Master Thesis in Finance
2019– 2020
Can A Bank’s Default Probability Predict
Its Future Non-Performing Loans?
Name Liangzi Ma
Student Number S3453065
Study Programme MSc Finance
Supervisor Asad Rauf
Abstract
The paper applies an option pricing-based credit risk model to estimate the credit default probability for a sample of banks both in emerging countries and western European countries. Due to market efficiency, this paper also contributes to the literature by linking the relationship between the bank’s default risk and non-performing loans (NPLs). The empirical results show that banks with low default probability will have low non-performing loans shortly. Moreover, this effect varies in countries and periods. In terms of panel regression analyses, there is a positive and significant association between non-performing loans and default probability. Thus the banks’ default probability can be a signal predicting non-performing loans.
1. Introduction:
Credit risk is the risk of loss arising from some credit events with a counterparty. It is an important and great value, especially in the banking sector. For example, according to Pop et al. (2018), in the Romanian banking sector in 2016, the most important risk was by far credit risk (over 60%), followed by liquidity risk (around 30%), while market, operational, and systemic risks scored a combined value of under 10%.
Banks have huge impacts on the economy as their basic business is making loans and absorbing deposits. This paper focuses on this basic banking business. On the one hand, banks produce credit risk in their deposits section, which is a liability account item in the banks’ balance sheets, when they, as ―borrowers,‖ risk not being able to repay their deposits. In this paper, measurement of credit risk is focused on default probability. For a bank, when its assets are insufficient to cover its debt payments, default occurs (Blavy and Souto, 2009) and thus, the probability of default exists. There are mainly two categories for estimating default probability. One type applies credit scoring models using financial data ratios. The other type uses capital market data with a standard Black-Scholes-Merton option pricing model for the asset value and the probability of ending out-of-the-money (default).
On the other hand, banks are exposed to credit risk that can be caused by the risk of borrowers’ failure to repay loans, which is an asset account item in banks’ balance sheets. When a borrower is identified as unable to repay the loan as scheduled, this means the borrower defaults. Then, a non-performing loan may be created and be regarded as a type of ex-post credit risk for banks (Berger and Young, 1997; Podpiera and Weill, 2008; Louzis et al., 2012; Chavan and Gambacorta, 2019).
FIGURE1
Assets Liabilities Loans (credit risk) Deposits (bank default) … …
A Non-Performing Loan (NPL) is defined as a sum of borrowed money which the debtor has not made scheduled payments for a specified period. Today, there are no specific criteria that define credits as non-performing loans (Tmava et al., 2018). The IMF’s Financial Soundness Indicators (FSIs) use delays in repayment of principal or interest over 90 days as a criterion for defining a non-performing loan.
NPL is accepted as a serious challenge for banks and is seen as the financial pollution of the banking system since it deteriorates banking liquidity and profit (Barseghyan, 2010; Ghosh, 2015; Makri et al., 2014). A majority of banks—e.g., Timberland Bank (TSBK)—devote their effort to credit reporting by discussing the important section labelled ―Credit Quality‖ or ―Asset Quality‖ in every quarterly report. In the banking system, loan quality is characterized as the share of non-performing loans to total loans. In later chapters, the words credit quality, NPL, and loan quality are used interchangeably but they all are represented as the ratio of non-performing loans to total loans.
Although Manz (2019) summarizes papers and argues that NPL is determined by bank-specific factors—e.g., bank size—and country-specific factors—e.g. employment rate—by investigating such determinants in different regions, there are few papers that raise indicators that can signal and predict non-performing loans.
(2012) used the dynamic method of panel data to examine the determinants of NPLs in the Greek banking sector for particular categories of credit (consumer loans, business loans, and mortgage loans). Sukanya and Vishwanatha (2015) also proved that NPLs is influenced by the creditworthiness of borrowers, their financial positions, past relationships with banks, and their credit servicing and repayment histories by analyzing the non-performing assets of commercial banking sector in India during the period 2000-2013.
However, there are two problems with the straightforward method above that investigating the borrowers’ default. One is that there are too many borrowers so it is complicated to verify each bill of the loan, while models meant to distinguish between ―good‖ and ―bad‖ borrowers can be complex. The other problem is that such straightforward method lacks a quantitative explanatory effect. Take one client company as the borrower for example—even if it is in high probability of default, but the fact is that it is more likely on a small scale and the non-performing loan it creates is limited. Then for the whole bank, high default probability may not be related to the high number of non-performing loans in the future. Thus, the particular characteristics of borrowers are needed as once the borrower defaults, non-performing loans occur. However, is there any easier way to use an indicator to predict non-performing loans?
