Publications

In recent years, a number of theoretical and empirical studies have examined the impact of regulatory capital standards on bank portfolio risk. For example, using the mean-variance framework, Kim and Santomero (1980) and Koehn and Santomero (1988) have shown that increasing regulatory capital standards may have the unintended effect of causing utility (shareholder value) maximising banks to increase portfolio risk. Under these conditions, changes in capital and portfolio risk would be positively correlated. In contrast, studies such as Benston et.al.(1986) have observed that bank capital and portfolio risk may be negatively correlated, as banks maximize the option value of deposit insurance by reducing capital and increasing risk. Furthermore, Furlong and Keeley (1989) have argued that the mean-variance approach is inappropriate because it ignores the option value of deposit insurance. Using a contingent claims model, their results suggest that increased capital standards will not cause banks to increase portfolio risk. This occurs because an increase in capital reduces the value of the deposit insurance put option, thereby reducing the incentive for banks to increase portfolio risk levels. However, one important limitation of the study is that banks continue to have an incentive to maximise risk in the model; an increase in capital merely reduces the magnitude of the gains from risk-taking activity. Gennotte and Pyle (1991) incorporated an adjustment for the value of deposit insurance as suggested by Keeley and Furlong (1990) and also allowed for the expected return on an asset to decrease as a bank increases its holdings. They found that even if an interior optimum for size and risk exists, then a rise in the capital level will lead to increased investment in the risky asset and a greater probability of failure.

Addressing the issue of risk-based capital regulations, Kim and Santomero (1988) examined how the design of risk-based capital standards influences the level of risk in bank portfolios. The results are particularly interesting, because they found that a risk-based rule designed to minimize the probability of bank failure will lead banks to choose high risk assets. Empirical evidence on the issue presents conflicting conclusions. The study by Haubrich and Watchel (1993) shows how the implementation of the Basle risk standards caused poorly-capitalised banks to reconfigure their portfolios away from high-risk and towards low-risk assets. This result, however, runs contrary to that of Hancock and Wilcox (1992) who found that, banks that had less capital than required by the risk-based standards, shifted their portfolios towards high-risk assets.

Capital adequacy has traditionally been regarded as a sign of strength of the financial system in India. In terms of Section 17 of the Banking Regulation Act, 1949, every banking company incorporated in India is required to create a reserve fund and transfer a sum equivalent to not less than 20 per cent of its disclosed profits, to the reserve fund every year. The Reserve Bank has advised banks to transfer 25 per cent and if possible, 30 per cent to the reserve fund.

Consequent upon the recommendations of the Committee on Financial Sector Reforms (Chairman: Shri M.Narasimham), a capital to risk-weighted assets system was introduced for banks in India since April 1992, largely in conformity with international standards, under which banks were required to achieve a 8 per cent capital to risk-assets ratio. Indian banks with branches abroad were given time till March 31, 1994 (subsequently extended to March 31, 1995) to achieve the norm of 8 per cent CRAR; the capital was to comprise of tier I plus tier II capital, of which tier II should not exceed 100 per cent of tier I (Appendix A3). Accordingly, the pattern of assigning risk weights and credit conversion factors were also delineated, broadly in line with those in the original Accord.^vi Although the switchover to stringent prudential regime did affect the banking system in the initial years, the system exhibited adequate resilience to record substantial improvements in financial strength through higher CRAR over the period. Data for public sector banks reveal that as on March 1996, while only 19 banks satisfied the CRAR of 8 per cent and above, the number increased to 26 in 1999.

In order to assess how banks responded to the capital requirements, we first note the following simple identity:

where C=capital, R=risk-weighted capital ratio, P=portfolio factor, and TA=total assets.

Using the superscript notation for proportionate changes (e.g., Z⁰=DZ/Z), we obtain, after some rearrangement,

(2.1)

Because the risk-adjusted capital requirements are a constraint on R, we see from the above equation that the relation descriptively allocates the adjustment of banks to three possible courses of action: raise capital (increase C), adjust the portfolio factor (lower P) or shrink total assets (lower TA).

Using the Basle standards as a benchmark for providing a basic framework of minimum capital standards, regulators in different countries have supplemented them with a range of other requirements designed to suit the country-specific requirements. Table 2.1 summarises the papers which examine this issue.

