Abstract
This Selected Issues paper highlights key policy challenges for accelerating growth on a sustainable basis and reducing external and financial vulnerabilities in the Republic of Croatia. A significant reduction in public expenditure would be needed to simultaneously provide room for cutting taxes, boost growth, and lower the budget deficit to help narrow the current account deficit. The analysis finds that Croatian banks are not necessarily passing on the higher risk of foreign exchange-linked loans to unhedged clients by charging higher interest rates.
V. Bank Stability and Credit Risk in Croatian Banks1
A. Introduction
1. The 2006 Article IV mission to Croatia takes place at a time when private sector credit growth has started to accelerate again. Both household and enterprise credit have recently been growing at 20 percent, year-on-year, with the share of household credit gaining prominence in total credit (Figure V.1).
Credit Growth in Croatian Banks—Selected Characteristics
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Source: CNB, IFS, Staff calculations.Credit Growth in Croatian Banks—Selected Characteristics
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Source: CNB, IFS, Staff calculations.Credit Growth in Croatian Banks—Selected Characteristics
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Source: CNB, IFS, Staff calculations.2. Credit booms can disguise underlying problems. This is because they can make bankse quality of their loan books, as they relax lending standards in their fight for market share during favorable macroeconomic conditions. At the same time, overall nonperforming loans, in percent of total loans, tend to look benign due to low default rates and high loan growth. Meanwhile, both good business and low provisioning (for loan-loss) levels tend to result in banks recording high profitability. But a key question remains: to what extent could banks be hurt by credit risk materializing once favorable macroeconomic conditions subside.
3. The purpose of this paper is to quantify the impact on Croatian banks’ capitalization of the possibility of macroeconomic conditions becoming less favorable. The impact depends on two important considerations. The first is the sensitivity of bank reserves for loan-losses (i.e., emerging credit risk) to changes in economic circumstances (proxied in this paper by real GDP growth and the unemployment rate). The second is the extent to which banks have already built up stability-enhancing buffers in anticipation of future credit risk. A related issue is whether fast-growing banks have higher credit risk.
4. Analysis based on annual macroeconomic and bank-by-bank data for Croatia (or, for that matter, any other single country in the region) would not give very reliable and meaningful results. This is due to an insufficient number of observations, spanning at least one business cycle. Therefore, this paper pools observations from Croatia and other countries in the region to quantify the average response of banks to changing macroeconomic conditions.2 Then, it uses this average response to calculate the effect of a downturn in economic conditions on the capitalization of Croatian banks. This would seem to be a reasonable approach as a large proportion of the banks in the countries in the sample are foreign-owned (especially by Austrian and Italian banks), like in Croatia.
5.We use a simple model, found elsewhere in the literature, for jointly estimatingcredit risk and bank stability in the countries in Emerging Europe (EE).3 In the model, loan-loss reserves proxy for credit risk and the so-called z-index measures bank stability. Box V. 1. describes the relationship between loan-loss reserves and credit risk in some detail. In estimating how changes in economic conditions affect loan-loss reserves and how these reserves then affect stability, the above-mentioned equations explaining bank stability and loan-loss reserves are estimated in a three-stage least squares framework to eliminate potential biases and to exploit efficiency-gains when there are feedback effects between the two equations. We repeat the exercise for EU15 countries—comprised of the Euro Area, Sweden, Denmark, and the United Kingdom—to see if EE banks behave differently and, if so, to explore how this affects their need to create additional reserves in the event of a downturn.
Loan-Loss Reserves and Credit Risk
Loan-loss provisions and loan-loss reserves: Loan-loss reserves are made against expected losses. Additions to loan-loss reserves are called loan-loss provisions. They are deducted from profits, and are made on a specific loan-by-loan basis and on pools of similar types of loans. These provisions are deducted from the loans on the asset side of the balance sheet, and are charged against profits, thus reducing retained earnings, hence capital. Both types of provisions adjusted for actual write-offs on loans are flow variables that add to the total loan-loss reserves. Any additional unexpected loss on loans is met from capital. In some countries, a percent of the loan-loss reserves is included under regulatory capital. Creating an adequate cushion is necessary if a capital crunch is to be avoided during high loan-loss events. Under International Financial Reporting Standards (IFRS), however, provisions can only be made when loans are actually impaired.
Loan quality, provisions and interest rate on loans:The current value of a loan is equal to the present discounted value (pdv) of the expected future cash flows generated by the loan, which is given by the contracted interest and principal payments less the expected value of losses from the non-repayment of the contracted amounts (Borio, Furfine and Lowe, 2001). However, the contracted interest rate can further be divided into the risk-free rate plus a default premium. Thus, the current value of the loan, Vt, can be written as:
where Ft is the face value of the loan, d is the default premium and E(l) is the expected loss from non-repayment of the contracted amounts. Loan-loss provisions try to provide a cushion to bridge the difference between the face value and the current value of the loan—Ft–V t. Thus, provisions are necessary to cover expected losses if the default premium charged to the borrower is insufficient to cover future losses.
But there can be a few reasons for the default premium to be underestimated, thus underpricing credit risk:
Banks competing for market share could relax lending standards.
Long-term lending or a multi-dimensional relationship with a borrower could cause banks to charge less.
In a partially euroized country with high fx lending and a closely managed exchange rate peg, banks expect the central bank to maintain the exchange rate in the future, thus underpricing fx-related credit risk.
In the event default premiums are underestimated, banks do not necessarily create the appropriate provisions: This is because of several reasons:
Additional provisions cut into current profits so banks can be reluctant to use them.
Accounting practices, especially International Accounting Standards, allow specific provisions to be made only for identified impairments. After adopting IFRS, although CNB reduced provisions to match identified losses, it raised provisions for unidentified losses.
Higher collateral values, buoyed by real estate booms, necessitate lower provisioning.
Croatian law allows banks to access borrowers’ wages in case of a credit event. However, such ‘wage collaterals’ could vanish if the borrower becomes unemployed.
One way of getting around the limitations posed by the accounting and tax regimes is to create buffers from capital instead. The provisions required to cover all future expected losses could be matched to the amount of capital buffers that can be set aside for the same purpose. Appropriately increasing risk weights is one way to do this.