In this paper, I focused on default probability based on the Merton Distance to Default model and Bharath and Shumway’s (2008) ―naive‖ estimation, which contains equity return and its volatility capturing timely information from the market. This calculated default probability is forward-looking, so it contains the bank’s future prospects. Furthermore, according to Pop et al. (2018), conversely, the impact of NPLs on bank stability and the measurement role it plays in credit risk is greater in emerging markets than in developed countries. Thus, I want to investigate the predictable effect of default probability, which is one aspect of stability, over non-performing loans in both emerging and developed counties.
I used data from 26 countries in total from 2011 to 2018 to conduct the back-testing. The results prove the default probability has a positive predictive effect over the bank’s non-performing loans within one year. I also found that the effect level varies depending on the region—e.g., emerging countries, represented by China, India, and Russia, and other 23 developed western European countries by studying the moderator effect. The positive effect is more substantial in developed markets.
This paper is organized as follows. Section 2 discusses the theoretical background and develops hypotheses. Section 3 illustrates the methodology and introduces the control variable. Section 4 is the sample selection and summary statistics. Section 5 presents empirical results. Section 6 draws conclusions.
2. Literature Review [1][2]
2.1 Empirical Literature Review
Loans are probably the oldest financial contracts. Given the long history of money
1
Default probability derived from the Merton model is one step further in distance to default. Since
2
lending, assessment and management of credit risk in loans is still one of the most urgent issues today since it has been observed that default rates have surged recently (Benkert, 2004). Once default happens, NPLs then become a serious issue. ―NPL-issue is attributed high relevance by policymakers such as the European Central Bank and is currently addressed with a variety of measures. And the availability of data in the NPL-sector expanded today, allowing for more sophisticated research‖ (Manz, 2019). As stated in introduction, Manz (2019) concludes that the NPL issue is determined by macroeconomic, bank-specific, and loan-specific factors in his literature summary. These elements will be considered in the panel regression models in Section 3 of this paper.
In the regression model, apart from the determinants of NPL, there is a left problem of an indicator signaling the NPL. The existing literature suggests that credit risk indicators can predict future non-performing loans. Actually, Blavy and Souto (2009) constructed credit risk indicators for 27 Mexican banks from December 1998 to March 2008, finding that the expected default frequency (EDF), which is the firm’s expected default probability within a one-year (ahead) period, is a useful indicator for predicting NPLs.
However, Adrian and Shin (2010) also explain that NPLs show a procyclical response to bank credit. In truth, measurements of bank credit from professional rating agencies are derived from models, of which default probability is one of the measurements. As for models, the classic Merton Distance to Default model offers clear advantages of updating in time by incorporating the banks’ equity volatility and market value of the equity. It was investigated and proven by Krainer and Lopez (2003) that equity prices contain more additional information than accounting variables by predicting BOPEC credit rating.
of borrowers and future financials, is already priced by the market. For example, when it became clear that Enron had serious accounting problems, Enron’s stock price began to fall and its distance to default in Merton model immediately decreased. Although the ratings given by the agencies are also based on similar models—e.g., the Moody’s KMV model—these updates are not on a timely basis. Stock market-based information has responded more quickly to changing financial conditions and prospects than credit risk agency ratings, and default probability (Shumway and Bharath, 2008) can reflect such information. Back to Adrian and Shin’s finding (2010) that NPLs may react in the same direction as bank credit, as ―procyclical credit policy‖ hypothesis asserts that successful loan performance increases future NPLs since banks tend to liberalize their credit policy partially. Then it means there is a negative relationship between NPLs and previous default probability.
In all, the difference between the two patterns comes out of the lending activities. However, this is not concerned with outside ratings.
2.2 Theoretical Literature Review
probability, but default probability is not publicly observable in the capital market. It can be conveniently calculated and largely influenced via lending activity information.
FIGURE2
Moral Hazard: Gomez and Ponce (2014) studied the relationship between bank
competition and loan quality, suggesting that default probability is unavoidably correlated with bank competition. In a bank screening model, Shaffer (1998) showed that the average creditworthiness of a bank's pool of borrowers declines as banking competition increases. In other words, low rated banks with big default probability have the incentive to lend money to riskier clients. The interest rate that banks charge their most creditworthy customers is known as the prime rate. Rating adjustments trigger asymmetric changes in loan terms: for instance, banks charge higher interest rate. As a consequence of this, the level of non-performing loans will increase. Keeton and Morris (1987) focused on the concept of moral hazard, finding that commercial banks with a low capital to asset ratio, which is comparatively unstable (as capital can be one measurement of the stability), are encouraged to taking on high risk loans in their portfolio. Thus, an unstable bank may incur more non-performing loans in the next period.
