A Framework for Macro Stress-Testing the Credit Risk of Commercial Banks: The Case of Vietnam

In this paper, we assess the capacity of Vietnamese commercial banks to withstand the effects of an increase in credit risk as a result of macroeconomic shocks. Firstly, VAR model is used to estimate the relationship among macro variables (real GDP, real exchange rate, lending interest rate and inflation rate) and from that, macroeconomic scenarios are set up. Next, we employ a GMM model to estimate the relationship between the non-performing loan ratio (credit risk) and macro variables involved in first step. Finally, the new capital requirement ratio (CAR) is recalculated, which is based on the increase in loan provision followed by the rise in non-performing loan. The results show that credit risk which the commercial banks have to face is relatively limited when their risk weighted assets are unchanged. If these numbers, however, increase as banks broaden their lending, all banks’ CAR will reduce remarkably and four large banks will be lack of capital seriously and cannot meet the requirement of Central Bank


Introduction
Stress testing refers to a range of techniques used to assess the vulnerability of a portfolio/bank to "exceptional but plausible" macroeconomic shocks.To assess the vulnerability, according to Basel, the adverse shock used in stress test must be "extreme and exceptional" but "plausible" (Basel, 2009).The impacts of stress test might be shown as financial impacts on capital, loss (solvency stress test), or solvency (liquidity stress test).By conducting stress test periodically, the supervisory body may be able to actively cope with the worst-ever scenarios.Stress testing applies appropriate techniques for credit risk, market risk, liquidity risk and other risks.Credit risk stress testing concentrates on the risk that a borrower is unable to pay debt in certain circumstances.

Figure 1. Credit Risk -A Typical Macro Stress Testing Process
The stress testing may be performed by an individual bank, group of banks with some similar characteristics or the system.At the micro level, stress test is designed to assess the sustainability of a financial institution for the purpose of risk management, while at the macro level, stress test is applied to test the sustainability of the system.The purpose of macro stress testing is to help the supervisory body define the sensitivity of the system to stressed events which might lead to the system break-down (Basurto, 2006).The macro stress testing, as a result, is mainly realized by Central Banks or international institutions as IMF, WB.The stress testing process involves several stages, as shown in Figure 1.
First step -Identify a stress event from exogenous factors.The stressed events might be a shock affecting to the domestic economy, which is large enough and still possible.The effects of the stressed scenario to macroeconomic conditions are measured by a structural econometric model, vector auto regression method and/or pure statistical approach (Foglia, 2009;Bank of Japan, 2007;Van den End, Hoeberichts & Tabba, 2006) Second step -Model the relationship between credit risks and other macroeconomic variables derived in stage 1.The macroeconomic model does not include credit risk measurement; therefore, in the second stage of stress testing, links between credit risk measurement and macroeconomic variables will be modeled by satellite model to assess the effects of endogenous shocks to the bank's asset quality.These models are estimated either for the whole system or for different levels of disaggregation as by industry, type of borrower, such as bank or individual borrowers.The regression models include loan performance variables as non-performing loan (NPL), or loan loss provisions (LLP) as dependent variables, the explanatory variables include a set of macroeconomic variables, bank/industry specific variables as measures of indebtedness… Variables as economic growth, unemployment, interest rate, exchange rate, equity prices… are all explanatory variables, in which, interest rate is a crucial variable as it is the direct cost of borrowing.(Castren, Dees, & Zaher, 2008;Blaschke, Jones, Majnoni, & Martinez-Peria, 2001) Third step (last step) -Assess the impacts on loan portfolio of the bank and assess the capacity of the bank to stand to given stressed events.This means the loss of the bank should be compared to a suitable benchmark.
The results of the simulation might be shown as loan loss provision or predicted default rate, which depends on types of employed credit risk models; or with a known recovery rate, the bank's projected loss might be estimated, loan loss provision is defined.
In the study of Pilinko and Romancenco (2014) in Latvia, the authors calculated the new CAR -CAR in stressed scenario considering the increase in provisions as the bank's bad loans increased in adverse condition.The authors compared the calculated CAR with the requirement of Basel -8% to assess the vulnerability of the bank in stressed case.

