MACROECONOMIC AND FINANCIAL MANAGEMENT INSTITUTE FOR EASTERN AND SOUTHERN AFRICA

Transcription

1 MACROECONOMIC AND FINANCIAL MANAGEMENT INSTITUTE FOR EASTERN AND SOUTHERN AFRICA A MACRO-FINANCIAL STRESS TESTING FRAMEWORK FOR MOZAMBICAN BANKING SECTOR Dércio Eunísio Mutimucuio BANCO DE MOÇAMBIQUE February, 2015

2 A Technical Paper submitted in partial fulfilment of the Award of MEFMI Fellowship II

3 LIST OF ABBREVIATIONS AND ACRONYMS BCBS BdM BOU CA CAMELS Basel Committee on Banking Supervision Banco de Moçambique Bank of Uganda Custody abroad Capital adequacy, Assets quality, Management, Earnings, Liquidity, Sensitivity to market risk CAR CBS CEBS CGFS EBA EDF ELR EUR FC FDI FSAP GDP IMF KMV LCR LGD LLP LR Capital adequacy ratio Central Bank of Swaziland Committee of European Banking Supervisors Committee on the Global Financial System European Banking Authority Expanded default frequencies Expanded liquidity ratio Euro Foreign currency Foreign-direct investment Financial System Assessment Program Gross domestic product International Monetary Fund Kealhofer, McQuown and Vasicek Liquidity coverage ratio Loss given default Loan loss provision Liquidity ratio I

4 LSDV LST MEFMI MZN NII NPL NR OLS PD PL RBZ REER RWA SCAP SLFR SRM UK USD VAR VEC ZAR Least square dummy variable Liquidity stress test Macroeconomic and Financial Management Institute for Eastern and Southern Africa Metical Net interest income Non-performing loan Non-resident Ordinary least square Probability of default Performing loan Reserve Bank of Zimbabwe Real effective exchange rate Risk-weighted asset Supervisory Capital Assessment Program Standing lending facility rate Systemic Risk Monitor United Kingdom American Dollar Vector autoregressive Vector error correction South African Rand II

5 ABSTRACT The study reviews the stress testing practices of some selected MEFMI member countries and the existing literature on credit risk determinants to put forward areas of improvement to the current micro-prudential stress testing framework of the Banco de Moçambique. The study also suggests a methodological approach to link a credit quality variable used in the micro-prudential stress testing framework with the macroeconomic environment. With regard to the former, the study found that the currently available dataset in the Banco de Moçambique does allow for the adoption of interest rate risk stress tests (using maturity gap analysis to measure NII exposures to interest rate risk, and duration to assess the impact of interest rate changes on the trading book), combined scenarios to assess the resilience of banks to multiple instantaneous shocks, and the inclusion of local regulation requirements to increase plausibility of the simulation exercises. The study found that the limited span of data series of NPL ratio used as a proxy for probability of default constrain the extent to which macro-to-micro stress tests can be implemented because no full economic cycle is present. Therefore, a proper identification of macroeconomic and bank specific factors that affect credit quality in Mozambique is crucial for using the proposed methodological approach that combines the forward-looking macroeconomic perspective, a focus on both the individual banks and the banking system as whole, and a uniform approach to the assessment of risk exposures across banking institutions. III

6 ACKNOWLEDGEMENT I owe my gratitude to the Banco de Moçambique and MEFMI for giving me the opportunity to undertake the MEFMI Fellows Development Programme. I would like to extend my sincere thanks to my mentor, Mr. Dirk Jan Grolleman for all his invaluable guidance in coming up with this technical paper. His mentorship made this research an interesting and enriching experience. I am grateful also for all the encouragement and support that I received from my colleagues in the Prudential Supervision and Human Resources Departments of the Banco de Moçambique, and I appreciate the MEFMI secretariat family for believing in me and offering me all the support during the research. Finally I would like to extend my gratitude to God and my family for their love and encouragement. IV

7 CONTENTS 1 INTRODUCTION Background Research Problem and questions Limitations Structure of the Study STRESS TESTING FRAMEWORK IN MOZAMBIQUE Introduction Risk Factors and Calibration of Shocks Credit risk Foreign Exchange Risk Direct Contagion Risk Liquidity Stress Testing Comparison of the BdM s Framework with the Original Čihák Model STRESS TESTING FRAMEWORKS IN SELECTED MEFMI COUNTRIES Introduction Swaziland and Zimbabwe Sensitivity tests Uganda From sensitivity analysis to stress tests based on scenarios LITERATURE REVIEW Introduction The Stress Testing Process Designing the Macroeconomic Stress Scenario Designing the Credit Risk Satellite Models Balance Sheet Implementation of Shock Scenario Impact Measures METHODOLOGY AND DATA V

8 5.1 Introduction Credit Risk Macro Stress Testing in Mozambique Modelling Approach and Specification Dependent Variable Explanatory Variables Expectations Limitations EMPIRICAL REVIEW OF FINDINGS Introduction Impulse Response of NPL Ratio After shock Macro financial Variables Development Results of the Panel Data Regression Capital Adequacy CONCLUSIONS AND RECOMMENDATIONS Summary of Conclusions Recommendations Further Research VI

