Estimating Deposit Insurance Premiums with Stochastic Interest Rates: The Case of Taiwan Hwei-Lin Chuang, Shih-Cheng Lee, and Min-Teh Yu * April 28, 2006 Abstract This study measures the deposit insurance premium under stochastic interest rates for Taiwan’s banks by applying the two-step maximum likelihood estimation method. The estimation results suggest that the current premiums charging 5, 5.5, and 6 basis points per dollar of insured deposits - are too low and are not proportional to the risks of the insured banks. Moreover, a comparison of the results estimated from different methods implies that the under-estimation issue of the deposit insurance premium due to the problems of statistical inconsistency and efficiency is more serious than that caused by the exclusion of the interest rate risk. An examination of the size effect indicates that large banks are more likely to have lower values of credit risk, asset volatility, and deposit insurance premiums. 1* Chuang: Department of Economics, National Tsing Hua University, Hsinchu 30013, Taiwan. Fax: 886-3-5722476, Email: hlchuang@mx.nthu.edu.tw. Lee : Department of Accounting, Yuan Ze University, Jung-Li 32003, Taiwan. Email: sclee@saturn.yzu.edu.tw. Yu: Department of Finance, Providence University, Taichung 43301, Taiwan. Tel.: 886-4-36320631, Fax: 886-4-26311170, Email: mtyu@pu.edu.tw. The authors acknowledge the support provided by the National Science Council, Taiwan. Corresponding author: Yu. -I- Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks 1. Introduction Ever since the pioneer work of Merton (1977) showing that deposit insurance can be modeled as a put option on the assets of insured institution, a lot of literature on pricing deposit insurance has been accumulated using this option framework. Progress has been seen in both the theoretical development in the estimation method and the empirical applications. As to the estimation method, the market-value-based approach developed by Ronn and Verma (1986) is extensively applied in the literature. Their method set up a two-equation system - one for the relation between the equity value and the bank’s asset value, and the other for the relation of the equity volatility and the asset volatility - to estimate the two unobserved variables in Merton’s pricing model - namely, the market value and volatility of assets. However, as shown in Duan (1994), the estimators of the Ronn-Verma method are statistically inconsistent, because their assumption of a constant equity volatility is incompatible with the nature of stochastic equity volatilities implied in Merton’s pricing model. Duan (1994) therefore introduces a maximum likelihood estimation (MLE) approach to deal with this problem by fully utilizing the distributional assumption in the specification of the theoretical model. The empirical estimation of the deposit insurance premiums for banks in Taiwan did not appear in the literature until the early 1990s.2 The fair pricing of the deposit insurance premium has been investigated by Yu and Hsu (1993) and Duan and Yu (1994). Yu and Hsu (1993) compute the premiums directly using the Ronn-Verma method. Duan and Yu (1994) estimate the deposit insurance value for a group of Taiwanese banks from 1985 to 1992 and find significant differences in the deposit 2 The Central Deposit Insurance Corporation (CDIC) of Taiwan was established in 1985. It was a voluntary plan at first, but became compulsory for all depository institutions in 1999. The CDIC charged a flat premium rate of 5 basis points (bps) in 1985 and 1986, 4 bps in 1987, and 1.5 bps in 1988 through 2000. A risk-based premium has been established since January 1, 2000, and the premium rates can be 5, 5.5, or 6 bps. -2- Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks insurance values estimated from the Ronn-Verma and MLE methods. Both studies conclude that Taiwan’s premium policy provides subsidies to financial institutions. In the deposit insurance literature, a typical way to model the dynamics of the bank asset value is as a lognormal diffusion process. This modeling approach fails to explicitly take into account the impact of stochastic interest rates on the asset value. This shortcoming is particularly important for modeling a bank’s asset value, because it is common for banks to hold a large portion of interest-rate-sensitive assets. Duan, Moreau, and Sealey (DMS) (1995) are the first to incorporate the dimension of the stochastic interest rate into the pricing of deposit insurance. However, the DMS method suffers the same inconsistent problem embedded in the Ronn-Verma method. To solve the inconsistent problem, Duan and Simonato (2002) apply the two-step MLE approach to the DMS model. According to their empirical results for a group of U.S. banks, the deposit insurance premiums using a maximum likelihood estimation are much higher than those using the modified Ronn-Verma method. Moreover, their Monte Carlo simulation supports the performance of the MLE method. In order to measure the effect of the interest rate risk on valuing deposit insurance in Taiwan, we intend to adopt the DMS deposit-insurance model and estimate the premiums using the two-step MLE procedure to measure the deposit insurance premiums with stochastic interest rates. We then compare our estimates with alternative methods and previous results. The structure of the paper is as follows. Section 2 reviews the DMS (1995) model for our empirical framework. The two-step MLE procedure and the description of the data are presented in Section 3. Section 4 presents the empirical findings and the discussion of the results. The final section concludes this study by a brief summary of our major findings. -3- Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks 2. Model As the purpose of this study is to reexamine the deposit insurance premium with the consideration of the interest rate risk for Taiwanese banks, our model is based on the framework developed in DMS (1995). DMS extend Merton (1977) and Ronn and Verma (1986) to incorporate stochastic interest rates into the deposit insurance