As is evident from Table 2.1, most of the studies on capital adequacy and the concomitant portfolio shifts have been with regard to the US experience. The early literature covering US bank behaviour prior to the introduction of formal requirements in that country in 1981 confirms the fact that capital requirements implemented by supervisors on banks were merely indicative in nature. Peltzman (1970) and Mingo (1975) regressed percentage growth in capital on a range of conditioning variables, including the banks’s lagged ABC ratio. Peltzman (1970), using state-wide averaged data found insignificant effects of ABC ratios on subsequent bank capital changes. Mingo (1975) using bank-level data found strong and statistically significant positive effects. Although the methodology used in these studies was quite naïve, they however, put in place the basic framework that most subsequent analyses followed: regressing a capital change variable on conditioning

a. The ';ABC'; ratio was the ratio of actual bank capital to the capital desired by the regulator.

b. In December 1981, the Federal Reserve and the Office of the Comptroller of Currency (OCC) announced a common set of standards to apply to all banks which they regulate. The Federal Deposit Insurance Corporation (FDIC) adopted these standards in 1985. The 1981 standards set a minimum capital ratio of 7 per cent for community banks and 6.5 per cent for regional banks. No formal standards were set for the multinational banks.

variables describing bank’s financial state and the nature of its business. Subsequent research has mainly focused on changes in capital ratios (either leverage ratios or ratio of equity to total risk-weighted assets) rather then merely changes in capital growth rates. Second, early literature made no distinction between the short and the long-run effects of capital requirements (since capital growth was regressed only on conditioning variable and not on lagged capital or capital growth). Most subsequent research has employed a partial adjustment specification in which if Y(t) is the actual capital ratio of the bank and YD(t) is the bank’s target capital ratio at time t, then Y(t) is assumed to be of the form

(2.2)

In this case, u(t) is a random error term and a is a positive parameter. When Y(t-1) exceeds (resp., is less than) YD(t), the sign of a implies than Y(t)-Y(t-1) is, on average, negative (resp., positive). Hence, in the long run, Y(t) will tend to converge towards YD(t) and the magnitude of a reflects the rate at which such convergence occurs. Since the bank’s desired capital ratio YD(t) is unobservable, researchers have employed a proxy, typically replacing YD(t-1) in equation (3.1) with a weighted sum, S_ibiX_i(t), where X_i are lagged conditioning variables describing the state of the economy and the bank’s financial situation and the b_i are the parameters to be estimated.

Among the main papers which investigated the impact of capital requirements using partial adjustment models are those of Shrieves and Dahl (1992) and Jacques and Nigro (1997). Shrieves and Dahl, using data on 1,800 FDIC insured banks for the period 1983-87 found that banks with CRAR of less than 7 per cent (applied by the US authorities at that time), increased their ratio on average by 140 basis points per annum more as compared to other banks. Subsequently, Jacques and Nigro (1997) using a 3SLS method of estimation for US banks for the period 1990-91, found that capital regulation had a significant impact on risk and vice versa.

As evident, most of the papers have examined the issue of capital regulation with regard to US banks for different points of time. Recent studies, in regard to the UK (Ediz et.al. 1998) and Switzerland (Rime, 1998) provide some useful evidence from non-US countries. Ediz et.al. (1998) employed quarterly data on 94 UK banks over the period Q4 1989 to Q4 1995, while Rime looks at annual data on 154 Swiss banks between 1989 and 1995. Both these papers adopt a broadly similar specification. In particular, both introduce among the X_i variables dummies for capital pressure which equal unity when a bank’s capital ratio falls into a zone starting above the regulatory minimum. The gap between the starting point of the zone and the regulatory minimum varies across banks and is taken to be proportional to the time-series standard deviation of the banks’ own capital ratio. The specification captures the idea that (a) banks prefer to maintain a buffer level of capital over and above the regulatory minimum and (b) the width of the buffer will reflect the variability of the banks’ ratio^vi. Both these papers find that regulation is effective in the sense that the dummy variables described above have statistically significant coefficients. In the case of Rime (1998), the impact of regulation (i.e., of a dummy for the capital ratio, which is less than one standard deviation above the regulatory minimum) is statistically significant at the 1 per cent level, while in the case of Ediz et.al.(1998), the ratio of capital to risk adjusted assets increased by 44 basis points per quarter more for banks in the regulatory pressure zone compared to adequately capitalised banks.