6. There are four main findings of the paper:
(1) For the EE, banks’ loan-loss reserves in percent of loans is procyclical—rising with a decline in real growth rates and higher unemployment rates and vice versa. This is in line with existing empirical evidence that many banks around the world delay provisioning for bad loans until cyclical downturns have already set in and it is too late (Laeven and Majnoni, 2003). Procyclicality exacerbates the business cycle—a recession is aggravated by a credit crunch stemming from a steep rise in loan-loss reserves affecting capitalization. The evidence of procyclicality is less strong in the case of the EU1 5: while loan-loss reserves react to the unemployment rate, they do not react to the real growth rate.4
(2) The results are consistent with more stable and better managed banks needing to provision less, a result with intuitive appeal that thereby adds to the credibility of the empirical work.
(3) Rapid credit growth does not necessarily lead to higher loan-loss reserves, unless credit growth is accelerating.
(4) The adverse effect of worsening macroeconomic conditions on the capitalization of Croatian banks could be quite high. If the Croatian banks were to behave more like the EU15 banks, the adverse effect would be much less.
The rest of the paper is organized as follows. Section B tracks various interest rates on customer loans to get a sense of the credit risk component built into Croatian interest rates. Section C outlines the data and the econometric model and Section D discusses the results. A back-of-the-envelope credit risk calculation, based on the econometric results, is made in Section E. Concluding remarks, including policy implications for Croatia, are made in Section F.
B. Interest Rate Spreads and Credit Risk Premium
7. In this section, we analyze trends in different interest rate spreads to try to extract information on credit risk perceptions by Croatian banks of lending to different sectors. In particular, while recognizing that household and corporate loan products can vary widely, this section looks at the difference in same-currency loan rates for these sectors. Differences between general household loans and mortgage loans to households are examined first.
8. Loans based on house mortgages have a lower interest rate due to their relatively low risk. This is especially the case when house prices are increasing and banks perceive risks to be lower due to higher collateral values. Indeed, the excess credit risk on non-housing household loans is close to 2 percentage points in 2006 for Croatia. However, the data does not distinguish loan rates by currency.
Spread between different household loans
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Spread between different household loans
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Spread between different household loans
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
9. Household credit risk seems underpriced when judged against corporate lending. Difference between loan rates for households and corporates should reflect two factors: excess risk in household lending, especially in fx or fx-indexed lending to unhedged customers; and the market power of banks over households since typically firms can access foreign funding sources but households cannot. The difference between fx-indexed loan rates to households and corporates has trended down, and is currently close to zero for short-term loans, and a little over 1 percentage point for long-term loans. A reasonable assumption is that more corporates are naturally hedged than households, so it is somewhat surprising that the long-term fx-indexed loan rate difference is so low. Moreover, the credit risk premium is especially low if allowance is made for a premium for banks’ market power over households.
This evidence suggests that banks might not be pricing in the higher credit risk embedded in household foreign exchange indexed loans.
Differences in long-term loan rates to household and enterprises
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Differences in long-term loan rates to household and enterprises
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Differences in long-term loan rates to household and enterprises
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Differences in short-term loan rates to household and enterprises
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Source: CNB, Staff calculationsDifferences in short-term loan rates to household and enterprises
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Source: CNB, Staff calculationsDifferences in short-term loan rates to household and enterprises
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Source: CNB, Staff calculationsC. The Model and Data
The model
10. Bank stability is defined by the z-index (De Nicolo 2000, Boyd, De Nicolo and Al Jalal 2006). A widely used measure of bank soundness, the z-index is directly related to the probability of loss exceeding equity capital and thus measures the risk of insolvency or distance to default. It can be summarized by:
where μ is the average return on assets (in percent), k is equity capital as a percent of assets, and σ is the standard deviation of the returns on assets as a proxy for return volatility.5 Statistically, z measures the number of standard deviations a return realization has to fall in order to deplete equity, under the assumption of normality of banks’ returns. A higher level of z implies a lower probability of insolvency risk, or higher stability.
11. Two alternative measures of the z-index are used. The measures differ only in the calculation of σ. One measure, log(z_md), uses the mean-deviation of profitability (i.e., the absolute deviation of μ from the bank-specific mean of μ) as a proxy for returns-variability. This is used in Boyd, De Nicolo and Al Jalal. The other, log(z_rol), uses a 3-year rolling standard deviation of μ for returns-volatility as used in Maechler, Mitra, and Worrell (MMW, forthcoming). The first allows us to use more observations than the second, but could result in a larger variability in returns if μ moves from negative to positive or vice versa. The second measure smoothes out such variability, but may induce serial correlation in the data.
12. Loan-loss reserves in percent of total loans is used as a measure of provisioning. The stock version of provisions is used instead of the flow version to take into account the net impact of provisioning flows and loan write-offs on the loan-loss reserves in any period at a particular bank (Box V. 1).6 Since the series loan-loss reserves/loans ranges from 0 to 100, it is logit-transformed to ensure a normal distribution.7
13. The econometric model assumes that loan-loss reserves are made against either expected or realized loan-losses, for which macroeconomic conditions are key factors. If loan-loss reserves are built up in advance in anticipation of a downturn, then they would not show up empirically as reacting procyclically—that is, reserves would not increase when the economic downturn sets in. We use the lagged z-index as a proxy for profit and capital buffers built up in advance and for a bank’s sound credit risk management policies. These considerations would seem to affect current loan-loss reserves. Noting that bank stability could, in turn, be affected by past loan-loss reserves, we set up a model that jointly estimates bank stability and loan-loss reserves in a systems framework adopting the methodology used in Tamirisa and Igan (2006) and Cihak and Tamirisa (2006).
14. The 3SLS framework allows us to estimate the equations jointly, even though neither of the equations seem to have endogenous variables on the right hand side. The endogeneity problem is taken care of by the inclusion of a lag of the other dependent variable on the right hand side. Still, given that the residuals of the 3SLS regressions are significantly correlated (albeit with a correlation coefficient < 0.1), we proceed with 3SLS rather than 2SLS because the former is more efficient. The inclusion of the lagged dependent variable on the right hand side could potentially give rise to endogeneity problems if the errors are serially correlated. But serial correlation was not found in the residuals.
15. The selection of the explanatory variables is based on the empirical literature and practical experience, and reflects those most likely to have an effect on the dependent variables. The two equations used in the systems estimation are:
Equation 1—Bank Stability:
Equation 2—Loan-loss reserves:
i = bank index
t = year index 1997-2004 for Emerging Europe; 1996-2004 for EU15.
j = country index, covering Emerging Europe, EU15 countries.
16. Equation 1 estimates a parsimonious representation of the models used in MMW and Tamirisa and Igan. Bank stability varies between banks, across countries, and through time. We use the natural logarithm of z. The expected signs of the explanatory variables, shown in the equation representations above, are explained below.
