Credit Risk Modelling For Assessing Deposit Insurance Fund

Credit Risk Modelling For Assessing Deposit Insurance Fund Adequacy: The Case Of Russia Sergey Smirnov, Higher School of Economics, Moscow Brief Description of the Project CIA Third Stochastic Modeling Symposium April 2006, Toronto 2 Main Goals of the Project BearingPoint - Deposit Insurance Agency  The main objective is the development of methodology and techniques for assessing deposit insurance fund sufficiency to meet its obligations of repayment against insured deposits in the Russian Federation for: – short-term (1 year)  These methodology and techniques should be based on best international practices and to a maximum extent adjusted to environment in the Russian Federation as well as situation in the Russian banking sector CIA Third Stochastic Modeling Symposium April 2006, Toronto 3 – medium-term (up to 5 year) period. Importance of the Project in Practice    The Deposit Insurance System is created to contribute to the financial stability of the national economy, first of all, by creating and maintaining public confidence towards the banking system. By reducing the risk of public discredit of the banking industry and, resultantly, by reducing the risk of a sudden deposit flow-out, a deposit insurance system contributes to the conditions required for a stable banking system and, more broadly, for a dynamically developing financial system. The whole economy benefits from the resultant stabilizing influence. The adequacy of the Fund is one of the main considerations in regards with the Deposit Insurance System efficiency, because it is the base of Fund’s financial stability. CIA Third Stochastic Modeling Symposium April 2006, Toronto 4 Requirements to the Fund’s financial stability   Fund should be large enough to assure the depositors that it can absorb the potential losses and that their deposits will be compensated as prescribed by the legislation in case that their bank fails. In order to meet its commitments, the Fund should remain stable even under unfavorable (stress) conditions that, however, do not result in a systemic banking crisis. CIA Third Stochastic Modeling Symposium April 2006, Toronto 5 Best practices   The international practice in assessing deposit insurance fund adequacy is still forming and the only interesting experience is present in USA, Federal Deposit Insurance Company (FDIC) Thus the present project enables Russian Deposit Insurance Agency to become one of the deposit insurers using state-of-the-art technologies. CIA Third Stochastic Modeling Symposium April 2006, Toronto 6 Some experience in assessing fund adequacy …It is difficult to propose a scientific reason explaining this target level. It is always possible to add that this level would be sufficient for one or two failures of small or medium size French banks, but, in fact, it was exactly 10 billions Francs in 1999, it is also valid reason, because 10 is a round number. My reply is not truly scientific but it is exact… Charles Cornut (excerpt from a reply to the request of Higher School of Economics) CIA Third Stochastic Modeling Symposium April 2006, Toronto 7 Requirements to the methodology and models A robust methodology is to be constructed, taking into account the influence on the fund balance evolution in time of the following significant factors:     the impact of probable insurance losses, which are primarily losses from failed institutions. deposit growth and related fee accumulation process. the amount of income that the fund will receive on asset portfolio. the compensation of fund‟s payments to the insured depositors, resulting from failed bank resolution CIA Third Stochastic Modeling Symposium April 2006, Toronto 8 Economic grounds CIA Third Stochastic Modeling Symposium April 2006, Toronto 9 Credit rating as a commonly recognized financial stability indicator   Based on the historical default frequencies of the rated issuers, agencies‟ credit ratings can be mapped to average default rates It should be mentioned that the average default rates calculated using the historical data depend largely on the content of the sample that was used for the computations. It is known that the average default rates for all ratings are lower during an economic boom and higher during a recession. CIA Third