Empirical insights from these studies can provide useful policy guidance to regulators in other countries to design their policy stance in accordance with the nature of the banking system in their countries.

The regulatory authorities in the UK, for instance, set two sets of capital requirements, a ';trigger'; ratio, which is the minimum ratio with which banks must comply and a ';target'; ratio, set above the trigger ratio. The purpose of having these dual capital standards serves two purposes: firstly, the gap between them acts as a ';buffer'; in the sense that regulatory pressure is initiated when the actual Capital to Risk Assets Ratio (CRAR) falls below the target. If the CRAR falls even below the trigger level, supervisory authorities initiate even more drastic action.

An important feature of such a practice is the specification of bank specific capital requirements^vii. Given the wide heterogeneity in terms of products and customer preferences among PSBs as well as the adjustment response of the PSBs, the regulatory framework should be designed so as to encourage individual banks to maintain higher CRAR than the stipulated minimum to reflect their differential risk profiles. Such adoption of bank-specific capital requirements has gained currency in recent times in view of the movement towards Risk-Based Supervision (RBS), which envisages inspection of institutions based on their risk profiles. Given that supervisory resources are scarce and different institutions have differential risk profiles, it would be useful if institutions were monitored according to their respective risk profiles. Riskiness, in such a situation would reflect supervisors’ evaluation of the banks’ loan book or possibly their perception about weaknesses in systems and controls. For most UK banks, for instance, capital requirements exceed the Basle minimum of 8 per cent. The ability to vary a bank’s capital requirements administratively provides the regulators with a useful tool for influencing the actions of the bank management.

In the light of the aforesaid discussion, the present paper seeks to address the following two issues: firstly, it seeks to examine, in the Indian context, whether capital requirements have been able to influence bank behaviour. The fact that capital requirements affect bank behaviour does not imply that the impact is undesirable. It is left to the discretion of bank supervisors to judge whether the induced levels of capital are adequate or not, given the broad goals of regulation. A second objective of the paper is to examine whether, consequent upon the introduction of the capital adequacy regime, there are any discernible shifts within each asset category towards riskier assets (or otherwise). Given that the two standard avenues of capital augmentation i.e., securitisation and shifting from banking book to trading book-is not available for banks in India^viii, it therefore follows that banks would have either moved away from riskier assets in order to boost their capital adequacy levels or alternately, would perforce have to access the market so as to maintain prescribed levels of capital^ix (equation 2.1). The paper focuses on the first of these two questions.

In this paper, we employ supervisory data for Indian public sector banks over the period 1997 Q1 to 1999 Q4 to address the issues outlined above. The data is made available by the Off-Site Monitoring and Surveillance Division (OSMOS) of the Reserve Bank. Several points about the data are in order. Firstly, consequent upon the introduction of off-site returns for banks since 1997, banks have been directed to submit data on mandated aspects of liquidity, solvency and asset quality on a quarterly basis^x. The range and extent of disclosures have gradually been enhanced over the years so as to give a clearer picture of bank behaviour to the regulators. To the extent that the data have to be submitted within a stipulated time frame (typically 1-month of the close of the quarter), the timeliness of the information obtained enables the authorities to monitor and understand trends in important banking variables, It however needs to be recognised that the data is unaudited. Notwithstanding the shortcoming, the short span of time (i.e., the close of banking business in every quarter) after which such data is obtained enables one to decipher, with a reasonable degree of accuracy several broad features of bank behaviour.

The panel data used in the above study comprises quarterly balance sheet and income data stretching from 1997 Q1 to 1999 Q4 on 27 public sector banks (PSBs). To the extent that PSBs constitute a sufficiently heterogeneous sample and comprise the bulk of the banking system in India^xi, a study confined to PSBs, in our view, suffices to draw broad inferences about shifts in the asset portfolio of the banking system as a whole. In particular, the two questions in which we are interested are (a) does pressure from supervisors affect bank capital dynamics when capital ratios approach their regulatory minimum, and (b) which items of their balance sheets bear the bulk of adjustment pressure when banks are subject to regulatory pressure?