A lagged z is used as a RHS variable to take into account that buffers built up in the previous period and good risk management policies help deliver stability through time. We expect the sign to be positive.
The variable lagged loan-loss reserves/loans reflects the reserves built in the previous period. It is expected to increase stability over and above the effect of better risk management policies and other buffers built up in the previous period as captured by the coefficient on lagged z. But previous empirical evidence (MMW) was ambiguous about the effect of loan-loss reserves on bank stability.
Higher real GDP growth in the last period reflects favorable macroeconomic conditions that could help bank stability…
…as does higher financial depth achieved through higher credit/GDP ratio.8
High bank-by-bank asset growth has an ambiguous effect on stability, depending upon the quality of such growth and its effect on volatility of earnings for the individual bank. Previous evidence (Tamirisa and Igan) has shown that even though credit growth has not deteriorated financial soundness in banks, future risks to bank stability could materialize due to the increase in the extension of credit by inherently weak banks.
The cost-to-income ratio is included as a bank-efficiency indicator: lower efficiency of a bank (represented by a higher cost-to-income) is associated with lower stability.9
17. The dependent variable for Equation 2 is loan-loss reserves/loans, with the various explanatory variables explained below.
Lagged loan-loss reserves/loans is expected to be a significant determinant of current loan-loss reserves/loans, because reserves are usually built up over time.
Higher bank stability in the previous period, represented by lagged z, would necessitate lower loan-loss reserves this period because less of a buffer is needed when a bank is comparatively more stable.
Deteriorating macroeconomic conditions are expected to result in higher loan-loss reserves, if loan-loss reserves are typically procyclical—that is, reserves were not built up in advance in anticipation of deteriorating macroeconomic conditions. Both real GDP growth rate and the unemployment rate are used to reflect macroeconomic conditions. Although there should be a high correlation between these two variables, both are included (with different lags) to reflect the almost equal presence of both household and corporate borrowers. While corporates are more sensitive to GDP growth rates, households are expected to be more sensitive to the unemployment rate. Thus the sensitivity of loan-loss reserves to lagged real GDP growth (current unemployment rate) is expected to be negative (positive).
Higher loan growth is expected to lower loan-loss reserves, due to the increasing base, but …
the rate of acceleration of loan growth—given by the coefficient on the square of loan growth—would necessitate higher loan-loss reserves.
Data
18. The paper uses bank-by-bank annual data from Croatia, Emerging European (EE) countries, and EU15 states.10 The data on the macroeconomic variables are obtained from the IMF’s International Financial Statistics. The bank-by-bank data comes from Bankscope for all the countries in the sample, for 1997–2004 (1996–2004 for EU15). Instead of looking at Croatia in isolation—given the short length of its time series—we look at the group of countries comprising Emerging Europe to estimate the model, in order to incorporate information from average macroeconomic cycles.11 Summary statistics of the dependent variables are provided in Table V.1.
Summary Statistics of the Dependent Variables 1/
EE refers to Emerging Europe. EU15 refers to the Euro Area countries, Denmark, Sweden and the U.K.
Summary Statistics of the Dependent Variables 1/
| Country groups | Obs | Mean | Sth. Dev. | Min | Max | |
|---|---|---|---|---|---|---|
| Log z_md | ||||||
| Croatia | 172 | 7.96 | 1.54 | 3.49 | 11.81 | |
| EE except | ||||||
| Croatia | 993 | 7.74 | 1.51 | 1.50 | 13.97 | |
| EU15 | 3582 | 6.05 | 1.52 | -1.38 | 21.22 | |
| Log z_rol | ||||||
| Croatia | 141 | 3.66 | 1.19 | 0.56 | 6.96 | |
| EE except | ||||||
| Croatia | 644 | 3.23 | 1.25 | -2.85 | 7.01 | |
| EU15 | 2803 | 8.50 | 1.14 | 1.99 | 12.35 | |
| loan-loss reserves/loans (%) | ||||||
| Croatia | 169 | 7.70 | 5.33 | 0.00 | 27.66 | |
| EE except | ||||||
| Croatia | 897 | 7.00 | 9.31 | 0.00 | 100.00 | |
| EU15 | 2046 | 3.82 | 5.84 | 0.00 | 93.51 | |
| Logit-transformed loan-loss reserves/loans | ||||||
| Croatia | 167 | -2.68 | 0.73 | -4.48 | -0.96 | |
| EE except | ||||||
| Croatia | 875 | -3.12 | 1.20 | -8.11 | 2.02 | |
| EU15 | 1977 | -3.77 | 1.25 | -9.21 | 2.67 | |
EE refers to Emerging Europe. EU15 refers to the Euro Area countries, Denmark, Sweden and the U.K.
Summary Statistics of the Dependent Variables 1/
| Country groups | Obs | Mean | Sth. Dev. | Min | Max | |
|---|---|---|---|---|---|---|
| Log z_md | ||||||
| Croatia | 172 | 7.96 | 1.54 | 3.49 | 11.81 | |
| EE except | ||||||
| Croatia | 993 | 7.74 | 1.51 | 1.50 | 13.97 | |
| EU15 | 3582 | 6.05 | 1.52 | -1.38 | 21.22 | |
| Log z_rol | ||||||
| Croatia | 141 | 3.66 | 1.19 | 0.56 | 6.96 | |
| EE except | ||||||
| Croatia | 644 | 3.23 | 1.25 | -2.85 | 7.01 | |
| EU15 | 2803 | 8.50 | 1.14 | 1.99 | 12.35 | |
| loan-loss reserves/loans (%) | ||||||
| Croatia | 169 | 7.70 | 5.33 | 0.00 | 27.66 | |
| EE except | ||||||
| Croatia | 897 | 7.00 | 9.31 | 0.00 | 100.00 | |
| EU15 | 2046 | 3.82 | 5.84 | 0.00 | 93.51 | |
| Logit-transformed loan-loss reserves/loans | ||||||
| Croatia | 167 | -2.68 | 0.73 | -4.48 | -0.96 | |
| EE except | ||||||
| Croatia | 875 | -3.12 | 1.20 | -8.11 | 2.02 | |
| EU15 | 1977 | -3.77 | 1.25 | -9.21 | 2.67 | |
EE refers to Emerging Europe. EU15 refers to the Euro Area countries, Denmark, Sweden and the U.K.