pricing model by generalizing Ronn and Verma’s two-equation system into a three-equation structure via dividing the second equation into two separate equations in order to capture interest rate risk. The instantaneous interest rate is assumed to follow the mean-reverting stochastic process as in Vasicek (1977) and can be expressed as: drt q (m rt )dt dZ rt , (1) where rt is the instantaneous spot rate at time t, m is the long-run mean of the instantaneous spot rate, is the volatility of the instantaneous spot rate, q is a positive constant measuring the magnitude of the mean-reverting force, and Z rt is a Wiener process. As to the bank’s assets, they are assumed to be characterized by a log-normal process as the following: dVt dt V dZ Vt , Vt (2) where Vt represents the value of bank assets at time t, and V respectively denote the instantaneous expected return and the volatility of the return on bank assets, and Z Vt is a Wiener process. The volatility of the assets of a depository institution, V , is decomposed into two parts - the interest rate risk and the credit risk as the following: V V2 2 2 , (3) -4- Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks where V is the instantaneous interest rate elasticity of the assets, and V V / . We define to be the volatility of assets that is orthogonal to the interest rate and 1 V (1 2 ) 2 . In addition, is a correlation coefficient between the Wiener processes Z Vt and Z rt , i.e., corr (dZVt , dZ rt ) / dt . Because the parameter V cannot be observed directly, DMS (1995) derive the following two equations for the observable variable: bank’s equity return, denoted by Qt as the substitutes. Q t [V B(t , T )] B(t , T ) , (4) Q Q2 2 t2 2 , (5) t and t where Qt t is the volatility of return on equity, t N (ht )Vt / Qt , and B(t , T ) [1 e q (t ,T ) ] / q , which denotes the negative of the instantaneous interest rate elasticity of the default-free zero-coupon bond. As shown in Vasicek (1977), given the assumption of a constant risk premium , the price of a zero-coupon bond with $1 face value and T-t periods of maturity, P(rt , t , T ) , is equal to: P(rt , t , T ) A(t , T )e B (t ,T ) rt , (6) where v 2 B 2 (t , T ) A(t , T ) exp ( B(T t ) (T t )) , 4q v v 2 m 2 . q 2q Let the face value of insured deposits be F. Since the deposits are insured, the full obligation to the depositors at time T is X Fe RT , where R is the fixed and continuously compounded rate of return on these deposits. We can relate the equity -5- Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks value, Qt , and the asset value, Vt , of a depository institution as the following: Qt Vt N (ht ) XP(rt , t , T ) N (ht t ) , (7) where ht t Vt ln , t P(rt , t , T ) X 2 1 (T ) 1 q (T ) 2 e 1 q q t2 (V2 v 2 2 )(T t ) 2V v 2 (T ) 2 1 e 2 q (T ) v 2 2 3 e q (T ) 1 3 q 2q q . Based on the stochastic process specification for the instantaneous interest rate (equation (1)) and for the bank’s assets (equation (2)), it can be shown that the market value of deposit insurance premium per dollar of insured deposits at time t, I t , is given by: I t XP(rt , t , T )[1 N (ht t )] Vt [1 N (ht )] . (8) By solving the simultaneous system of Equations (4), (5), and (7), we can obtain the values of V , , and Vt . We then can solve for the premium of deposit insurance by plugging the values of Vt , V , and into Equation (8). 3. Estimation Procedure and Data 3.1 MLE for Vasicek’s Term Structure Model The log-likelihood function of the instantaneous spot rate of Vasicek (1977) as shown in equation (1) is developed in Duan (1994) as follows: Lr (rt , t 1,..., n; q, m, v) n n 1 n 1 1 ln( 2 ) ln V (v, q ) [rt m (rt 1 m)e q ]2 , 2 2 2V (v, q) t 2 (9) where V (v, q ) is the conditional variance of rt . However, rt is unobservable and therefore the likelihood function shown in equation (9) cannot be applied. Given that -6- Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks the price of a zero-coupon bond with $1 face value and a maturity of T-t period is specified as equation (6), at a certain period of time t, the continuously compounded interest rate on the zero-coupon bond with maturity T-t, Rt (T t ) , can be written as: Rt (T t ) B(T t; r ) 1 1 ln Pt (T t; r ) ln A(T t; r ) rt , T t T t T t (10) where r denotes the set of interest rate parameters q , m , , and . As Rt (T t ) is observable, equation (10) is the key for transformation. The transformation from unobservable rt to the observable Rt (T t ) through equation (10) leads to the following log-likelihood function. L( Rt (T t ), t 1,..., n; r ) (n 1) ln( T t ) (n 1) ln B(T t; r ) ^ where r t ( r ) n 1 n 1 ln( 2 ) ln V (v, q ) 2 2 n ^ ^ 1 [r t ( r ) m (r t 1 ( r ) m)e q ] 2 , 2V (v, q ) t 2 1 [(T t ) Rt (T t ) ln A(T t ; r )] . B(T t ; r ) This (11) likelihood function based on the observable Rt (T t ) can then be used to estimate the interest rate parameters r . 3.2 Two-Step MLE for the Deposit Insurance Premiums D&S (2002) develops a MLE procedure to estimate the three-equation system of DMS (1995) in order to avoid the inconsistency problem due to confusing the parameters with random variables. This MLE procedure is briefly described as below. Let wt be a 2×1 vector of unobserved variables at time t and wt [rt , ln Vt ]' , and yt is a 2×1 vector of observed variables at time t and yt [ P(rt , t , T ), Qt ]' . Let be a parameter vector, in which elements are related to the two unobserved variables and [q, m, v, , , V ,]' . The mapping relation between the observable and the unobservable variables is Y M (W ; ) , where Y [ y1' ,..., y n' ]' and -7- Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks W [ w1' ,..., wn' ]' . Given this mapping relation, it can be shown that the density function of W is a multivariate normal distribution (D&S (2002), p. 115), and therefore we can derive an expression for the likelihood function of W . V Define ut rt , ln t Vt 1 Vˆt ( ) . Applying the and uˆ t ( ) rˆt ( ), ln Vˆ ( ) t 1 above mapping relation, the logarithm of the full-information likelihood function for the DMS (1995) model is then given by: L( ; P(), Qt , t 1,, n) n 1 1 n ln