9 LIST OF FIGURES FIGURE 3.1: A SCHEMATIC OVERVIEW OF THE BOU'S STRESS TESTING FRAMEWORK FIGURE 4.1: A TYPICAL MACRO SCENARIO STRESS TESTING PROCESS FIGURE 5.1: STATIONARY MACRO FINANCIAL VARIABLES FIGURE 6.1: IMPULSE RESPONSE OF NPL RATIO (INCREASE IN QRGDPG, CG, INFL, AND SLFR) FIGURE 6.2: MODEL FORECASTING ABILITY USING HISTORICAL DATA LIST OF TABLES TABLE 2.1: CREDIT RISK SENSITIVITY SHOCKS... 7 TABLE 2.2: SECTORAL SHOCKS ASSUMED TABLE 2.3. FOREIGN EXCHANGE RISK SENSITIVITY SHOCKS TABLE 3.1. CALIBRATION OF CREDIT RISK SHOCKS TABLE 3.2. CALIBRATION OF INTEREST RATE RISK SHOCKS TABLE 3.3. CALIBRATION OF FOREIGN EXCHANGE RATE RISK SHOCKS TABLE 3.4. CALIBRATION OF LIQUIDITY RISK SHOCKS TABLE 3.5. LIQUIDITY RISK SHOCK LEVELS TABLE 5.1: A PRIORI AUTHOR'S EXPECTATION TABLE 6.1: FORECASTED 2014 CAR BY BANK VII

10 1 INTRODUCTION 1.1 Background The 2008/09 financial crisis has highlighted the relevance of financial stability assessment and the need for developing improved analytical tools to better quantify the systemic risk. Especially, the stability of the financial system and its ability to withstand unanticipated shocks has become the centre of attention of various studies in recent years. Quite a bit of work has been done recently to develop an operational macroprudential policy toolkit that may be employed to assess the stability of the financial sector (see Osinski, et al. (2013), CGFS (2012), Mitra, et al. (2011), Borio & Drehmann (2009)), and other work building upon Borio (2003) for general theory and/or best practices on the range of macroprudential instruments that can be considered as possible components of the toolkit). Even though the set of indicators that would be necessary and sufficient for operationalizing a macroprudential policy to attain the tangible goal of detecting both the slow build-up and the sudden materialization in systemic risk, as argued by the International Monetary Fund (IMF) in its Global Financial Stability Report, is still in phase of design and definition, some of its main elements have already been applied with wide acceptability. In particular, stress tests have recently been widely used to assess the resilience of banking systems (Mitra, et al., 2011). Stress tests were first carried out by internationally active banks for strategic purposes as part of their risk management tools. The banks would design and implement stress tests for business analysis to ensure that management policies are synchronised with bank risks for better allocation of funds and to improve the quality of business in anticipation of potential adverse shocks (Montes & Artigas, 2013). Based on the initial impulse provided by banks, bank supervisors realized that they could also use stress tests as a tool to assess the overall resilience of individual entities and more recently to produce an overall picture of the resilience of the banking sector as a whole. Therefore, despite their shortcomings as a tool for detecting vulnerabilities during the lead-up period of the 2008/09 global financial crisis, stress tests 1

11 reclaimed their credibility as a core macroprudential tool as they made a key contribution in restoring confidence in the financial system, as demonstrated by the successful Supervisory Capital Assessment Program (SCAP) exercise carried out by the United States authorities in 2009 (Bernanke, 2010), or crises management exercises performed by the Committee of European Banking Supervisors (CEBS)/European Banking Authority (EBA) in countries like Greece, Ireland, Portugal, Spain and Cyprus. Moreover, a recent study by Borio, et al. (2014) shows that stress testing could be used in effective way as a crisis management and resolution tool and it could raise the discipline of thinking about financial stability. In addition to the reasons considered above, the design and implementation of stress tests has taken on particular importance in recent years due to the recommendations of the IMF and the World Bank. These institutions recommend running stress tests regularly and these exercises have been assigned a major role in the Financial Sector Assessment Program (FSAP) to assess the stability of international financial systems in both developed and emerging economies. 1 Moreover, principle 20 of the Basel Committee on Banking Supervision s (BCBS) Principles for sound stress testing practices and supervision requires supervisors to perform stress test exercises based on common scenarios for banks in their jurisdiction as a complement for the stress testing exercises performed by banks (BCBS, 2009). The recommendations from the above mentioned institutions and the concerns in relation to financial system stability reviewed above highlight the need to enhance the rigour of the micro- and macro-prudential stress tests performed by the Central Banks. Since June 2013, Banco de Moçambique (BdM) adopted a simple sensitivity analysis stress testing model adapted from Čihák (2007) with additional techniques derived from Worrell (2008). This model focuses on testing the impact of stressed variables on the solvency and liquidity of banks based on ad-hoc shocks on 1 IMF, Integrating Stability Assessments Under the Financial Sector Assessment Program into Article IV Surveillance (2010) details the incorporation of the financial stability tests to the FSAP exercises. IMF, Macro-financial Stress Testing - Principles and Practices (2012) provides guidelines for this exercise and a summary of the experience of the IMF in its application. 2