19. Bank-by-bank data and macroeconomic data show wide variations between yearly averages for Croatia and yearly averages for the rest of Emerging Europe (Figure V.2) and the EU15 (Figure V.3).
In Croatia, bank stability, as measured by z, has increased over the years, with the peak (in 2001) coinciding with the average Emerging Europe peak. Bank stability seems to have declined for the EU15 banks over the sample period probably reflecting higher volatility of profits.12
The variable loan-loss reserves/loans in Croatia has gradually declined since 2000; it has also declined almost continually in other Emerging and EU1 5 countries since 1998. In the EU15, however, this variable had started from and declined to a lower level than the Emerging Europe averages.
While bank-by-bank loan growth in Croatia has generally moved in parallel with other parts of Emerging Europe, the growth in Croatia has mostly been lower. Loan growth in EU15 banks took off quite dramatically following recovery from the 2000 U.S. tech bubble burst—a take off that coincided with loan growth in Emerging European and Croatian banks. The loan growth was helped by easing global liquidity conditions and low interest rates. But, even at its peak, loan growth in EU15 banks was only a fraction of that in Croatian and Emerging European countries. The difference is largely due to higher levels of intermediation already achieved by the EU15 banks compared to the others.
Bank efficiency—measured by cost/income ratio—has continually improved since 2000, but this ratio in Croatia remains above Emerging Europe averages.
Finally, there are two observations about the real economic cycle: First, although real GDP growth rates bounced back from adverse developments in 1999 in Croatia, the unemployment rate only started improving from 2002 onwards. This suggests that changes in the unemployment rate may have been mostly structural and therefore unrelated to cyclical fluctuations in real GDP. Second, the real GDP growth in the EU1 5 was adversely affected by the tech bubble burst in 2000; growth in Emerging Europe, especially Croatia, was not. The depressed growth rates in the EU15 that followed, along with easing global liquidity conditions, could have been instrumental in pushing EU15 funds into their Emerging European bank subsidiaries to exploit their relatively favorable business conditions, thus partly contributing to rapid credit growth in the region.
Emerging Europe and Croatia—Mean of Key Variables by Year 1/
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Source: Bankscope, Staff estimates.1/ Pooled mean of bank-specific variables by year for Emerging Europe and Croatia.Emerging Europe and Croatia—Mean of Key Variables by Year 1/
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Source: Bankscope, Staff estimates.1/ Pooled mean of bank-specific variables by year for Emerging Europe and Croatia.Emerging Europe and Croatia—Mean of Key Variables by Year 1/
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Source: Bankscope, Staff estimates.1/ Pooled mean of bank-specific variables by year for Emerging Europe and Croatia.
EU1 5—Mean of Key Variables by Year 1/
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Source: Bankscope, Staff estimates.1/ Pooled mean of bank-specific variables by year for the EU1 5.EU1 5—Mean of Key Variables by Year 1/
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Source: Bankscope, Staff estimates.1/ Pooled mean of bank-specific variables by year for the EU1 5.EU1 5—Mean of Key Variables by Year 1/
Citation: IMF Staff Country Reports 2007, 082; 10.5089/9781451817454.002.A005
Source: Bankscope, Staff estimates.1/ Pooled mean of bank-specific variables by year for the EU1 5.D. Results
20. The results for Emerging Europe are given in Table V.2 and those for EU15 are in Table V.3. Columns 1 and 2 use one measure of z (log z_md) and columns 3 and 4 use the other measure (log z_rol). The key results are as follows:
More stable banks are susceptible to lower credit risk. A bank that was more stable last period—already having built a buffer against risk through good risk management policies—could do with less loan-loss reserves this period; this is especially significant if log(z_rol) is used as a stability measure. This finding is given by the negatively significant coefficient for lagged z (column 4) in Equation 2, both for Emerging Europe and EU15. Past loan-loss reserves have an ambiguous effect on stability (similar to the findings in a single equation framework in MMW).
Loan-loss reserves increase with adverse macroeconomic fluctuations for the Emerging Economies, but the increase is less for EU15.13 While the EE banks’ loan-loss reserves increase with a fall in lagged real GDP growth and a rise in the current unemployment rate, the EU 15 banks only respond to the unemployment rate. The reason for the latter response of the EU15 banks could be their better preparation—in terms of credit risk management and provisioning—for cyclical downturns.
Higher loan growth is associated with lower loan-loss reserves, but accelerating loan growth eventually leads to higher loan-loss reserves. This quadratic relationship is robust across country groups and specifications.
There is weak evidence of higher asset growth in the previous period lowering bank stability. This result could be driven by higher volatility of profits associated with higher asset growth, which in turn lowers stability indicators.
Efficiency enhances stability. This observation emerges from the negative relationship between cost-to-income and the z-index.
3SLS Estimates of Bank Stability and Credit Risk—Emerging Europe 1/
Absolute value of z statistics in parentheses+ significant at 10%; * significant at 5%; ** significant at 1% Adding dummies for specific country clusters--like south eastern European or the ten recently acceded EU member states--did not change the results.
3SLS Estimates of Bank Stability and Credit Risk—Emerging Europe 1/
| Emerging Europe (EE) Dependent variables | Emerging Europe (EE) Dependent variables | |||
|---|---|---|---|---|
| Explanatory variables | Equation 1 bank stability | Equation 2 loan-loss reserves | Equation 1 bank stability | Equation 2 loan-loss reserves |
| 1 | 2 | 3 | 4 | |
| Zijt-1 (log z_md) | 0.426** | -0.01 | ||
| (12.04) | (0.33) | |||
| Zijt-1 (log z_rol) | 0.613** | -0.049* | ||
| (17.61) | (2.19) | |||
| Loan-loss reserves/loansijt-1 | -0.01 | 0.750** | 0.078+ | 0.711** |
| (0.23) | (30.15) | (1.90) | (24.60) | |
| Credit/GDP jt-1 | 0.00 | 0.00 | ||
| (0.63) | (1.12) | |||
| Total Asset Growthijt-1 | -0.003+ | 0.00 | ||
| (1.74) | (0.54) | |||
| Real GDP growthjt-1 | 0.01 | -0.025* | 0.02 | -0.033** |
| (0.43) | (2.57) | (1.22) | (2.94) | |
| Unemployment ratejt | 0.015** | 0.016** | ||
| (3.50) | (3.47) | |||
| loan-growthijt-1 | -0.003** | -0.002** | ||
| (4.26) | (3.01) | |||
| (loan-growthijt-1)2 | 0.000** | 0.000** | ||
| (4.24) | (3.64) | |||
| Cost-to-income | -0.009** | -0.005** | ||
| (4.48) | (4.17) | |||
| Constant | 4.998** | -0.869** | 1.872** | -0.859** |
| (13.45) | (5.53) | (9.08) | (6.40) | |
| Observations | 550 | 550 | 452 | 452 |
| “R2” | 0.26 | 0.73 | 0.49 | 0.69 |
Absolute value of z statistics in parentheses+ significant at 10%; * significant at 5%; ** significant at 1% Adding dummies for specific country clusters--like south eastern European or the ten recently acceded EU member states--did not change the results.