Stochastic Modeling Symposium April 2006, Toronto 10 Issuer Moody’s Rating Baa1 Baa2 Baa3 Ba1 Ba2 Ba3 B1 B2 B3 Caa - C Historic average default rate, % In 1 year In 5 years 0,17 0,12 0,41 0,66 0,62 2,23 3,03 5,93 10,77 22,24 1,46 2,11 3,60 6,76 8,82 19,14 25,27 31,24 43,55 60,40 11 CIA Third Stochastic Modeling Symposium April 2006, Toronto Expected and unexpected Losses    In order to work out an economically grounded Fund adequacy level, one should consider the impact on the stability of the Fund of both expected (EL) and unexpected (UEL) losses In order to ensure long term stability, it is required that the expected incomes of the Fund (assessed premiums and investment income) exceed the expected payouts in a given period of time. However, this is not sufficient. The Fund should be sufficient not only to pay out typical (moderate) depositor compensations, but also to offset the unexpected loss, i.e. the compensations that will be paid out in the unlikely event of substantial liabilities before the depositors of failed institutions. CIA Third Stochastic Modeling Symposium April 2006, Toronto 12 Implied Solvency Standard    The target level of the Fund should correspond to some implied solvency standard that can be represented by a certain credit rating. This does not imply that the Agency should obtain a rating from some credit rating agency. The model will allow one to estimate the probability of the Fund defaulting given the level of the Fund and the time horizon. In term, this estimate can be mapped to an implied credit rating based on the historic average default frequencies Thus, financial stability of the Fund will have an explicit representation that can be compared to the correspondent indicators of Russian and foreign banks, as well as the Russian Federation and other countries. CIA Third Stochastic Modeling Symposium April 2006, Toronto 13 Choosing the target Fund stability level   that it would be unreasonable to set the Fund solvency standard above the solvency standard of the sovereign debt of the Russian Federation At the same time, if the Fund‟s solvency standard is below that of some insured bank, one can reasonably doubt the Agency ability to enhance the stability of the banking system. Therefore, it is reasonable to suggest that the Fund‟s stability should be not lower than the stability of the most reliable Russian banks. CIA Third Stochastic Modeling Symposium April 2006, Toronto 14 Credit rating of banks (June 2005) Moody’s Rating PD1, % Russian Federation Alfa Bank Standard&Poors Rating BBBB PD1, % 0,39 8,34 Baa3 Ba2 0,41 0,62 Bank Zenith Bank of Moscow Vneshtorgbank MDM Bank Rosbank B1 Ba1 Ba1 --B1 3,03 0,66 0,66 --3,03 ----BB+ B B- ----0,56 8,34 12,15 Sberbank Ba1 0,66 --- --15 CIA Third Stochastic Modeling Symposium April 2006, Toronto Economically grounded target level of the Fund  In accordance with the proposed principle, it is assumed that an economically grounded target level of the Fund should correspond to a probability of Fund deficit over one year lying in the range of 0.4% to 0.6%  This correspond to practice of some countries: in Hong Kong a probability of Fund deficit over one year is required to be at the level of 0.2% CIA Third Stochastic Modeling Symposium April 2006, Toronto 16 Special Status of Sberbank    The largest Russian bank, Sberbank, accounts for 1.2 trln rubles of deposits at the end of 1st quarter of 2005, equivalent to over 60% of the total personal deposits. Even in case of a reasonable decline of its insured deposits, withdrawal of Sberbank license or a moratorium imposed on its payments (an insurance incident for the Agency) will result in an excessive deficit of the Fund given any reasonable Fund balance. Thus it is reasonable to treat Sberbank as a „too big to fail‟ bank. CIA Third Stochastic Modeling Symposium April 2006, Toronto 17 Estimate of Sberbank default is of relative meaning    This means that given the great social and political importance of Sberbank, in case of a critical situation with Sberbank the government and the supervisory authorities will intervene early and will not allow for the license to be withdrawn or a moratorium to be placed on Sberbank payments, i.e. the insurance incident for the Agency will not occur. The arising problems will likely be resolved through some form of