Macroeconomic Variable and Credit Risk
The effects of macroeconomic variables to credit risk in commercial banks has been widely discussed in many research.Some macroeconomic variables which have great impacts on credit risk are listed in the followings.
Economic growth, usually represented by Gross Domestic Product (GDP), is a basic indicator of the cyclical position of the economy.A rise or high growth in GDP affect credit risk via a positive effect on corporate earnings, wage growth, which, inturn, increase the borrower's capacity to pay their debts leading to the increase in loan portfolio quality (Jakubik, 2007).Garr (2013), by using unbalanced panel data set from 33 commercial banks covering the period from 1990 to 2010, suggests that credit risk in Ghana is significantly influenced by management efficiency, GDP per capita, government borrowing and the financial sector development, in which GDP per capita have a positive relationship.Besides, many other researches on credit risk proves the emperical impacts of economic growth (GDP) on credit risk, as Lowe (2002), Koopman and Lucas (2005).
Inflation is another macroeconomic variable which affects credit risk.Casto (2013) in his research concludes that inflation has positive effects on credit risk.When inflation increases, though the real interest rate of the borrower decreases, their real income decrease, leading to the deteriation of the borrower's capacity to pay their debt, which increase the credit risk for banks.High inflation rates also ask the borrowers to pay more for their essential commodities, which negatively affects credit risk (Mkukwana, 2013).
Interest rate is the price of borrowing, which affects credit risk via debt burden.The rising interest rate, followed by the increase in debt burden, leads to a higher rate of nonperforming loan.Research of Poudel (2013) concludes that interest rate has positive effects on credit risk in banks.
Exchange rate has great impacts on foreign trade of an economy, the exchange rate fluctuation is one of the main source of economic growth and economic stability (Zameer & Siddiqi, 2010).Bozovic, Urosevic and Zivkovic (2009) suggest that the appreciation of foreign currencies against the local currency directly increase the cost of foreign currency borrowings which increase the debt burdens of the borrowers and increase the credit risk as a result.Hoggarth et al. (2005) analyzes the impacts of exchange rate via inflation and nominal interest rate and finds out that when foreign currency appreciates against the local currency, importing cost increases, inflation increases which burdens the borrowers and increases credit risk.The same relationship between exchange rate and credit risk is also found by Wong et al. (2010), Vogiazas and Nikolaidou (2011).

Macroeconomic variables estimation model
In order to estimate the relationship among macroeconomic variables and to make forecast about trends next time in Vietnam, the authors apply Vector Auto Regression model (VAR).Four variables including RGDP (Real Gross Domestic Product), INF (inflation rate), ITR (lending rate) and REER (Real effective exchange rate) are employed in the VAR model.All the data of four variables which are collected quarterly from statistical reports of Vietnam General Statistics Office, the State Bank of Vietnam and International Financial Statistics (IFS) of International Monetary Fund (IMF) during the period from 1994 to 2015.Since VAR model employs time-series data, the Unit Root Test (Table 1) and the Lag Length Criteria Test (Table 2) need to be conducted to assure the stationary of macro variables and appropriate lag for the model.The results of model proved that the GMM model is proper determination.The coefficient of lagged dependent variable (NPLs) was significant at 1% showed that the level of adjustment to the balance point of dependent variable was relatively fast.Note.*,**,***: the significance level at 10%, 5% and 1% respectively The regression result shows the positive relationship between nonperforming loan ratio and inflation and lending rate.This ratio and real GDP and real exchange rate are proved to be negatively related.Unfortunately, the coefficient of inflation rate variable is not significant.

Stress Testing Credit Risk for Vietnamese Banks
In order to assess capacity of bank to stand to given stressed events in banking business activities, the Capital adequacy ratio (CAR) which is often used as a proxy to determine the adequacy of banks' capital keeping in view their risk exposures.Banking regulators require a minimum capital adequacy ratio so as to provide the banks with a cushion to absorb losses before they become insolvent.
The research groups use CAR to assess the vulnerability of Vietnamese banks in stressed case.According to Circular No.36/2014/TT-NHNN issued by State bank of Vietnam, which applies to all Vietnamese banks, the CAR index is calculated as follow: Based on the results of predicted changes in NPLs and others assumptions, the authors calculate banks' new capital adequacy ratios and compare them to regulatory minimum of 9% (according to circular No36) to determine how resilient are Vietnamese banks to adverse macro-financial shocks and in case the cushion-own capital-cannot protect the bank from potential credit risks then how much capital do the banks need to hold in the following year to ensure the regulatory minimum of 9% required by State bank of Vietnam.

Assumptions
The new capital adequacy ratios of banks are calculated as following equation: * ∆ (3) -RWA: Risk-weighted assets -∆P: Expected increase in loan loss provisions in response to an increase in non-performing loans.It is calculated as ∆P = ∆NPLs x Loss given default.This formula means that the amount of the additional loan loss provisions that the bank is supposed to make will equal to the amount of the loans that it doesn't expect to recover.
Before estimating expected credit risks, the study bases on the following assumptions: Firstly, Loss Given Default (LGD) is calculated as a percentage of losses to exposure at default.In other word, it reflects the percentage loss suffered by banks on a credit exposure if the borrowers default.According to Basel's figure, recovery as a percentage of exposure is either relatively high (around 70-80%) or low (around 20-30%).
In the case of Vietnam, the authors assume that LGD of financial institutions is at 70%.
Secondly, it is assumed that a bank's profit for the forecasted year equals zero, thus, own capital is not increased by the amount of profit.
Thirdly, in order to analyze the effects on CAR in different conditions, the study takes into account two cases of changes in RWA: (i) RWA remains stable, which aims to assess partly impact of adverse macroeconomic scenario on Vietnamese bank's CAR; (ii) RWA increases by 18% which is based on forecasted credit growth rate in 2016 being around 18 -20%.The purpose of this case is to assess the simultaneous effects of growth in RWA and adverse macroeconomic scenarios on Vietnamese bank's CAR.

