## _cr1507

## Source details

**Canonical URL:** [_cr1507](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/scr/2015/_cr1507.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/scr/2015/_cr1507.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/scr/2015/_cr1507.pdf.json)

---

### EXECUTIVE SUMMARY — INTRODUCTION AND OVERVIEW
- Exercise measured solvency, liquidity, and contagion risks under three scenarios: a baseline and two adverse scenarios.  
- ST framework based on the modified “The Next Generation Balance Sheet Stress Testing” framework (Schmieder et al, 2011) and a simple cash-flow-based liquidity model.  
- Banking system dominated by banks: number of commercial banks declined from 102 (1995) to 21 banks in 2013. The two largest banks account for 58 percent of assets.  
- NBFIs (70 microfinance institutions, 14 insurance companies, and five pension funds as of June 2014) combined have shares of less than 5 percent of GDP and were excluded.

### STRUCTURE OF BANKS' BALANCE SHEETS AND INCOME
- Assets: loans are the largest item; about 22 percent of assets are very liquid (cash and correspondent accounts), covering almost 40 percent of all deposits.  
- Liabilities: deposits dominate funding; a majority of deposits are below maturity of one month.  
- Income: net interest income constitutes about 70 percent of banks’ income; fees and commissions account for 10 percent. Trading income is small and mainly from currency conversion operations.  
- Dominance of fixed-rate loans of longer maturity implies vulnerability to sudden increases in funding costs.

### KEY FINANCIAL SOUNDNESS INDICATORS (selected figures from 2010–13)
- Nonperforming loans (national definition, in % of total loans): 2010 Dec. 12.5; 2011 Dec. 8.6; 2012 Dec. 9.3; 2013 Dec. 7.5.  
- Nonperforming loans (90-day overdue definition, in % of total loans): 2010 Dec. 5.4; 2011 Dec. 4.6; 2012 Dec. 3.7; 2013 Dec. 3.1.  
- Loans in foreign exchange (in % of total loans): 2010 Dec. 74.0; 2011 Dec. 68.8; 2012 Dec. 67.5; 2013 Dec. 62.1.  
- Specific provisions (in % of total loans): 2010 Dec. 6.5; 2011 Dec. 4.6; 2012 Dec. 4.6; 2013 Dec. 3.8.  
- Deposit dollarization (residents and non-residents, in %): 2010 Dec. 68.6; 2011 Dec. 63.3; 2012 Dec. 66.0; 2013 Dec. 63.6.  
- Deposit dollarization (residents, in %): 2010 Dec. 65.0; 2011 Dec. 58.6; 2012 Dec. 60.4; 2013 Dec. 57.8.  
- Capital adequacy ratio (national definition, in %): 2010 Dec. 17.4; 2011 Dec. 17.1; 2012 Dec. 17.0; 2013 Dec. 17.2.  
- Tier 1 Capital Ratio: 2010 Dec. 13.7; 2011 Dec. 11.3; 2012 Dec. 13.4; 2013 Dec. 13.0.  
- Liquidity ratio (liquid assets to all deposits plus other liabilities with 6-month and shorter maturity, in %): 2010 Dec. 38.7; 2011 Dec. 37.3; 2012 Dec. 39.8; 2013 Dec. 41.8.  
- Net open foreign exchange position (in % of regulatory capital): 2010 Dec. 8.1; 2011 Dec. 5.9; 2012 Dec. 3.3; 2013 Dec. 2.1.

### MAIN VULNERABILITIES AND RISKS
- Credit and funding risks related to dollarization, concentration in the banking sector, and reliance on nonresident deposits.  
- Dollarization-specific challenges:
  - Most USD borrowers are unhedged (income and expenditures in national currency), especially households.  
  - NBG has limited ability to provide liquidity support in USD and other foreign currencies.  
  - Current risk weights for FX loans are topped at 175 percent (changes cited: reduced from 200 to 175 percent in September 2008, to 150 percent in August 2009, and raised to 175 percent in January 2011).  
- Concentration: default by the largest three borrowers would require additional capital of GEL 50 million for five banks.  
- Funding risks: 15 percent of bank funding came from nonresident deposits and 25 percent from wholesale funding as of end-2013; domestic banks without parent support might face higher USD liquidity needs.  
- High cyclicality of income and dominance of fixed-rate long-maturity loans increase vulnerability to funding-cost shocks.

### STRESS TESTS — DESIGN, SCENARIOS, AND MACRO ASSUMPTIONS
- Data and horizon:
  - Solo data as of end-December 2013; historical data for 2006–13 used for solvency ST models.  
  - ST time horizon: three years (quarterly results during 2014–2016) for top-down; two years (2014–2015) for bottom-up.
- Approaches:
  - Top-down STs by FSAP team and NBG; bottom-up STs by the four largest banks (73 percent of assets).  
  - FSAP built a satellite model for credit-risk-related losses, a sensitivity-based income model, and a modified Next Generation Balance Sheet ST framework.
- Hurdle rates and Basel phasing (as presented):
  - Capital adequacy assessment in 2014: 8 percent rate for Tier I capital and 12 percent for total capital.  
  - Basel II/III phasing table (as presented):
    - 2014 (in percent): Tier I Capital 88.58.5 ; Total Capital 1210.510.5 ; Phasing-out (Total CAR) 100
    - 2015 (in percent): Tier I Capital 8.5 ; Total Capital 10.5 ; Phasing-out (Total CAR) 95
    - 2016 (in percent): Tier I Capital 8.5 ; Total Capital 10.5 ; Phasing-out (Total CAR) 90
    - 2017 (in percent): Tier I Capital 8.5 ; Total Capital 10.5 ; Phasing-out (Total CAR) 80
- Scenarios (three-year horizon 2014–2016; baseline, mild recession, extreme recession) — key macro assumptions (summary from Table 4):
  - GDP growth
    - Baseline: 2014: 5% ; 2015: 5% ; 2016: 5%
    - Mild recession: 2014: -5% ; 2015: 0% ; 2016: +3%
    - Extreme recession: 2014: -10% ; 2015: -3% ; 2016: +1%
  - Interest rates
    - Baseline: Unchanged (10%)
    - Mild recession: 2014: +5 p.p. (15%) ; 2015: Unchanged compared to 2014 ; 2016: Unchanged compared to 2015
    - Extreme recession: 2014: +8 p.p. (18%) ; 2015: Unchanged compared to 2014 ; 2016: Unchanged compared to 2015
  - Exchange rate
    - Baseline: Unchanged (10%)
    - Mild recession: Depreciation by 10%
    - Extreme recession: Depreciation by 30%
  - Change in property prices
    - Mild recession: 2014: -10%
    - Extreme recession: 2014: -35%
  - Credit growth
    - Baseline: 2014: 14% ; 2015: 9% ; 2016: 9%
    - Mild recession: 2014: 0% ; 2015: 0% ; 2016: 0%
    - Extreme recession: 2014: -5% ; 2015: -5% ; 2016: -5%
  - Liquidity shock
    - Mild recession: Increase in average funding costs by 1 p.p.
    - Extreme recession: Run on deposits and increase in average funding costs by 2 p.p.
- Exchange-rate calibration note:
  - Daily USD/GEL rate volatility around 7 percent; multiplied by 3 gives 21 percent under 99.7 percent confidence; additional 9 percentage points added to arrive at 30 percent depreciation scenario.

### SOLVENCY STRESS TEST METHODOLOGY AND RESULTS
- Methodology:
  - Top-down FSAP model links macro developments to credit risk via: satellite NPL model, sensitivity-based income model, and balance-sheet CAR model (modified Next Generation framework). Static balance-sheet assumption, tax rate 20 percent, no dividends, no external capital in scenarios. Provisioning conservatively fixed at 100 percent for adverse scenarios (alternative 66.4 percent noted).  
  - Complementary Basel II standardized and quasi-IRB approaches used for cross-checks.
- Satellite model (FSAP specification):
  - LNPLS = -1.5 + 0.7*LNPLS(-1) + 0.056*LINT - 0.033*GDP_R(-1) + CX_i
  - Estimation: Panel GMM; sample 2006Q2–2013Q4; periods 31; cross-sections 18; observations 477; R-squared 0.824392.
- Top-down quantitative findings:
  - Baseline: system Tier I capital increases reflecting high profitability and declining NPLs.  
  - Mild scenario: four banks would need additional capital; total recapitalization needs close to GEL 120 million, corresponding to 0.4 percent of GDP.  
  - Extreme scenario: eleven banks would not meet minimum Tier I CAR; total recapitalization needs close to GEL 600 million (1.7 percent of GDP).  
  - Single-factor shocks (interest, exchange rates) have limited direct impact on the banking book; direct losses from GEL depreciation are not large due to small net open positions.
- Bottom-up results (four largest banks; B-U ST):
  - Three banks out of four would maintain minimum Tier I and total CAR under all scenarios.  
  - Under current regulatory minimum (8 percent Tier I), one bank falls below minimum Tier I CAR marginally (by less than 85 basis points) under extreme recession in 2015. Under new Tier I minimum of 8.5 percent (starting 2015), difference 135 basis points and still only one bank fails.  
  - B-U results contrast with top-down results mainly due to optimistic income assumptions and different treatment of accrued interest on NPLs.
- Basel II standardized and IRB cross-checks:
  - Basel II Standardized: even under severe shock, total Tier I CAR would be above minimum; however, four individual banks would fall below regulatory minimum and recapitalization needs close to GEL 100 million.  
  - IRB-based: adequate buffers under baseline and mild scenarios; under severe shock and conservative LGD (100 percent) recapitalization needs double compared to Basel I; with LGD 66.4 percent losses about half of 100 percent LGD scenario. Constraints: absence of real PDs and LGDs, proxy methods used (write-offs and recovery rates).
- Concentration risk:
  - Default of three largest exposures would require GEL 50 million recapitalization for five banks (assumes independence of largest borrowers across banks).