Adverse Selection: Bank-Firm possibly Matched in the Loan Market (Ho et al., 2019)
suggest “several studies show that small firms borrow from small banks while large firms borrow from large banks (Stein, 2002; Hubbard, Kuttner, and Palia, 2002; Berger et al., 2005)”.
Passive Choice for Banks The clients of banks analyze offers from the banking sector
in order to assess every offer from the banks’ portfolios (Pop et al., 2018). The equity Ex-antee
exeegjk
Low rating
monitoring hypothesis predicts low credit quality firms will borrow from well-capitalized banks since it adds value and for banks, such monitoring increase return (Schwert, 2018). However, for loan interest and commitment affairs, the financial commitment hypothesis generally acknowledges that large and reputable clients may choose banks that are less likely to default first, since they want the cash flow from the bank loan ensured, because informational frictions is costly for them to switch sources of capital (Ivashina and Scharfstein, 2010; Schwert, 2018). Also, the point is that such large firms can afford a higher cost to borrow money from more stable banks. Based on this point, stable banks may face less risk of loan defaults. At the same time, large and reputable borrowers usually carry considerable large amounts of loan. Thus, with low default probability and a significant portion of loan shares, the level of non-performing loans they cause may be low.
Active Choice for Banks Inderst and Mueller (2008) show banks conduct a credit
suggested by Berger et al. (2005) that large banks are less willing than small banks to lend to ―informationally difficult‖ credits. On the other hand, Froot and Stein (1998), as low capital makes a bank more risk averse, so it will charge riskier borrowers a higher rate. Higher interest rates charged to risky loan customers make it harder for them to repay loans. When market participants observe the clients’ information and the banks’ choices, the stock price of a bank will be influenced, which is thus reflected in the default probability.
However, there are findings giving explanations for abnormalities. And they may cause results consistent with ―procyclical credit policy‖ hypothesis mentioned in empirical literature review. For some small companies, the effort needed to repay a loan could be greater than for a large public firm (Dermine and Carvalho, 2006). Moreover, Pastor and Dymski (1990) also suggest that it is also not guaranteed that borrowers will make a ―good faith‖ effort to repay their obligations even though lenders assessed their credit worthiness beforehand. Another abnormality of ―too big to fail‖ explained by Oleg and Martin (2016). They found that country-specific factors can influence results since some banks, especially in China and Russia, are governed by the government. Even when they lend to risky borrowers, there is market confidence resulting in a stable equity return and low default probability when the level of non-performing loans is actually high. This can be the foundation of the risk management hypothesis that predicts that well-capitalized banks lend to low credit quality firms (Schwert, 2018).
predict non-performing loans, instead of bankruptcy, is more direct and reliable. Since bankruptcy data is comparatively limited, the relationship between default probability and non-performing loans should be studied, thus leading to the null hypothesis:
Ho: The lower the default probability a bank has at present, the lower level of non-performing loans it will face in the future.
As most of the literature focus on particular districts, as mentioned in the introduction, banks in different markets are to be studied to see the predictive effect. The mechanism behind the hypothesis is: The default probability is highly determined by equity volatility calculated based on the Merton Distance to Default model. Such market information is inherently more forward looking than accounting data. Bank supervisors use this signal to monitor bank fragility and the future state of the bank (Gropp et al., 2006) since market-based indicators fully reflect past accounting information and forward-looking expectations about the prospects of the bank (Harada et al., 2013). This can also be proven through the default probability calculation formulas. If the capital market can distinguish or be informed about the categories and industries of the loan borrowers and their creditworthiness, and since market participants know about the bank’s future prospects, its future credit risk can be reflected in the equity price. To be more specific, the expectation of market participants for the bank’s future non-performing loans is correctly priced in today’s equity return and is thus reflected in the bank’s default probability considering the calculation of the Merton model.
3. Methodology
3.1 Default probability calculation
3.1.1 Merton DD model
The Merton DD model is an intelligent application of classic finance theory-call option pricing, but how well it performs in forecasting depends on how realistic its assumptions are. The model relies on the primary assumption: market value follows geometric Brownian motion, and it is logarithm normal distributed.
The Merton model is for measuring the credit default probability. The model originates from the BSM model of option pricing while the market value is similar to the value of a call option. When the market value is over the debt, the shareholders get the excessive part. The more the rest, the more they can get, which is consistent with the call options operating between stock price and the strike price. Otherwise if the market value is below the debt, the shareholders lose nothing because they have the right not to exercise.