uˆt ( ) Et 1 (ut ) 1 uˆt ( ) Et 1 (ut ) 2 2 t 2 n ln[ P(rt , t , T ) B(t , T )Vt N (ht )] , (12) t 2 where Et 1 (ut ) is the expected value and is the covariance matrix of u t . Given this likelihood function it seems straight forward to estimate the parameter values in and then substitute these estimates into equation (8) to solve for the deposit insurance premium It. However, there are two issues for directly applying the likelihood function of equation (12) to obtain the parameter estimates. First, Equation (12) is specified for a single bank only. If there is more than one bank, then this likelihood function should be expanded for joint estimation. This will become inpractically complex as the number of banks grows. Second, the time structure effect of the interest rate is a long-run phenomenon, as Vasicek’s mean-reverting process for the interest rate suggests that it usually takes a long time for the interest rate to return to its mean level. On the contrary, the movement of asset value (and the movement of equity value) changes rapidly, or in other words, it is a short-run phenomenon and should be captured in a short period of time. Therefore, it is inappropriate to set the same time horizon for estimating parameters of the bond pricing model to capture the mean reversion in the interest rate and for estimating the equity valuation function to -8- Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks yield the volatility parameters. D&S (2002) proposes a two-step estimation procedure to deal with these problems. The first step is to estimate the interest rate parameters only, which is similar to the approach in Duan (1994). The second step is to estimate the equity valuation parameters in Equation (12) taking the interest rate parameters obtained in the first step as given. This two-step estimation procedure not only produces consistent estimators, but also allows for setting different time horizons in the estimation of the interest rate parameters and the parameters of the bank’s equity value function. In sum, our estimation procedure can be described as a two-step maximum likelihood approach. The first step estimates the interest rate parameters for each year using daily commercial paper observations during the ten-year preceding period. A set of ten-year interest rate data is used in order to capture the long-run mean-reverting characteristics. For example, the daily observations of 1982 to 1991 are taken to obtain the interest rate parameters of 1991. The same procedure is run to yield interest rate parameters for each year from 1991 through 2002. Hence, we have twelve sets of interest rate parameter estimates. The second step estimates equity valuation parameters of the objective year with the interest rate parameters treated as given - that is, in the second step the interest rate parameters are set at the values calculated from the first step, and then the equity valuation parameters are obtained based on the data of daily equity and quarterly updated debt observations for each bank at each available year. This procedure is repeated twelve times for each bank at the available years in our sample period. 3.3 Data The sample in this study includes twenty-eight banks chosen from domestic -9- Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks CDIC-insured financial institutions in Taiwan. These sample banks are all listed in the TSE and therefore their daily stock return data are available. Their listed codes in the TSE are 2801, 2802, 2803, 2806, 2807, 2808, 2809, 2811, 2812, 2815, 2824, 2826, 2828, 2829, 2830, 2831, 2834, 2836, 2837, 2838, 2839, 2840, 2842, 2843, 2844, 2845, 2847, and 2849. The daily stock return and the quarterly equity and debt data of balance sheets are from the TEJ database. These observations span from 1991 to 2002. The dividend-adjusted market value of equity is measured by multiplying the daily price with the common shares outstanding. The debt is updated quarterly according to the quarterly balance sheet. Because the purpose of this study is to examine the deposit insurance premium, it would be better to construct a debt series including the deposits only. However, the TEJ database did not record the detailed data of deposits until recent years, and as such it is impossible to construct a complete debt series for total deposits for our entire sample period. Therefore, the debt variable is defined to be total debt from the balance sheets in our debt series. One advantage of using the total debt information is that historically all debts, insured and uninsured, are treated the same when banks fail. The interest rate data used in this study are drawn from the three-month commercial paper collected from the TEJ database. The daily observations of the primary 90-day commercial paper are drawn for years from 1982 to 2002 to estimate the interest rate parameters. 4. Results and Discussion 4.1 Interest Rate The estimation results of the interest rate parameters from the first step are - 10 - Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks reported in Table 1. The estimates of the long-run mean m based on a 10-year sample period are all significant for the years from 1991 to 2002. This estimate increases gradually from 0.686 in 1991 to the maximum of 0.808 in 1997. It then reduces at a faster rate from 0.808 to the minimum of 0.0498 in 2002. This inverse U shape of a first increasing pattern and then a decreasing pattern is also shown in the estimated values for the volatility parameter . The maximum value of estimates for is 0.0513 for the year 1997 and the minimum value occurs at 0.0282 for the year 2002. The estimates of the mean-reverting force parameter q vary a great deal from the minimum of 0.4597 in 1991 to the maximum of 1.8930 in 2000. This greater variation in the estimation results for q is similar to those found in D&S (2002). Moreover, there is no clear pattern for the changes in the estimated values of q . Finally, the estimates of risk premium are all insignificant with an undetermined sign. 4.2 Deposit Insurance Premium The second step maximum likelihood estimation