12 balance sheet items that directly or indirectly affect capital adequacy, not establishing any link between banks losses and the macroeconomic environment. This study discusses how the existing Čihák/Worrell based micro-prudential stress testing framework of the BdM can be improved and investigates the possibilities of linking the variables stressed in this framework to the macroeconomic environment, while focusing on the variables related to the main risk in the Mozambique banking system, credit risk, with the objective of bringing the practises of the BdM more in line with the internationally developed practises regarding stress testing. 1.2 Research Problem and questions The problem of this paper is set out as follows: How can the existing Čihák/Worrel based stress testing framework of the BdM be improved taking into account literature and regional stress testing practises? In order to answer the research problem, the following more detailed research questions have been formulated: 1. What is the BdM s current micro-prudential stress testing framework? a. What is the risk spectrum covered by the stress testing framework? b. What approaches are used for the different risks covered by the existing framework? c. How does the implemented framework compare to the Čihák/Worrell framework as described in literature, and what improvements could be made? 2. How do the BdM stress testing practices compare with regional practices? a. What are the stress testing practices of MEFMI region member countries? b. How would these regional practices be usable by BdM to improve its stress testing framework? 3

13 3. How does the literature suggest linking the variables used in the micro-prudential stress testing framework with macroeconomic environment? a. How does literature suggest to link the Čihák/Worrell based micro-prudential framework to the macroeconomic environment? b. When focusing on the main risk for the Mozambican banking system (credit risk) how could this link be established/modelled? 1.3 Limitations The main shortcoming of stress tests mainly the macroeconomic stress tests are the frequent data limitations. Severe historical shocks are rare in many countries including Mozambique, which limits the predictive power of historical data. Therefore, adjustment of stress testing models by additional assumptions that are set by expert judgment is needed. Other limitations are related to the inability of models to capture correlation of risks and risk measures over time and across institutions, and the limitations to interpret results in longer time horizon, which originates the problem of disregarding endogenous responses of the system under simulation. Lastly, the incorporation of stress testing models implications in policy decision-making is only partial Structure of the Study This study consists of seven sections. In this section an introduction to the research is presented, and the research purpose and questions are stated. The research questions and sub questions are discussed in the subsequent sections. Section two describes the current micro sensitivity stress testing framework of the BdM and points out some improvements that could be made. Section three discusses the stress testing practices of some selected MEFMI countries, and highlights possible enhancements that could be imported to the BdM s stress testing framework. In the fourth section relevant literature on macro-to-micro credit risk 2 Comprehensive discussion of the limitations and challenges of stress tests can be found in Hardy & Schmieder (2013), Drehmann (2008), Čihák (2007), or Sorge & Virolainen (2006). 4

14 stress tests, exploring the possible link between Čihák s microprudential stress test framework with macrofinancial variables, is presented. In section five, the methodology and data used to build a satellite model to link macro-financial drivers of stress with credit risk in Mozambique is discussed. In the sixth section empirical results are analysed and discussed. In section seven conlusions and recommendations of the study are discussed. 5

15 2 STRESS TESTING FRAMEWORK IN MOZAMBIQUE 2.1 Introduction This section describes the current micro sensitivity stress testing framework of the BdM in terms of the risks covered, shock types, and deviations from the Čihák/Worrell framework. The section ends by pointing out some improvements that could be made to the BdM s current stress testing framework. 2.2 Risk Factors and Calibration of Shocks The BdM s stress testing framework is implemented through an Excel-based tool that captures bank-bybank data to assess the impact of adverse instantaneous shocks in some risk factors, keeping the others constant. Four risk categories are considered: credit risk, foreign exchange risk, interbank contagion risk, and liquidity risk. After impacts are calculated on individual institutions, the results are compared and aggregated to get the system resilience picture Credit risk Under credit risk, the model assesses the impact of increases in non-performing loans (NPLs 3 ) and its effect on the required provisions for loan losses, which are deducted from current regulatory capital in line with international stress test practice. Two main assumptions are considered, which exacerbate the effect of the assumed shocks: 100 percent provisioning rate for additional NPLs regardless of the collateral associated with the loan and its classification 4 ; and profits are not taken into account. The nature and magnitude of the shocks are shown in Table From first quarter 2014, by NPLs BdM means loans whose repayments are more than 90 days overdue. From third quarter 2008 to fourth quarter 2013, by NPLs BdM meant impaired loans. 4 BdM has five classes of quality to describe bank loans and determine specific provisions: Class I for loans repayments which are not more than 30 days overdue; Class II for loan repayments in arrears between days; Class III for loan repayments in arrears between days; Class IV for loan repayments which are in arrears between days; and Class V for loans at least 361 days in arrears. The probability of loss increases with each class of loan quality, up to 100 percent in Class V. If loans are assumed with no collateral, BdM requires provisions of 5 percent, 15 percent, 50 pecent, 85 percent, and 100 percent for the five classes of loans, in that order. 6

16 In terms of presentation and analysis of results, firstly the results are presented in terms of the magnitude of the change in the capital adequacy ratio (CAR). Secondly, the results are presented in terms of the effective solvency levels before and after the shocks, allowing the analysis of the banks capacity to withstand the shocks. Table 2.1: Credit risk sensitivity shocks Shock (i) Adjustment for underprovisioning (ii) Proportional increase in the nonperforming loans (NPLs) (iii) Increase in the level of default in some economic sectors Sectoral shocks (iv) Default of the five largest borrowers Concentration risk Magnitude Gradual increases varying from 10% to 350% Varying dimensions of shocks depending on the economic sector. Range [5% to 20%] (i) Adjustments for underprovisioning The objective of this shock is to determine provisioning shortfall to be adjusted in the assets value and capital so that the exercise is focused on the economic value of the asset. Performing loans are assumed to have 0 percent provisioning rate. Provisions needed are determined by summing the reported values on the NPLs category, since these are assumed to have 100 percent provisioning rate. Provisions held are taken from banks balance sheets. New capital adequacy ratios are calculated using the following formulas: (2.1),0 (2.2),0 (2.3) (2.4) 7