3SLS Estimates of Bank Stability and Credit Risk—Emerging Europe 1/
| Emerging Europe (EE) Dependent variables | Emerging Europe (EE) Dependent variables | |||
|---|---|---|---|---|
| Explanatory variables | Equation 1 bank stability | Equation 2 loan-loss reserves | Equation 1 bank stability | Equation 2 loan-loss reserves |
| 1 | 2 | 3 | 4 | |
| Zijt-1 (log z_md) | 0.426** | -0.01 | ||
| (12.04) | (0.33) | |||
| Zijt-1 (log z_rol) | 0.613** | -0.049* | ||
| (17.61) | (2.19) | |||
| Loan-loss reserves/loansijt-1 | -0.01 | 0.750** | 0.078+ | 0.711** |
| (0.23) | (30.15) | (1.90) | (24.60) | |
| Credit/GDP jt-1 | 0.00 | 0.00 | ||
| (0.63) | (1.12) | |||
| Total Asset Growthijt-1 | -0.003+ | 0.00 | ||
| (1.74) | (0.54) | |||
| Real GDP growthjt-1 | 0.01 | -0.025* | 0.02 | -0.033** |
| (0.43) | (2.57) | (1.22) | (2.94) | |
| Unemployment ratejt | 0.015** | 0.016** | ||
| (3.50) | (3.47) | |||
| loan-growthijt-1 | -0.003** | -0.002** | ||
| (4.26) | (3.01) | |||
| (loan-growthijt-1)2 | 0.000** | 0.000** | ||
| (4.24) | (3.64) | |||
| Cost-to-income | -0.009** | -0.005** | ||
| (4.48) | (4.17) | |||
| Constant | 4.998** | -0.869** | 1.872** | -0.859** |
| (13.45) | (5.53) | (9.08) | (6.40) | |
| Observations | 550 | 550 | 452 | 452 |
| “R2” | 0.26 | 0.73 | 0.49 | 0.69 |
Absolute value of z statistics in parentheses+ significant at 10%; * significant at 5%; ** significant at 1% Adding dummies for specific country clusters--like south eastern European or the ten recently acceded EU member states--did not change the results.
3SLS Estimates of Bank Stability and Credit Risk—EU15 1/
Absolute value of z statistics in parentheses; + significant at 10%; * significant at 5%; ** significant at 1%. EU15 comprises the Euro Area, Denmark, Sweden and the United Kingdom.
3SLS Estimates of Bank Stability and Credit Risk—EU15 1/
| EU15 Dependent variables | EU15 Dependent variables | |||
|---|---|---|---|---|
| Equation 1 | Equation 2 | Equation 1 | Equation 2 | |
| Explanatory variables | bank stability | loan-loss reserves | bank stability | loan-loss reserves |
| 1 | 2 | 3 | 4 | |
| Zijt-1 (log z_md) | 0.299** | -0.003 | ||
| (10.54) | (0.31) | |||
| Zijt-1 (log z_rol) | 0.650** | -0.027+ | ||
| (29.99) | (1.78) | |||
| Loan-loss reserves/loansijt-1 | -0.008 | 0.876** | -0.034+ | 0.835** |
| (0.23) | (61.09) | (1.72) | (57.27) | |
| Total Asset Growthijt-1 | -0.004* | -0.0004 | ||
| (2.37) | (0.42) | |||
| Real GDP growthjt-1 | -0.002 | 0.006 | -0.012+ | 0.001 |
| (0.18) | (1.20) | (1.72) | (0.25) | |
| Unemployment ratejt | 0.016** | 0.016** | ||
| (3.51) | (3.47) | |||
| Unemployment ratejt-1 | 0.008 | 0.004 | ||
| (0.74) | (0.73) | |||
| loan-growthijt-1 | -0.001 + | -0.002* | ||
| (1.87) | (2.51) | |||
| (loan-growthijt-1)2 | 0.000** | 0.000* | ||
| (3.07) | (2.20) | |||
| Cost-to-income | -0.013** | -0.003** | ||
| (5.40) | (3.60) | |||
| Constant | 5.224** | -0.602** | 3.059** | -0.487** |
| (16.84) | (5.92) | (13.60) | (3.22) | |
| Observations | 1235 | 1235 | 1161 | 1161 |
| “R2” | 0.13 | 0.79 | 0.47 | 0.77 |
Absolute value of z statistics in parentheses; + significant at 10%; * significant at 5%; ** significant at 1%. EU15 comprises the Euro Area, Denmark, Sweden and the United Kingdom.
3SLS Estimates of Bank Stability and Credit Risk—EU15 1/
| EU15 Dependent variables | EU15 Dependent variables | |||
|---|---|---|---|---|
| Equation 1 | Equation 2 | Equation 1 | Equation 2 | |
| Explanatory variables | bank stability | loan-loss reserves | bank stability | loan-loss reserves |
| 1 | 2 | 3 | 4 | |
| Zijt-1 (log z_md) | 0.299** | -0.003 | ||
| (10.54) | (0.31) | |||
| Zijt-1 (log z_rol) | 0.650** | -0.027+ | ||
| (29.99) | (1.78) | |||
| Loan-loss reserves/loansijt-1 | -0.008 | 0.876** | -0.034+ | 0.835** |
| (0.23) | (61.09) | (1.72) | (57.27) | |
| Total Asset Growthijt-1 | -0.004* | -0.0004 | ||
| (2.37) | (0.42) | |||
| Real GDP growthjt-1 | -0.002 | 0.006 | -0.012+ | 0.001 |
| (0.18) | (1.20) | (1.72) | (0.25) | |
| Unemployment ratejt | 0.016** | 0.016** | ||
| (3.51) | (3.47) | |||
| Unemployment ratejt-1 | 0.008 | 0.004 | ||
| (0.74) | (0.73) | |||
| loan-growthijt-1 | -0.001 + | -0.002* | ||
| (1.87) | (2.51) | |||
| (loan-growthijt-1)2 | 0.000** | 0.000* | ||
| (3.07) | (2.20) | |||
| Cost-to-income | -0.013** | -0.003** | ||
| (5.40) | (3.60) | |||
| Constant | 5.224** | -0.602** | 3.059** | -0.487** |
| (16.84) | (5.92) | (13.60) | (3.22) | |
| Observations | 1235 | 1235 | 1161 | 1161 |
| “R2” | 0.13 | 0.79 | 0.47 | 0.77 |
Absolute value of z statistics in parentheses; + significant at 10%; * significant at 5%; ** significant at 1%. EU15 comprises the Euro Area, Denmark, Sweden and the United Kingdom.
