of view, was not possible due to the complexity of defining individual categories and transactional systems. 4 Empirical Analysis 4.1 Starting Point for Model Estimation and Implementation of Regression Analysis We used ordinary least squares (OLS), which is considered the most commonly used method of determining the regression coefficients. OLS is often called the "queen" of the assessment methods of regression coefficients and is considered the "best linear unbiased estimator" (BLUE). Regression analysis must meet certain assumptions that the estimator of regression coefficients will be BLUE. In our regression analysis, we considered the assumptions unconditionally (Greene 2003, Gujarati 1988, Pfajfar 1998 and Schwert 2011). In our analysis, we used two types of data: quantitative and qualitative data. Quantitative data were determined by time series of daily data from September 2013 to March 2014 (DPRS 2013 and 2014, EURIBOR 2013 and 2014, SURS 2013 and 2014). Qualitative data were included through a dummy variable, with which we indicated the presence or absence of certain properties. Using regression analysis, 15 NAŠE GOSPODARSTVO / OUR ECONOMY Vol. 61 No. 5 / October 2015 we studied the movement of retail deposits separately for KAT01, KAT02, and KAT03 and evaluated models that incorporate the selected parameters. Based on the results, we conducted a 30-day forecast (or 22 working days) and calculated the stability of retail deposits for each category in the form of higher outflow rates. Regression analysis was performed using the Econometric Views 7 (EViews 7, 2010a and 2010b) software package. for ST_REG_BREZPOS (registered unemployment rate) a negative sign. Increased unemployment usually means that people receive just an unemployment compensation for the certain period of time and afterwards social support. Both amounts are lower than the salary. for DUMMY_INFOR a negative sign. for CONSTANT a positive sign because of an increase in or the preservation of KAT01 retail deposits' balance. 4.2 Regression Model for Category 1 (KAT01) 4.3 Regression Model for Category 2 (KAT02) The mathematical form of the model for KAT01 is as follows: The mathematical form of the model for the medium risky category, KAT02, is as follows: KAT011 = P1 + P2KAT01(t-1)t + P3EURIBORt + + PiNETO_PLACAt + + PsST_REG_BREZPOSt + + PfDUMMY_INFORt + ut, KAT021 = P1 + P2KAT02(t-1)t + P3EURIBORt + + PNETO_PLACAt + + PsST_REG_BREZPOSt + + PfDUMMY_INFORt + ut, where: where: KAT01t retail deposits for Category 1 for t day (dependent variable) KAT02t retail deposits for Category 2 for t day (dependent variable) KAT01(t-1) t retail deposits for Category 1 for t-1 day, hereafter referred to as KAT01(-1), (explanatory variable) KAT02(t-1) t retail deposits for Category 2 for t-1 day, hereafter referred to as KAT02(-1) (explanatory variable) EURIBOR t reference interest rate 6M Euribor for t day (explanatory variable) EURIBOR t reference interest rate 6M Euribor for t day (explanatory variable) NETO_PLACA t net salary in Slovenia for t day (explanatory variable) NETO_PLACA t net salary in Slovenia for t day (explanatory variable) STReg BREZPOS unemPl°yment rate in Slovenia for t day - - t (explanatory variable) Streg BREZPOS unemployment rate in Slovenia for t day - - f (explanatory variable) DUMMYJNFOR dummy variable due to negative t information about the bank in media for t day (explanatory variable) DUMMY_INFOR dummy variable due to negative t information about the bank in media for t day (explanatory variable) ut stochastic disturbance (or stochastic error term) for t day ut stochastic disturbance (or stochastic error term) for t day P i regression coefficient of the constant (or an intercept) P i regression coefficient of the constant (or an intercept) P2 do p5 regression coefficients of the explanatory variables P2 do P5 regression coefficients of the explanatory variables Pe regression coefficient of the dummy variable Pe regression coefficient of the dummy variable time series from 1 September 2013 until 31 March 2014 time series from 1 September 2013 until 31 March 2014 t t We expect the following signs of regressions' coefficient estimators: • for KAT01(-1) a positive sign. As this is a lagged dependent variable, the only reasonably outcome is a positive sign. • for EURIBOR a positive sign. Higher interest rates should attract deponents to put their funds into the bank.; • for NETO_PLACA (net salary) a positive sign. Higher salaries should lead to higher savings/deposits. 