17 Where: i is an individual bank k is the loan class RWA are the risk-weighted assets (ii) Proportional increase in NPLs The objective of this shock is to assess the extent to which the deterioration in asset quality affects the level of capitalization of institutions in the banking system. General NPL shocks from 10 percent to 350 percent to the stock value of existing overdue loans in steps of 10 percent are made. New NPLs lead to new provisions which impact both capital and risk weighted assets (RWA). Results are measured in the form of stressed capital adequacy ratios (CARs) as follow:, 1 (2.5),,,,, (2.6) If total provisions calculated based on the new NPLs are less than the existing amount of specific provisions already held by the bank, no extra provisions are required. System results are measured and presented in the form of a chart. For each shock j, assets of banks who have stressed CARs below the regulatory minimum of 8 percent are summed and compared to the system s total assets. These are called Noncompliant Banks. The same methodology is used to compare insolvent banks within the system. Banks are considered insolvent if their stressed CARs are below zero. The following formulas are used: (2.7) 8

18 (2.8) Where: i is an individual bank j is the shock size n is the total number of banks considered in the stress test k is a bank which has its stressed CAR below 8 percent is a bank which has its stressed CAR below 0 (iii) Sectoral shocks This shock aims to determine how each bank will be affected by ad-hoc adverse events in different economic sectors where it has exposures. Shocks on specific economic sectors using loans by sector and by bank are simulated. For each sector a specific impact factor, so that new provisions have to be made to cover new NPLs, is considered. Table 2.2 shows the magnitude of shocks assumed per sector. Results are measured in the form of stressed CARs by bank and shown in the form of a chart representing CAR changes by bank. 9

19 Table 2.2: Sectoral shocks assumed (percent of performing loans in the sector becoming NPLs) Agriculture 20 Livestock 5 Forestry 5 Fishing 5 Mining 5 Manufacturing 5 Electricity, Gas and Water 5 Construction and Public Infrastructure 20 Tourism 20 Commerce 20 Transport and Communication 5 Non-monetary Institutions 5 Household 20 Mortgage 20 Other 5 The following formulas are used:,,, (2.9), (2.10),, (2.11) Where: i is an individual bank s is the sector 10

20 (iv) Concentration risk Concentration risks are assessed by large exposures default tests, which are captured from risk concentration returns 5. Since there is no information on collateral, one hundred percent loan given default (LGD) is considered. The Top 5 large exposures by bank are simulated to default one by one. Results are shown in a chart where decreasing CARs can be compared. The following formulas are used:,, (2.12),.,,, (2.13) Where: i is an individual bank b is the borrower The contagion effect that results from the fact that some top borrowers have loans in more than one institution is not taken into account Foreign Exchange Risk The objective of this stress test is to assess the impact of foreign exchange changes in capital and therefore in the solvency ratio. The transmission channels are the foreign exchange positions institutions hold at the end of the month under analysis. Increases in foreign exchange may result in losses or gains depending on the net position of the institution, whether it is long or short. In cases of losses, these are directly deducted from capital, meanwhile gains are considered in capital with 50 percent regulatory haircut. 5 The risk concentration returns take into consideration provision calculation exemptions and reductions as established by articles 16 and 17 of Governor s Notice No. 16/GBM/2013, of 31 st December. 11

21 It is assumed that there is a positive correlation between the three main currencies (American Dollar, Rand and Euro), so that an exchange rate variation of one of those currencies results in a similar variation on the exchange rates of the remaining currencies. The nature and magnitude of the shocks are shown in Table 2.3 Table 2.3. Foreign exchange risk sensitivity shocks Shock Nature Increase in foreign exchange rate MZN/USD, ZAR, EUR: Magnitude a) Direct foreign exchange rate risk Gradual depreciation of MZN varying from 10% to 200% in the actual value. b) Indirect foreign exchange rate risk Gradual depreciation of MZN varying from 10% to 200% in the actual value with assumed increase of NPLs in foreign denominated loans with half of the assumed percentage point exchange rate depreciation. The impact on each bank s capital is estimated by simulating shocks from 10 to 200 percent to the exchange rate level and using reported data on net open positions in foreign exchange. Indirect induced credit risk is simulated by assuming an increase of the NPLs of foreign exchange denominated loans with half of the percentage point exchange rate depreciation assumed. Results are shown by a chart with the same measures used for the sensitivity stress tests for credit risk. Average system CARs are also shown. The following formulas are used: (i) Direct shock, (2.14) 12

22 , (2.15) if, 0 then, 0.5 (2.16), (2.17) Where: are the gains or losses arising from the foreign exchange variation; i is the currency (USD, ZAR, EUR) j is the assumed Metical depreciation rate. j=1.1, for 10% depreciation rate n represent the number of the main currencies is the foreign exchange position per currency i ; and (ii) Indirect shocks, (2.18),, (2.19),,,0 (2.20),,, (2.21),, (2.22) Where: 13

