The impact of bank asset composition on profitability.

Brown, Christopher L.

INTRODUCTION

The purpose of this paper is to investigate the impact of bank asset composition on accounting returns. Since the risk-based capital standards were implemented in January of 1993, most research on bank asset composition has focused on the impact of the risk-based capital requirements on bank asset composition.

Haubrich and Wachtel (1993) point out that the risk-based capital standards have increased the attractiveness of government securities as investments. They find that government securities increased from 15 percent of assets to 22 percent of assets over the period from 1989 to 1992. They find a shift in bank portfolio composition away from commercial loans and into government securities and attribute the portfolio changes to the introduction of risk-based capital standards.

Berger and Udell (1994) state that the risk-based capital standards might function as a regulatory tax that increases the cost of making loans with 100% risk weighting (like commercial and consumer loans) relative to the costs of carrying government securities with a risk-weighting of 20 percent or less. They also indicate that the risk-based capital standards could have a substantially negative impact on commercial lending due to the regulatory tax.

Calem and Rob (1996) find the risk-based capital standards do not reduce risk-taking of well-capitalized banks. If banks significantly exceed their capital requirements, the regulatory tax on commercial loans effectively disappears. Recent trends in bank asset composition support the findings of Calem and Rob (1996). In 1993, government securities accounted for 22.58 percent of total assets; by the end of 1996 they accounted for only 17.49 percent of total assets (FDIC Quarterly Banking Profile). During the same period, commercial and industrial loans increased slightly from 14.53 percent in 1993 to 15.31 percent in 1996. Commercial and industrial loans peaked at 16.9 percent in 1999 and fell to 12.89 percent by December 2002.

This portfolio shifting appears to be more of a flight to safety during difficult economic times than the result of risk-based capital standards. Since December 1999, commercial lending portfolios have shrunk significantly, but it is highly unlikely that the reduction in commercial lending is the result of risk-based capital standards. The average capitalization rate of commercial banks increased from 8.36 percent of assets in 1999 to 9.16 percent of assets in December 2002. It appears that bank managers are acting rationally, by moving into safer investments during difficult economic times.

Bank managers rebalance their asset portfolios based on economic conditions because they want to put assets on the books that have the best risk/return combinations. In a healthy economic environment, commercial and industrial loans and real estate loans should provide superior returns over securities. In a healthy economy with low interest rates, loan demand will be high, so bank managers can choose the best from the pool of loan applicants. The result of a healthy economy with strong demand for loans is that loan losses should be very low. In an economic downturn, loan demand will decline, as fewer business opportunities are available for entrepreneurs. The risk of commercial and industrial loans and real estate loans increase during an economic downturn even if the pool of loan applicants remain the same. However, the quality of the pool of applicants will decline, because most high quality borrowers will be very cautious during an economic downturn. Therefore, banks should have higher loan losses on commercial and industrial loans and real estate loans during an economic downturn.

This paper focuses on the contribution of securities, commercial and industrial loans, and real estate loans to banks' return on assets (ROA) over time. The next section of the paper outlines the data and methodology, followed by the findings and conclusions.

DATA AND METHODOLOGY

The data used in this study is call report data obtained from the Federal Reserve Bank of Chicago's Bank Holding Company Database. The data includes information about the size of the institution, the dollar amount of holdings in various asset accounts, and the bank's net income. The call reports used are the end of year reports for the years 1997 through 2002. The total number of observations for all years is 53,904 (See Table 1). The period 1997 through 2002 is used because there were significant changes in the economy over that time period. The economy was strong during the period from 1997 through 1999, it weakened significantly in 2000 and the economic downturn continued to impact business throughout 2001 and 2002. The contribution of securities, commercial and industrial loans, and real estate loans to bank ROA may be different during the strong economy period (1997 through 1999) compared to the period after the economic downturn (2000 through 2002).