restructuring. Respectively, the model estimate of Sberbank default is of relative meaning and is rather illustrative. CIA Third Stochastic Modeling Symposium April 2006, Toronto 18 Scenarios of the Fund adequacy modeling Scenario Favorable Baseline Unfavorable Default correlations Sberbank probability of default 0 + 0 + >0 19 CIA Third Stochastic Modeling Symposium April 2006, Toronto Selection of banks – participants of the Deposit Insurance System CIA Third Stochastic Modeling Symposium April 2006, Toronto 20 Models CIA Third Stochastic Modeling Symposium April 2006, Toronto 21 The economic sense of the model input is of major importance According to Thomas Henry Huxley, "Mathematics may be compared to a mill of exquisite workmanship, which grinds your stuff to any degree of fineness; but, nevertheless, what you get out depends on what you put in..." CIA Third Stochastic Modeling Symposium April 2006, Toronto 22 Quality of Econometric Models   One should use the econometric approach to estimating the default probabilities cautiously, since in fact such models estimate the probability of default of a group of banks, whose set of explanatory variables (risk indicators) is similar to those of the actual bank Using different methods of assessing the probabilities of default we empirically verified that econometric estimates of the probability of default comply with those obtained by another ways: based on credit spreads on Eurobonds and credit rating (for several banks, when this data is available) CIA Third Stochastic Modeling Symposium April 2006, Toronto 23 Case of Alfa Bank CIA Third Stochastic Modeling Symposium April 2006, Toronto 24 Sensitivity to macroeconomic environment    In USA the averaged default intensity of insured commercial banks were about 0.06% during relatively stable time period: 1995 – 2004. To compare with intensity of 1.37% in Thus it differs 22 times ( and distinction is even more pronounced for thrifts) The averaged default intensity of Russian commercial banks was 1.48% during relatively stable time period: 2000-2004 To compare with intensity of 14.5% in1998. Thus it differs only 10 times It is clear that in 1998 there was more stress for Russian banking system, then in 1990 for American one. CIA Third Stochastic Modeling Symposium April 2006, Toronto 25 Bank and Thrift Failures in USA BANK AND THRIFT FAILURES, 1920 - 2002 Total Failures, 1920-33: 14,977; Total Failures, 1934-2002: 3,588 4,000 3,500 3,000 2,500 2,000 1,500 1,000 500 0 19 20 19 23 19 26 19 29 19 32 19 35 19 38 19 41 19 44 19 47 19 50 19 53 19 56 19 59 19 62 19 65 19 68 19 71 19 74 19 77 19 80 19 83 19 86 19 89 19 92 19 95 19 98 20 01 CIA Third Stochastic Modeling Symposium April 2006, Toronto 26 Bank Failures in Russia 350 300 250 200 150 100 50 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 CIA Third Stochastic Modeling Symposium April 2006, Toronto 2005 0 27 Distribution of Exposures Exposure Distribution 1 0.9 0.8 0.7 Percent of Exposure 0.6 0.5 0.4 0.3 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 0.6 Percent of Banks 0.7 0.8 0.9 1 CIA Third Stochastic Modeling Symposium April 2006, Toronto 28 Lorenz curves for the different models ROC Curves 1 0.9 0.8 0.7 Tree Linear logit Linear probit Bayesian Default Cases 0.6 0.5 0.4 0.3 0.2 0.1 0 0 0.2 0.4 0.6 Non-Default Cases 0.8 1 29 CIA Third Stochastic Modeling Symposium April 2006, Toronto Comparison with the results of other researchers CIA Third Stochastic Modeling Symposium April 2006, Toronto 30 Power curves: Russian vs. American banking systems ROC Curves 1 0.9 0.8 0.7 Tree Linear logit Linear probit Bayesian Default Cases 0.6 0.5 0.4 0.3 0.2 0.1 0 0 0.2 0.4 0.6 Non-Default Cases 0.8 1 CIA Third Stochastic Modeling Symposium April 2006, Toronto 31 Estimate of default probability of Promeximbank Dynamics of Probability of Default 25 20 15 PD, % 10 5 0 1998 1999 2000 2001 2002 Month 2003 2004 2005 CIA Third Stochastic Modeling Symposium April 2006, Toronto 32 Averaging of estimates of default probability    A Simplified Approach is used in the project, that can be regarded as a first approximation of the technique of Jarrow et.al (2003) that was recommended to the FDIC as a core methodology for measuring and valuing the risk of the FDIC deposit insurance funds by McKinsey in 2003 (still under