Appendix I. Macroeconomic Model with the Banking Sector
23. The model consists of a set of over 20 econometric equations, covering the household, real, external, monetary, and fiscal sectors. Most of the estimation uses quarterly year on year change data for the period of 1998–2010, using OLS.
24. The core part of the macro-financial model consists of linking the aggregate NPL ratios for the corporate sector loans and for the household sector loans to macroeconomic variables. The models are estimated with data from 2000 to 2010, using monthly year-on-year growth rate of each variable. The model for the household sector loans includes CPI, household income, exchange rate (vis-à-vis the US dollar), lending rate to the household, and M0. The model for the corporate sector loans includes exchange rate (vis-à-vis the U.S. dollar), production, fixed capital investment, PPI, oil price, and lending rate to the sector.
Prepared by Hiroko Oura (MCM), building on substantial inputs from the stress testing team at the CBR.
Information on the country composition of the loan exposures is kept by the CBR in a limited manner, which is one of the major regulatory deficiencies (see detailed assessment on the compliance with Basel Core Principles).
Some declines in the capital ratio in 2010 partly reflect the repayments of subordinated debts by major state owned banks, originally injected for capital support purposes by the government.
The data on restructured loans include loans with any changes with the original terms of loans, including changes with interest rates, payment frequency, and maturity. The restructuring records are collected only for large loans.
The actual estimation and simulation work for this framework is implemented by an external consulting company with which the CBR has been working closely for a long time on macroeconomic modeling. The core of the framework is a system of about 30 econometric equations covering the real, external, fiscal, and financial sectors, estimated using 1998-2010 data.
Recent FSAPs to the U.S. and major EU countries often covered five years, and European Banking Authority’s annual exercise covers two years.
Established macroeconomic models that are often used by central banks for macroeconomic assessment (especially real business cycle models) usually ignore the financial sector, reflecting the tradition of macroeconomics that implicitly assume Arrow-Debreu type perfect and complete market for financial risks. Such assumption reduces the role played by the financial intermediation sector. For instance, the IMF’s Dynamic Stochastic General Equilibrium (DSGE) model, Global Integrated Monetary and Fiscal model (GIMF) includes interest rates, but does not include banks’ capitalization measures or NPL ratio. For this reason, FSAP teams often use separate satellite models to establish linkages between macro variables that come out of GIMF or other macro models used for building macroeconomic scenario and risk parameters needed for stress testing, including NPL ratio or probability of default. See, for instance, the technical note on U.S. FSAP stress testing http://www.imf.org/external/pubs/cat/longres.cfm?sk=24101.0
Another macroeconomic modeling approach that is often used for stress testing is time-series models such as a vector auto-regression (VAR) models and other time-series variations, including vector error correction model (VECM). These statistical models can build various scenarios originating from structural shocks easily. For instance, standard VAR or VECM packages easily allow projecting the macroeconomic impact of a 1 standard (or other) deviation decrease in oil prices on other variables such as GDP. Also, an overall impact of a shock on GDP while keeping some external variables (such as foreign interest rate and oil prices) at a given level could be projected fairly easily. For instance, STATA command such as var and fcast could be set up in a loop where, in each loop, exogenous variables and endogenous variables set by outside-of-the model priors (e.g. exchange rate or policy rate) are inserted to overwrite or add onto variables forecasted using the VAR/VECM.
Such as the breakdown of liquid assets by major issuers.
Russian banks are currently regulated with Basel II, standardized approach. No specific dates are given for the adoption of more advanced approach. As of April 2010, the CBR does not have any plan to adopt Basel III framework. Therefore, we did not examine capital adequacy with Basel III metrics in detail.
Deposits with all maturities are included, as maturity ladder data are not available. In any case, in the past, deposits were withdrawn before maturity at the time of liquidity stress events as there are little legal and financial costs to do so.
Corporate and household loan NPL ratios increases to 1.65 standard deviation + historical average (i.e. an increase to the highest 5 percentiles of historical distribution of NPL ratios).
The model could not be expanded to include multiple years within the current FSAP schedule. However, the CBR and Prognoz indicated the possibility to expand the horizon with sufficient time.
Credit shock (A) in table 5 is what the CBR typically tests. In shock (A), 5 percent tail point of the empirical distribution of NPL ratio is taken by applying 1.65 standard deviation shock on historical average NPL ratio. The marginal size of the shock differs depending on where the current actual NPL ratio is compared to historical average. When the actual data is already 1 standard deviation above the average, the marginal shock size is a 0.65 standard deviation. When the actual data is 1 standard deviation below the average, the marginal shock size is a 2.65 standard deviation. While such a calibration has reasonable foundation, it makes time-series comparison of stress test results difficult, by varying the size of the shock each year. Therefore, a different type of shocks is calibrated. Shock (B) gives a 1.65 standard deviation shock onto actual NPL ratio data, maintaining the size of the marginal shock. Using the current data, shock (A) increases headline NPL ratio by 5 percentage points and shock (B) increases it by 8 percentage points (Figure 2).
The adjustment included for the stress testing is based on a rough proxy measure taking the difference of provisions between July 2010 (the last month when forbearance measure was applied) and August 2010. Apparently, this is a very noisy measure, and it could be influenced by a range of other factors such as change in underlying credit quality. In addition, it might not capture the effects with loans carrying grandfathering effects of the forbearance. Having said that, the estimated impact with this methodology—about 2 percent of capital (Table 6)—is in line with the other “bottom-up” estimate built on internal estimates by each major bank (last bullet, paragraph 7), giving some comfort with the estimate.
Table 6 shows only the impact of each type of shocks in percent of capital for single factor tests, without showing post-shock CAR, reflecting the view that the resilience of the system (adequacy of CAR) is better be addressed in using macro scenario shock. Therefore the corresponding CAR and capital shortfall data are provided only for macro scenario tests and test of combined single-factor shocks.
The assumed LGD is extremely conservative at 100 percent, contributing to these severe results.
The losses from liquidity risks are quite different between macro scenario and sensitivity tests. This mainly reflects the difference of time horizon considered in the two tests, and resulting difference regarding the severity of the assumptions and policy measures incorporated in the tests (see Appendix for details).