23 , is the NPL variation arising from the assumed j Metical depreciation rate for each bank i is the NPL for foreign denominated loans, is the capital variation from the foreign exchange variation; n represent the banks assessed is the foreign currency denominated loans value for each bank i ; and j is the Metical depreciation rate Direct Contagion Risk While the other stress simulations assume that there is no direct interbank contagion between the banks in a failure situation, this stress test assumes that the failure of one institution has implications in other institutions through interbank exposures. Through the interbank exposures matrix the net position of each institution against the others is determined, and only the net borrowers are considered. The direct contagion is simulated by removing a bank from the market and measuring the domino effect on the system. For this exercise, interbank exposures received from the Financial Markets Department are used to draw a chart 6 in which it is possible to see an interconnectivity diagram. Capital below zero is the measure chosen to trigger insolvency and propagate the domino effect Liquidity Stress Testing Liquidity Stress Tests (LST) aim to produce information for the analysis of liquidity risks arising from foreign currency cash flows, non-resident customers cash flows, and a proxy for the stress scenario for the Liquidity Coverage Ratio (LCR) established in the Basel III liquidity paper, Basel III: The Liquidity Coverage Ratio and liquidity risk monitoring tools, issued in January An Excel based network analysis and visualization tool called NodeXL is used to draw and analyse the graphs. 14

24 The tests are applied to balance sheet data and to data on main foreign positions (by currency and by customer s country of origin) monthly received from the banks. These stress tests are run in Excel spreadsheets and cover Currency Stress Scenario, Non-Resident Stress Scenario and LCR Stress Scenario. The tests measure the banks capacity to withstand specific scenarios of stressed cash flows over the following 30 days period in all scenarios, in order to allow comparative analysis among results. The metrics consist of comparing the amount of unencumbered liquid assets with the stressed cash flow. Liquidity Shortfall and Liquidity Surplus, expressed in Meticais units, are the amount of liquidity under or above the amount of resources needed to settle the 30-day stressed cash flow, respectively. Liquidity Ratio is the ratio between the unencumbered liquid assets and the stressed cash flow. Ratios 7 under 100 percent indicate lack of liquidity to face the adverse situations hypothesized in the scenario. Most Mozambican banks are subsidiaries of foreign banks hence the exposure to foreign currency funding is substantial, mainly from their parent banks. Also, not only do banks accept deposits but give credit in domestic and foreign currencies, to residents and non-residents. In the case of LST, liquidity risk is assessed by currency and by customer s country of origin. Additionally, the LST tool offers a proxy for the calculation of the Basel III LCR, as a means of comparing the results among the scenarios. The objective and the premises of each scenario are: Currency Stress Scenario this test estimates the bank s capacity to settle its obligations in foreign currency under adverse circumstances that involve (a) partial loss of foreign currency capacity; (b) early settlement of non-maturing obligations in foreign currency; (c) unexpected withdrawal of credit and liquidity lines in foreign currency; (d) difficulties in attaining foreign currency inflows (maturing assets and loans payment); and (v) difficulties in performing foreign currency exchange operations. 7 Ratios are comparable among banks. 15

25 Non-Resident Stress Scenario this test estimates the bank s capacity to settle its obligations with foreign counterparties under adverse circumstances that involve (a) partial loss of foreign customer funding capacity; (b) difficulties in attaining assets issued by foreign entities or in receiving loan payments from foreign customers; (c) early settlement of non-maturing obligations with foreign counterparties; (d) unexpected withdrawal in irrevocable credit and liquidity lines to foreign customers. LCR Stress Scenario based on BCBS (2013, p. 19) this test estimates the banks capacity to survive under a significantly severe liquidity stress scenario, which entails a combined idiosyncratic and market-wide shock that would result in (a) the run-off of a proportion of retail deposits, including the run-off of one to a hundred of the top 100 deposits; (b) a partial loss of unsecured wholesale funding capacity; (c) a partial loss of secured, short-term financing with certain collateral and counterparties; (d) additional contractual outflows that would arise from a downgrade in the bank s public credit rating by up to and including three notches, including collateral posting requirements; (e) increases in market volatilities that impact the quality of collateral and thus require larger collateral haircuts or additional collateral, or lead to other liquidity needs; (f) unscheduled draws on committed but unused credit and liquidity facilities that the bank has provided to its clients; and (g) the potential need for the bank to buy back debt or honor non-contractual obligations in the interest of mitigating reputational risk. Defining the LCR scenario for Mozambican banks involves some proxies to adjust the Basel metrics to the available data. For the metrics components, each item from the balance sheet is classified as a liquid asset, outflow or inflow component, in domestic or foreign currency. Resident or non-resident counterparties are classified as retail, corporate, financial institution, central bank, etc.; for every position taken, it is indicated if it was in Mozambique or abroad, and the possibility of settlement in 30 calendar days. The stressed amount is calculated by multiplying the amount in the balance sheet with the stress factor. 16

26 (2.23) Stress factors estimate the behaviour of each component under a stressed environment. The calibration of the stress factors is the estimation of the position behaviour during stress, for example, haircut factors indicate loss of value of liquid assets positions, as well as run-off factors applied on deposit positions, to estimate deposits runs. (i) Currency Stress Scenario The Currency Scenario approach estimates the amount of domestic currency needed to face foreign currency obligations in a currency stressed environment. The scenario compares liquid assets with stressed cash flow, in terms of both foreign and domestic currency. In the event that foreign currency liquid assets are not sufficient (i.e. liquid assets are less than the stressed cash flow amount), the test results compare the foreign currency stressed cash flow with the total amount of liquid assets. To calculate the amounts of the stress components (liquid assets, outflows and inflows), Stressed Amount is summed by Item and Currency. (2.24) If 0 => no further analysis is undertaken, else: If 0 there is a Liquidity Gap in foreign currency stressed cash flow; If 0 the Liquidity Gap in foreign currency stressed cash flow cannot be covered by domestic currency liquid assets; If 0 there is a Liquidity Surplus in foreign currency stressed cash flow. (2.25) 17