### LIQUIDITY STRESS TESTS, NETWORK/CONTAGION, AND RESULTS
- Liquidity ST scenarios and assumptions:
  - Liquidation values: cash and funds at central bank, interbank deposits, deposits at foreign banks, foreign government securities, and foreign bonds at 100 percent; domestic bank deposits, domestic government securities, domestic bonds, other liquid assets at 60 percent. Interbank market assumed closed; 50 percent of committed credit lines drawn.
  - Scenario 1 (Withdrawal of nonresident funding and closure of foreign funding markets): overnight withdrawal of wholesale funding with variants (A) 10 percent of liabilities coming due within three months; (B) 10 percent of all liabilities; (C) 50 percent of liabilities coming due within three months; assume no customer deposit withdrawals.
  - Scenario 2 (Bank run): deposit withdrawals over 5- and 30-day horizons with variants:
    - (A) 10 percent residents; 40 percent nonresidents; 100 percent domestic interbank deposits.
    - (B) 30 percent residents; 80 percent nonresidents; 100 percent domestic interbank deposits.
  - Scenario 3 (Combination): combinations of scenarios 1(A/B/C) with 2(A).
- Historical calibration:
  - Outflow without interbank deposits from July 2008 to 2009 was 31.9 percent; nonresident deposits were most volatile with close to half withdrawn in 2008–09.
- Liquidity ST results (selected numeric outcomes; as presented):
  - Scenario 1(B): Sum of Liquidity Shortages — (84,562,201) GEL; (10,356,295) FX; Number of Banks with Shortages — GEL: 4; FX: 1.
  - Scenario 2(B): Sum of Liquidity Shortages — (2,148,134) GEL; (568,108,734) FX; (114,909,709) Total 1/; Number of Banks with Shortages — GEL: 1; FX: 12; Total 1/: 6.
  - Scenario 3(A): Sum of Liquidity Shortages — (85,151,709) GEL; (10,356,295) FX; Number of Banks with Shortages — GEL: 4; FX: 1.
  - Scenario 3(B): Sum of Liquidity Shortages — (88,964,799) GEL; (10,356,295) FX; Number of Banks with Shortages — GEL: 5; FX: 1.
  - Scenario 3(C): Sum of Liquidity Shortages — (87,509,744) GEL; (10,356,295) FX; Number of Banks with Shortages — GEL: 4; FX: 1.
  - Note 1/: "Based on total liquidity, i.e., including domestic and FX reserves."
- Key liquidity findings:
  - Market funding shocks (overnight wholesale withdrawal) produced no banks with liquidity shortages in both currencies in tested variants.  
  - Retail funding shock (Scenario 2(B)): 12 banks might need additional liquidity support in FX, while only one bank would fail under GEL liquidity shock; considering total liquidity, only six banks would fail.  
  - Basel III LCR assessment: Only in the most extreme cases do banks not meet LCR requirement (100 percent) after stress.
- Interbank contagion and network:
  - Interbank market is shallow; two systemically important banks are widely connected.  
  - Interbank ST: four small banks might not meet minimum CAR if one systemically important bank fails; single counterparty risk for some small banks exceeds 100 percent of their regulatory capital.  
  - Overall, contagion via balance-sheet exposures is limited given current network and exposures.

### MODELING, DATA LIMITATIONS, AND CALIBRATION NOTES
- Time-series limitations: data cover at least one business cycle and the 2008–09 crisis, but recovery from 2008–09 is incomplete; forecasting losses three years ahead is challenging.  
- FX shocks are large and non-linear; actual deterioration of loan portfolio quality after a currency shock could be higher than forecasted.  
- PD/LGD estimation constraints: no bank applies IRB; proxy PDs calibrated from write-offs and NPLs with average through-the-cycle ratio 10.88 percent; proxy LGD based on 2012 recovery rate RR = 33.6 percent implying LGD = 66.4 percent; alternative conservative LGD = 100 percent used.  
- Income modeling: simplified sensitivity model used due to limited data; assumptions include net spread declines of 1 p.p. (mild) and 2 p.p. (extreme); non-interest income declines of -15 percent (mild) and -30 percent (extreme); OPEX unchanged in shocks.  
- Static balance-sheet assumption: nonperforming loans not replaced by new loans except via credit growth assumptions; accrued interest treatment differs across bottom-up and top-down approaches (FSAP top-down excludes accrued income from NPLs).

### POLICY FINDINGS AND RECOMMENDATIONS (prioritized)
- High, Immediate:
  - Discuss with banks their ST models and assumptions to ensure bottom-up ST results are conservative and consistent with scenarios.  
  - Continue to build-up capital buffers to address non-linearity of shocks related to lari depreciation.  
  - Closely monitor foreign currency-related funding risks and continue de-dollarization policies; collect additional cash-flow-based data.
- Medium, Near-term (1–3 years):
  - Develop a comprehensive, multi-period ST framework that incorporates the income model and assumptions about interbank market contagion and links between liquidity and solvency STs; calibrate existing models.  
  - Start preparing for advanced Basel II/III approaches and collect needed data (PDs, LGDs, collateral valuation).  
  - Refine liquidity ST framework by including more detailed cash-flow-based liquidity reporting data (in addition to LCR).  
  - Ensure links between solvency and liquidity risks include scenarios with higher funding costs due to retail deposit withdrawals.
- Other policy options and context:
  - Macroprudential measures making FX lending more expensive (e.g., higher risk weights to FX loans) are being implemented.  
  - If crisis materializes, NBG could lower capital requirements to maintain countercyclical policy.  
  - Further diversification of funding sources and de-dollarization could help minimize vulnerabilities.  
  - Prepare contingency plans to address liquidity and solvency problems at individual bank and system levels.

### RISK ASSESSMENT MATRIX — MAIN THREATS (selected)
- Increasing geopolitical tensions (e.g., Ukraine): Likelihood: High; Expected impact: High.  
- Capital outflow shock: Likelihood: Medium; Expected impact: High.  
- Foreign exchange shock: Likelihood: Medium; Expected impact: High.  
- Global financial conditions and slower growth abroad: Likelihood: Medium; Expected impact: Medium.

*Source: IMF FSAP mission materials (EXECUTIVE SUMMARY (Georgia) and FSAP chapter excerpt, solo data as of end-December 2013).*

### EXECUTIVE SUMMARY ___________________________________________________________________________ 5

### EXECUTIVE SUMMARY

### INTRODUCTION AND OVERVIEW
- The exercise measured solvency, liquidity, and contagion risks under three scenarios: a baseline and two adverse scenarios.  
- The ST framework applied in Georgia was based on the modified “The Next Generation Balance Sheet Stress Testing” framework (Schmieder et al, 2011) and a simple cash-flow-based liquidity model.  
- The Georgian financial system is dominated by banks: number of commercial banks declined from 102 (1995) to 21 banks in 2013. The two largest banks account for 58 percent of assets.  
- Nonbank financial institutions (NBFIs), 70 microfinance institutions, 14 insurance companies, and five pension funds (as of June 2014) combined have shares of less than 5 percent of GDP and were excluded from the exercise.

### STRUCTURE OF BANKS' BALANCE SHEETS AND INCOME
- Assets: loans are the largest item; about 22 percent of assets are very liquid (cash and correspondent accounts), covering almost 40 percent of all deposits.  
- Liabilities: deposits dominate funding; a majority of deposits are below maturity of one month.  
- Income: net interest income constitutes about 70 percent of banks’ income; fees and commissions account for 10 percent. Trading income is small and mainly from currency conversion operations rather than securities trading.  
- Dominance of fixed-rate loans of longer maturity implies vulnerability: a sudden increase in funding costs would squeeze profit margins.

