banks during the last crisis: macroeconomic conditions or risky business

Transcription

1 Anna Pestova Mikhail Mamonov What was the key determinant of loan qualy deterioration of Russian banks during the last crisis: macroeconomic condions or risky business strategies? Objectives During the latest crisis Russian banking system was faced wh a significant deterioration in the loan qualy. The growth of bad loans caused a sharp increase in banks loan loss reserves. The later reduced opportunies for banks to earn profs and exerted pressure on banks' capal adequacy ratio. Under these condions Russian government was forced to carry out massive recapalization of the major banks to maintain their stabily. In Vnesheconombank (Russian Bank for development), granted Russian cred instutions more than 400 billion rubles in the form of subordinated loans. Besides, Bank of Russia provided Sberbank the largest Russian bank wh the subordinated loan of 500 billion rubles. The question inevably arises, how justified was the decision of Russian government to provide financial support to cred instutions? To evaluate the effectiveness of such a policy measure is necessary to distinguish among the cred risk sources of affected banks. In particular, state capal injections into banks wh aggressive pre-crisis business strategies creates a distorting incentives for other banks and may lead to an exacerbation of moral hazard in middle-term perspective. On the contrary, government support is reasonable for those banks who suffer decrease in loan qualy mainly due to the worsening of macroeconomic condions. The main purpose of our research is to separate the influence of macro- and microeconomic factors that led to an increase in bad loans of Russian banks using wide tools of panel data econometrics. The main hypothesis to be tested is the predominance of microeconomic factors of 1

2 cred risk realization for most banks that received government support. Testing such a hypothesis involves estimation of panel data regression models explaining the dynamics of bad loans of Russian banks. We ll use as the independent variables those that reflect macroeconomic condions and specific business-strategies of banks. Practical contribution of research The resulting findings can be applied for remote stress testing of banks and improve efficiency of government decisions on providing financial support to cred instutions. It will be possible to assess the impact of deterioration of macroeconomic condions on loan qualy of individual banks on the basis of our econometric models for bad loans. The fact is that current stress tests, published by reputable organizations, including the Bank of Russia, Moody's, Standard & Poor's and others, do not disclose the structure of cred instutions, which are exposed to higher risk. The proposed methodology in our research helps to fill this gap and to provide the analytical and scientific communy wh more detailed information about the existing problems in different banks in Russia. It is necessary in order to provide an early response to possible threats and to prevent growth of bad loans in future. In addion, the resulting conclusions may be of great importance when government makes decisions about the financing lims to different cred instutions, who have suffered decrease in loan qualy. If such a decrease was mainly due to macroeconomic factors and bank can prove that government should raise the individual lim of financial support and vice versa. Lerature review There are two main areas of modeling cred risk - integrated (Boss 2002), (Virolainen 2004), (Jacobson, Linde, Roszbach 2005) and piecewise approach (Sorge, 2004). The methodology used in the first group is not of interest for our study because implies estimation of the entire densy function of losses due to the realization of cred risk, while is sufficient to 2