The most critical inputs to the model are the market value of equity, the book value of debt, and the volatility of equity, which is the volatility of the daily stock market prices in one year in this paper.
The setup is similar to the option pricing approach. The Merton model stipulates that the equity value of a firm satisfies
= rT ( 2)
E VN（d1)-e DN d (1)
where E is the market value of the firm’s equity,
D is the face value of the firm’s debt,
r is the average rate of U.S. 30-year treasury bond yield, N (·) is the cumulative standard normal distribution function,
d1 is given by 2) ln( / ) ( 0.5 1 v v V D r T d T (2) Where T=1 ;
model) and to be solved.
d2 = d1 − 𝛿𝑉√𝑇 (3)
Another important equation relates v with the volatility of the equity value. It is derived from the Ito’s lemma as
𝛿𝐸 = (𝑉𝐸) 𝑁(𝑑1)𝛿𝑉 (4)
From the equations above, firm value and 𝛿𝑉 can be solved and the default probability is defined as N(-d2).
3.1.2 Naïve model
This approach contains almost the same information and captures the structure but avoids solving the equation in the Merton model. This naive model has significant meaning since it is more practical compared with bank rating models. This advantage is consistent with the purpose of this paper's study since the default probability, other than agency ratings, requires timely effect.
naiveD 0.05 0.25* E
(5)
This model includes the five percentage points in the debt volatility calculation to represent term structure of volatility, and also shows a proportion relationship between debt and equity volatility suggested in formula (6).
naive V E E naiveD naive D
E naiveD E naiveD
(6)
Next, the expected return on the firm’s assets is set to equal to the firm’s stock return over the previous year.
1
it
naiveur
(7)
The iterative procedure is able to condition on an entire year of equity return data. By allowing the µ of naïve model equal to past returns, the same information is incorporated. The naive distance to default is then defined as
2 1 ln[( ) / ] 0.5 naive it V V E F F r naive T naiveDD T （ ） (8)
Merton DD model but also captures the structure of Merton DD model. The naive default probability is defined as
naive N(naiveDD) (9)
In this naïve model, the decision to use past returns for µ is arbitrary and makes a difference with the risk-free rate from the Merton model. But if it has significant predictive power, considering the convenient way for calculation, it is more practical and efficient.
3.2 Default probability and nonperforming loans panel regression
Gropp et al. (2006) point out that the equity market signals are leading indicators of bank fragility. They investigate a sample of between 59 and 86 US and European banks and employ the distance to default in predicting the downgrading of banks to a C-rating, finding that the distance to default has some forecasting power at a horizon of 12–18 months. Similarly, in this paper, the horizon of 1-2 years is applied to see the predicting effect over non-performing loans.
3.2.1 Control variable
Us (2017) finds that NPLs are mostly shaped by bank-specific variables before the crisis, and macroeconomic variables after the crisis, by investigating a balanced panel of 21 deposit banks from 2002–2013 in Turkey. Depending on this, more macroeconomic variables are included since this study is from 2011-2018.
Macroeconomic control variable
Unemployment Louzis et al. (2012) focus on unemployment the most, even indicating
that it is a primary macroeconomic determinant of NPL. Moreover, Klein (2013) asserts that higher unemployment translates into less income, and therefore borrowers face difficulties to repay their debt.
Interest rate According to the summary of Manz (2019), works of literature widely
(Beck et al., 2015; Us, 2017). Thus loan default may happen. Also, it is imperative since rising interest rates may be passed through immediately to the creditors; thus, it is dominant also as a loan-specific factor and included in this paper.
Bank-specific control variable
Louzis et al. (2012) examine the link between lagged measures of performance and problem loans and approve that worse performance serves as a proxy for lower quality of skills concerning lending activities, which means better performance signals lower non-performing loans. Consistently Peric and Konjusak (2017) prove that ROA has a negative and statistically significant influence on NPL, which means the larger ROA, the lower NPLs. Thus the lagged value of ROA will be put into consideration. While ROA is regarded as one measurement of banks’ performance, the negative effect is opposite to the procyclical credit policy stated in the theoretical literature review.
3.2.2 Markets as moderator
In this paper, the emerging countries and EU developed countries are both investigated. I tested the moderating effect of emerging market and developed market, respectively. A dummy variable Country was employed to differentiate banks in emerging countries and developed countries. Indicator ―1‖ represents banks in the emerging countries and indicator ―0‖ represents banks in the developed countries.