results of the deposit insurance premium are presented in Tables 2 and 3. Table 2 reports the results for one specific bank - the First Commercial Bank. We also estimate the deposit insurance premium without the consideration of the interest rate risk as in Duan and Yu (1994) and the results are presented in the lower panel of Table 2. The estimated values of the deposit insurance premium for this specific bank vary from the minimum of 0.0397 for 1996 to the maximum of 120.7303 for the year 2000. The two-step maximum likelihood estimated deposit insurance premium for each year is higher than the premiums estimated in Duan and Yu (1994). This result is intuitively plausible as the two-step MLE approach include the interest rate risk while Duan and Yu do not consider the - 11 - Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks interest rate risk in their model. In order to have a general idea about the results of all the banks in our sample, the value-weighted averages of the estimation results for the 28 banks in our study are reported in Table 3. We also estimate the deposit insurance premium based on DMS’s (1995) method and Duan and Yu’s (1994) approach for comparison. These estimation results are all reported in Table 3. The value-weighted average of the two-step maximum likelihood estimates of the deposit insurance premium, denoted by IPPmle in Table 3, ranges from 5.0125 to 396.9562. It is plausible to have such a big variation in the estimated values of the deposit insurance premium given the volatile market risk. These value-weighted average estimates are all much higher than the flat-rate assessments at 1.5 basis points from 1997 to 1999 and the three-level flat-rate premiums (5, 5.5, and 6 basis points) charged by CDID from 2000 onwards. Table 3 also reports the year-end market value of equity (denoted by Equity), the year-end book value of debt (denoted by Debt) and the ratio of Equity and Debt (denoted by E/D). According to Table 3, the E/D ratio and the value of assets have a positive relationship. Since the equity is viewed as a call option on the value of assets in which the strike price is its debt value, the inputs of equity and debt value through a proper transformation can yield the estimates of the asset values and volatility, and then can be used to calculate the deposit insurance premiums. In addition, the E/D ratio and the insurance premium are negatively related since the asset values and volatility are the inputs of the deposit insurance premium through their put-option relationship. As the equity is calculated by multiplying the stock price with the outstanding common shares, the denominator of the E/D ratio is highly positively related to the weighted stock market index and the weighted bank stock index. The result of the two-step MLE approach shows that the maximum likelihood estimates of instantaneous expected return on assets, mle , are statistically - 12 - Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks insignificant with un-determined signs. The estimates of credit risk, mle , are significant and positive. All the estimates of the interest rate elasticity of a bank’s assets, mle , are negative. This finding is consistent with the negative relationship between the interest rate and the asset value. The estimates of the bank asset value, Vmle , and the value of insurance premium per dollar of insured deposits, IPPmle , are significant. A larger estimate of Vmle is not necessary to predict a smaller estimate of IPPmle as shown in Table 3. This is because the determinants of IPPmle include not only Vmle , but also the estimates of volatility, mle and mle . The estimates of the bank asset value, Vmle , are higher than the book value of debts. However, they are smaller than the sum of the book value of debts and the market value of equity. Their differences are 20169, 17170, 16990, 17503, 18221, 15990, 17555, 19537, 20568, 31238, 28196, and 42857 million dollars (NT$) from 1991 through 2002 respectively. From the results of the modified Ronn-Verma method, the estimates of credit risk of the non-MLE method, mrv , are smaller than those of the two-step MLE method, mle , which explains for the vast difference between their estimated deposit insurance premiums. The deposit insurance premiums of the modified Ronn-Verma method are highly under-estimated. This finding is similar to that of D&S (2002) for ten U.S. banks. Both the empirical studies of Duan and Yu (1994) for Taiwanese banks and D&S (2002) for the U.S. banks point out that the modified Ronn-Verma method seriously underestimates the insurance premium, because of the problems of inconsistency and asymptotic inefficiency. In other words, our findings provide further evidence to support the arguments in Duan and Yu (1994) and D&S (2002). According to the results estimated based on the MLE of the Ronn-Verma method, the maximum likelihood estimates of instantaneous expected return on assets without considering the stochastic interest rate, DY , are not significant, and neither are those - 13 - Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks of mle . The Ronn-Verma method takes the volatility as a whole, whereas the modified Ronn-Verma method decomposes the volatility into two parts. Hence, the estimates of asset value differ between these two methods. The estimates of asset value based on the Ronn-Verma method are higher than those estimated from the modified Ronn-Verma. Comparing the results of the two-step MLE for the deposit insurance premium with those obtained from the modified Ronn-Verma method (denoted by IPPmrv) and Duan and Yu’s approach (denoted by IPPDY), it is found that the two-step MLE yields much higher deposit insurance premium estimates than IPPmrv, while the two-step MLE and Duan and Yu’s approach have much closer estimation results. These results imply that the under-estimation issue due to the problem of statistical inconsistency and efficiency is more serious than the exclusion of the interest rate risk. Moreover, IPPmle is somewhat larger than IPPDY in every year except for the years of 1991, 1996, 1997 and 1998. According to D&S (2002), the two-step MLE procedure yields higher values of the deposit insurance premium, because of the larger estimates of the credit risk and the lower estimated values for banks’ assets than the corresponding book values of debts. The volatility of assets is divided into two parts: interest-rate risk and credit risk - and their relation can be expressed as V V2 2 2 . The estimation results in Table 3 suggest that neither the interest rate risk nor the credit risk dominates in explaining the volatility of asset values. These two components jointly affect the volatility of asset values. 