Results
Based on empirical results of panel data regressions between NPLs and others macroeconomic variables under these above assumptions, the authors estimated the new CAR for 10 banks in 2 cases and 4 different scenarios.
Case I: Risk weighted assets remain stable.In this case, the empirical results show that in 3 adverse scenarios, most of banks' CAR witness a slightly decrease.To be more specific, there is only 1 (BIDV) out of 10 banks which has CAR being under 9% in modest and severe scenarios.Furthermore, the amount of capital that BIDV needs to be added in these two scenarios to ensure the regulatory minimum of 9% is not significant (being at 1.39% and 048% respectively, which are considerably less than the average own capital growth rate of banking system in the previous year) The results show that when RWA is assumed to be increased by 18%, in all 4 scenarios considered, the banks' CAR reduce significantly with 4 out of 10 banks (Vietinbank, BIDV, Vietcombank and Sacombank) having their forecasted CARs under the regulatory minimum of 9% according to Circular No36.BIDV is the bank which has lowest CAR among the others and in severe scenario, its CAR is estimated to be reduced to 7.31%.
According to SBV, the banks with CAR being under regulatory minimum of 9% will be required to increase their capital.The authors estimated the amount of capital that 4 banks (Vietinbank, BIDV, Vietcombank and Sacombank) need to be added so that their CARs can ensure the minimum of 9%.The amount of capital that In both two assumed cases, it is noticed that there is only a slight fluctuation in NPL ratio and even in severe scenario, banks' NPL ratio are still less than 3%.

Conclusion
In order to assess credit risk tolerance in given macroeconomic stressed events, the study develops a methodology which consists of three main stages to conduct credit risk stress test for commercial banks in Vietnam against unfavorable macroeconomic scenarios.First, a relationship among macro variables (RGDP -Real Gross Domestic Product, INF -inflation rate, ITR -lending rate of commercial banks, REER -Real exchange rate) was estimated through Vector Auto Regression (VAR) model, and based on this relationship, macroeconomic scenarios was set up in case there is any change in one of the variables.These scenarios include baseline, mild, modest and severe cases.Next, this study suggested a model which estimates the relationship between credit quality (expressed through nonperforming loan ratio) and the macro variables which are involved in first step.Simultaneously, through the change in bad debt ratio, the study assesses bank's ability to react against unfavorable macroeconomic shocks in the following year.Last, the new capital requirement ratio (CAR) will be recalculated, which is based on the increase in loan provision followed by the rise in nonperforming loan.
With several assumptions, the new CAR of banks are calculated in two specific cases for the next year: (i) RWA remains unchanged and (ii) RWA grows by 18%.In the first case, there are slight changes in the banks' CAR in which only BIDV's CAR falls under 9% in medium and severe scenarios.In the case that RWA increase by 18%, the empirical results indicate that most of banks' CARs witnessed significant reductions in all of the scenarios.
The estimated CARs of Vietcombank, BIDV, Vietinbank and Sacombank were being under regulatory minimum of 9% (according to Circular No36).Therefore, in order to achieve the credit growth rate of 18% and ensure the CAR of 9%, it is advisable that the banks should have specific plans for increasing their own capital.
This paper has developed the stress test methodology for the Vietnamese banking system with regards to its credit risk.However, it is also subject to several limitations.First, it fails to take non-linearity between the variables of interest into account.Second, it doesn't distinguish between the boom-and-bust events (fails to consider the business cycle).Third, it disregards the feedback effects, which are infeasible to incorporate due to data restrictions.
Besides, although the study used a complex, multi-step approach to study the relationships between banking system's credit risk and macro-financial variables and to examine its resilience to adverse shocks, further exploration of relevant variables could improve the fit of the model; more bank-specific variables such as deposit-to-loan ratio or loan-to-GDP ratio could be added.Furthermore, the significant contribution to existing literature could be made by developing a model that accounts for feedback loops between banking system and real economy.

Figure
Figure 3. F e different deg In fact, this m we assumed th 8.GDP growth ra espectively.

Table 1 .
Result from the Augmented Dickey-Fuller Test Note. *,**,***: the significance level at 10%, 5% and 1% respectively Table3and Figure2indicate that real GDP mostly unaffected by other macroeconomic variables whereas this indicator has strong influence on others.In contrast, the inflation rate has a minor impact on interest rate and real exchange rate but is affected relatively by other variables.Meanwhile, real exchange rate is relatively sensitive to the changes in macroeconomic conditions whereas interest rate nearly has no response to these changes.Through empirical analysis, it is concluded that real GDP is the most important variable which has significant impacts on the remaining variables in the model.ass.ccsenet.2.1.3Macr Based on the reality of credit risk in the case of Vietnam, four factors including real GDP growth rate, CPI, real exchange rate and lending rate are taken into account to the model to conduct the credit stress testing.

Table 8 .
Results of regression model

Table 9 .
Estimated 2016 CAR* by Banks in Different Scenarios in Case of Unchanged RWA