Thus, the model can be summarized as 𝑁𝑃𝑟𝑖𝑡 = 𝑎 + 𝑏𝐷𝑃𝑖𝑡−1+ 𝜀𝑖𝑡 +𝑈𝑟𝑖𝑡+ 𝑅𝑓𝑡+ 𝑅𝑂𝐴𝑖𝑡−1 +𝐶𝑜𝑢𝑖 ∗ 𝐷𝑃𝑖𝑡−1 (10) 𝑁𝑃𝑟𝑖𝑡 = 𝑎 + 𝑏𝐷𝑃𝑖𝑡−2+ 𝜀𝑖𝑡 +𝑈𝑟𝑖𝑡+ 𝑅𝑓𝑡+ 𝑅𝑂𝐴𝑖𝑡−2 +𝐶𝑜𝑢𝑖 ∗ 𝐷𝑃𝑖𝑡−2 (11) Where i indexes for banks and t for time;
NPr stands for the non-performing loans ratio;
DP is default probability, Ur is the unemployment rate;
developed countries.
4. Data
4.1 data source
The data are from BankFocus and Compustat database. The valid data are about 791 banks from China, India, Russia and other 23 western European countries in 8 years, 3 years after financial crisis of 2008. The quantity and period are matched with limited stock price information from Compustat. The United Kingdom has 93 banks’ data available, which is the most among countries. China, India and Russia have a quantity of 31.35% of the total banks as emerging markets. The other western European countries are studied as developed markets. The non-performing loans for both markets do not differ too much every year between 2011 and 2018. More detailed distribution is summarized in Appendix Table 2.
4.2 data summary
The data ranges from the year 2011 to the year 2018. Then there are supposed to be 6328 observations, but it turns out there are 1614 observations instead. This suggests that the panel data is unbalanced. The risk-free rates are represented by the annual average yield rates of U.S. 30-year Treasury bond, ranging from 0.026 to 0.039. The banks are with a size of assets ranging from 2 million USD to 4006242 million USD. More detailed summaries are in Appendix Table 3.
From Figure 3, the average non-performing loans in the two markets are very close in the year 2013. China had substantial non-performing loans in the year 2013. Furthermore, NPLs change mostly with time for each emerging country, which does not violate with Ouhibi and Hammami’s finding (2015) that the credit quality of the loan portfolio in most countries remained relatively stable until the financial crisis hit the global economy in 2007-2008. The average quality of the bank assets deteriorates sharply due to the global economic recession. However, this is not observable in the dataset starting from 2011.
between the DP and interaction term dummy*DP. But it is less than 0.5, so there is no collinearity between variables.
TABLE1 Matrix of correlations
DP Cou*DP Ur Rf ROA DP 1.0000 Cou*DP 0.4396*** 1.0000 Ur 0.0207 -0.4787*** 1.0000 Rf -0.0802*** -0.0678*** 0.0980*** 1.0000 ROA -0.0929*** -0.0537** 0.0272 0.0258 1.0000
Notes: This table presents the correlations between variables. ***,**,and* indicate significance at the 1%,5%, and 10% levels respectively.
FIGURE3
5. Results
The yearly default probability is summarized in Appendix Table 5.
Figure4 shows the yearly default probability both in emerging markets and western EU markets. The default probability seems close to each other market and stable with
0 1000 2000 3000 4000 5000 6000 7000 8000 2011 2012 2013 2014 2015 2016 2017 2018
Non-performing loans:yearly distribution (million USD)
time. The main reason is that the numbers are small and do not show much fluctuation in the figure. However, the standard deviation is 9.13%. Other statistics are summarized in Appendix Table 3.
FIGURE4
The regression results are shown in table 2 and table 3 for the period 2012 to 2018 and 2013 to 2018. The R squares are both small due to the limited observations. There are year gaps for observed banks.
TABLE2 Regression results for period t-1
VARIABLES 𝑁𝑃𝑟𝑖𝑡 𝑁𝑃𝑟𝑖𝑡 𝑁𝑃𝑟𝑖𝑡 𝐷𝑃𝑖𝑡−1 0.0386 0.0395 0.0511 (0.0269) (0.0270) (0.0327) 𝑈𝑟𝑖𝑡 0.117 0.0841 (0.0860) (0.101) 𝑅𝑓𝑡 0.412 0.417 (0.965) (0.965) 𝑅𝑂𝐴𝑖𝑡−1 0.00617 0.00825 (0.0553) (0.0554) 𝐶𝑜𝑢𝑖 ∗ 𝐷𝑃𝑖𝑡−1 -0.0317 (0.0504) Constant 0.0391*** 0.0170 0.0194 (0.00352) (0.0299) (0.0302) R-squared Observations 0.004 539 0.008 539 0.009 539 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 2011 2012 2013 2014 2015 2016 2017 2018
yearly default probability
The t-statistics based on standard errors adjusted for firm-year are shown in the table below the coefficient estimates. ***, ** and * indicate significance at the 1%, 5% and 10% levels respectively.