4.3 Size Effect In order to examine whether bank size has any implication in determining the - 14 - Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks deposit insurance premium, we further classify our samples into three categories 3 and conduct the second-step estimation procedure to investigate the size effect on the deposit insurance premium. Table 4 presents the estimation results of the subsamples for large, medium, and small banks, respectively. Panel A presents the results based on Duan and Yu’s method (1994) which treats the total asset risk as a whole, and Panel B reports the results based on the two-step MLE method which includes the interest rate risk as well as the asset risk. As shown in Panel A, the estimates of total risk, DY , decease as the firm size grows. The estimates of DY are in general the largest for small banks and the smallest for large banks. The estimates of the deposit insurance premium for the large banks are smaller than the estimates of medium banks for all sample periods except for the year 1992. However, there is no consistent pattern of the difference in the estimates of the deposit insurance premium between medium banks and small banks. Comparing the large bank group with the small bank group, the estimates of insurance premium for large banks are smaller than those of the small banks except for 1991. In other words, large banks tend to have smaller estimates of the deposit insurance premium due to their relatively smaller estimates of total risk. According to Panel B, all the estimates of credit risk, mle , are smaller for the large banks than that for the small banks. However, there is no unanimous direction of pattern both between the large and medium banks and between the medium and small banks. 3 The criteria to classify the size of our sample banks are: A bank is classified as a large bank if its asset value is higher than NT$700 billion and a small bank if its asset value is lower than NT$100 billion at the end of 1991 and 1992. As for 1993, 1994, and 1995, a large bank is defined to have assets larger than NT$800 billion, and a small bank has assets lower than NT$150 billion. For the period of 1996-2002, a large bank has assets larger than NT$1,000 billion. A small bank has assets lower than NT$150 billion for 1996, NT$200 billion for 1997, NT$300 billion for 1998, and NT$350 billion for the period of 1999-2002. If a bank is neither a large nor a small one according to the above definition, then it will be classified as a medium bank. - 15 - Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks As for the estimates of interest rate elasticity of bank assets, mle , the absolute value of estimates is smaller for the large banks than that for the medium banks, and the absolute value of estimates is smaller for the medium banks than that for the small banks. The absolute value of the estimates for the interest rate elasticity of bank assets, in general, decreases with banks. Larger banks have smaller interest rate responses. Given that large banks tend to have smaller credit risk and smaller absolute value of interest rate elasticity of bank assets, their estimated deposit insurance premium, IPPmle, is smaller than that for the medium and small banks except for a few years. However, there is no apparent clear-cut difference in the estimated deposit insurance premium between medium and small banks. In sum, both Panel A and Panel B show the risks are lower for large banks, and therefore the estimated deposit insurance premiums are smaller for them. Based on the estimation results of this study, the current risk-based premiums of 5, 5.5, and 6 basis points are too low and are not proportional to the risks of the insured banks. These under-priced premiums as well as not being able to close down undercapitalized banks promptly together may encourage insured banks to take excessive risk, which destabilizes the deposit insurance scheme and the banking system. The regulators should at least enhance the supervision to contain their risk exposure and charge risk-based premiums according to their economic risk exposure measured by market-based data. 5. Conclusion This study applies the two-step maximum likelihood estimation approach derived in D&S (2002) to estimate the interest rate and equity pricing parameters, which are then used to measure the deposit insurance premiums for banks in Taiwan. The first step of our empirical procedure is to estimate the interest rate parameters by MLE based on the Vasicek term structure model. The interest rate parameters are - 16 - Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks brought into the full-information likelihood function as fixed in the second step to estimate the equity pricing parameters. The deposit insurance premium is calculated through the option relationship between the bank asset values and the deposit insurance premium. This two-step MLE method can solve the inconsistent and asymptotic inefficient problems of the DMS (1995) method Our empirical results indicate that the present premiums set up by the CDIC - 5, 5.5, and 6 basis points - are too low and are not proportional to the risks of the insured banks. The bank characteristic of market value of equity to book value of debts ratio is positively related with the asset value and negatively related with the deposit insurance premium. Moreover, the weighted stock market index and the weighted bank stock index have a positive relation with the market value