### KEY FINANCIAL SOUNDNESS INDICATORS (selected figures from 2010–13)
- Nonperforming loans (national definition, in % of total loans): 2010 Dec. 12.5; 2011 Dec. 8.6; 2012 Dec. 9.3; 2013 Dec. 7.5.  
- Nonperforming loans (90-day overdue definition, in % of total loans): 2010 Dec. 5.4; 2011 Dec. 4.6; 2012 Dec. 3.7; 2013 Dec. 3.1.  
- Loans in foreign exchange (in % of total loans): 2010 Dec. 74.0; 2011 Dec. 68.8; 2012 Dec. 67.5; 2013 Dec. 62.1.  
- Specific provisions (in % of total loans): 2010 Dec. 6.5; 2011 Dec. 4.6; 2012 Dec. 4.6; 2013 Dec. 3.8.  
- Deposit dollarization (residents and non-residents, in %): 2010 Dec. 68.6; 2011 Dec. 63.3; 2012 Dec. 66.0; 2013 Dec. 63.6.  
- Deposit dollarization (residents, in %): 2010 Dec. 65.0; 2011 Dec. 58.6; 2012 Dec. 60.4; 2013 Dec. 57.8.  
- Capital adequacy ratio (national definition, in %): 2010 Dec. 17.4; 2011 Dec. 17.1; 2012 Dec. 17.0; 2013 Dec. 17.2.  
- Tier 1 Capital Ratio: 2010 Dec. 13.7; 2011 Dec. 11.3; 2012 Dec. 13.4; 2013 Dec. 13.0.  
- Liquidity ratio (ratio of liquid assets to all deposits plus other liabilities with 6-month and shorter maturity, in %): 2010 Dec. 38.7; 2011 Dec. 37.3; 2012 Dec. 39.8; 2013 Dec. 41.8.  
- Net open foreign exchange position (in % of regulatory capital): 2010 Dec. 8.1; 2011 Dec. 5.9; 2012 Dec. 3.3; 2013 Dec. 2.1.

### MAIN VULNERABILITIES AND RISKS
- Credit and funding risks related to dollarization, concentration in the banking sector, and reliance on nonresident deposits.  
- Dollarization-specific challenges:
  - Most USD borrowers are unhedged (income and expenditures in national currency), especially households.  
  - The NBG has limited ability to provide liquidity support in USD and other foreign currencies.  
  - Current risk weights for FX loans are topped at 175 percent (note: risk weight changes cited in table footnote: reduced from 200 to 175 percent in September 2008, to 150 percent in August 2009, and raised to 175 percent in January 2011).  
- Concentration: default by the largest three borrowers would require additional capital of GEL 50 million for five banks.  
- Funding risks: domestic banks without access to parent support might experience higher liquidity needs, especially in USD; NBG’s ability to provide USD liquidity is limited given level of foreign reserves.  
- High cyclicality of income and dominance of fixed-rate long-maturity loans increase vulnerability to funding-cost shocks.

### STRESS TESTS: DESIGN AND RESULTS
- Separate stress tests were performed by the NBG and by the FSAP mission. Tests covered scenarios ranging from slow growth to severe macroeconomic shocks, and included solvency, liquidity, and contagion tests.  
- Overall system resilience:
  - The banking system as a whole is able to withstand severe shocks; most banks maintain healthy capital buffers well above regulatory minimum.  
  - Taking into account current profits and introduction of Basel II, major banks would generally remain adequately capitalized.  
  - In adverse scenarios, recapitalization needs are manageable in terms of GDP: 1.6 percent for the worst-case scenario.  
  - Market risks are very limited; trading books do not exist.  
- Vulnerable banks: some banks exceed minimum capital requirement by only a few percentage points, limiting loss-absorption capacity. These banks need to strengthen capital buffers and mitigate funding risks.  
- Liquidity stress test highlights:
  - A majority of deposits are short-term (below one month), elevating funding risk.  
  - Loan-loss provisions (LLPs) cover around 90 percent of NPLs, higher than international average of 60 percent.  
  - Crisis experience (2008–09) showed deposit withdrawals and the need for contingency plans to address liquidity and solvency at both individual bank and system levels.

### MODELING AND DATA LIMITATIONS
- STs are subject to data limitations and modeling challenges relating to FX-related risks.  
- Time-series limitations: data cover at least one business cycle and the 2008–09 crisis, but recovery from the 2008–09 crisis is still not complete; the length of the recovery period is uncertain. Forecasting losses three years ahead is challenging given short time series relative to forecast horizon.  
- FX shocks are large (initial overshooting) and non-linear; actual deterioration of loan portfolio quality after a currency shock could be higher than forecasted.  
- Recommendation in analysis: develop a comprehensive, multi-period ST framework that incorporates the income model and assumptions about interbank market contagion and links between liquidity and solvency STs; calibrate existing models.

### POLICY FINDINGS AND RECOMMENDATIONS (summarized from Table 1)
Priority: High, Immediate
- Discuss with banks their ST models and assumptions to ensure bottom-up ST results are sufficiently conservative and consistent with scenarios.  
- Continue to build-up capital buffers to address non-linearity of shocks related to lari depreciation.  
- Closely monitor foreign currency-related funding risks and continue de-dollarization policies; collect additional cash-flow-based data.

Priority: Medium, Near-term (1–3 years)
- Develop a comprehensive, multi-period ST framework that incorporates the income model and assumptions about interbank market contagion and links between liquidity and solvency STs; calibrate existing models.  
- Start preparing for advanced Basel II/III approaches and collect needed data.  
- Refine liquidity ST framework by including more detailed cash-flow-based liquidity reporting data (in addition to LCR).  
- Ensure links between solvency and liquidity risks are based on scenarios that involve higher funding costs due to withdrawal of retail deposits.  

Other policy context and options:
- Macroprudential measures are being implemented to make FX lending more expensive for banks (e.g., higher risk weights to FX loans).  
- If a crisis were to materialize, the NBG would have the option to lower capital requirements to maintain countercyclical policy.  
- Further diversification of funding sources and de-dollarization could help minimize identified vulnerabilities.  
- Preparation of contingency plans to deal with liquidity and solvency problems of individual banks and the whole system is justified by past crisis experience.

*Source: EXECUTIVE SUMMARY (Georgia), IMF Financial Sector Assessment Program mission materials.*

### 5.      Notwithstanding positive macroeconomic developments, the banking system faces

### _cr1507 - 5.      Notwithstanding positive macroeconomic developments, the banking system faces

### Key risks and vulnerabilities
- Credit risks related to dollarization.
- Liquidity risks given weaknesses of safety net and crisis management framework.
- Concentration in the banking sector due to dominance of two banks.
- Key identified shocks to the financial system:
  - i) Regional geopolitical turmoil coupled with global recession due to fall in international trade and subsequent depreciation of lari.
  - ii) Loss of confidence in the banking system by nonresidents, which yields to a large withdrawal of nonresident deposits.

### Main transmission channels
- Banks’ exposure to FX unhedged domestic clients is the key transmission channel for GDP and lari exchange rate shocks.
- Mechanism:
  - Decline in export revenue → fall in exchange rate → increased debt burden for the corporate sector → increasing defaults.
- Dollarization and borrower exposure:
  - Roughly 60 percent of both deposits and loans are in foreign currency.
  - Peaks of around 75 percent in 2008–2009.
  - Over 90 percent of foreign currency borrowers have income sources in local currency and are unhedged against lari depreciation.
  - Household debt to GDP ratio is 27 percent.
- Central bank limitations:
  - NBG can only provide limited liquidity support in foreign currencies, complicating crisis management.

### Nonresident funding and liquidity transmission
- Increasing reliance on nonresident funding is the main transmission channel for liquidity shock from potential loss of confidence.
- Funding shares as of end-2013:
  - 15 percent of bank funding came from nonresident deposits.
  - 25 percent of bank funding came from wholesale funding.
- Historical stability: credit lines and nonresident deposits have proved relatively stable in past shock episodes, but some smaller banks remain vulnerable due to exposure to a few large depositors.