Column 1 from Table 2 shows that ignoring other control variables and moderator, DP is positively related with the NPLs next year. Furthermore, such a positive relation is at 5% significance level. Since there are limited observations and the R-squared is extremely small, higher level of significance is required then. However, the positive result is consistent with the null hypothesis.
Column 2 shows with control variables added the coefficient of DP is 0.0395 at a 5% significance level. The coefficients of unemployment and risk-free rate are positive as assumed. High level of risk-free rate makes loan rate high. Thus, high loan rate and unemployment rate both make borrowers less likely to afford the loan, and non-performing loans occur, which is consistent with Klein’s finding (2013). However, as for lagged value for ROA, the coefficient is not as assumed negatively. Then the effect is influenced by the good performance’s lending behavior according to procyclical credit policy.
Column 3 shows that the negative coefficient of the interaction term is -0.0317 at 5% significance level. When banks are in emerging countries, where Country=1, the coefficient is smaller compared with banks in developed countries, this means for developed countries, the effect of default probability over next year’s non-performing loans is more substantial, and prediction effect is more transparent. This can be explained as the developed capital markets, market efficiency is strong. Other coefficients have been discussed through former columns1 and 2.
TABLE3 Regression results for period t-2
VARIABLES 𝑁𝑃𝑟𝑖𝑡 𝑁𝑃𝑟𝑖𝑡 𝑁𝑃𝑟𝑖𝑡
(0.0410) (0.0417) (0.0469) 𝑈𝑟𝑖𝑡 -0.130 -0.103 (0.144) (0.161) 𝑅𝑓𝑖𝑡 -0.488 -0.505 (1.524) (1.526) 𝑅𝑂𝐴𝑖𝑡−2 0.0151 0.0149 (0.107) (0.107) 𝐶𝑜𝑢𝑖∗ 𝐷𝑃𝑖𝑡−2 0.0293 (0.0783) Constant 0.0449*** 0.0703 0.0686 (0.00581) (0.0471) (0.0474) R-squared Observations 0.001 472 0.003 472 0.003 472
The t-statistics based on standard errors adjusted for firm-year are shown in the table below the coefficient estimates. ***, ** and * indicate significance at the 1%, 5% and 10% levels respectively.
Column 1 from Table 3 shows that ignoring other control variables and moderator, DP is negatively related with the NPL after two years. It means that banks with large default probabilities will have less non-performing loans after two years. This can be related to lending activities in longer time horizons explained by procyclical credit policy hypothesis. Banks take measures with adjusting loan policies according to performance. Performance may be related with default probability at the moment. There is the same problem with the situation of one lagged time. Since the R square is extremely small, and such a negative relation is at 5% significance level, the effect is not too convincing.
Column 3 shows that the positive coefficient of the interaction term is 0.0293 at 5% significance level. When banks are in emerging countries, where Country=1, the coefficients are larger compared with banks in developed countries, although they are still negative. This means for developed countries, the negative effect of default probability over non-performing loans after two years is still more substantial, and prediction effect is more obvious. Other coefficients have been discussed through former columns1 and 2.
6. Robust check (1) (2) (3) (4) VARIABLES 𝑁𝑃𝑟𝑖𝑡 𝑁𝑃𝑟𝑖𝑡 VARIABLE 𝑁𝑃𝑟𝑖𝑡 𝑁𝑃𝑟𝑖𝑡 𝐷𝑃𝑖𝑡−1 0.0533*** 0.0710*** 𝐷𝑃𝑖𝑡−2 0.00495 -0.0116 (0.0192) (0.0225) (0.0151) (0.0166) 𝐶𝑜𝑢𝑖∗ 𝐷𝑃𝑖𝑡−1 -0.0430 𝐶𝑜𝑢𝑖∗ 𝐷𝑃𝑖𝑡−2 0.0558** (0.0287) (0.0240) Constant 0.0418*** 0.0410*** Constant 0.0436*** 0.0441*** (0.00524) (0.00526) (0.00440) (0.00439) Observations 538 538 Observations 469 469 R-squared 0.014 0.018 R-squared 0.000 0.012
The t-statistics based on standard errors adjusted for firm-year are shown in the table below the coefficient estimates. ***, ** and * indicate significance at the 1%, 5% and 10% levels respectively.