of equity to book value of debts ratio. The two-step maximum likelihood estimates of deposit insurance premiums are much higher than those estimated from the modified Ronn-Verma method after solving the simultaneous equations directly. This is due to higher values of credit risk estimated from the former method. This finding coincides with that of Duan and Yu (1994) and D&S (2002). A comparison of the results among the two-step MLE approach, the modified Ronn-Verma method, and Duan and Yu’s (1994) procedure further implies that the under-estimation issue of the deposit insurance premium due to the problem of statistical inconsistency and inefficiency is more serious than that caused by the exclusion of the interest rate risk. After classifying the sample banks by their size to investigate the size effect on the deposit insurance premium, it is found that small banks have higher estimates of credit risk, asset volatility, and deposit insurance premiums. On the contrary, the large banks have smaller estimates of credit risk, asset volatility, and deposit insurance premiums. In sum, this study estimates deposit insurance premiums with - 17 - Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks incorporation of the interest rate risk and provides further evidence that there is a serious underestimation of the deposit insurance premium due to the problems of inconsistency and asymptotic inefficiency in the Ronn-Verma method. References Black, F. and M. Scholes, 1973, The pricing of options and corporate liabilities, Journal of Political Economy 81, 637-659. Cox, J. C., J. E. Ingersoll, and S. A. Ross, 1985, A theory of the term structure of interest rates, Econometrica 53 (2), 385-402. Duan, J. C., 1994, Maximum likelihood estimation using price data of the derivative contract, Mathematical Finance 4 (2), 155-167. Duan, J. C., A. Moreau and C. W. Sealey, 1995, Deposit insurance and bank interest rate risk: Pricing and regulatory implications, Journal of Banking and Finance 19, 1091-1108. Duan, J. C. and J. G. Simonato, 2002, Maximum likelihood estimation of deposit insurance value with interest rate risk, Journal of Empirical Finance 9 (1), 109-132. Duan, J. C. and M. T. Yu, 1994, Assessing the cost of Taiwan’s deposit insurance, Pacific Basin Finance Journal 2, 73-90. Hull, John C., 2000, Options, futures, and other derivatives, 4th edition, Prentice-Hall, Inc.. Laeven, Luc, 2002, Bank risk and deposit insurance, The World Bank Economic Review 16 (1), 109-137. Laeven, Luc, 2002, Pricing of deposit insurance, World Bank Policy Research Working Paper 2871. Merton, R. C., 1977, An analytic derivation of the cost of deposit insurance and loan guarantees, Journal of Banking and Finance 1, 3-11. Ronn, E. I. and A. K. Verma, 1986, Pricing risk-adjusted deposit insurance: An option-bases model, Journal of Finance 41, 871-895. Vasicek, O., 1977, An equilibrium characterization of the term structure, Journal of Financial Economics 5, 177-188. Yu, M.-T and L.-L. Hsu, 1993, Regulatory Constraints and Deposit Insurance Pricing, Taiwan Economic Review, Vol. 21, No. 1, pp. 45-59. - 18 - Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks Table 1. Estimation Results of the Interest Rate Parameters Vasicek (1977) assumes the instantaneous spot rate follows a mean-reverting process: drt q(m rt )dt dZ t , where rt is the instantaneous risk-free rate of interest at time t, m is the long-run mean of the interest rate, is the volatility of the interest rate, q is a positive constant measuring the magnitude of the mean-reverting force, and Z t is a Wiener process. denotes the risk premium. Paramet 1982-199 1983-19 1984-19 1985-19 1986-19 1987-19 1988-19 1989-19 1990-19 1991-20 1992-20 1993-20 1982-20 ers 1 92 93 94 95 96 97 98 99 00 01 02 02 m 0.0686 0.0724 0.0706 0.0739 0.0749 0.0758 0.0808 0.0784 0.0715 0.0678 0.0608 0.0498 0.0603 (0.0327) (0.0236) (0.0239) (0.0235) (0.0216) (0.0179) (0.0123) (0.0112) (0.0120) (0.0078) (0.0140) (0.0252) (0.0187) q 0.4597 0.6408 0.6409 0.6726 0.7602 0.9162 1.4440 1.6239 1.3557 1.8930 1.0678 0.6554 0.4625 (0.3227) (0.3431) (0.3449) (0.3699) (0.3841) (0.3877) (0.4394) (0.4634) (0.4526) (0.6946) (0.5822) (0.4952) (0.2068) v 0.0429 0.0446 0.0448 0.0468 0.0481 0.0469 0.0513 0.0501 0.0428 0.0374 0.0328 0.0282 0.0348 (0.0018) (0.0019) (0.0019) (0.0022) (0.0023) (0.0022) (0.0027) (0.0028) (0.0023) (0.0030) (0.0023) (0.0017) (0.0009) λ 0.0140 0.0060 -0.0109 -0.0467 -0.1004 0.0583 0.0860 0.3575 0.3361 0.4001 0.4380 0.3581 0.1834 (0.5486) (0.5360) (0.5407) (0.5357) (0.5448) (0.5512) (0.5522) (0.5589) (0.5523) (0.5500) (0.5419) (0.5410) (0.3811) - 19 - Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks Table 2. Estimation Results for the First Commercial Bank Equity is the year-end market value of shareholders’ equity and Debt is the year-end total debt from the quarterly balance sheet obtained from the TEJ database. Equity and debt are in million NT$. E/D is the ratio of Equity over Debt. mle is the two-step MLE estimate of instantaneous expected return on assets. mle is the two-step MLE estimate of credit risk. mle is the two-step MLE estimate of interest rate elasticity of assets. Vmle is the two-step MLE estimate of year-end assets. IPPmle is the two-step MLE estimate of the insurance premium per dollar of insured deposits in basis points. mrv , mrv , Vmrv , and IPPmrv are the solutions of the modified Ronn-Verma method (with interest rate risk). DY , DY , VDY , and IPPDY are estimated from the MLE of the Ronn-Verma method as in Duan and Yu (1994). The standard errors are shown in parentheses. Year end 1991 Equity 118973 Debt E/D 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 84206 216530 263846 181535 302862 246303 141624 128036 75110 82547 87897 767118 719899 808414 811526 835293 888069 977003 1024201 1096443 1147297 1192000 1252581 0.1551 0.1170 0.2678 0.3251 0.2173 0.3410 0.2521 0.1383 0.1168 0.0655 0.0693 0.0702 Two-Step Maximum Likelihood Estimation Results Year end 1991 1992 mle 0.0048 -0.0706 (0.1017) mle mle mle 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 0.2113 0.0291 -0.0368 0.1489 