Towards achieving our objective, a formal regression analysis is performed towards understanding the impact on capital changes of regulatory pressure, holding other influences on capital constant. This latter aspect is important because when a bank falls into financial distress, it might seek to adjust its capital in line with its own internally generated capital targets, even without intervention by regulators (Hancock and Wilcox, 1993). We, therefore, formulate a dynamic, multivariate panel regression model in which changes in capital ratios depend on the lagged level of the ratio, a range of conditioning variables describing the nature of the bank’s business and its current financial health (these proxy for the bank’s internal capital target), and variables that may be regarded as measuring regulatory pressure. Formally, our model may be stated as:

(2.3)

where E(e_n,t)=E(X_n,t,jen,t)=0, t indicates the time period and where X_n,t,j (j=1,2,…, N) are a set of regressors.

where E(z_n,t)=0 for all n,t and E(z_{n,t zm,s})=0 for all t, s, n, m except when t=s and n=m. To include random effects, we suppose that for any bank, E(z²_{n, t})=s_n².

Our conditioning variables are designed to proxy the bank’s own internal capital target and include the following: net interest income over total risk-weighted assets (NIIRWA), fee income over total risk-weighted assets (FIRWA), bank deposits over total risk-weighted assets (BDRWA), total off-balance exposures over total risk-weighted assets (OBSRWA), profits over total risk-weighted assets (PFRWA), provisions over total risk-weighted assets (PVRWA) and 100-percent risk-weighted assets over total risk-weighted assets (HRRWA). The net interest income, fee income and 100-percent risk weighted asset variables reflect the nature and riskiness of the banks’ operations. Bank deposits and off-balance sheet exposures reflect the vulnerability to runs on deposits, although they may also reflect the degree of financial sophistication of the bank and its consequent ability to economise on capital. Total profits and loan loss provisions variables indicate the bank’s state of financial health.

Intuitively, while higher NIIRWA is expected to raise the capital adequacy ratio, similar is the case with FIRWA. Likewise, higher OBSRWA is also expected to raise the capital adequacy standards and the same is the case with PFRWA. Provisions, on the other hand, to the extent it represents an outflow, would lower the capital adequacy ratio. Finally, higher the level of deposits, higher would be capital required to sustain an eventuality of a run on deposits.

The variables and their expected signs is presented in Table 2.2

Before embarking on a formal analysis, we present the correlation matrix of the variables of interest in the analysis. This is presented in Table 2.3.

Of particular interest for the present exercise are the regulatory pressure variables. We measure regulatory pressure in two ways. First, we incorporate a dummy variable that equals one if the bank has experienced an upward adjustment in its trigger ratio (the minimum CRAR that a bank must comply with) in the previous three quarters.^xii This we refer to as the ';trigger'; variable (TRIGD). The second dummy variable we employ is referred to as ';target'; variable (TARGD). This dummy variable equals unity if the CRAR falls close to the regulatory minimum. As we argue above, the degree that a bank is close to the ';trigger'; depends not just on the absolute percentage difference between the current CRAR and the trigger, but also on the volatility of the CRAR. Hence, we calculate this dummy in such a way that it is unity if the CRAR is less than one bank-specific standard deviation (s.d.) above the bank’s trigger. Thus, our hypothesis is that there exists a zone above the trigger in which the bank’s capital ratio choices are constrained by regulatory pressure. In this sense, out study is comparable to Jacques and Nigro (1997).^xiii

The dummy variable associated with one-standard deviation above the trigger may be regarded as introducing a simple regime switch in the model for low levels of the CRAR. To generalise this regime switch, we also estimate a switching regression model in which all the parameters on the conditioning variables (and not just the intercept) are allowed to change when the CRAR is less than one-standard deviation above the trigger. This specification allows for the possibility that all the dynamics of the capital ratio change when the bank is close to its regulatory minimum level of capital.