7. Conclusions
In all, this paper proves the credit risk indicator can be a signal to predict the non-performing loans, which is consistent with the guidance of working paper of the International Monetary Fund. The default probability derived from Merton DD model and its naïve model is valid in predicting the non-performing loans in a positive relationship within one year. Prediction in two years is not reliable since there are many uncontrolled activities and policies among banks and clients.
Default probability is a precise way to predict the non-performing loans compared with loans analysis. Such predictive effects of DP with future prospects also show the efficiency in the capital market. As for the markets moderating effect, however positive or negative effect in the regression results, the developed markets from western European countries show strong reaction with a higher absolute value of coefficients. This can be explained that the markets in developed countries are mature and can detect and react more efficiently.
Reference
Abid, L., Masmoudi, A., Zouari, G. S. 2018.The consumer loan’s payment default predictive model: An application of the logistic regression and the discriminant analysis in a tunisian commercial bank. Journal of the Knowledge Economy v9 n3, 948-962.
Adrian,T., Shin, H.S.2010. The changing nature of financial intermediation and the financial crisis of 2007–09. Annual Review of Economics v2, 603-618.
Barseghyan, L. 2010. Non-performing loans, prospective bailouts, and Japan’s slowdown. Journal of Monetary Economics v57 n7, 873-890
Beck, R., Petr, J., Anamaria, P. 2015. Key determinants of non-performing loans: new evidence from a global sample. Open Economies Review v26 n3, 525–550.
Benkert, C.2004. Default risk in bond and credit derivatives markets. Dissertation, Springer. Universit t, Frankfurt.
Berger, A., Young, R., 1997. Problem loans and cost efficiency in commercial banks. Journal of Banking and Finance 21, 849-870.
Bharath, S.T., Shumway, T. 2008. Forecasting default with the Merton Distance to Default Model. Review of Financial Studies v21 (3) n13, 39–69.
Biswas, R., Michaelides, M. 2019. Essays in financial economics. Research in Finance v35
Blavy, R., Souto, M., 2009. Estimating default frequencies and macro financial linkages in the Mexican banking sector. International Monetary Fund working paper
Chavan, P., Gambacorta, L. 2019. Bank lending and loan quality: An emerging economy perspective. Empirical Economics : Journal of the Institute for Advanced Studies v57, 1-29.
Claessens, T., Law, A., Wang, Teng. 2018. How do credit ratings affect bank lending under capital constraints?BIS Working Papers No 747
Dangl, T., Zechner, J., 2004. Credit risk and dynamic capital structure choice. Journal of Financial Intermediation vol. 13(2), 183-204.
Dermine, J., de Carvalho, C. N. 2006. Bank loan losses-given-default: A case study. Journal of Banking & Finance vol. 30(4), 1219-1243.
Froot, K., Stein, J. 1998. Risk management, capital budgeting, and capital structure policy for financial institutions: An integrated approach. Journal of Financial Economics 47, 55–82.
Ghosh, A. 2017. Sector-specific analysis of non-performing loans in the US banking system and their macroeconomic impact. Journal of Economics and Business
93:29-45.
Gomez, F., Ponce, J. 2014. Bank competition and loan quality. Journal of Financial Services Research vol. 46(3), 215-233.
Greenidge,K., Grosvenor, T. 2010.Forcsting non-performing loans in Barbados business. Finance& Economics In Emerging Economies v5 n1
Gropp, R., Vesala, J., Vulpes, G. 2006. Equity and Bond Market Signals as Leading Indicators of Bank Fragility. Journal of Money, Credit and Banking, v38 n2, 399-428.
Güner, A.B. 2008. Bank lending opportunities and credit standards. Journal of Financial Stability 4(1), 62–87.
Harada, K., Ito, T., Takahashi, S.2013. Is the Distance to Default a Good Measure in Predicting Bank Failures? A Case Study of Japanese Major Banks. Japan & the World Economy 27, 70–82.
Hubbard, R.G., Kuttner, K.N., Palia, D.N.2002. Are there bank effects in borrowers' costs of funds? Evidence from a matched sample of borrowers and banks. Journal of Business 75,559–581.
Inderst, R., Mueller, H.M. 2008. Bank capital structure and credit decisions. Journal of Financial Intermediation 17 (3): 295–314.
Ivashina, V., David S. 2010. Bank lending during the financial crisis of 2008. Journal of Financial Economics 97, 319–338.
Jacobson, T., Linde, J., Roszbach, K.2013. Firm default and aggregate fluctuations. Journal of the European Economic Association 11(4), 945–972.