### Overview of Stress Tests (ST)
- Data and horizon:
  - ST exercise based on solo data as of end-December 2013.
  - Historical data for 2006–13 used for solvency ST models.
  - ST time horizon: three years (quarterly results during 2014–2016).
- Approaches applied and cross-checks:
  - Top-down STs by FSAP team and NBG; bottom-up STs by the four largest banks.
  - FSAP team built a satellite model for credit-risk-related losses, a sensitivity-based model for income before loan-loss provisions, and cross-checked with Hardy and Schmieder (2013) cross-country results.
  - Limitations noted for quantitative models based on relatively short time series.
- Hurdle rates and regulatory frameworks:
  - Capital adequacy assessment in 2014: 8 percent rate for Tier I capital and 12 percent for total capital.
  - CAR ratios for 2015 and 2016 based on new regulations.
  - Additional calculations compared results with upcoming Basel II standardized and quasi-IRB frameworks.
- Table of Basel II/III minimum capital adequacy ratio introduction (as presented):
  - 2014 (in percent): Tier I Capital 88.58.5 ; Total Capital 1210.510.5 ; Phasing-out (Total CAR) 100
  - 2015 (in percent): Tier I Capital 8.5 ; Total Capital 10.5 ; Phasing-out (Total CAR) 95
  - 2016 (in percent): Tier I Capital 8.5 ; Total Capital 10.5 ; Phasing-out (Total CAR) 90
  - 2017 (in percent): Tier I Capital 8.5 ; Total Capital 10.5 ; Phasing-out (Total CAR) 80

### Scenarios for solvency stress tests
- Time horizon: three-year horizon (2014–2016); scenarios: baseline, mild recession, extreme recession.
- Calibration: adverse scenarios use historical episodes reflecting regional geopolitical tensions; extreme scenario exceeds the 2008–09 shock by a wide margin.
- Exchange-rate considerations:
  - Historical USD/GEL volatility around 1.6-2 GEL per USD boundaries.
  - One-time devaluation in November 2008: USD/GEL depreciated by 16 percent.
  - Daily USD/GEL rate volatility is around 7 percent; multiplying by 3 gives 21 percent under 99.7 percent confidence; additional 9 percentage points added to arrive at 30 percent depreciation scenario.
- Summary of macroeconomic assumptions (Table 4 reproduced):
  - GDP growth
    - Baseline: 2014: 5% ; 2015: 5% ; 2016: 5%
    - Mild recession: 2014: -5% ; 2015: 0% ; 2016: +3%
    - Extreme recession: 2014: -10% ; 2015: -3% ; 2016: +1%
  - Interest rates
    - Baseline: Unchanged (10%)
    - Mild recession: 2014: +5 p.p. (15%) ; 2015: Unchanged compared to 2014 ; 2016: Unchanged compared to 2015
    - Extreme recession: 2014: +8 p.p. (18%) ; 2015: Unchanged compared to 2014 ; 2016: Unchanged compared to 2015
  - Exchange rate
    - Baseline: Unchanged (10%)
    - Mild recession: Depreciation by 10%
    - Extreme recession: Depreciation by 30%
  - Change in property prices
    - Baseline: Unchanged
    - Mild recession: 2014: -10% ; 2015: Unchanged compared to 2014 ; 2016: Unchanged compared to 2015
    - Extreme recession: 2014: -35% ; 2015: Unchanged compared to 2014 ; 2016: Unchanged compared to 2015
  - Credit growth
    - Baseline: 2014: 14% ; 2015: 9% ; 2016: 9%
    - Mild recession: 2014: 0% ; 2015: 0% ; 2016: 0%
    - Extreme recession: 2014: -5% ; 2015: -5% ; 2016: -5%
  - Liquidity shock
    - Baseline: Not included
    - Mild recession: Not included. Increase in average funding costs by 1 p.p.
    - Extreme recession: Run on deposits and increase in average funding costs by 2p.p.

### Impact channels in scenarios
- Credit risk drivers in mild and extreme scenarios:
  - Depreciation of the lari → higher NPLs → higher provisioning ratios due to fall in property prices.
  - High dollarization and mostly unhedged borrowers lead to sharp increase in NPLs.
  - Increased interest rates raise banks’ funding costs and lower profitability.
- Calibration:
  - Shocks on NPLs and credit growth calibrated using historical data.
  - In extreme recession, IMF team macro model output exceeds NPLs observed in 2009 peak.
- Cross-check against Hardy et al. (2013):
  - Applied emerging markets GDP elasticity parameters for Default Rate/NPL: Moderate shock -0.6 ; Medium shock -0.8 ; Severe Shock -1.5.
  - Extreme recession scenario treated as severe shock; extreme 10th percentile of elasticity also considered as a conservative approach.

### Profitability and funding-cost effects
- Net interest margins:
  - World Bank data: margin in 2009 was 7.1 p.p.; in 2012 it dropped to 4.1 p.p. on average.
- Funding-cost shocks:
  - Combination of solvency and liquidity risks produces an increase in average funding costs by 0.5 p.p. in mild shock scenarios and by 1 p.p. in extreme shock scenarios.

### Liquidity stress tests: scenarios and assumptions
- Key liquidity risks tested:
  - (i) dollarization; and (ii) significant share of nonresident funding.
- Scenario descriptions:
  - Withdrawal of wholesale funding and closure of foreign funding markets: wholesale funding not rolled over per shock parameters.
  - Bank run: deposit withdrawals of up to 30 percent; wholesale funding withdrawal of up to 30 percent; fire sales of liquid assets with haircuts up to 40 percent; 50 percent of committed credit lines drawn down; interbank market closed.
  - Combined shocks: withdrawal of nonresident funding plus bank-run shocks.
- Historical calibration:
  - 2008–09 external shocks and domestic tensions triggered large deposit outflows; nonresident deposits were the most volatile with close to half withdrawn in 2008–09.
  - Outflow without interbank deposits from July 2008 to 2009 was 31.9 percent.
- Table 5 summary of assumptions for liquidity STs (as presented):
  - Liquidation value (in percent) of cash and funds held at the central bank, interbank deposits, deposits held at foreign banks, foreign government securities, and foreign bonds: 100 percent.
  - Liquidation value (in percent) of deposits held at domestic banks, domestic government securities, domestic bonds, and other liquid assets: 60 percent.
  - Other assumptions for all scenarios:
    - Interbank market is closed.
    - Percent of committed credit lines that are drawn down: 50 percent.
  - Assumptions on liabilities withdrawals:
    - Scenario 1 (Withdrawal of nonresident funding and closure of foreign funding markets): Overnight withdrawal of wholesale funding.
      - (A) 10 percent of liabilities coming due within three months.
      - (B) 10 percent of all liabilities.
      - (C) 50 percent of liabilities coming due within three months.
      - In all cases, assume no withdrawals of customer deposits.
    - Scenario 2 (Bank run): Withdrawal of resident and nonresident deposits over 5- and 30-days horizons.
      - (A) A withdrawal of 10 percent of residents deposits; 40 percent of nonresidents deposits; and 100 percent of domestic interbank deposits.
      - (B) A withdrawal of 30 percent of residents deposits; 80 percent of nonresidents deposits; and 100 percent of domestic interbank deposits.
    - Scenario 3 (Combination): Combination of scenarios 1A and 2(A), 1B and 2(A), and 1C and 2(A).
      - (A) Combination of scenarios 1A and 2 (A).
      - (B) Combination of scenarios 1B and 2 (A).
      - (C) Combination of scenarios 1C and 2 (A).

### Solvency stress tests: methodology and approaches
- Framework:
  - ST methodology based on current capital adequacy requirements in Georgia, i.e., Basel I.
  - Loan-loss provisions are subtracted from capital and risk-weighted assets (RWAs).
- Complementary approaches:
  - Basel II standardized approach and quasi-IRB approach (quasi-foundation IRB) used to account for transition to Basel II.
  - For Basel I STs, main risk drivers were increases in NPLs; provisioning rates fixed at 100 percent (no collateral taken into account).
  - For Basel II STs, increases in NPLs complemented by migration in risk weights: migration from higher (100 percent for loans in GEL and 175 percent for loans in USD) to lower risk weights (50 percent) for the baseline scenario; migration from 50 percent of risk weights to 100 percent risk weights for adverse scenarios.
  - Quasi-IRB approach main driver: quasi-probabilities of default (PDs) that lead to higher capital requirements.
- Time coverage:
  - STs covered a two-year period (bottom-up) and a three-year period (top-down).
- Cross-checks and model choice rationale:
  - Inclusion of various cross-checks and an economic capital-based IRB model to address volatility and potential nonlinearity of losses in emerging markets.
  - Basel I and Basel II standardized approaches are less risk-sensitive compared to IRB-based approach; IRB reveals vulnerabilities more quickly.

*Source: IMF FSAP mission material (solo data as of end-December 2013; scenario parameters and assumptions as presented in the chapter).*

### 24.      While there is a danger that banks did not fully disclose the true condition of their

### _cr1507 - 24.      While there is a danger that banks did not fully disclose the true condition of their

### Data quality, provisioning, and asset-quality review
- While there is a danger that banks did not fully disclose the true condition of their loan books (due to ever-greening), the FSAP mission did not appear to face data quality and consistency issues.
- No additional assumptions about latent loan portfolio deterioration were made because the majority of banks experience a relatively high increase in NPLs (up to 3 or 4 times the original level).
- Loan collateral is not taken into account for provisioning purposes.
- NBG’s asset quality review system is well-suited to capture ever-green loans.
- Regulatory provisioning of NPLs varies from 30 percent to 100 percent.
- NBG does not collect data on loan collateral valuation; hence that data were not available.

### Solvency stress test (ST) general setup
- Solvency STs were based on static balance sheet and zero behavioral assumptions (static balance sheet assumes nonperforming loans are not replaced by new loans during the ST horizon).
- It is assumed that banks pay no dividends in all scenarios. Capital increase is based on internally generated sources (profits).
- During the baseline and adverse scenarios, banks do not increase capital from external sources.
- Profit is taxed at the current applicable tax rate (20 percent).