implementation) It is based on heuristic engineering idea that the future intensities can be simulated as time- independent and approximated using the EWMA estimated intensities that typically fluctuate from month to month due to fluctuations in explanatory (state) variables The averaging of monthly estimates of default probability is performed parameter of geometric series equal to 0.286 and with truncating data older then one year CIA Third Stochastic Modeling Symposium April 2006, Toronto 33 Fluctuations of estimates of default probability, Bank Olympiysky Dynamics of Probability of Default 70 60 50 PD, % 40 30 20 10 0 1998 1999 2000 2001 2002 Month 2003 2004 2005 2006 CIA Third Stochastic Modeling Symposium April 2006, Toronto 34 Default correlations    Default correlations are calculated for banks that are grouped according to the intensity of their operations on the inter-bank lending market. The most appropriate approach is to single out a cluster of banks who are active participant of the inter-bank lending market (a total of 250 banks, including 160 largest banks by total assets), where the correlation within the cluster is considered constant and same for all banks, and a cluster of all other banks, for which a zero correlation is assumed. Deterioration of the macroeconomic and industrial environment results not only in an increase of the probabilities of default, but also in an increase of default correlations. Further specification of the proposed model is possible based on the clusters identified within the cluster of 250 banks. However, such model would be rather important for development of an early warning system rather than the Fund adequacy model CIA Third Stochastic Modeling Symposium April 2006, Toronto 35 Deposit growth model   The model assumes that total deposits growth is described with the help of geometric Brownian motion and there is four independent groups of banks In order to estimate the parameters of the motion, the banks were grouped into 4 categories based on the size of total assets. The following categories were used: – – – – Sberbank; 50 largest bank; 200 large banks; all other banks. CIA Third Stochastic Modeling Symposium April 2006, Toronto 36 Annual deposit growth rates CIA Third Stochastic Modeling Symposium April 2006, Toronto 37 Crude model of investment returns CIA Third Stochastic Modeling Symposium April 2006, Toronto 38 Modeling Total Return Bond Index CIA Third Stochastic Modeling Symposium April 2006, Toronto 39 Indirect method of modeling LGD and time to recovery   For each bank that defaulted before January 1, 1999, the maximum debit balance of account 30102 („Correspondent accounts of credit institutions with the Bank of Russia‟) following the withdrawal of the license was used as the estimate of remaining assets after resolving . The number of months from the moment the license was withdrawn to the moment the value was observed was used as the estimate of time to recovery. It should be observed that the results do not change substantially if the debit balance of account 30102 is replaced with the amount of obligatory reserves held with the bank of Russia, defined as the sum of debit balances of accounts 30202 („Obligatory Ruble reserves of credit institutions held with the Bank of Russia‟) and 30204 („Obligatory foreign currency reserves of credit institutions held with the Bank of Russia‟). CIA Third Stochastic Modeling Symposium April 2006, Toronto 40 Distribution of the time to recovery CIA Third Stochastic Modeling Symposium April 2006, Toronto 41 Analysis of the remaining assets and time to recovery   The statistical analysis of the remaining assets and time to recovery, both in absolute and relative values with respect to different financial indicators was performed, 6 month before revocation of bank‟s license. The percentage of banks with the remaining assets less the total deposits (i.e. when insurance incidents incurred loss to the Fund) equals 74%. CIA Third Stochastic Modeling Symposium April 2006, Toronto 42 Maximal amount on account in Central Bank and Obligatory Reserves at the moment of license revocation CIA Third Stochastic Modeling Symposium April 2006, Toronto 43 Software developed in course of the project     All software developed in course of the project can be divided into the following