16 We expect the same signs of regressions coefficient estimators for KAT02 retail deposits as we described for KAT01 retail deposits. Aleksandra Murks Bašič: Model for Determining the Stability of Retail Deposits with Higher Outflow Rates 4.4 Regression Model for Category 3 (KAT03) The mathematical form of the model for the riskiest category, KAT03, is as follows: KAT031 = P1 + P2KAT03(t-1)t + P3EURIBORt + P4NETO_PLACAt + + P5ST_REG_BREZPOSt + + ppUMMY_INFORt + ut, where: KATÜ3„ retail deposits for Category 3 for t day (dependent variable) retail deposits for Category 3 for t-1 day, hereafter referred to as KAT03(-1) (explanatory variable) KAT03{t-1) t reference interest rate 6M Euribor for t day (explanatory variable) EURIBOR NETO PLACA. net salary in Slovenia for t day (explanatory variable) Streg BREZPOS unemployment rate in Slovenia for t day - - f (explanatory variable) dummy variable due to negative DUMMY-INFOR t information about the bank in media for t day (explanatory variable) ß2 do ß5 stochastic disturbance (or stochastic error term) for t day regression coefficient of the constant (or an intercept) regression coefficients of the explanatory variables regression coefficient of the dummy variable time series from 1 September 2013 until 31 March 2014 We also expect the same signs of regressions' coefficient estimators for KAT03 retail deposits as described for KAT01. 4.5 Testing Stationary Time Series 4.6 Economic and Statistical Interpretation of the Results Econometric model for Category 1 Table 1 Results of Econometric Model for Retail Deposits Category 1 (KAT01), 1 September 2013 to 31 March 2014 Dependent Variable: LOG(KAT01) N: 150 Constant 1.77 t-stats (2.12) LOG(KAT01(-1)) 0.91 t-stats (21.03) LOG(EURIBOR) 0.15 t-stats (1.89) LOG(ST_REG_BREZPOS) -0.28 t-stats (-1.98) R2: 0.877 Adjusted R2: 0.874 Note: Statistically insignificant variables are omitted from the table. The estimated regression coefficients are statistically significant and their signs are in accordance with our expectations after the elimination of statistically insignificant regression coefficients. Regression constant C in our model represents an increase or preservation of the retail deposits balance for KAT01. As evident in Table 1, we eliminated two explanatory variables: the net salary and the dummy variable. They were defined as statistically insignificant. The other three explanatory variables have an impact on the KAT01 retail deposits, as shown in Table 1. The estimated equation suggests that the increase of KAT01 retail deposits by one percentage point during the previous day will lead to an increase of KAT01 by 0.91 percentage points the next day. The series used in the model for the estimation of outflow rates for the three categories of retail deposits are KAT01, KAT01(-1), KAT02, KAT02(-1), KAT03, KAT03(-1), Euribor, net salary, and registered unemployment rate. Using augmented Dickey-Fuller (ADF) testing, we can conclude that series KAT01, KAT01(-1), KAT02, and KAT02(-1) are non-stationary in their absolute form. Series KAT03, KAT03(-1), Euribor, net salary, and registered unemployment rate are stationary on the first difference. Therefore, we used the logarithm transformation. The results of diagnostic tests in Table 2 show that the regression model is very good and suitable for predicting the movement of Category 1 retail deposits. The results of the econometric model for retail deposits of Category 1 were used to estimate the volatility and expected movements of these deposits within the next 22 working days under stress conditions. The estimated stability of retail deposits (KAT01, KAT02, and KAT03) were calculated as the ratio between the lowest predictive value of the deposits in the next 22 working days minus two standard errors and the maximum predictive value of 17 u ß ß t NAŠE GOSPODARSTVO / OUR ECONOMY Vol. 61 No. 5 / October 2015 Table 2 Diagnostic Test Results of Regression Model for Retail Deposits Category 1 (KAT01) (number of observations = 150) Test Critical value (c)* Calculated value * Fulfillment Yes/No & Results description