### Bottom-up stress tests (B-U ST)
- Bottom-up STs were performed by the four largest banks based on scenarios provided by the IMF and the NBG; the four banks constitute 73 percent of total banking system’s assets.
- Banks used their own ST methodology and financial projections, including assumptions about dividend payout; each bank performed STs under scenarios that differed from final top-down scenarios (addition of a third year and adjusted GDP growth rates).
- Scenario specifics for Bottom-Up Stress Tests (as provided):
  - Baseline: GDP growth 2014: 5%; 2015: 5%. Interest rates unchanged (10 percent). Exchange rate unchanged (10 percent). Change in property prices unchanged. Credit growth 2014: 14%; 2015: 9%. Liquidity shock: Not included.
  - Mild recession: GDP growth 2014: -5%; 2015: -3%. Interest rates 2014: +5 p.p. (15%); 2015: Unchanged compared to 2014. Exchange rate depreciation by 10%. Change in property prices 2014: -10%; 2015: Unchanged compared to 2014. Credit growth 2014: 0%; 2015: 0%. Liquidity shock: Run on deposits and increase in average funding costs by 1 p.p.
  - Extreme recession: GDP growth 2014: -10%; 2015: -0%. Interest rates 2014: +8 p.p. (18%); 2015: Unchanged compared to 2014. Exchange rate depreciation by 30%. Change in property prices 2014: -35%; 2015: Unchanged compared to 2014. Credit growth 2014: -5%; 2015: -5%. Liquidity shock: Run on deposits and increase in average funding costs by 2 p.p.
- Bottom-up ST results:
  - Three banks out of four would maintain minimum Tier I and total CAR under all scenarios.
  - Under current regulatory minimum (8 percent of Tier I), one bank falls below minimum Tier I CAR only marginally (by less than 85 basis points) under the extreme recession scenario, and only in 2015.
  - Under the new Tier I minimum of 8.5 percent (starting from 2015), the difference would be 135 basis points, and still only one bank fails under this higher threshold.
  - The same bank reported negative net income after loan-loss provisions and a substantial fall in net interest income; other banks reported positive income even after loan-loss provisioning in the STs.
- Limitations and concerns:
  - Bottom-up results sharply contrast with top-down results, largely due to ST methodologies and assumptions.
  - Analysis indicates banks are overly optimistic about their net interest and other income; in most cases net income after loan-loss provisions remains positive even under the extreme recession scenario.
  - Differences most likely stem from assumptions about accrued interest income: FSAP top-down STs explicitly assumed all accrued income from NPLs is excluded from profit calculations, whereas many banks include income only when received as cash.
  - Bottom-up results are of limited use given limited ST experience and inability to fully trace losses and income assumptions; FSAP recommended NBG discuss banks’ models and assumptions to ensure conservative, scenario-consistent ST results.

### Market risk stress tests (B-U ST)
- Banks have very limited exposure to market risk; largest losses come from net open positions in USD.
- Among the four banks: two have short positions in FX and two have long positions; overall positions do not exceed 8 percent of total capital (largest).
- Securities portfolios among the four banks are limited; however, the largest exposure in the system is more than 110 percent of capital.
- Most securities are debt securities and are held to maturity (HTM); no bank reported available for sale (AFS) or trading book positions.
- Shock effects from market risks are limited on the four banks’ income over the ST horizon.
- Some very small banks (excluded from bottom-up ST) might experience higher impact on CAR but pose no systemic risk.

### FSAP top-down stress test (Modeling approach)
- ST model links Georgian macro developments to credit risk losses via three structural components:
  - (i) satellite model linking NPLs with macro variables;
  - (ii) sensitivity-based income model;
  - (iii) balance sheet model calculating CAR (modified and simplified Next Generation Balance Sheet Stress Testing framework).
- Framework allows application of Basel I, Basel II standardized, and IRB approaches. Banks must comply with Basel I now and Basel II Standardized by end-2014; none use IRB during the ST horizon.
- Model includes domestic exposure only (98.5 percent of Georgian banks’ exposure is domestic).
- Domestic loan portfolio composition: 55 percent to corporates and 43 percent to households.
- FX risk: 60 percent of loans are in foreign currencies, mostly in USD.
- Georgia experienced one significant depreciation episode based on official devaluation (16 percent); FX rate was managed by NBG to avoid excess volatility.

### Satellite model, forecasting, and provisioning
- NPLs ratio used as proxy for PD: share of total loan portfolio that became nonperforming (value of NPLs loans divided by total loan portfolio).
- GDP growth and interest rates were significant explanatory variables; exchange rate dynamics showed relatively little direct impact on NPLs.
- Because almost 60 percent of the loan portfolio is in FX, calculating direct shock of currency devaluation can yield significant non-linearity and underestimate total effect.
- Real GDP growth rate assumed as a systematic risk factor; devaluation usually follows decline in GDP and, as a result, interest rates go up.
- An alternative specification based on Hardy and Schmieder (2013) was used for robustness and was broadly in line with satellite model results.
- Forecast NPLs for baseline and adverse scenarios were calculated using macro data on GDP and long-term interest rates; quarterly modeling applied annual y-o-y shocks divided by 4 to reflect equal realization in each quarter (example: if annual drop in GDP is -10 percent, each quarter y-o-y real GDP growth rate is reduced by -2.5 p.p.). The same rule applied to long-term interest rates.
- Loan-loss provisions (credit losses) were calculated as differences (flows) between the stock of NPLs in two consecutive quarters multiplied by the provisioning ratio and exposure:
  - Flow of additional loan-loss provisions t+1_i = (NPLs_t+1_i - NPLs_t_i) * Provisioning ratio * Loan portfolio
- Provisioning assumptions: provisioning was based on current regulatory requirements; for ST purposes a fixed 100 percent provisioning rate was used (the most conservative approach).
- Note: It is possible to use different fixed provisioning ratios: 66.4 percent (Ratio was obtained as 1-Recovery ratio) for the baseline scenario and 100 percent for the adverse scenarios.

### Income before loan-loss provisions and other assumptions
- Net income is the first line of defense against loan portfolio losses.
- ST model included the following assumptions affecting net income before loan-loss provisions:
  - a) Interest rate shock, which affects net interest income (including inverse yield curve);
  - b) Credit growth: increases net interest income;
  - c) NPLs growth: declines net interest income.
- Further modeling assumptions are explained in Appendix V (not reproduced here).
- ST based on static balance sheet approach: over the ST horizon, composition changes according to scenario, and nonperforming/defaulted loans are not replaced by new loans except via credit growth assumptions.
- No distinction between trading and banking books for net interest income; trading book is very small because most assets are held until maturity.
- STs assumed all banks are price takers rather than price setters; margin changes assumed the same for all banks.
- No second-round effects were included (no disproportionate increase in funding costs from individual bank CAR deterioration).

### CAR calculation and concentration risk
- CAR change accounts for changes in both regulatory capital (numerator) and risk-weighted assets (denominator); additional loan-loss provisions reduce capital, while subtraction of the same provisions from RWAs reduces the denominator (in line with regulatory requirements).
- If net income after loan-loss provisioning is positive, 20 percent income tax rate is applied before this income is added to CAR.
- Portfolio concentration risks are relatively small and manageable:
  - Under the simple assumption that the largest exposure defaults and using current 100 percent provisioning rule, no bank needs additional capital.
  - Assuming default of three largest exposures, only five banks would need GEL 50 million recapitalization (assumes largest borrowers are independent for each bank, i.e., no contagion).
- Note on risk weights: Georgian banks operate under Basel I. Majority of GEL loans have 100 percent risk weight (exceptions: loans to financial institutions and sovereign entities); loans in FX have 175 percent risk weight. Based on this, explicit modeling of risk-weight migration was not needed.

### Top-down ST results and system resilience
- Solvency top-down ST results suggest banks are able to withstand considerable economic downturn; resulting capital shortfall during distress is relatively modest in terms of GDP.
- Key quantitative results and findings:
  - Baseline scenario: with all other factors fixed, the system’s Tier I capital would increase from 3 p.p. to 11 p.p. up to 16–24 percent range. This reflects high profitability, capital buffers, moderate credit growth, and declining levels of NPLs. Were credit growth to increase, some banks would need additional capital to support loan portfolio growth.
  - Adverse scenarios: banking system resilience is unevenly distributed; aggregated system-wide results mask vulnerabilities of some banks.
    - Mild scenario: four banks would need additional capital to be in line with minimum CAR. Total recapitalization needs are close to GEL 120 million, corresponding to 0.4 percent of GDP.
    - Extreme shock scenario: reveals vulnerabilities related to indirect FX shock–related credit risks and decline in profitability. Eleven banks would not meet minimum Tier I CAR. Total recapitalization needs are close to GEL 600million (1.7 percent of GDP).
    - Lowering risk weights (countercyclical policy) as a result of introducing Basel II and exclusion of collateral for provisioning purposes would unfreeze additional capital and lower recapitalization needs by almost half.
  - Single-factor shocks (interest, exchange rates) show very limited direct impact on the banking book. Impact on the trading book is negligible because most assets are held until maturity.
  - Direct losses from GEL depreciation are not large due to small net open positions.
  - Credit portfolio concentration risk ST: overall largest exposures are 16 percent of regulatory capital; only several small banks are particularly vulnerable to default on their three largest borrowers and may fail the test. No bank fails due to default of their largest borrower.