27 (2.26) Where: FC is the foreign currency Total Liquid Assets = Liquid Assets in foreign currency + Liquid Assets in domestic currency "LR FC " is the Liquidity Ratio in foreign currency " = Expanded Liquidity Ratio in foreign currency (ii) Non-Resident Stress Scenario Non-Resident Scenario approach estimates the amount of domestic assets needed to meet obligations to foreign counterparties in a foreign-country-stressed environment. The scenario compares liquid assets with stressed cash flow, both held or due to foreign customers and counterparties. In the event that liquid assets from the bank s positions abroad are not sufficient (i.e. liquid assets are less than the stressed cash flow amount), the test results compare the stressed cash flow of foreign customers and counterparties with the amount of total liquid assets. To calculate the amounts of the stress components (liquid assets, outflows and inflows), Stressed Amount is summed by Item and Counterparty Country. (2.27) If 0 => no further analysis is undertaken, else: 18

28 If 0 there is a Liquidity Gap in nonresidents cash flow; If 0 the Liquidity Gap in non-residents cash flow cannot be covered by domestic currency liquid assets; If 0 there is a Liquidity Surplus in nonresidents cash flow. (2.28) (2.29) Where: NR is non-resident CA is the custody abroad => liquid asset custody is not in Mozambique Total Liquid Assets = Liquid Assets (custody abroad) + Liquid Assets (custody in Mozambique) " " is the Liquidity Ratio for non-residents " is the Expanded Liquidity Ratio for non-residents (iii)lcr Stress Scenario LCR Scenario approach estimates the amount of liquid assets needed to survive for 30 days in a systemic and idiosyncratic stressed environment. According to BCBS (2013 para. 144) recommendations, even banks with negative liquid cash flows (inflows < outflows) are expected to maintain a minimum level of liquid assets equal to 25 percent of the total cash outflows. An adjustment to the LCR is done to test a sudden withdraw of one to a hundred of the top 100 deposits (demand and time deposits). To calculate the amounts of the stress components (liquid assets, outflows and inflows), Stressed Amount is summed by Item. 19

29 ; 25% (2.30) % (2.31) Where: " % " is the Liquidity Coverage Ratio 2.3 Comparison of the BdM s Framework with the Original Čihák Model The current BdM s stress testing framework covers four individual risk factors, namely credit risk, foreign exchange, interbank contagion risk, and liquidity risk. Though possible to simply determine the impact of interest rate changes on NII using maturity gap analysis, the efforts of the BdM were towards assessing the impact of interest rate changes on capital using the Macaulay duration, more specifically the modified duration. However, the quality of the data provided by the banks on bonds structure prompts BdM to make significant approximations on portfolio maturities, which ends up producing results that need to be interpreted very carefully. On the risks to solvency that are tested, the most relevant difference from the original Čihák s model stems from the inclusion of Worrel s concept of gradually increasing the magnitudes of shocks to put the system on a trajectory to reach its breaking point as a means to visualize the inflection point as stress intensifies. Further, in the BdM s framework profits are not taken into account as the first line of defence before eroding the banks capital; data on collateral value is not easily available, and when available it is not reliable, which vindicates the application of 100 percent LGD on the credit risk stress test. Moreover, provisioning regulation is not taken into account in the assumptions that are considered for credit risk stress test, i.e. performing loans (Class I and Class II) are assumed to be provisioned at 0 percent rate, while non performing loans, regardless of their classes are provisioned at 100 percent level. With regard to the foreign exchange risk, the severity of the indirect shock, i.e. the severity of the effect of foreign exchange rates changes on foreign positions taken by borrowers, is highly dependent on the magnitude of the actual NPLs. Furthermore, 20

30 hedged foreign currency denominated borrowers (e.g. exporters) will keep performing in case of local currency depreciation. This means that if banks mostly have hedged foreign currency denominated borrowers, the indirect effect will be on the domestic currency denominated portfolio as the depreciation will result in more expensive imports and increasing prices. The current BdM s stress testing framework does not cover combined scenarios, which are explained in the Čihák s model. Another major difference between the BdM s stress testing framework and the Čihák s model is on the liquidity risk stress testing. The BdM s framework measures the banks ability to withstand a liquidity run during a period of 30 calendar days based on three scenarios of stress: (i) foreign currency stressed cashflow, (ii) non-resident customers stressed cashflow and (iii) a proxy for the stress scenario for LCR, which includes a test for large deposit concentration risk. For the LCR, Basel metrics or some proxies (due to data limitations) are considered after classifying each balance sheet item as liquid asset, inflow, or outflow. In all scenarios it is determined a liquidity ratio of unencumbered liquid assets over stressed cashflow, which should be equal or over 100 percent to indicate bank s ability to face the adverse scenario under test. This models differs from the Čihák s one on the metrics considered and on the time horizon of the test. Čihák provides a tool to analyse the survival period considering an outflow of deposits and the liquidation of liquid assets. From the aforementioned BdM s credit risk sensitivity analysis is limited in several ways. Required data are not easily available and part of the assumptions used make the results of the test less realistic; besides, the instant shocks considered assume that the market agents do not change their behaviour in the light of a crisis, which in reality is usually not valid. However, since the BdM s framework was implemented in a modular manner, implementing a satellite model to translate macroeconomic shocks into an impact on financial risks, mainly the credit risk, would contribute to bring more plausibility to the exercise. 21