E. A Back-of-the-Envelope Calculation of Credit Risk in Croatian Banks
21. The estimates obtained in Tables V.2. and V.3. can be used to calculate the effect of a down cycle on Croatian banks’ capitalization. To keep the analysis simple, a downturn is defined as a reduction in real GDP growth in isolation or in combination with an increase in the unemployment rate.
22. The following steps are followed. First, based on the regression estimates for the loan-loss reserve equation for Emerging Europe in Table V.2., we calculate the sensitivity of loan-loss reserves/total loans of Croatian banks to a 1-unit or 1-percentage point adverse change in the real GDP growth rate and/or the unemployment rate. Second, we take total assets for each bank, as a proxy for total loans, and calculate the amount of increase in reserves based on total assets of each bank.14 Third, this increase in nominal reserves is deducted from both regulatory capital and risk-weighted assets to come up with the new capital adequacy ratio (CAR). Fourth, we repeat the previous steps based on the EU15 estimates shown in Table V.3. The results of this exercise are shown in Tables 4 and 5. The Appendix provides more details.
23. Notwithstanding the potential overestimation of credit risk, the results raise the prospect of Croatian banks being very sensitive to adverse changes in the economic cycle.15 As Table V.4 shows, even a one standard deviation change in real GDP growth and unemployment rates could push several banks, accounting for almost 49 percent of total banking system assets, below the 10 percent minimum CAR. The adverse effect is manifold if the extreme historical realizations—a very low-probability event—for Croatia are considered.
Change in Loan-Loss Reserves/Loans in Response to Adverse Economic Cycle—Based on Emerging Europe Regression Estimates
One-standard deviation of real GDP growth rate and unemployment rate are 2.06 percent and 2.66 percent respectively.
Minimum real GDP growth in Croatia -0.86; Maximum unemployment rate 22.3. The 2006 baselines for the two variables are taken as 4.5 and 13.3 respectively.
Change in Loan-Loss Reserves/Loans in Response to Adverse Economic Cycle—Based on Emerging Europe Regression Estimates
| Change (from 2006 baseline) in | Change in loan- loss reserves/loans due to | ||||||
|---|---|---|---|---|---|---|---|
| Type of change in the real GDP growth rate and the unemployment rate | Real GDP growth rate | Unemploym ent rate | Change in Real GDP growth rate | Change in Unemployment rate | Old aggregate CAR (as of June 30, 2006) | New aggregate CAR | Banks below 10% CAR (in % share of total assets of banking system) |
| Unit change | -1.00 | 1.00 | 0.65 | 0.39 | 12.88 | 11.70 | 22.50 |
| 1 s.d. change 1/ | - 2.06 | 2.66 | 1.34 | 1.04 | 12.88 | 10.00 | 48.90 |
| Historical extreme 2/ | - 5.36 | 9.10 | 3.49 | 3.55 | 12.88 | 3.90 | 97.65 |
One-standard deviation of real GDP growth rate and unemployment rate are 2.06 percent and 2.66 percent respectively.
Minimum real GDP growth in Croatia -0.86; Maximum unemployment rate 22.3. The 2006 baselines for the two variables are taken as 4.5 and 13.3 respectively.
Change in Loan-Loss Reserves/Loans in Response to Adverse Economic Cycle—Based on Emerging Europe Regression Estimates
| Change (from 2006 baseline) in | Change in loan- loss reserves/loans due to | ||||||
|---|---|---|---|---|---|---|---|
| Type of change in the real GDP growth rate and the unemployment rate | Real GDP growth rate | Unemploym ent rate | Change in Real GDP growth rate | Change in Unemployment rate | Old aggregate CAR (as of June 30, 2006) | New aggregate CAR | Banks below 10% CAR (in % share of total assets of banking system) |
| Unit change | -1.00 | 1.00 | 0.65 | 0.39 | 12.88 | 11.70 | 22.50 |
| 1 s.d. change 1/ | - 2.06 | 2.66 | 1.34 | 1.04 | 12.88 | 10.00 | 48.90 |
| Historical extreme 2/ | - 5.36 | 9.10 | 3.49 | 3.55 | 12.88 | 3.90 | 97.65 |
One-standard deviation of real GDP growth rate and unemployment rate are 2.06 percent and 2.66 percent respectively.
Minimum real GDP growth in Croatia -0.86; Maximum unemployment rate 22.3. The 2006 baselines for the two variables are taken as 4.5 and 13.3 respectively.
24. The adverse effect on the capitalization of the Croatian banks would be much less if these banks behaved more like EU15 banks. To run this counterfactual analysis, the regression estimates from the EU1 5 banks (Table V.3) are used to calculate the effect on Croatian banks’ capitalization in the event of adverse economic conditions. The results (Table V.5) show that for one-percentage point changes in the real GDP growth rate and the unemployment rate, the post-shock CAR is much higher than when the estimates from the EE case are used (Table V.4)—only banks accounting for a little over 1 percent share of assets fall below the minimum CAR. This result—and other results in Table V.5.—is driven by the fact that EU1 5 banks seem to create larger buffers in advance of real GDP downturns, lessening the need to increase them when the actual downturn sets in. However, the EU15 banks do react significantly to changes in the unemployment rate, perhaps due to the relatively recent boom in mortgage credit to households, the sector that would be more sensitive to the unemployment rate. This is why Croatian banks, in this latter exercise, would still be highly affected if the extreme historical realization for unemployment were to occur. But even then, the post-shock CAR for the aggregate banking sector would still be much higher than the post-shock CAR in Table V.4.
Change in Loan-Loss Reserves/Loans in Response to Adverse Economic Cycle—Based on EU15 Regression Estimates
One-standard deviation of real GDP growth rate and unemployment rate are 2.06 percent and 2.66 percent respectively.