Table 2.4 reports the regression results for the case in which the dependent variable is the CRAR. Our analysis suggests that the capital requirements significantly affect banks’ capital ratio decisions. The coefficient on the regime dummy is positive and significant. The point estimate implies that banks decrease their CRAR by more than 100-percentage points per quarter when the capital ratio approaches the regulatory minimum. In addition, we find that banks raise their CRAR by roughly 60 percent per quarter following an increase in the trigger ratio by the supervisors.

In Column 3 of Table 2.4, we present the results for the regressions of changes in 100-per cent weighted assets as a ratio of total risk-weighted assets on a lagged level of this ratio and on the same conditioning variables as those included in the CRAR regressions. Although the parameter on the trigger dummy has the expected sign, it is insignificant. The only significant coefficient is the off-balance sheet activity, suggesting the possibility that increasing diversification by public sector banks into off-balance sheet activity is engendering a significant change in 100- percent risk weighted

Figures in brackets indicate t-ratios.

*significant at 1 per cent

# significant at 10 per cent

assets. However, in general, t-statistics are low, suggesting that 100-percent weighted asset ratio does not behave in a statistically stable way over time and across public sector banks. In summary, it seems fair to conclude that banks do not significantly rely on asset substitution away from high-risk-weighted assets to meet their capital requirements as they approach the regulatory minimum.

Table 2.5 reports results similar to the CRAR regressions, but using tier-I and tier-II capital ratios. As the table reveals, banks lower both the tier-I and the tier-II ratios when they come close to their triggers, the decline being significant in case of the latter. In addition, we find that banks raise their tier-I ratios by around ½ per cent per quarter following an increase in trigger ratios.

The second and more interesting question would be to understand how exactly banks achieve changes in their capital ratios if they are subject to regulatory pressure. The most obvious possibilities are either they adjust the asset side of their balance sheet, thereby substituting low-risk government securities for high-risk loans, or alternately raise additional capital from the market by issuing securities or by means of retained earnings (equation 2.1 and the earlier discussion on securitisation).

Figures in brackets indicate t-ratios.

** significant at 5 per cent

# significant at 10 per cent

Table 2.6 shows regressions of changes in 100 per cent weighted assets as a ratio of total risk-weighted assets on the lagged level of this ratio and on the same set of conditioning variables as those included in the CRAR regressions. Although the parameter on the regulatory intervention dummies have the right sign, they are insignificant. The magnitude of the point estimates is fairly small when the level of 100 per cent risk-weighted assets is above the threshold^xv, but is quite substantial, vice versa. However, the lagged level of the variable is fairly significant, suggesting that higher level of hundred per cent weighted assets might be inducing banks to lower the same, both above and below the threshold value of the variable. In general, however, t-statistics on all the conditioning are fairly small, suggesting that this ratio does not behave in a statistically significant way, both over time and across banks. Based on the available data, it appears that banks have not resorted to asset substitution in a significant degree to meet their capital requirements.

The Table also reports results for the case in which the coefficients on all the conditioning variables are allowed to change when the CRAR is greater than or less than one bank-specific standard deviation above the trigger. Not surprisingly, the variables which are significant in the simple model are also important in the switching regression case. While the net interest income and off-balance sheet activity variables are significant, the magnitude of the point estimates are fairly small. In contrast, the estimate on the regulatory pressure dummy is positive, significant and of a reasonable magnitude, indicating that the response of banks to changes in regulatory pressure is more significant when their capital ratios are above the threshold. Another variable which appears to have an important bearing on bank capital ratios is profits. It may be recalled from our earlier discussion that

Figures in brackets indicate t-ratios.

** significant at 5 per cent

# significant at 10 per cent

Section 17 of the Banking Regulation Act, 1949 stipulates that every banking company incorporated in India transfer a sum equivalent to not less than 20 per cent of its disclosed profits to a reserve fund created for the said purpose. Clearly, banks with higher capital ratios could be taking recourse to this measure to shore up their capital adequacy standards.