31 3 STRESS TESTING FRAMEWORKS IN SELECTED MEFMI COUNTRIES 3.1 Introduction This section highlights the stress testing practices of the Central Bank of Swaziland (CBS), Bank of Uganda (BOU) and Reserve Bank of Zimbabwe (RBZ) with the purpose of exploiting aspects that could improve the BdM s experience. 8 In particular, the section starts by describing the micro sensitivity tests implemented by CBS and RBZ, and the credit risk macro scenario stress test implemented by BOU, and finishes by highlighting the possible enhancements that could be adopted by BdM. 3.2 Swaziland and Zimbabwe Sensitivity tests Stress testing in Swaziland and Zimbabwe is a supervisory tool that complements other tools used by these central banks such as the risk-based supervision, CAMELS rating system, and off-site monitoring. It is based on Čihák s framework, and tests the impact of stressed variables on the solvency and liquidity of banks based on ad-hoc shocks on balance sheet items that affect capital adequacy or regulatory liquidity requirements, not establishing any link between banks losses and the macroeconomic environment. The risks covered are credit risk, interest rate risk, foreign exchange risk and liquidity risk. Additionally, CBS and RBZ conduct a combined instantaneous stress test by looking at the collective effect of a percentage of performing loans (PL) becoming NPL, depreciation of the local currency (Emalangeni, for Swaziland only) by a certain percentage, and an increase by a certain percentage of interest rates CBS (2012) and RBZ (2011). (i) Credit risk stress testing Under credit risk, CBS and RBZ simulate the following common shocks: a deterioration in quality of impaired loans; and a sudden migration of a percentage of PL to NPLs categories. CBS also shocks large exposures; and industries and/or sectors, while RBZ also simulates a general downgrade of a percentage of loans by one grade. 8 Efforts to get other countries stress testing frameworks encountered reluctance from the respective authorities. 22

32 The shock levels considered by both central banks are shown in Table 3.1. Table 3.1. Calibration of credit risk shocks Shocks Minor Moderate Major CBS RBZ Deterioration in quality of NPLs 5% 10% 20% A sudden migration of PL to NPL/downgrade of a 5% 10% 20% percentage of loans Impairment of large exposures Default of government debt Source: CBS Stress Testing Guidelines, 2012 & RBZ Stress Testing Framework, 2011 In both central banks, minimum provisioning requirements for the different loan categories are taken into account on the determination of additional provisions resulting from each shock. However, profits are not taken into account, which means that any increase in provisions directly reduces capital that in turn results in a deterioration of a bank s capital ratios. Equation (3.1) is used by both central banks to determine additional provisions resulting from deterioration in quality of the three categories of loan impairment. 1 % 1 % 1 %,0 (3.1) On the migration of loans, CBS assumes that a percentage of all performing loans become special mention. Special mention loans become substandard, and all NPLs migrate to the next category of impairment (special mention to substandard, substandard to doubtful, and doubtful to loss). CBS expresses this stress using Equation (3.2). % % % %,0 (3.2) 23

33 RBZ measures a general downgrade of a percentage of loans by one grade using Equation (3.2) and, on the other hand, a sudden increase in NPLs caused by a direct shift from PLs (sum of pass and special mention loans) using Equation (3.3). % % %,0 (3.3) For the large exposures test, CBS assumes the impairment of one or more large exposures that absent the stress are classified as PLs. The variations in implementation would reflect particular concerns that CBS may have. The RBZ also assesses the effect of a certain percentage default on government debt. Shocks levels are defined according to RBZ s assessment of the current sovereign debt management situation. The amount of loss is calculated using Equation (3.4). %*(Treasury bills+investments in Other Government Securities) (3.4) In both central banks, the additional provisions arising from the stress tests described above are fed into the capital to risk-weighted assets ratio calculation using Equation (3.5). For the Zimbabwean government default shock, capital adequacy ratio is determined using Equation (3.6) 100 (3.5) 100 (3.6) 24

34 (ii) Interest rate risk stress testing To measure the interest rate risk on banks balance sheets, CBS and RBZ have adopted the repricing gap model through the distribution of interest sensitive assets and liabilities into buckets according to their time to re-pricing. These central banks use the gap between assets and liabilities in each bucket to determine the NII exposure to interest rate changes. The shock factors they use to calculate sensitivity of income to changes in interest rate are shown in Table 3.2. Table 3.2. Calibration of interest rate risk shocks Shock Level CBS RBZ Low +/- 1% +/- 3% Moderate +/- 2% +/- 5% High +/- 5% +/- 10% Source: CBS Stress Testing Guidelines, 2012 & RBZ Stress Testing Framework, 2011 Using Equation (3.7), CBS and RBZ assess the impact of changes in interest rates on earnings, for the one year cumulative repricing gap. And for the impact on capital and capital adequacy, they use Equations (3.8) and (3.9). % 12 (3.7) % (3.8) 100 (3.9) The model of CBS and RBZ only look at the direct interest rate risk, not taking into account the impact that an increase in interest rates is likely to have on credit. 25