3 model the expectation of cred risk measure (the proportion of bad loans). However, the factors used in these studies can be successfully used for the purpose of our study. Whin the second group of studies (based on piecewise approach) one can distinguish between studies at the level of the banking system as a whole (macroeconomic approach - see (Hoggarth, Sorensen, Zicchino 2005), (Babihuga 2007), Pesola (2005)), as well as at the level of individual banks (microconomic approach - see (Jimenez, Saurina 2005), (Espinoza, Prasad 2010), (Quagliariello, 2007). We utilize methods and factors used as the macro and micro level, because the former can be adapted to the level of individual banks. Whin the piecewise approach indicators of cred risk measure are calculated on basis of the balance sheet data of the banking system or individual banks. One possible measure of cred risk models based on the balance sheet data can be ratio of loans wre-offs to total loan portfolio - (Hoggarth, Sorensen, Zicchino 2005), (Pesola 2005). The disadvantage of this indicator is the presence of the lagged response to the real size of the problems. Another indicator which is frequently used in empirical work is the ratio of loan loss provisions to total loan portfolio (Quagliariello 2007), (Głogowski 2008). Unfortunately has a high proportion of noise compared to the real size of the cred risk (because of differences in management policies of banks over the cred cycle). We plan to use the share of nonperforming loans (NPL) in the aggregate loan portfolio of each bank as the dependent variable (a similar variable is used in Babihuga (2007), Jimenez, Saurina (2005), Espinoza, Prasad (2010), because properly indicates the level of banks cred risk and is free from the above-mentioned drawbacks. Overview of the cred risk determinants used in existing empirical studies allowed us to identify two groups of factors included in bad loans equations: bank-specific factors: growth of lending activies, loan-to-depos ratio, interest rates, cost and operational efficiency, loan market concentration and individual market shares, capal adequacy ratio, ownership structure, bank profile, etc.; 3

4 macroeconomic condions, including Internal factors: GDP growth, inflation, unemployment, the ratio of consumption and investment to GDP, bubbles in asset markets; External factors: dynamics of exchange rates, international reserves, net exports, terms of trade. As noted in Salas, Saurina (2002), the share of bad loans is closely related to s values in previous periods, because the bad loans are not immediately wrten-down and can remain on banks' balance sheets over several years. In other words NPL shows a tendency to persist over time. This necessates the use of the dynamic specification of equation. Such specifications were also estimated in Quagliariello (2007), Espinoza, Prasad (2010), Jimenez, Saurina (2005). Most empirical studies support the hypothesis pro-cyclicaly of the banking sector. The influence of macroeconomic condions on the loan qualy of banks is statistically significant in most specifications. Some studies also found the influence of the pre-crisis cred boom in the depth of the bad loan problem in banking sector (see, for example, Quagliariello (2007), Espinoza, Prasad (2010), Jimenez, Saurina (2005)). These empirical works also confirm the influence of management strategies on the sensivy of banks to the cred risk. Management strategies are proxied by return-on-asset ratio (ROA) in Quagliariello (2007), loans interest rates in Espinoza, Prasad (2010), capal adequacy ratio in Boudriga, Boulila, Jellouli (2009) and the profile of banks in Głogowski (2008). Methodology Data description 1. bank specific factors collected from Bank of Russia webse ( macroeconomic factors collected from Federal State Statistics Service webse ( Forms 101 and 102: 2 Data on GDP, inflation, exchange rates, unemployment, etc: 4

5 We use public data from 101 and 102 forms of the consolidated financial statements of Russian commercial banks whin the period Q Q (Table 1). Earlier data is unavailable. Table 1. Sources and data structure for Russian commercial banks Variables Data Sources: 101 Form Assets and liabilies Assets: households and corporate loans, granted inter-bank loans, purchased securies, total assets, etc. 102 Form Profs and losses Income: interest income, operating income, etc. Liabilies: funds, deposs, attracted inter-bank loans, foreign liabilies, etc. Expense: interest expenses, operating expenses, etc. Capal and performance: total equy, loan loss provision, prof Duration monthly quarterly Availabily from January 2004 from 1Q 2004 Size of bank sample sample includes 525 banks and covers about 85% of Russian banking system assets. Wh 90% of sample is 51 large banks (wh total assets more than 50 bln rubles) The reporting of the Forms 101 and 102 in Russia has one special feature: is not necessary for each bank to make s individual Forms publicly available. Accordingly, from quarter to quarter the size of the sample of banks, who agreed to publish their accounts, can vary widely. In order to avoid possible negative impact of such a volatily in the sample of banks on the proportion of bad loans in total loans of each bank in each quarter we undertook a standard technique. We fixed the sample of 525 banks, who 5