Finally, Table 2.7 provides results for regressions similar to the above, but using different capital ratios. In particular, referring to the discussion on the switching regression estimate pertaining to CRAR above (Table 2.6), it would be interesting to understand which of tier I capital or tier II capital is relatively more important for effecting changes in the capital of banks. The results, as evidenced from the table (Table 2.7) seems to suggest that the response of banks to increases in their triggers is much higher for tier I capital than tier II capital, suggesting that bulk of the adjustments comes through increases in the former. Also, while profits play a critical role in impacting tier I ratios, especially at higher levels of the ratio, bank deposits appear to be more prominent in the case of tier II capital. In other words, banks with higher levels of core capital could have been ploughing back their profits into reserves, which has been instrumental in raising overall capital levels. This evidence runs contrary to the recent belief of a high amount of cross-holdings of sub-ordinated debt among banks. In view of the above, it seems fair to conclude that, in view of the importance of core capital, profits does seem to have an important role in determining adjustments in the same.

Figures in brackets indicate t-ratios.

*significant at 1 per cent;

** significant at 5 per cent;

# significant at 10 per cent

ii. As observed by Jackson et.al.(1999), the average ratio of capital to risk-weighted assets of major banks in the G-10 countries increased from 9.3 per cent in 1988 to 11.2 per cent in 1996.

iii. For recent studies addressing the impact of risk-based capital standards on bank lending and the credit crunch, see Berger and Udell (1994) and Shrieves and Drew (1995).

iv. The 12 countries are Belgium, Canada, France, Germany, Italy, Japan, Luxembourg, Netherlands, Sweden, Switzerland, United Kingdom and the United States.

v. The risk-based capital standards also incorporated off-balance sheet activities. This is done by converting the value of the off-balance sheet item to an on-balance sheet credit exposure equivalent. The on-balance sheet equivalent is then multiplied by the corresponding risk weight and added to the bank’s total risk-weighted assets.

vi. Keeping the realities of the Indian situation in mind, the risk weights on several on- and off balance sheet items were adjusted to reflect market realities.

vii. A detailed background on trigger and target ratios in the UK is provided in Richardson and Stephenson (2000).

viii. The Report of the Committee on Banking Sector Reforms (Chairman: Shri M.Narasimham) had observed that, ';the RBI should also have the authority to raise [the minimum capital to risk assets ratio] further in respect of individual banks if in its judgement the situation with respect to their risk profile warrants such an increase'; (pp.21, para 3.15).

ix. Another way of capital augmentation is by raising average spreads (popularly termed as net interest margin). However, spreads have shown a decline in recent times for all bank groups.

x. The trading book comprises both the short-term proprietary position taken by the bank in financial instruments for its own account, and its exposures relating to the provision of financial services to customers-for example, agency business. On the other hand, banking book comprises all other transactions, for example, lending and other types of credit activities and long-term investments.

xi. The second tranche of DSB returns covering the aspect of asset liability management has been introduced in July 1999.

xii. As at end-March 1997, end-March 1998 and end-March 1999, PSBs accounted for 80, 82 and 81 per cent, respectively, of the total assets of Scheduled Commercial Banks.

xiii. Note that, our data begins from 1997 Q1 and therefore, for fixing the dummy ';trigger'; variable for the first and second quarters of 1997, we consider the capital adequacy position of the respective bank as on March 31, 1995 and March 1996, respectively, since quarterly data for earlier periods in not available. Similar is the procedure adopted for fixing dummy ';target'; variables.

xiv. In Jacques and Nigro (1997), the regulatory pressure variables are defined in relation to the 8 per cent risk-based capital ratio. Since banks with total risk-based capital ratios above and below the 8 per cent regulatory minimum may react differently, the study partitioned regulatory pressure into two variables: RPG and RPL. RPL equals (1/RBCj-1/8) for all banks with a total risk-based capital ratio less than 8 per cent, and zero otherwise. A second regulatory pressure variable, RPG equals (1/8-1/RBC_j) for all banks with total risk-based ratio greater than or equal to 8 per cent, zero otherwise. The econometric exercise then seeks to examine how the behaviour of these two sets of banks in terms of capital requirement and risk-taking activity is affected by the regulatory stipulations.

xv. In the rest of the Chapter, TRIGD is the trigger dummy and TARGD is the CRAR trigger (less than 1 s.d.)

xvi. The threshold level of 100 per cent risk-weighted assets is selected as the mean level of the variables across all banks across all quarters. Removing the 7 outlier observations (with hundred per cent risk weighted of 100 and above), does not substantially alter the results.