35 (iii)foreign exchange rate risk stress testing To stress test foreign exchange risk, CBS assesses the direct impact of exchange rate movements on each bank s net open positions. To model the impact of foreign exchange rate risk the overall net open position of one or more currencies is separately assessed using the shocks shown in Table 3.3. Table 3.3. Calibration of foreign exchange rate risk shocks Low level shock 5% Moderate level shock 10% High level shock 15% Source: CBS Stress Testing Guidelines, 2012 The impact of shocks in Table 3.3 on income and on capital adequacy is assessed using Equations (3.10) and (3.11). % (3.10) % 100 (3.11) Loss stemming from appreciation of the Emalangeni when a bank has surplus of assets over liabilities in foreign currencies is not assessed. Moreover, the effect of foreign exchange rate changes on foreign positions taken by borrowers or counterparties is also not taken into account 9. (iv) Liquidity risk stress testing CBS liquidity risk stress testing assesses the banks resilience to funding liquidity by determining a ratio of liquid assets over short term liabilities based on four shocks: a sudden withdrawal of top depositors; a sudden termination of contractual long-term deposits; a sudden non-availability of credit line; and a decline of the 9 This would serve to stress test the indirect foreign exchange risk. 26

36 deposit base across the banking sector. These shocks are separately assumed using the calibration shown in Table 3.4. Table 3.4. Calibration of liquidity risk shocks Shocks Minor shock Moderate shock Major shock Sudden withdrawal of top top 5 Top 20 depositors top 10 depositors depositors depositors Sudden termination of contractual long-term deposits. 10% 15% 30% Sudden withdrawal/nonavailability of credit lines 20% 50% 75% A decline of the deposit base across the banking sector 10% 20% 50% Source: CBS Stress Testing Guidelines, November 2012 According to local regulations Savings and Development Banks are expected to maintain a minimum level of liquid assets equal to 17 percent of total liabilities to the public, while all other banks are expected to maintain a minimum level of liquid assets equal to 20 percent of total liabilities to the public. So, the liquid liabilities coverage ratio under the four shocks is assessed in light of these minimum liquidity requirements using Equation (3.12). % 100 (3.12) The Zimbabwean liquidity stress test model considers the effect of both a systemic and an idiosyncratic liquidity crisis on banks. It shocks a daily run-off of deposits, and the assessment of a banking institution s resilience to the liquidity risk shock is done considering the existence of a stock of liquid assets than can sustain it for at least 5 calendar days. Table 3.5 shows the shock levels on liabilities daily run-off rates applied. Table 3.5. Liquidity risk shock levels Deposits shock Minor shock Moderate shock Major shock Demand 2% 3% 5% Savings 2% 3% 5% Interbank 5% 10% 20% Other 1% 1% 1% 27

37 Source: RBZ Stress Testing Framework The assessment assumes a static balance sheet with liquid assets being considered with different haircuts, based on the nature of the crisis i.e. bank specific or systemic. From day one, the model assesses if deposits held and available liquid assets satisfy the Equation (3.13). On subsequent days remaining deposits and liquid assets are considered. % % % % 0 (3.13) Where p, q, r, and s are the shock levels for minor, moderate and major shocks. Stress testing frameworks of CBS and RBZ are similar, and cover interest rate risk and combined scenarios, which are not covered in the BdM s framework. However, the stress testing frameworks of CBS and RBZ do not cover contagion effects and do not consider the indirect effects of changes of foreign exchange rate and interest rate on positions taken by borrowers or counterparties. Furthermore, these frameworks do not establish a link between banks losses and the macroeconomic environment. 3.3 Uganda From sensitivity analysis to stress tests based on scenarios BOU carries out quarterly stress tests to assess the resilience of the banking sector to systemic risks. The Commercial Banking Department of BOU uses sensitivity stress tests based on Čihák s model to assess credit risk, interest rate risk, and foreign exchange risk. The shocks included in these sensitivity stress tests are: increase in non-performing loans; 100 percent loan loss of each bank s largest borrower; decline in net interest margin; decrease in interest income on government securities; and depreciation of the Ugandan Shilling against the US dollar. The Worrel s concept of applying incremental magnitude of shocks to selected variables until banks fail to meet minimum requirements is used in these sensitivity stress tests. Since July 2013, BOU introduced, through its Financial Stability Department, a macro-financial stress testing framework that can be used for both micro- and macro-prudential purposes. This stress testing framework is modular, and comprises three-pillars: the first pillar is the scenario design, which involves the 28

38 design of the macro-financial scenario to be imposed on the Ugandan banking sector; the second pillar is the credit risk satellite model, which translates the scenarios designed on the first pillar into variable affecting the banks loss absorption capacity; the third pillar is the balance sheet module, which applies the projected losses derived from the satellite model to individual bank balance sheets with the objective of calculating the resulting impact on each bank s solvency position BOU (2013). Figure 3.1 shows an overview of the BOU s macro-financial stress testing framework. The description below focus on the BOU s macro-financial stress testing framework for credit risk. Figure 3.1: A schematic overview of the BOU's stress testing framework Source: Background paper on macro stress testing at BOU, pp. 4 (i) Pillar 1 - Designing a macro-financial scenario and shock calibration The first part of the BOU s framework is the design of a macro-financial scenario, which serves as a basis for defining a set of adverse macroeconomic shocks to apply to the banks. In its paper on macro stress testing, BOU considered a price shock, where effects of an increase in international prices on food and energy are assessed in terms of direct impact on economic output through increased commodity prices and production costs, which in turn triggers domestic inflation to rise above the policy target, causing interest rates to increase, and as a consequence of the rising interest rates credit defaults soar, generating a severe credit tightening, as supply falls far below demand. 29