t-statistics 1.655 t. > 1.894 Yes. We can reject the null hypothesis that individual regression coefficients are zero at significance level a = 0.10. F-statistics 2.667 347.889 Yes. The regression model is overall statistically significant (F > Fc). R2/Adjusted R2 0 2.240 Yes. We can reject the null hypothesis that individual regression coefficients are zero, except for d_infor. F-statistics 2.667 326.992 Yes. The regression model is overall statistically significant (F > Fc). R2/Adjusted R2 0 x2c), we reject the null hypothesis of no heteroscedasticity, and it is corrected with White test. Multicollinearity (VIF test) VIF < 10 VIF. < 3 Yes. No multicollinearity present. Model specification (Ramsey-Reset test) 2.667 0.376 Yes. As F < Fc we cannot reject the null hypothesis; therefore, we conclude that our model is correctly specified. Note: * at the a = 0.05 significance level (confidence interval 95%). ** i = regression coefficients/variables. *** VIF = Variance Inflation Factor. the regression coefficient estimator of 6M Euribor is 1.332; therefore, we can reject the null hypothesis about its statistical significance only at 80% probability (tc = 1.287). Yet we need to emphasize that due to data complexity only a short time series was included. We decided to keep the impact of 6M Euribor on the bank's retail deposits within Category 3 because some contracts included in this category are based on the reference rate movement. As shown in Table 5, we again eliminated two explanatory variables: net salary and the dummy variable. The other three explanatory variables have an impact on the KAT03 retail deposits. An increase in 6M Euribor by one percentage point will lead to an increase in KAT03 retail deposits by 0.10 percentage points (taking into account the 80% probability). The results of diagnostic tests are presented in Table 6. They show that the regression model is good and suitable for predicting the movement of Category 3 retail deposits. The results of the econometric model for retail deposits of Category 3 were used to estimate the volatility and expected movements of these deposits within the next 22 working days under the assumption of a combined idiosyncratic and market-wide stress scenario. Retail deposits within Category 3 are treated as the riskiest category, taking into account the predetermined number and type of risk factors (retail deposits who meet two criteria from very high-risk Group 2 factors or two criteria from Group 1 and one criterion from Group 2 risk factors or any other risk factor combination). For the LCR reporting, the outflow rate for retail deposits that fall into Category 3 was calculated at 34.40% using the historical and expected volatility assessment. taking into account the predetermined number and type of risk factors (retail deposits that meet three criteria from high-risk Group 1 factors or one criterion from Group 1 and one criterion from Group 2 risk factors). For the LCR reporting, the outflow rate for retail deposits that fall into Category 2 was calculated at 22.57% using historical and expected volatility assessment. Econometric model for Category 3 Table 5 Results of Econometric Model for Retail Deposits Category 3 (KAT03), 1 September 2013 to 31 March 2014 Dependent Variable: L0G(KAT03) N: 150 Constant 1.94 t-stats (2.92) L0G(KAT03(-1)) 0.89 t-stats (22.96) LOG(EURIBOR) 0.10 t-stats (1.33) L0G(ST_REG_BREZP0S) -0.26 t-stats (-1.82) R2: 0.800 Adjusted R2: 0.797 Note: Statistically insignificant variables are omitted from the table. Table 5 shows that the estimated regression coefficients are statistically significant and their signs are in accordance with our expectations after the elimination of statistically insignificant regression coefficients. The calculated t statistic for 19 NAŠE GOSPODARSTVO / OUR ECONOMY Vol. 61 No. 5 / October 2015 Table 6 Diagnostic Test Results of Regression Model for Retail Deposits Category 3 (KAT03) (number of observations = 150) Test Critical value (c)* Calculated value * Fulfillment Yes/No & Results description t-statistics 1.655 t. > 1.815 Yes. We can reject the null hypothesis that individual regression coefficients are zero, except for Euribor. F-statistics 2.667 195.791 Yes. The regression model is overall statistically significant (F > Fc). R2/Adjusted R2 0