Minimum real GDP growth in Croatia-0.86; Maximum unemployment rate 22.3. The 2006 baselines for the two variables are taken as 4.5 and 13.3 respectively.
Change in Loan-Loss Reserves/Loans in Response to Adverse Economic Cycle—Based on EU15 Regression Estimates
| Change (from 2006 baseline) in | Change in loan- loss reserves/loans due to | ||||||
|---|---|---|---|---|---|---|---|
| Type of change in the real GDP growth rate and the unemployment rate | Real GDP growth rate | Unemploym ent rate | Change in Real GDP growth rate | Change in Unemployment rate | Old aggregate CAR (as of June 30, 2006) | New aggregate CAR | Banks below 10% CAR (in % share of total assets of banking system) |
| Unit change | -1.00 | 1.00 | 0.00 | 0.47 | 12.88 | 12.30 | 1.20 |
| 1 s.d. change 1/ | -2.06 | 2.66 | 0.00 | 1.25 | 12.88 | 11.40 | 22.50 |
| Historical extreme 2/ | -5.36 | 9.10 | 0.00 | 4.29 | 12.88 | 7.60 | 94.50 |
One-standard deviation of real GDP growth rate and unemployment rate are 2.06 percent and 2.66 percent respectively.
Minimum real GDP growth in Croatia-0.86; Maximum unemployment rate 22.3. The 2006 baselines for the two variables are taken as 4.5 and 13.3 respectively.
Change in Loan-Loss Reserves/Loans in Response to Adverse Economic Cycle—Based on EU15 Regression Estimates
| Change (from 2006 baseline) in | Change in loan- loss reserves/loans due to | ||||||
|---|---|---|---|---|---|---|---|
| Type of change in the real GDP growth rate and the unemployment rate | Real GDP growth rate | Unemploym ent rate | Change in Real GDP growth rate | Change in Unemployment rate | Old aggregate CAR (as of June 30, 2006) | New aggregate CAR | Banks below 10% CAR (in % share of total assets of banking system) |
| Unit change | -1.00 | 1.00 | 0.00 | 0.47 | 12.88 | 12.30 | 1.20 |
| 1 s.d. change 1/ | -2.06 | 2.66 | 0.00 | 1.25 | 12.88 | 11.40 | 22.50 |
| Historical extreme 2/ | -5.36 | 9.10 | 0.00 | 4.29 | 12.88 | 7.60 | 94.50 |
One-standard deviation of real GDP growth rate and unemployment rate are 2.06 percent and 2.66 percent respectively.
Minimum real GDP growth in Croatia-0.86; Maximum unemployment rate 22.3. The 2006 baselines for the two variables are taken as 4.5 and 13.3 respectively.
F. Concluding Remarks and Policy Implications
25. This paper adds a new dimension to analyzing bank stability in Croatia. Using data on other emerging European countries, it quantifies the effect of an economic downturn on loan-loss reserves, and uses this quantification to calculate the adverse effect of higher loan-loss reserves on the capitalization of the Croatian banking system. While caveats should be mentioned, there are a number of aspects of the analysis that help to give new insights.
26. On caveats, the data from Bankscope covers the systemically important banks in each country, but does not cover all the banks. In addition, although panel data is used, the estimation technique, to preclude complications of using lagged dependent variables in systems regressions, does not exploit panel characteristics by taking fixed or random effects. Finally, the analysis does not capture the indirect credit risk stemming from foreign exchange or foreign exchange indexed loans to unhedged borrowers.
27. That being said, the number of loan-loss episodes related to large exchange rate depreciation events is few in the countries in the sample, given their closely managed exchange rate regimes. Thus an econometric analysis on foreign exchange induced credit risk would be difficult to make under such limited information. The estimation of the system of equations does benefit from efficiency gains by taking into account feedback effects between loan-loss reserves and bank stability.
28. Novel features of the analysis include the use of the regression results to quantify the potential deterioration of capitalization under scenarios for an economic downturn. Furthermore, using data on EU15 banks, the analysis goes further to demonstrate that the potential deterioration of capitalization would be less if Croatian banks behaved more like the EU15 banks.
29. To close, some key policy-related observations and policy implications are summarized below.
The analysis confirms that more stable banks are susceptible to lower credit risk. Banks with higher z-indices required less loan-loss reserves, ceteris paribus.
Accelerating credit growth increases credit risk, especially if credit growth is already high. Accelerating credit growth in previous periods along with an economic downturn could have severe consequences on credit quality, and hence on profitability and capitalization. The negative effect of these latter factors is accentuated in banks with a lower stability or z-index.
A downturn could have a large and negative effect on capitalization, thus larger buffers would be helpful in Croatia. These buffers could be built either in the form of higher provisions for unidentified loan-losses or higher risk weights on risky loans. This policy implication is similar in spirit with recommendations made in Kraft and Jankov (2005) that argue for higher capital requirements for fast growing banks to prevent future asset quality problems from turning into bank failures.16
Provisions on loans, especially those made under the incorrect assumption that the banks have access to the borrowers’ income in case of default, should be increased. This is because such income would be unavailable if the defaulted borrower becomes unemployed.
Croatian banks would also be less vulnerable if they behaved more like EU15 banks. If they did, counterfactual analysis suggested that the adverse effect of a downturn on capitalization would be less.
The possibility that the risk premium embedded in loan interest rates is too low makes it all the more important to evaluate the need for building up provisions or increasing risk weights. Croatian banks are not necessarily passing on the higher risk of foreign exchange denominated or indexed loans to unhedged clients by charging higher interest rates on these loans. A CNB publication (CNB, 2006, pp 49) also suggests that Croatian banks tend to underestimate credit risk during periods of high economic growth. This behavior could be due to the high competition among the top banks. Loan policies could vary between banks, with some banks perhaps better than others in passing on credit risk to their customers. The recent increase in risk weights on loans to unhedged clients by 25 percentage points is a welcome move. It is also encouraging in this context that the CNB has further enhanced supervision of banks’ credit risk management policies by issuing a couple of guidelines on monitoring fx-induced credit risk and household credit risk.17
The stability of Croatian banks could be further enhanced by improving efficiency. The econometric estimates show that banks with lower cost/income ratio enjoys higher stability. And the ratio of cost to income in Croatia was above average in the EU context.