In order to directly determine the returns attributable to various asset accounts, it would be necessary to have the interest income, loan loss provisions, salary and employee expense, equipment expense, etc. associated with each type of account. This detailed information is not available on the call reports. Therefore, this paper measures the relative returns on different types of assets by determining the contribution that each asset category makes to bank ROA. In order to minimize the impact of extreme observations, a dummy variable, HIGHROA, is created. HIGHROA equals 1 if the institution's ROA was above the median ROA for banks in that year and 0 otherwise. For each year in the sample period, a logit regression model is estimated.

The dependent variable in the logit regression model is HIGHROA. The independent variables are (1) the ratio of securities to total assets (PCTSEC), (2) the ratio of commercial and industrial loans to total assets (PCTCI), and (3) the ratio of real estate loans to total assets (PCTRE). If the coefficient on an independent variable is positive and significant, higher proportions of that asset category are associated with higher ROA levels. If the coefficient is negative, increases in that asset category are associated with lower ROA levels.

Two model specifications are used. The first specification is to run six different logit regression models, one for each year during the sample period. The second method is to include all of the observations in one sample and to use dummy variables to account for the different years. In the regression using the year dummies, the base year for the regression is 2000.

Based on the theory that the performance of commercial and industrial loans and real estate loans depend on the state of the economy, the coefficients on these variables are expected to be negative and significant during the economic downturn (2000 through 2002) and positive and significant during the healthy economy period (1997 through 1999). The coefficient on securities should be positive and significant during the economic downturn and negative and significant during the healthy economy period. The coefficient should be negative during the healthy economy period because the net yield on securities should be significantly below the net yield on risky loans during a healthy economic period. Therefore, banks with high concentrations of securities during these periods should earn below average returns.

FINDINGS

The findings are presented in Tables 2 through 4. The results of the logit regression models for the bad economy period are shown in Table 2. As expected, the PCTSEC is positive and significant in all three years (2000 through 2002). The coefficient on PCTCI is negative and significant at the 1 percent level in 2001 and 2002 and insignificant in 2002. PCTRE is negative and significant at the 1 percent level in 2000, at the 5 percent level in 2001, and insignificant in 2002. Therefore, there is some evidence that banks with higher proportions invested in securities and lower proportions invested in commercial and real estate loans perform better in bad economic times.

The results of the logit regression models for the good economy period are shown in Table 3. As expected, the coefficients on PCTCI and PCTRE are positive all three years (1997 through 1999). However, PCTRE is not statistically significant in any of the three years. PCTCI is significant at the 10 percent level in 1997, at the 1 percent level in 1998, and is insignificant in 1999. The surprising finding is that the coefficient on PCTSEC is positive and significant at the 1 percent level for all three years. Business conditions were good during these years. During years when business conditions are good, it would seem reasonable that banks with higher proportions invested in loans (and less invested in securities) would do better than banks with higher proportions invested in securities.

The results of the logit regressions for the entire sample period are shown in Table 4. In the first model specification that excludes the year dummies, the PCTSEC is positive and significant at the 1 percent level and the PCTRE is negative and significant at the 1 percent level. The PCTRE is insignificant in the model that includes the year dummies. With 2000 as the base year, the coefficient on YR1997, YR1998, and YR1999 were expected to be positive and significant. YR1997 and YR1998 are positive and significant at the 1 percent level, but YR1999 is insignificant. One finding that is consistent across all of the model specifications is that PCTSEC is positive and highly significant.

CONCLUSIONS

The findings support the hypothesis that securities perform better in difficult economic times. There is also some support for the hypothesis that commercial and industrial loans and real estate loans don't perform well during difficult economic times. The empirical evidence indicates the movement by bank managers to safer portfolios was a prudent move. Banks with higher proportions of assets invested in commercial and industrial loans and real estate loans have lower ROA's than banks with higher proportions invested in securities.

The findings for the strong economy period were somewhat perplexing. The coefficients on PCTCI and PCTRE were positive, but the coefficient on PCTSEC was also positive. This indicates that banks do better in a healthy market if they have higher proportions invested in securities. Since securities have the lowest yields of all bank assets, this is surprising. One possible explanation is that the cost of making commercial loans and real estate loans is significantly higher than the cost of investing in securities, and the increased yield on commercial and real estate loans is not sufficient to overcome the higher costs.