*IMF FSAP extract as provided in the source content*

### 44.      Basel II Standardized ST results reveal significant capital buffers that can be used to

### _cr1507 - 44.      Basel II Standardized ST results reveal significant capital buffers that can be used to

### Basel II Standardized stress test (STD) — key findings
- Results reveal that even under severe shock scenario, total Tier I CAR would be above minimum level (Figure 14).
- At the same time, when it comes to individual banks, four banks would fall below the regulatory minimum and total recapitalization needs are close to GEL 100 million.
- Footnote: "The Figure 15 already incorporates jump of Tier I CAR under Basel II: from 13 percent in Q4 2013 to almost 27 percent."

### IRB-based stress test results
- IRB-based ST results reveal that banks have adequate capital buffers to cover losses under baseline and mild scenarios; however, they will need additional capital under severe shock scenario.
- Constraints: "The absence of real PDs and LGDs hinders calculation of potential losses, but, nevertheless, it is possible to use proxy data to arrive at a very approximate estimation."
- LGD scenarios and implications:
  - Using 66.4 percent LGD, baseline and mild shocks results are more or less in line with Basel I ST results.
  - Under the severe scenario and the most conservative LGD approach (100 percent), recapitalization needs double compared to results under the Basel I.
  - "ST results under 66.4 percent of LGD provide considerably lower amount of potential losses—about twice as low as under the severe scenario using Basel I."
- Footnote on PD/LGD proxy methodology: write-offs are flow-based measure of PDs; although 100 percent LGD is too conservative (average LGD under Basel IRB approaches varies between 20–50 percent), proxy PD formula assumes 100 percent LGD due to flow of write-offs.

### NBG top-down stress tests — framework and differences with FSAP
- NBG framework: satellite model linking NPLs to macro variables (specification in Appendix IV).
- Key differences from FSAP satellite model:
  - Modeling approach: FSAP uses dynamic panel GMM with White correction; NBG uses simple OLS panel model.
  - Cross-section unit: FSAP uses banks as cross-sections; NBG uses economic sectors as cross-sections and system-wide NPLs.
  - Correction for different NPLs levels: FSAP uses individual NPL-level correction; NBG does not use individual correction. Correction coefficients used differently under baseline vs mild/adverse scenarios.
  - Explanatory variables: NBG model covers a wider array (GDP, interest rates, exchange rate, etc.); FSAP includes just interest rates and GDP.
- NBG top-down ST followed same macro scenarios and logic as FSAP ST but differs in:
  - Assumptions on net interest income before loan-loss provisions (FSAP more conservative, assuming larger share of loan portfolio would not generate interest income).
  - Satellite model specification: NBG includes FX rate and GDP directly; FSAP models assume implicit effect of devaluation via interest rates and GDP.

### NBG top-down ST — quantitative results
- Capital adequacy and recapitalization needs:
  - Baseline: No one bank needs additional capital for baseline scenario (the one bank that did not meet minimum CAR in Q4 2013 is already recapitalized).
  - Mild recession scenario: four banks might need up to GEL 30 million in additional capital.
  - Extreme recession scenario: 10 banks might need around GEL 550 million in additional capital.
  - "The ability to reduce risk weights for FX loans further lowers recapitalization needs."
- Countercyclical buffer and policy response are considered in scenarios (Figures 19–21 referenced).

### Reconciliation of top-down and bottom-up results
- Both top-down approaches reveal similar amounts of losses under all three scenarios.
- Bottom-up ST results show much lower risks and are inconsistent in some cases; bottom-up STs are unsatisfactory, attributable to lack of experience and sophistication in ST models.
- Differences in profitability assumptions among ST approaches are reflected in CAR evolution; FSAP top-down ST incorporates more severe assumptions about decline in banks’ profitability during shock period (decline in lending margins and rigidity of operational expenses).
- Basel II Standardized vs IRB results:
  - "Additional STs under Basel II Standardized and IRB approaches reveal conflicting messages; while, compared to main Basel I based model, capital needs under Basel II Standardized approach are small and capital needs under IRB approach are much higher."
  - Interpretation: "Basel II Standardized lowers risk weights, while the IRB approach is highly pro-cyclical."
  - Policy implication: This justifies current NBG policy to maintain higher risk weights for FX-denominated loans. Transition to Basel II Standardized involves risks if higher risk weights are fully abandoned and credit growth is fuelled. NBG plans either to reduce risk weights or introduce countercyclical buffer.

### Recommendations on solvency stress testing and policy
- High-level solvency assessment:
  - "Solvency ST results suggest that the Georgian banking system has strong capital buffers and is adequately capitalized against severe macroeconomic shocks."
  - Ongoing vulnerabilities: stock of FX loans and increasing dependency on nonresident funding.
  - ST results reveal banks would suffer most from decline in GDP coupled with increase in interest rates and shock to USD/GEL rate.
  - Policy example: "it might be straightforward prohibition of granting new FX loans to non-hedged borrowers."
- Need to continue building capital buffers as global and regional risks grow; probability of extreme scenario realization is growing due to regional tensions affecting trade and remittances.
- Enhancements to NBG solvency ST framework:
  - Develop models for forecasting income and credit growth — "Profit before loan portfolio losses have a high impact on banks’ CAR."
  - Recalculate ST model using either static panel and OLS or dynamic panel and GMM; exclude NPLs from loans to large corporates from satellite model to avoid double counting.
  - Link credit risk ST model with liquidity STs (e.g., via inverse yield curve and funding cost assumptions).
  - Include assumptions about migration of risk weights and concept of economic capital following Basel II/III implementation; address data gaps on PDs and LGDs and start collecting relevant information to support model validation, ICAAP, SREP, and ICG procedures.
- Bottom-up ST capacity: Banks need additional investments and guidance to develop macro STs, calibrate models, and use them for capital allocation and risk management; NBG in-house framework does not replace banks’ need to have their own models.

### Liquidity and network (contagion) stress tests — models
- Liquidity ST coverage:
  - Examine resilience to liquidity shocks including foreign currency and external funding.
  - Based on gross and net liquidity mismatch positions; NBG carried out liquidity STs under IMF-designed scenarios including all banks.
- Model approach and limitations:
  - Simple implied cash flow balance sheet approach; NBG reporting framework lacks full maturity ladder contractual and behavioral cash flows.
  - Market liquidity assumptions: liquidation value (in percent) of deposits held at domestic banks, domestic government securities, domestic bonds, and other liquid assets is 60 percent (haircut of 40 percent).
  - Funding liquidity shocks modeled via:
    - (i) retail funding — deposit withdrawals of residents and nonresidents; and
    - (ii) wholesale funding — residents and nonresidents, including funding from parent banks.
- Contagion risks: estimated using a simple balance sheet exposure-based network model; interbank market is shallow; no second round or higher round spillover effects beyond direct credit exposure losses.

### Liquidity stress test — results and numeric outcomes
- Market funding shocks:
  - Scenario 1 (overnight wholesale withdrawal): none of the banks face liquidity shortages in both domestic and foreign currencies (Table 7 summary indicator).
- Retail funding shock (Scenario 2(B)):
  - Assumptions: withdrawal of 30 percent of residents deposits, 80 percent of nonresidents deposits, including foreign bank deposits and 100 percent of domestic interbank deposits; 5 day withdrawal period; based on a shock higher than the largest withdrawal observed in the last decade.
  - Findings:
    - 12 banks might need additional liquidity support in FX, while only one bank would fail under GEL liquidity shock.
    - Taking into account total liquidity, only six banks would fail (i.e., liquidity shortages in FX would be offset by liquidity in GEL).
- Table 7 excerpts (as presented):
  - Scenario 1(B): Sum of Liquidity Shortages — (84,562,201) GEL; (10,356,295) FX; Number of Banks with Shortages — GEL: 4; FX: 1.
  - Scenario 2 (B): Sum of Liquidity Shortages — (2,148,134) GEL; (568,108,734) FX; (114,909,709) Total 1/; Number of Banks with Shortages — GEL: 1; FX: 12; Total 1/: 6.
  - Scenario 3 (A): Sum of Liquidity Shortages — (85,151,709) GEL; (10,356,295) FX; Number of Banks with Shortages — GEL: 4; FX: 1.
  - Scenario 3 (B): Sum of Liquidity Shortages — (88,964,799) GEL; (10,356,295) FX; Number of Banks with Shortages — GEL: 5; FX: 1.
  - Scenario 3 (C): Sum of Liquidity Shortages — (87,509,744) GEL; (10,356,295) FX; Number of Banks with Shortages — GEL: 4; FX: 1.
  - Note 1/: "Based on total liquidity, i.e., including domestic and FX reserves."
- Basel III LCR assessment:
  - LCR calculated before and after stress (Basel III 100 percent requirement). "Only in the most extreme cases, banks do not meet LCR requirement (100 percent) after stress."
- Interbank contagion:
  - "The shallow interbank market limits possible spillovers among Georgian banks, although two systemically important banks are most widely connected with smaller banks."
  - Exposure patterns: one small bank has relatively high exposure in terms of capital to one of the largest banks; the rest of exposures are within prudential limits.
  - Impact of hypothetical foreign financial market shock on Georgian banks is low due to low exposures to other countries and foreign financial institutions.