four building blocks: Econometric block is used to construct and test econometric models. Portfolio block is used to model the Fund‟s dynamics and forecast the distributions of the relevant variables. Alternative probability of default computation block is used to estimate Eurobond spreads and map them to econometric and credit rating probabilities of default. Miscellaneous items block consists of auxiliary programs and scripts aimed at solving smaller tasks, i.e. estimating the deposit growth ratio, LGD etc. CIA Third Stochastic Modeling Symposium April 2006, Toronto 44 Scenario modeling of fund balance CIA Third Stochastic Modeling Symposium April 2006, Toronto 45 Parameters of the Fund  Insurance premiums of the member banks are    assumed to be assessed at maximum rate constituted by the Deposit Insurance Agency Act that is 0.15% of the average amount of deposits made by natural persons per quarter. Investment Yields are approximated by the Total Return Index of government ruble-nominated bonds. Operating Expenses of the Fund are assumed to equal zero. Time Horizons used in the model are December 31, 2006 and December 31, 2010 (i.e. 1.5 and 5.5 years). CIA Third Stochastic Modeling Symposium April 2006, Toronto 46 More detailed scenarios     We split Initial set of scenarios (favorable scenario, baseline, and an unfavorable scenarios) using additional parameters and run them using the models and software developed. In addition to Sberbank considered as “too big to fail”, we run simulations including or excluding another major bank, Vneshtorgbank, that has a substantial government stake (thus multiplying the number of scenario by two). One year time period (January 1st to December 31st 2006) modeling results are of major interest, since they are essential for the Agency planning needs. For 5 year simulation, one of the principal questions is what will the deposit growth be. We consider two possibilities: (1) the deposit growth rate remains unchanged, or (2) the deposit growth ratio decreases at constant rate from 40% annual growth for the first year to 10% annual growth for the last year. CIA Third Stochastic Modeling Symposium April 2006, Toronto 47 Typical cash flows of the fund CIA Third Stochastic Modeling Symposium April 2006, Toronto 48 Distribution of fund balance (end of 2006, baseline scenario, Vneshtorgbank excluded) CIA Third Stochastic Modeling Symposium April 2006, Toronto 49 Fat tails of distributions CIA Third Stochastic Modeling Symposium April 2006, Toronto 50 Estimate of density with the help of approximation by Pareto distribution CIA Third Stochastic Modeling Symposium April 2006, Toronto 51 Evaluation of accuracy of Monte-Carlo simulations Baseline scenario, Vneshtorgbank excluded, deposit growth rate remains unchanged, number of simulations : 100 000 confidence interval confidence interval Quan(99.73%) (95.0%) tile Length of interval / Length of interval / value value 1.0% 3.6% 2.3% 0.9% 0.8% 0.7% 0.6% 0.5% 0.4% 3.1% 2.9% 2.6% 2.4% 2.3% 2.1% CIA Third Stochastic Modeling Symposium April 2006, Toronto 2.1% 1.9% 1.7% 1.6% 1.5% 1.4% 52 Structure of returns (end of 2010, baseline scenario, Vneshtorgbank excluded) CIA Third Stochastic Modeling Symposium April 2006, Toronto 53 Decision making approach  The provision level should be set by Agency at the level that would correspond with the implied solvency standard within the range between the credit rating of the Russian Federation and the credit rating of the most reliable bank in the Deposit Insurance System (BBB- and BB+ respectively in the S&P scale, as of June 2005), using the results of scenario modeling. CIA Third Stochastic Modeling Symposium April 2006, Toronto 54 Fund parameters used to make decision on adequacy For each of scenarios the following parameters are calculated  The key parameters are the quantiles of fund balance distribution for the different significance levels, corresponding to the current ratings (e.g. 0.4% and 0.6% as of June 2005)  Conditional Expected Losses (Shortfall) for given significance levels, as well as mathematical expectation, median and standard deviation are also calculated  Sensitivity analysis is performed ( for example, with respect to default intensity, default correlations) CIA Third Stochastic Modeling Symposium April 2006, Toronto 55