6 constantly publish their statements whin the period from 2004 to This sample of banks is representative, as covers most (about 85%) of the total assets of Russian banking system (Table 1.). All the quarterly indicators from 102 Form that will be considered in this study (interest income and expenses, etc.) will be taken in annual terms, i.e. as moving sum of each indicator values for the four previous quarters, to avoid the problems associated wh seasonaly. following: Estimation According to the lerature review, we specify the static form of bad loans model as the OL = α + β BSV + γ MV + ε, where OL - ratio of overdue loans to total loans BSV - bank-specific variables MV - macroeconomic variables ε = µ + v - the error term, sum of individual effect and idiosyncratic shock i i = 1,, 525 t = 1q2004 2q2011 This model will be estimated using well-known fixed and random effects estimators and also pooled OLS approach. The best error term specification will be chosen according to F-test on individual effects, Breush-Pagan LM test on random effects and Hausman test. The dynamic model of bad loans is the following: OL = + δ OL 1 α + β BSV + γ MV + ε The inclusion of lagged dependent variable into the list of regressors leads to inconsistency of fixed effects estimator (most often used for bank-level panel data). That is why we will estimate this model in first differences and use lags of the dependent variable as instrumental variables (Arellano, Bond, 1991) in a Generalized Method of Moments (GMM) estimator. 6

7 The estimation of dynamic specification of bad loans equation can also solve the problem of endogeny of some variables included in the right side of the equation. In case of panel data at level of the individual banks endogenous variables are those bank-specific factors, which are simultaneous determined wh dependent variable. The set of bank-specific explanatory variables will be included into bad loans equation: rate of cred growth. The cred policy of the banks is expected to be highly correlated wh share of overdue loans in total loan portfolio. The possible explanation of this phenomenon is that rapid cred growth of loans is achieved by decreasing lending standards and lowering interest rates. This inevably leads to borrowers adverse selection and may result in rise of bad loans during the economic contraction. capal to assets ratio. From the one hand high value of this indicator may reflect bank s capacy to absorb loan losses in the future, thus may be posively correlated wh dependent variable. From the other hand, low capal adequacy means higher probabily of bank s default, thus may induce management to involve in more risky projects. cost efficiency indicators. Low values of this variable may induce banks to take on more risk to improve profabily. income diversification measure corresponds wh a lower ratio of bad loans due to a lower cred risk of earning non-interest income average lending rate. High value of this indicator may increase debt service costs and lead to rise of bad loans bank profile dummy variables and ownership structure dummy variables explain differencies in risk profile of different groups of banks. The macroeconomic condions include internal and external condions and reflect the abilies of borrowers to repay their debts. 7

8 Preliminary research results We have already evaluated some different specifications of bad loans models based on static panel data. Significant explaining factors are the concentration of banking, average funding rate, stabily of banks, capal adequacy ratio - all variables wh time lags (from one to four quarters). Among the macroeconomic factors, which have a significant impact on bad loans, we found GDP growth, inflation and stock market index (RTSI) Bibliography Arellano M., Bond S. (1991): Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations // Review of Economic Studies. 58. P Babihuga R. (2007): Macroeconomic and Financial Soundness Indicators: An Empirical Investigation. IMF Working Paper 07/115. Boss M. (2002): A Macroeconomic Cred Risk Model for Stress Testing the Austrian Cred Portfolio. Financial Stabily Report. 4. Oesterreichische Nationalbank. Boudriga, A., Boulila N., Jellouli, S. (2009) Bank specific, business and instutional environment determinants of nonperforming loans: Evidence from MENA countries. Paper for ERF сonference on «Shocks, Vulnerabily and Therapy», Cairo, Egypt, November 7-9, Espinoza R., Prasad A. (2010): Nonperforming Loans in the GCC Banking System and their Macroeconomic Effects. IMF Working Paper 10/224. Głogowski A. (2008) Macroeconomic determinants of Polish banks loan losses results of a panel data study. National Bank of Poland working paper No. 53. November Hoggarth G., Sorensen S., Zicchino L. (2005): Stress Tests of UK Banks Using a VAR Approach. Bank of England. Working Paper No