*IMF staff compilation from the cited stress testing section.*

### 64.      The interbank market serves two very important functions in a financial system

### The interbank market serves two very important functions in a financial system

### Interbank market functions and contagion channels
- The interbank market performs two functions:
  - a) distributes liquidity from banks with excess reserves to banks that need funding; and
  - b) provides opportunity for banks to invest excess liquidity for returns higher than government bonds.
- During crises the interbank market may not work properly and can transmit liquidity shocks via three mechanisms:
  - market prices (market contagion),
  - behavior of depositors (behavioral), and
  - balance sheet exposure (direct losses).

### Georgia: structure, past episode, and network findings
- In transition economies like Georgia, the two latter channels are usually important: contagion via behavior of depositors, and direct losses if systemically important banks fail.
- In 2008 the banking system experienced loss of confidence that led to almost 13 percent deposit withdrawal; the interbank market was “literally shut down.”
- NBG data indicate two systemically important banks that attract excess liquidity from smaller market players.
- The interbank structure shows these two banks have relationships with almost all banks in the system.
- Interbank market stress test (ST) assumptions: consider what happens if one of the systemically important banks faces liquidity/solvency issues.
  - Results revealed:
    - four small banks might not be able to meet minimum CAR were one of the systemically banks to fail;
    - there is little balance sheet contagion effect between systemically important banks;
    - in one case, single counterparty risk for small banks exceeds 100 percent of their regulatory capital.
- While PD for systemically important banks is low, some smaller banks need to improve risk management and impose prudential exposure limits to a single counterparty (as a percentage of their own capital).

### Liquidity stress testing findings and constraints
- Liquidity ST results reveal that banks’ liquidity is resilient to the extreme shocks, albeit additional precautionary liquidity buffers might be needed to cope with FX funding risks.
- In case of a system-wide deposit run, the NBG has limited ability to support banks in USD.
  - In Q1 2014 NBG had US$2.3 billion as official reserves. Hence, it is not obvious how much of these reserves can actually be used to support banks without severely affecting FX rate.
- The NBG has an adequate liquidity ST and monitoring framework and is able to calculate LCR for individual banks, but the liquidity ST framework needs refinement.
  - An effective framework should be based on an extensive maturity ladder-based reporting system including behavioral and contractual cash flows in both GEL and FX (mainly USD).
  - The NBG indicated it is working on this reporting framework, but it is not part of the liquidity ST framework yet.
  - The new reporting framework might be based on EU example and would allow NBG to better detect individual bank liquidity issues.
- Cash flow based liquidity stress testing shall evaluate banks’ funding risks in different currencies and allows testing of scenarios not available under concise LCR or balance-sheet-based reporting.
- Links between solvency and liquidity risks should include scenarios with higher funding costs due to withdrawal of retail deposits to capture tail-risk events observed in countries relying on wholesale and retail funding from nonresidents.

### Recommendations (from the document)
- Impose prudential exposure limits to a single counterparty as a percentage of banks’ own capital, particularly for smaller banks.
- Consider additional precautionary liquidity buffers to cope with FX funding risks.
- Refine liquidity ST framework toward maturity-ladder and cash-flow-based reporting covering behavioral and contractual cash flows in GEL and FX (mainly USD).
- Incorporate scenarios that link solvency and liquidity risks through higher funding costs from retail deposit withdrawals to capture tail-risk events.

### Stress test design and methodological details (selected)
- Solvency stress tests covered:
  - Bottom-up by banks (4 largest banks: Bank of Georgia, TBC Bank, ProCredit Bank, Liberty Bank) with market share 73 percent.
  - Top-down by NBG and Top-down by FSAP Team (All banks; market share 100 percent).
  - Stress test horizons: 2 years (2014–2015) for bottom-up; 3 years (2014–2016) for top-down.
  - Hurdle rates & Basel III: Basel I/NBG minimum CAR rates: 12 percent for total CAR and 8 percent for Tier I CAR. For 2015 and 2016: 8.5 percent for Tier I CAR (top-down specifications).
- Liquidity risk stress tests (NBG top-down):
  - Institutions included: All 21 banks; Market share 100 percent.
  - Stress test horizon: 5- and 30-day outflow.
  - Methodology: Implied cash-flow-based; bank-run type tests; separate stress tests for GEL and USD.
  - Risks: Bank runs: deposit withdrawals of up to 30 percent; wholesale funding withdrawal of up to 30 percent; fire sales with haircuts of up to 40 percent.
  - Presentation: Number of banks that fail; liquidity shortage in each currency (GEL and USD).
- Interest and exchange rate (market) risk tests:
  - Impact of 15 percent, 20 percent, and 25 percent sudden devaluation of the Lari against the US$ on bank profitability assessed in sensitivity analyses.

### Top-down satellite model for credit risk: specification and estimation results
- PD proxy options: (i) NPLs ratio (flow-based via differences of NPLs ratios between period t-1 and t0), (ii) loan-loss provisions (LLPs), (iii) loan write-offs.
- By definition: a loan is counted as nonperforming if it is past-due more than 90 days. LLPs vary from 30 percent to 100 percent based on days past due and collateral. Loan write-offs are net after recoveries.
- For econometric work, NPLs ratios transformed into logit format to bound PDs between 0 and 1.
- Final regression specification obtained:
  - LNPLS = -1.5 + 0.7*LNPLS(-1) + 0.056*LINT - 0.033*GDP_R(-1) + CX_i
  - Where LNPLs = logit transformed NPLs ratio; LINT = long-term interest rate; GDP_R = real GDP rate growth; CX_i = individual fixed effect.
- Estimation details:
  - Method: Panel Generalized Method of Moments (GMM) with fixed effects.
  - Sample (adjusted): 2006Q2 2013Q4.
  - Periods included: 31; Cross-sections included: 18; Total panel (unbalanced) observations: 477.
  - Instrument specification: C LNPLS(-2) LINT(-1) GDP_R(-2). White cross-section instrument weighting matrix.
- Coefficients and statistics (as reported):
  - C coefficient: -1.50348; Std. Error: 0.469758; t-Statistic: -3.20055; Prob.: 0.0015.
  - LNPLS(-1) coefficient: 0.704018; Std. Error: 0.063139; t-Statistic: 11.15028; Prob.: 0.
  - LINT coefficient: 0.056502; Std. Error: 0.027949; t-Statistic: 2.021597; Prob.: 0.0438.
  - GDP_R(-1) coefficient: -0.0335; Std. Error: 0.012568; t-Statistic: -2.66516; Prob.: 0.008.
  - Effects specification: Cross-section fixed (dummy variables).
  - R-squared: 0.824392; Adjusted R-squared: 0.816689.
  - Mean dependent var: -2.5949; S.D. dependent var: 1.395594.
  - S.E. of regression: 0.59752; Sum squared resid: 162.806.
  - Durbin-Watson stat: 1.820923; J-statistic: 2.26E-20; Instrument rank: 21.

### Simplified income sensitivity model and scenario calibrations
- FSAP used simplified back-of-the-envelope calculations for forecasted net interest income due to limited data.
- Net interest income and net interest spread assumptions:
  - Baseline: net spread assumed not to change.
  - Mild scenario: net interest rate spread will fall by 1 p.p.
  - Extreme scenario: net interest rate spread will fall by 2 p.p.
  - Scenario calibration based on events observed during the 2008-09 crisis.
- Assumptions regarding non-interest income and OPEX:
  - Non-interest income assumed to decline by -15 percent for the mild shock and -30 percent for the extreme shock.
  - Calibration note: during 2008–09 crisis non-interest income in the Georgian banking system declined by -31 percent for the month of July 2009 vs. July 2008 (average annual decline was 15 percent).
  - OPEX left unchanged for shock scenarios and set at 12.5 percent growth for the baseline (half the average growth rate for 2005–2013).
- Additional simplifying assumptions:
  - Deposits assumed not to change during a stress test horizon.
  - Change in loan amounts equals either credit growth for baseline or decline equal to proportionate increase in NPLs for adverse scenarios.
  - Instant pass-through of shock in loan and/or deposit interest rates given short average durations.