0.0372 -0.0391 0.0309 0.0005 0.0331 0.0088 (0.1223) (0.1363) (0.1133) (0.0731) (0.0911) (0.1293) (0.0841) (0.0830) (0.0796) (0.0610) (0.0687) 0.0960 0.0741 0.0958 0.1024 0.0631 0.0755 0.1127 0.0689 0.0628 0.0595 0.0418 0.0468 (0.0037) (0.0035) (0.0036) (0.0035) (0.0020) (0.0017) (0.0041) (0.0021) (0.0027) (0.0020) (0.0015) (0.0021) -0.5409 -0.5379 -0.7004 -0.4023 -0.6237 -0.3885 -0.3194 -0.1680 -0.5214 -0.4951 -0.4724 -0.6991 (0.3057) (0.1805) (0.4162) (0.3203) (0.2377) (0.1830) (0.1664) (0.3478) (0.2930) (0.3829) (0.0967) (0.2685) - 20 - Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks Vmle IPPmle 860858 780860 1000523 1050957 991754 1164288 1193079 1134293 1190484 1182313 1237627 1300818 (400) (385) (74) (25) (7) (0) (226) (142) (254) (460) (222) (493) 58.1601 55.0445 5.3707 2.2540 0.8850 0.0397 19.2049 23.1253 30.7240 120.7303 44.3697 59.7137 (8.1812) (9.2860) (1.9328) (0.6786) (0.3335) (0.0137) (3.9380) (2.9650) (4.9364) (4.7320) (8.3119) (6.2641) MLE of the Ronn-Verma Method, Duan & Yu (1994) Year end 1991 1992 DY 0.0001 -0.0803 (0.1119) DY VDY IPPDY 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 0.2211 0.0152 -0.0275 0.1569 0.0279 -0.0320 0.0479 0.0039 0.0608 0.0298 (0.0530) (0.1579) (0.0957) (0.0699) (0.0928) (0.1359) (0.0999) (0.0845) (0.1092) (0.0879) (0.0577) 0.1048 0.0498 0.1077 0.0944 0.0653 0.0800 0.1145 0.0755 0.0688 0.0742 0.0562 0.0444 (0.0030) (0.0011) (0.0013) (0.0029) (0.0016) (0.0013) (0.0019) (0.0012) (0.0018) (0.0011) (0.0011) (0.0009) 882469 803928 1024464 1075336 1016806 1190928 1222109 1164310 1222552 1210362 1270060 1338793 (445) (30) (42) (13) (6) (1) (126) (128) (247) (534) (332) (155) 47.2174 2.4675 5.9447 0.4433 0.2633 0.0273 12.2447 14.7955 17.5799 104.9920 37.6452 13.4592 (5.7990) (0.4108) (0.5203) (0.1560) (0.0744) (0.0073) (1.2893) (1.2504) (2.2502) (2.7881) (1.2401) - 21 - (4.6524) Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks Table 3. Estimation Results of the Deposit Insurance Pricing Model Equity is the year-end market value of shareholders’ equity and Debt is the year-end total debt from the quarterly balance sheet obtained from the TEJ database. Equity and debt are in million NT$. E/D is the ratio of Equity over Debt. mle is the two-step MLE estimate of instantaneous expected return on assets. mle is the two-step MLE estimate of credit risk. mle is the two-step MLE estimate of interest rate elasticity of assets. Vmle is the two-step MLE estimate of year-end assets. IPPmle is the two-step MLE estimate of the insurance premium per dollar of insured deposits in basis points. mrv , mrv , Vmrv , and IPPmrv are the solutions of the modified Ronn-Verma method (with interest rate risk). DY , DY , VDY , and IPPDY are estimated from the MLE of the Ronn-Verma method as in Duan and Yu (1994). The standard errors are shown in parentheses. Year end 1991 Equity 111133 Debt E/D 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 68884 164753 208183 135684 167443 132923 72200 73069 47727 50051 50437 591577 530772 561230 578402 600364 530395 544026 581093 611020 648430 689040 731614 0.1879 0.1298 0.2936 0.3599 0.2260 0.3157 0.2443 0.1242 0.1196 0.0736 0.0726 0.0689 Two-Step Maximum Likelihood Estimation Results Year end 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 mle 0.0290 -0.0235 0.2132 0.1081 -0.0272 0.1447 0.0700 -0.0161 0.0520 -0.0110 0.0387 0.0130 (0.1482) (0.1303) (0.1446) (0.1374) (0.1122) (0.1215) (0.1807) (0.1472) (0.1417) (0.1826) (0.1909) (0.2644) 0.1318 0.0864 0.1032 0.1177 0.0819 0.0882 0.1140 0.0780 0.0826 0.0947 0.0802 0.0972 (0.0049) (0.0038) (0.0038) (0.0040) (0.0027) (0.0024) (0.0040) (0.0020) (0.0033) (0.0030) (0.0038) (0.0045) -0.5942 -0.5437 -0.6710 -0.4600 -0.6860 -0.4706 -0.3900 -0.2535 -0.5074 -0.4631 -0.4817 -0.8213 (0.3504) (0.2194) (0.3749) (0.2339) (0.3017) (0.2306) (0.2026) (0.4133) (0.3600) (0.4159) (0.3416) (0.9111) mle mle mle - 22 - Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks Vmle IPPmle 682541 582486 708993 769082 717827 681848 659394 633756 663521 664919 710895 739194 (353) (283) (56) (43) (47) (10) (190) (204) (340) (718) (808) (2323) 70.9280 48.0653 5.0125 7.4044 16.3508 10.2450 65.0354 98.0546 120.7537 311.6521 245.1668 396.9562 (9.5399) (8.1635) (1.5731) (1.7792) (2.0295) (1.0192) (7.1751) (6.6884) (11.9247) (16.9328) (20.2827) (37.5695) The Modified Ronn-Verma Method, DMS (1995) Year end 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 mrv mrv 0.0595 0.0413 0.0955 0.1048 0.0476 0.0512 0.0817 0.0436 0.0416 0.0423 0.0366 0.0317 -0.6007 -0.7774 -0.9080 -0.4697 -0.9143 -0.4644 -0.5428 -0.5083 -0.3448 -0.4012 -0.7354 -0.7484 Vmrv 684968 583843 709149 769148 717595 681597 659115 634445 663532 674476 714572 758914 IPPmrv 0.4713 1.0717 2.0661 2.7767 0.0943 0.1873 5.3700 8.3037 8.7880 48.1875 40.8671 40.1631 MLE of the Ronn-Verma Method, Duan & Yu (1994) Year end 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 DY 0.0183 -0.0357 0.2204 0.0913 -0.0188 0.1552 0.0576 -0.0093 0.0660 -0.0063 0.0633 0.0399 (0.1667) (0.1036) (0.1504) (0.1331) (0.1072) (0.1385) (0.1947) (0.1752) (0.1542) (0.1736) (0.1830) (0.1522) 0.1438 0.0799 0.1104 0.1178 0.0862 0.0965 0.1204 0.0892 0.0945 0.1028 0.0887 0.0870 (0.0032) (0.0013) (0.0017) (0.0027) (0.0016) (0.0017) (0.0023) (0.0017) (0.0017) (0.0018) (0.0019) (0.0018) 698668 599897 726043 786550 735965 697847 675685 651739 687169 685866 734626 773161 (339) (45) (25) (19) (22) (14) (122) (183) (263) (666) (598) (709) 72.3888 19.7417 4.3238 6.0281 14.4990 15.0560 67.3298 99.6722 109.0419 244.7238 189.7599 255.3895 (6.0486) (1.3654) (0.4248) (0.4973) (0.6971) (1.0464) (3.8054) (5.7464) DY VDY IPPDY - 23 - (5.5437) (9.9918) (9.7099) (9.7841) Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks Table 4. Size Effect on Deposit Insurance Premium The sample banks are grouped into three categories by their size in order to examine the size effect on deposit insurance premium. Panel A reports the results estimated from the MLE of the Ronn-Verma method and Panel B presents the two-step MLE results. The results reported are the value-weighted average of each subsample. The portfolios of each