Klein. N. 2013. Non-performing loans in CESEE; Determinants and impact on macroeconomic performance. IMF working paper WP/13/72,
Krainer, J., Lopez, J. 2003. Forecasting bank supervisory ratings using securities market information. Proceedings of the Federal Reserve Bank of Chicago
Conferences on Bank Structure and Competition, 278-289.
Louzis, P., Vouldis, A., Metaxas, V., 2012. Macroeconomic and bank specific determinants of non-performing loans in Greece: A Comparative study of mortgage, business and consumer Loan Portfolios. Journal of Banking and Finance 36(4), 1012-1027.
Makri, V., Tsagkanos, A., Belles, A. 2014. Determinants of non-performing loans: the case of Eurozone. Panoeconomicus 61, 193–206.
Manz, F., 2019. Determinants of non-performing loans: what do we know? A systematic review and avenues for future research. Management Review 69(4), 351– 389.
Marquez, R. 2002. Competition, adverse selection, and information dispersion in the banking industry.The Review of Financial Studies summer v15 n3, 901-923.
Nagel,S., Purnanandam, A. 2019. Bank risk dynamics and distance to default. NBER Working Paper No. 25807
Oleg, D., Martin, H. 2014. Sovereign default risk and state-owned bank fragility in emerging markets. Post-Communist Economies v28 n2, 232-248.
Ouhibi, S., Hammami, S. 2015. Determinants of nonperforming loans in the Southern Mediterranean countries. International Journal of Accounting and Economics Studies 3 (1), 50-53.
Pastor, M., Dymski,G.A. 1990. Debt crisis and class conflict in Latin America. Review of Radical Political Economics 22(1), 155–178.
Peric, B.S., Konjusak, N. 2017. How did rapid credit growth cause non-performing loans in the CEE countries? South East European Journal of Economics and Business 12, 73–84.
Podpiera, J., Weill, L. 2008. Bad luck or bad management? Emerging banking market experience. Journal of Financial Stability v4 n2, 135-148.
.
Salas, V., Saurina, J. 2002. Credit risk in two institutional regimes: Spanish
commercial and savings banks. Journal of Financial Services Research 22, 203-224.
Santos, J.A.C., Winton, A. 2019. Bank capital, borrower power, and loan rates. Review of Financial Studies 32(11), 4501-4541.
Schwert, M. 2018. Bank capital and lending relationships. The Journal of Finance 73 (2), 787–830.
Shaffer,S. 1998. The winner’s curse in banking. Journal of Financial Intermediation 7, 359-392.
Stein, J.C.,2002. Information Production and Capital Allocation: Decentralized versus Hierarchical Firms. Journal of Finance 57,1891–1921.
Sukanya, N., Vishwanatha. 2015. The non-performing assets of commercial banks in India: A bank group-wise analysis. Journal of Commerce and Management Thought 6 (4), 724–24.
Tmava, Q., Avdullahi, A., Sadikaj, B. 2019. Loan portfolio and nonperforming loans in western Balkan countries. Journal of Finance & Banking Studies 7(4),10-20.
Us, V. 2017. Dynamics of non-performing loans in the Turkish banking sector by an ownership breakdown: the impact of the global crisis. Finance Res Letters v20,109– 117.
Vassalou, M., Xing, Y.2004. Default risk in equity returns.Journal of Finance v59 n2, 831-868
Wu, H.K.1969. Bank examiner criticisms, bank loan defaults, and bank loan quality. The Journal of Finance v24 n4, 697-705
APPENDIX
Table 1
List of available data and sources
Table 2
Bank samples distribution
Country Abbreviation Freq. Percent Emerging country
China CN 90 11.38
India IN 89 11.25
Russia RU 69 8.72
Western European country
Table 3
Descriptive statistics of variables (USD million)
Table 4
Non-performing loans: yearly distribution (million USD)
Markets 2011 2012 2013 2014 Emerging markets 2644.707 3795.109 4432.82 4138.891 China 2504.9 1976.5 7113.9 4885.65 India 2810.87 4599.258 2479.19 4953.958 Russia 2664.714 4807.25 2296.5 866.625 Developed markets 4095.236 4028.418 4447.862 5358.228 Markets 2015 2016 2017 2018 Emerging markets 4443.985 3048.078 1587.369 3240.974 China 6135.095 6115.961 827.0541 3813.967 India 2748.621 1179.192 2743.765 2484.461 Russia 3306.286 290.5 1245.375 5805.8 Developed markets 3928.48 4212.075 4162.398 3417.786 Table5
Yearly default probability