*Source: IMF mission materials as presented in the provided content.*

### Introduction of Basel II would allow banks to unfreeze significant amounts of capital. The

### Introduction of Basel II would allow banks to unfreeze significant amounts of capital. The

### Basel II Standardized Approach: impact on banks’ CAR
- The ST Approach was based on the assumption that Georgia implements Basel II by the end of this year.
- Basel II standardized approach would allow banks to save capital by having lower risk weights on most exposures.
- Assumption made that most asset classes would fall under 50 percent risk-weight category.
- This represents a significant reduction compared with current risk weights:
  - 100 percent weights for loans in GEL
  - 175 percent for loans in U.S. dollars and other foreign currencies
- Off-balance sheet exposures were included in credit loss calculations.
  - Average CCF factor used for Basel II Standardized approach is 35 percent.
  - Off-balance sheet exposure varies bank by bank from 0 percent to almost 50 percent from total credit exposure.
- All other assumptions (income, balance sheet projections, hurdle rates and other factors) are the same as in the Basel I based ST model.

### IFRS and provisioning implications
- Full introduction of the IFRS 39th standard (IFRS 9th standard in the future) would allow banks to reduce provisioning levels by deducting fair value of collateral (fair value of hedge).
- This change would be an additional source of capital (and decline in provisioning levels).

### Basel II IRB Approach: estimation of economic capital and procyclicality
- Basel II IRB approach allows estimation of procyclicality of capital requirements and credit portfolio sensitivity using an unexpected losses approach.
- Economic capital of a bank is estimated as the difference between unexpected and expected losses, taking into account non-linearity between macroeconomic shocks and losses (PDs).
- Unexpected losses (ULs) are risk-based capital adequacy measures; provisions cover expected losses and capital should cover ULs.
- It is reasonable to assume banks should hold capital equal to 99.9 percent confidence level under which losses would not exceed a certain amount.
- Using Basel II formula, capital requirements K are equal to:
  - K=[LGD*PDs(99.9%)-LGD*PDs]*Maturity Adjustment
- PDs(99.9%) is defined as:
  - PDs(99.9%) = Φ(Φ^-1(PD) + √ρ Φ^-1(0.999) ) / √(1-ρ)
  - where Φ is the standard normal cumulative distribution, PD is probability of default, Φ^-1 is inverse standard normal cumulative distribution at average and 99.9 percent confidence levels, and ρ is correlation coefficient between a systematic factor and asset returns.

### PD and LGD assumptions and calibration
- PDs were defined as the amount of write-offs divided by total amount of NPLs.
- None of the banks apply or are allowed to apply IRB models; “true” PDs are not available.
- FSAP mission calibrated proxy PDs using loan write-off data.
  - Based on historical data, average through-the-cycle ratio between NPLs and write-offs was calculated as 10.88 percent:
    - 10.88% = (∑ WR_i / ∑ NPL_i) over quarters k=2006Q1 to 2013Q4 (as presented)
  - Result: on average, 10.88 percent of NPLs are written-off per quarter.
  - Proxy PDs calculated for each bank i for each forecasted quarter t using:
    - PD_i,t = (WR_i,t / NPL_i,t) * 10.88%  (presented formulaic relationship)
  - To avoid zero write-offs, Basel II minimum transition PDs of 0.03 percent were used.
- LGD definition:
  - Based on World Bank data (recovery rate) and NBG conservative approach.
  - 2012 recovery rate (RR) for Georgia was 33.6 percent, implying proxy LGD = (1 - RR) = 66.4 percent.
  - Alternative NBG assumption: banks need to provision 100 percent of NPLs and do not take collateral into account.
  - Write-offs used as proxy for PDs already include assumptions about 100 percent LGD (i.e., write-offs are net of collateral value).
  - Both LGD approaches were included in the report.

### Portfolio and parameter choices for IRB estimation
- Portfolio correlation, maturity adjustments and other parameters were included based on bank-by-bank data availability and loan portfolio structure (corporate, retail, and SME positions) and average duration.
- Granularity adjustment and correlation between PDs and LGDs are not included in the calculations.
- All other assumptions are the same as in Basel I based ST: income, balance sheet projections, hurdle rates, and other factors.

### Key differences between Basel I, Basel II Standardized and Basel II IRB approaches (summary)
- PDs (Proxy PDs):
  - Basel I: NPLs/Total portfolio
  - Basel II Standardized: NPLs/Total portfolio
  - Basel II IRB: Write-Offs (flows)/Total portfolio
- LGDs (provisioning ratios):
  - Basel I: 100% for Baseline and 100% for adverse
  - Basel II Standardized: 100% for Baseline and 100% for adverse
  - Basel II IRB: 66.4% or 100% for both Baseline and adverse
- Risk weights:
  - Basel I: Current (100% and 175% for FX loans)
  - Basel II Standardized: 50% baseline scenario; 100% adverse scenarios
  - Basel II IRB: Model based
- Exposure at default:
  - Basel I: Net loan portfolio
  - Basel II Standardized: Net loan portfolio + CCF of 35%
  - Basel II IRB: Net loan portfolio + CCF of 75%

### Top-Down Satellite Model for Credit Risk: NBG Approach (Appendix IV)
- Model output (Panel Least Squares):
  - Dependent Variable: LNPL
  - Sample (adjusted): 1999Q2 2013Q2
  - Periods included: 55
  - Cross-sections included: 7
  - Total panel observations: 373
- Estimated coefficients (selected):
  - Constant: -0.372200 (Std. Error 0.086119, t-Statistic -4.321925, Prob. 0.0000)
  - D08: 1.045368 (Std. Error 0.189871, t-Statistic 5.505677, Prob. 0.0000)
  - NPL(-1)/[1-NPL(-1)]: 0.552825 (Std. Error 0.048268, t-Statistic 11.453280, Prob. 0.0000)
  - NPL(-2)/[1+NPL(-2)]: 0.245319 (Std. Error 0.047614, t-Statistic 5.152210, Prob. 0.0000)
  - GDP_gr: -0.020003 (Std. Error 0.007711, t-Statistic -2.594221, Prob. 0.0099)
  - IR_diff_gr (-1): 0.000461 (Std. Error 0.000148, t-Statistic 3.101952, Prob. 0.0021)
  - Depr(-1)-inf(-1): 0.013544 (Std. Error 0.004929, t-Statistic 2.747813, Prob. 0.0063)
- Model diagnostics:
  - R-squared 0.744591; Adjusted R-squared 0.736078
  - S.E. of regression 0.578104
  - F-statistic 87.45887; Prob(F-statistic) 0.000000; Durbin-Watson stat 1.902538
- Elasticities (Short Term / Long Term):
  - GDP_gr: -1.86 / -9.89
  - IR_diff_gr: 0.04 / 0.23
  - Depr(-1)-inf(-1): 1.26 / 6.7
- Variable definitions:
  - NPL = Non Performing Loan
  - D08 = Dummy variable for 3rd and 4th quarter of 2008
  - GDP_gr = Real GDP growth rate
  - Depr = Depreciation rate (when there is appreciation this becomes 0)
  - Inf = CPI inflation rate
  - IR_diff_gr = Growth rate of interest rate difference between interest rates in national and foreign currency

### Risk Assessment Matrix (Appendix V): main threats and assessments
- Global financial conditions
  - Likelihood: Medium
  - Expected impact: Medium — foreign-bank-dominated system vulnerable to funding pressures, parent bank deleveraging, nonresident deposit outflows.
- Slower growth in advanced and emerging economies
  - Likelihood: Medium
  - Expected impact: Medium — hurts exports and GDP, raises NPLs, lowers profitability, potential solvency pressures; currency depreciation could boost unemployment and widen current account deficit.
- Increasing geopolitical tensions (e.g., Ukraine)
  - Likelihood: High
  - Expected impact: High — trade, remittances, commodity disruptions; reduced investment and GDP; higher NPLs; reversal of capital inflows and funding/liquidity difficulties.
- Capital outflow shock
  - Likelihood: Medium
  - Expected impact: High — large nonresident deposits and external funding create vulnerability; sharp drawdown could create funding and liquidity problems.
- Commodities price shock
  - Likelihood: Medium
  - Expected impact: High — oil price increases could hurt corporate profitability and increase NPLs.
- Foreign exchange shock
  - Likelihood: Medium
  - Expected impact: High — dollarized liabilities with lari income expose unhedged borrowers to higher NPLs, lower capital, reduced profitability, and potential nonresident deposit outflows.
- Asset quality shock (household credit uptick)
  - Likelihood: Medium
  - Expected impact: Medium — higher household credit could raise NPLs and affect bank balance sheets.
- Bank concentration risk
  - Likelihood: Medium
  - Expected impact: Medium — system dominated by three major banks with similar business models/funding could transmit stress rapidly; absence of deposit guarantee scheme raises fiscal contingency needs.

*International Monetary Fund — Georgia FSAP chapter excerpt*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/scr/2015/_cr1507.pdf_