9 Jacobson T., Linde J., Roszbach K. (2005): Exploring Interactions Between Real Activy and the Financial Stance // Journal of Financial Stabily. 1. P Jimenez G., Saurina J. (2006): Cred Cycles, Cred Risk, and Prudential Regulation // International Journal of Central Banking. 2(2). P Pesola J. (2007): Financial Fragily, Macroeconomic Shocks and Banks Loan Losses Evidence from Europe. Bank of Finland Research Discussion Papers 15. Quagliariello M. (2007): Banks' Riskiness over the Business Cycle: A Panel Analysis on Italian Intermediaries // Applied Financial Economics. 17: 2. P Salas V., Saurina J. (2002): Cred Risk in Two Instutional Regimes: Spanish Commercial and Savings Banks // Journal of Financial Services Research 22 (3). P Sorge M. (2004): Stress-testing Financial Systems: An Overview of Current Methodologies. BIS Working Papers No 165. Virolainen K. (2004): Macro Stress Testing wh a Macroeconomic Cred Risk Model for Finland. Bank of Finland discussion papers 18. Participants Our research team consists of two persons: Anna Pestova and Mikhail Mamonov. Both researchers received a master's degree at the Faculty of Economics of Lomonosov Moscow State Universy. In addion, currently both researchers are training in the postgraduate study at the faculty of Economics of Higher School of Economics National Research Universy. The main reason for the team structure is experience of effective cooperation at the Center for Macroeconomic Analysis and Short-Term Forecasting (CMASF) during the past four years. This experience includes realization of scientific and research projects (private and public) and wring of six joint papers, which were published in Russian leading scientific journals on the macroeconomic and, in particular, on the banking problems. 9

10 The division of responsibilies whin the research is the following. Anna Pestova is a project manager, Mikhail Mamonov is an ordinary member of the project. The role of each researcher is as follows: 1. Lerature review and formulation of research problems Pestova Anna; 2. Constructing a database of bank specific factors whin Russian banking system and a database of Russian macroeconomic factors Mamonov Mikhail; 3. Econometric estimation of bad loans models of Russian banks: Static models (OLS, FE, RE estimators) Mamonov Mikhail; Dynamic models (IV-estimator Andersen-Hsiao, GMM-estimator of Arellano and Bond, Arellano and Bover) Pestova Anna; 4. Selection of the final specifications of the models and their interpretation Pestova Anna and Mamonov Mikhail 5. Conclusion: description of the practical relevance of research and development proposals for monetary authories and management of different banks, both aimed at improving the qualy of banks' loan portfolios Pestova Anna Alternative/addional sources of funding Analysis of current condion and different types of forecasts for the Russian financial sector are all directly linked to the core job responsibilies of Anna Pestova and Mikhail Mamonov, because both researchers are the financial sector analysts at the Center for Macroeconomic Analysis and Short-Term Forecasting (CMASF). However, the analysis of loan qualy of Russian banks are not included in the permanent responsibily of researchers, so CMASF will not fund the project. Grants from other charies and organizations have not been received. Accordingly, we expected to carry out all stages of the project at the expense of the EERC. Project timetable Project stage Execution period An advanced lerature review on determinants of loan december april

11 portfolio qualy (bad loans) whin the banking industry microeconomic level (specific business strategies) december february 2012 macroeconomic level (general condions) march - may 2012 Methodology: specification of econometric equations for bad loans may 2012 Building a database on Russia (over the period from 2004 to 2011): june 2012 bank specific factors (panel-series database) june 2012 macroeconomic condions (time-series database) june 2012 Empirical results: econometric estimation of different specifications for bad loans equation july - september 2012 Analysis and interpretation of achieved results october 2012 Practical recommendations for monetary authories october 2012 Making the final report november - december