References
Borio, Claudio, Craig Furfine, and Philip Lowe, 2001, “Procyclicality of the financial system and financial stability: issues and policy options,” BIS Papers 1 http://www.bis.org/publ/bispap01a.pdf.
Boyd, John H., Gianni De Nicolo, and Abu Al Jalal, 2006, “Bank Risk Taking and Competition Revisited: New Theory and New Evidence,” mimeo, April, forthcoming IMF Working Paper.
Croatian National Bank, 2006, “Macroprudential Analysis,” II(3), September. http://www.hnb.hr/publikac/makrobonitetna-analiza/e-mba-03.pdf?tsfsg=3cf2637317b63c68ec1028d90f843a03.
De Nicolo, Gianni, 2000, “Size, Charter Value and Risk in Banking: An International Perspective,” International Finance Discussion Papers 689, FED Board, December.
De Nicolo, Gianni, and Alexander Tieman, 2006, “Economic Integration and Financial Stability: A European Perspective,” mimeo, September, forthcoming IMF Working Paper.
Laeven, Luc, and Giovanni Majnoni, 2003, “Loan-loss provisioning and economic slowdowns: too much, too late?” Journal of Financial Intermediation, Vol. 12, pp. 178–197.
Kraft, Evan, 2004, “Dynamic Provisioning: results of an initial feasibility study for Croatia,” Paper prepared for the Banking Supervisors of Central and Eastern Europe Conference, Dubrovnik, Croatia, May 27–28.
Kraft, Evan, and Ljubinko Jankov, 2005, “Does Speed Kill? Lending Booms and their Consequences in Croatia,” Journal of Banking and Finance, Vol. 29, pp. 105–121.
Maechler, Andrea, Srobona Mitra, and Delisle Worrell, 2006, “Exploring Financial Risks and Vulnerabilities in New and Potential EU Member States,” IMF WP, forthcoming. The 2005 DG-ECFIN conference version available at http://ec.europa.eu/economy finance/events/2005/bxlforum1005/maechler_mitra_en.pdf.
Tamirisa, Natalia, and Martin Cihak, 2006, “Credit, Growth and Financial Stability,” Selected Issues, IMF Country Report No. 06/392. http://www.imf.org/external/pubs/ft/scr/2006/cr06392.pdf.
Tamirisa, Natalia, and Deniz Igan, 2006, “Credit Growth and Bank Soundness in the New Member States,” IMF Country Report No. 06/414. http://www.imf.org/external/pubs/ft/scr/2006/cr06414.pdf.
Appendix. Credit Risk Calculation
It should be noted that the dependent variable is the logit transformed loan-loss reserves to total loans, and the specification has a lagged dependent variable. Let loan-loss reserves/total loans be Y in percent and the unemployment rate or real GDP growth rate (or any variable in question) be X in percent. Then the (medium-term) change in Y in response to changes in X is given by
We assume that this change in Y is entirely due to a change in loan-loss reserves, and that there is no change in total loans (the denominator). Thus, the nominal change in loan-loss reserves of bank i is
The resulting change in the capital adequacy ratio (CAR) of bank i is:
The calculations (not shown) were based on bank-by-bank data, as of June 2006, published by the CNB on their website.19 The aggregate (weighted by risk-weighted assets) CAR of the banking system is reported in Tables V.4 and V.5.
Prepared by Srobona Mitra.
In Croatia, almost 80 percent of loans are in foreign currency or in kuna indexed to foreign currency. A substantial portion of these loans are made to unhedged clients, raising implications for credit risk in the event of a large depreciation. Thus, while an analysis of the foreign exchange induced credit risk implications of large movements in the exchange rate would have been desirable, the closely managed exchange rates maintained in most countries in our sample precluded such analysis.
Maechler, Mitra, Worrell (2006) discusses provisions and bank stability in a single equation framework, and Tamirisa and Igan (2006) and Cihak and Tamirisa (2006) discuss the relationship between credit growth and bank stability in a two-equation three-stage least squares framework.
The paper discusses later (in Section C) why changes in real growth rate and the unemployment rate can be expected to generate different responses.
Typically, the market values of equity and assets and shareholders’ profits should be taken to calculate this index. However, due to lack of data on market capitalization of most of the banks in our sample, we have taken the book values of all variables derived from balance sheet data.
Loan-loss reserves(t)=loan-loss reserves(t-1) + new charges to provisions(t) through the profit and loss accounts-(write-offs(t)-recoveries(t)) + currency and other adjustments(t).
The logit transformation of x (in percent) is log(x/(100-x)).
Note that this variable varies between countries but not between banks within the same country; thus it does not have an i subscript.
The cost-to-income ratio, a flow concept, includes provisioning charges. This is the reason for its lagged response to recessions.
For the EE countries, we use the dataset used in MMW.
In order to include episodes involving more traditional macroeconomic cycles, we tried including data on banks from Spain, Portugal and Greece, countries that are so-called non-core EU members. However, including these countries did not affect the estimates significantly. Thus the results including these three countries are not reported separately.
This is consistent with the finding in De Nicolo and Tieman (2005) who find, using market based indicators, that financial risk in large European banks has not declined in the past 15 years.
Kraft (2004) finds a negative association between (lagged) real GDP growth rates and loan-loss provisions over 1998-2003 for a panel of Croatian banks.
The analysis can easily substitute total loans for total assets, once bank-by-bank data on total loans is available. We use bank-by-bank data published on the CNB website as of June 30, 2006.
Some overestimation could be due to the following: (i) total assets are used instead of total loans, which inflates the nominal amount of increase in reserves; and (ii) banks could vary widely in the quality of their loan portfolio in light of their individual provisioning policies and existing buffers. In other words, banks with a higher z-index would have to make less reserves than shown in Table V.1.
However, Kraft (2004) had suggested that dynamic provisioning—similar to the model used in Spain—would not be feasible for Croatia mainly due to unavailability of data over at least two business cycles.
These documents can be obtained from http://www.hnb.hr/supervizija/e-smjernice-za-upravljanjeinformacijskim-sustavom.pdf?tsfsg=45967eb2b3b4ec8a77b2bfb2e369b12e and http://www.hnb.hr/supervizija/e-smjernice-za-upravljanje-kreditnim-rizikom.pdf?tsfsg=65e6f2fc58 11 56cf65704e28f1 e3282b.
The calculations are sensitive to values of
See “Indicators of Banking Institutions Operations” at http://www.hnb.hr/supervizija/esupervizija.htm?tsfsg=7860b148848e3eafbe66832697b67f32.