subsample change every year, according to their asset values. The standard errors of estimates are in parentheses. Panel A: MLE of the Ronn-Verma Method Large Banks Year end 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 DY 0.0240 -0.0830 0.1819 0.0715 -0.0455 0.1482 0.0215 -0.0295 0.0491 -0.0099 0.0645 0.0330 (0.1453) (0.0625) (0.1407) (0.1029) (0.0774) (0.0983) (0.1357) (0.1129) (0.1060) (0.1307) (0.1382) (0.0687) 0.1216 0.0567 0.1033 0.1011 0.0708 0.0834 0.1163 0.0831 0.0769 0.0818 0.0693 0.0477 (0.0026) (0.0009) (0.0014) (0.0029) (0.0015) (0.0014) (0.0022) (0.0015) (0.0015) (0.0014) (0.0014) (0.0009) 854628 786256 964034 1047889 979022 1132566 1154029 1113205 1155216 1143538 1207731 1273303 (405) (41) (34) (12) (11) (3) (134) (196) (297) (755) (709) (250) 71.0024 7.1945 4.8535 0.4899 0.7751 0.1978 12.6135 24.4135 37.3837 161.7324 119.4969 34.9320 (5.4228) (0.5721) (0.4322) (0.1560) (0.1400) (0.0328) (1.4643) (2.0167) (2.8563) (2.1226) DY VDY IPPDY (6.9176) (6.2233) Medium Banks Year end 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 DY -0.0109 0.0497 0.2872 0.1231 0.0208 0.0737 0.0548 -0.0203 0.0953 -0.0055 0.0761 0.0210 (0.2350) (0.1774) (0.1654) (0.1829) (0.1422) (0.1030) (0.1486) (0.1075) (0.1560) (0.2475) (0.2308) (0.2380) 0.2132 0.1224 0.1212 0.1457 0.1048 0.0832 0.0976 0.0735 0.1042 0.1406 0.1126 0.1336 DY - 24 - Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks VDY IPPDY (0.0052) (0.0020) (0.0022) (0.0024) (0.0016) (0.0016) (0.0019) (0.0010) (0.0018) (0.0023) (0.0023) (0.0027) 210398 225182 310485 369731 396949 461215 522415 635415 691400 709108 772932 784233 (136) (55) (10) (34) (40) (12) (91) (130) (319) (1015) (820) (1512) 78.9295 39.6629 2.9116 16.3949 17.5113 3.5949 34.5983 46.2498 87.7126 362.2927 225.6542 506.4189 (8.1495) (2.6072) (0.3638) (1.1125) (1.0456) (0.3158) (1.8386) (2.0696) (4.6658) (13.8843) (11.5629) (17.0255) Small Banks Year end 1991 1992 1993 1994 1995 1996 DY 0.1065 0.2051 0.3271 0.1408 0.0278 0.2583 (0.2443) (0.3173) (0.1967) (0.1954) (0.2440) 0.2363 0.1927 0.1442 0.1474 (0.0058) (0.0035) (0.0023) 53169 67786 (29) DY VDY IPPDY 1998 1999 2000 2001 2002 0.1179 0.0241 0.0481 -0.0038 0.0467 0.0718 (0.2542) (0.3539) (0.3174) (0.2012) (0.1284) (0.1685) (0.1301) 0.1516 0.1361 0.1602 0.1135 0.1007 0.0785 0.0786 0.0683 (0.0027) (0.0022) (0.0024) (0.0029) (0.0027) (0.0019) (0.0016) (0.0018) (0.0017) 98311 106236 117491 132526 154847 196361 205581 214440 228144 226550 (42) (7) (3) (23) (37) (149) (231) (162) (168) (222) (157) 56.2615 113.4706 7.8279 (7.2305) (7.4842) (0.8875) 5.7124 199.0056 (0.7057) (6.0885) 1997 55.9473 200.2612 238.6419 207.5962 186.9166 214.5289 165.0040 (3.7792) (10.3196) (13.8209) - 25 - (9.3334) (8.3947) (10.8516) (8.5624) Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks Panel B: Two-Step Maximum Likelihood Estimation Results Large Banks Year end 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 mle 0.0315 -0.0737 0.1758 0.0860 -0.0550 0.1395 0.0309 -0.0373 0.0326 -0.0164 0.0384 0.0069 (0.1255) (0.1232) (0.1516) (0.1116) (0.0819) (0.0952) (0.1336) (0.0936) (0.0916) (0.1113) (0.0881) (0.1015) 0.1086 0.0751 0.1028 0.1038 0.0693 0.0789 0.1160 0.0755 0.0666 0.0699 0.0490 0.0564 (0.0038) (0.0035) (0.0037) (0.0036) (0.0023) (0.0017) (0.0042) (0.0021) (0.0025) (0.0024) (0.0017) (0.0025) -0.6169 -0.5287 -0.6868 -0.4217 -0.6395 -0.4105 -0.2728 -0.2034 -0.5213 -0.4603 -0.4551 -0.6701 (0.2895) (0.2044) (0.4101) (0.2314) (0.2459) (0.1933) (0.1650) (0.3566) (0.2541) (0.3554) (0.1756) (0.3483) 835125 763938 941547 1024633 954949 1106951 1126635 1084870 1125891 1117092 1181364 1233958 (408) (390) (84) (17) (22) (2) (224) (194) (322) (812) (440) (1039) 68.8946 56.9835 7.1224 1.4823 2.2543 0.3140 21.0571 34.4036 47.2083 203.5687 114.4723 141.0544 (8.1048) (9.5315) (2.1983) (0.4862) (0.7499) (0.0610) (4.0746) (3.7483) (5.7756) (10.0301) mle mle mle Vmle IPPmle (6.5891) (13.2048) Medium Banks Year end 1991 1992 1993 1994 mle 0.0095 0.0664 0.2777 0.1430 (0.2196) (0.1370) (0.1286) 0.2041 0.1050 (0.0085) mle mle mle 1995 1996 1997 1998 1999 2000 2001 2002 0.0118 0.0614 0.0657 -0.0288 0.0856 -0.0106 0.0461 0.0003 (0.1799) (0.1443) (0.1057) (0.1428) (0.1303) (0.1410) (0.2608) (0.2232) (0.4325) 0.1020 0.1411 0.0962 0.0825 0.0935 0.0714 0.0908 0.1266 0.0985 0.1386 (0.0040) (0.0038) (0.0049) (0.0032) (0.0028) (0.0031) (0.0015) (0.0041) (0.0029) (0.0042) (0.0054) -0.5090 -0.5726 -0.6346 -0.5048 -0.7903 -0.4964 -0.4561 -0.2117 -0.4290 -0.5568 -0.5557 -1.0625 (0.5466) (0.2401) (0.2998) (0.2403) (0.3694) (0.2193) (0.1803) (0.3906) (0.3985) (0.5069) (0.4078) (1.4699) - 26 - Deposit Insurance Premium under Stochastic Interest Rate: An Empirical Study of Taiwan’s Banks Vmle IPPmle 206307 221001 303909 362026 387686 450542 510365 618002 677404 697148 755237 762009 (191) (61) (5) (95) (92) (11) (156) (265) (391) (920) (1356) (4795) 79.3668 25.9871 0.9766 18.5905 21.2002 2.4264 33.4220 87.2828 73.8300 367.1022 246.9313 625.6219 (14.3258) (4.5640) (0.3514) (4.2682) (3.8316) (0.5890) (4.3235) (5.4912) (9.7988) (14.9990) (20.6099) (58.5161) Small Banks Year end 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 mle 0.1334 0.2305 0.3205 0.1685 0.0401 0.2462 0.1371 0.0200 0.0322 -0.0063 0.0304 0.0346 (0.2386) (0.2337) (0.1622) (0.1911) (0.2823) (0.1889) (0.3086) (0.2211) (0.1914) (0.1602) (0.2494) (0.2221) 0.2289 0.1612 0.1239 0.1420 0.1437 0.1119 0.1408 0.0879 0.0886 0.0816 0.0881 0.0874 (0.0099) (0.0074) (0.0061) (0.0039) (0.0026) (0.0034) (0.0051) (0.0025) (0.0032) (0.0036) (0.0052) (0.0054) -0.6406 -0.5991 -0.7170 -0.7161 -0.4395 -0.5563 -0.4754 -0.3523 -0.5841 -0.3590 -0.4204 -0.6777 (0.5428) (0.3614) (0.4202) (0.2162) (0.5361) (0.3136) (0.2931) (0.4969) (0.4186) (0.3688) (0.4195) (0.7912) 52085 66573 96140 103785 114604 129681 151914 192794 198480 206354 220365 215490 (38) (60) (6) (3) (18) (24) (188) (144) (299) (401) (514) (585) 58.5569 69.6343 3.0767 (12.4242) (13.7900) (1.2581) mle mle mle Vmle IPPmle 5.9846 188.9044 (1.0203) (5.0945) 37.7015 179.1598 175.0916 246.1383 349.4127 365.0345 373.5168 (3.3115) (16.1267) (11.0448) (20.3413) (25.5736) (32.6762) (36.3473) - 27 -