REFERENCES

Berger, A. N. & G. F. Udell. (1994). Did Risk-Based Capital Allocate Bank Credit and Cause a 'Credit Crunch' in the United States?, Journal of Money, Credit and Banking, 26(3), 585-628.

Caleb, P.S. &R. Rob. (1996). The Impact of Capital-Based Regulation on Bank Risk-Taking: A Dynamic Model. Board of Governors of the Federal Reserve System, Finance and Economics Discussion Series, Working paper no. 96-12.

FDIC Quarterly Banking Profile. FDIC Home Page, http://www.fdic.gov. (Accessed August 2003.

Haubrich, J.G. & P. Wachtel. (1993). Capital Requirements and Shifts in Commercial Bank Portfolios. Federal Reserve Bank of Cleveland Economic Review, 29(3), 2-15.

Christopher L. Brown, Western Kentucky University

Table 1: Sample Size by Year

 Year No. of Banks in Sample
 1997 9,696
 1998 9,308
 1999 9,109
 2000 8,824
 2001 8,590
 2002 8,377
 Total Sample 53,904

Table 2: Logit Results for the Bad Economy Period (2000 - 2002)

 Year Variable Parameter Standard Chi- p-Value
 Estimate Error Square

2002 Intercept -0.0750 0.1111 0.4562 .4994
 PCT 0.9665 0.1877 26.5184 .0001
 PCTCI -0.8499 0.3079 7.6183 .0058
 PCTRE -0.0520 0.1555 0.1116 .7358
2001 Intercept -0.1624 0.1069 2.3071 .1288
 PCTSEC 0.8212 0.1867 19.3565 < .0001
 PCTCI -0.9007 0.2991 9.0689 .0026
 PCTRE -0.3768 0.1509 6.2319 .0125
2000 Intercept 0.1222 0.1108 1.2173 .2699
 PCTSEC 0.7551 0.1927 15.3498 < .0001
 PCTCI -0.2507 0.2888 0.7538 .3853
 PCTRE -0.5859 0.1514 14.9854 < .0001

Table 3: Logit Results for the Good Economy Period (1997 - 1999)

 Year Variable Parameter Standard Chi- p-Value
 Estimate Error Square

1999 Intercept -0.2128 0.1075 3.9152 .0479
 PCTSEC 0.9240 0.1856 24.7738 < .0001
 PCTCI 0.2837 0.2877 0.9717 .3242
 PCTRE 0.0277 0.1487 0.0346 .8524
1998 Intercept -0.2778 0.0995 7.7900 .0053
 PCTSEC 1.1663 0.1760 43.8985 < .0001
 PCTCI 0.7934 0.2950 7.2334 .0072
 PCTRE 0.2356 0.1469 2.5724 .1087
1997 Intercept 0.0521 0.1046 0.2481 .6185
 PCTSEC 0.9147 0.1810 25.5316 < .0001
 PCTCI 0.5486 0.3093 3.1462 .0761
 PCTRE 0.1922 0.1480 1.6855 .1942

Table 4: Logit Results for All Periods (1997 - 2002)

 Variable Parameter Standard Chi- p-Value
 Estimate Error Square

Model 1 Intercept -0.0710 0.0432 2.6972 .1006
 PCTSEC 0.9735 0.0751 167.8739 < .0001
 PCTCI -0.1140 0.1202 0.8989 .3429
 PCTRE -0.1660 0.0607 7.4743 .0063
Model 2 Intercept -0.1279 0.0482 7.0411 .0079
 PCTSEC 0.9334 0.0754 154.1477 < .0001
 PCTCI -0.0530 0.1208 0.1919 .6610
 PCTRE -0.0965 0.0612 2.4861 .1148
 YR2002 0.0003 0.0306 0.0002 .9925
 YR2001 -0.2538 0.0305 69.2797 < .0001
 YR1999 -0.0082 0.0300 0.0738 .7852
 YR1998 0.1055 0.0299 12.4137 .0004
 YR1997 0.3323 0.0299 123.6842 < .0001