“Duration Ratio” as a New Risk Measure in Bank Risk Management Yoon K. Choi Department of Finance University of Central Florida Orlando, FL 32816-1400 (407) 823-5023; FAX: (407) 823-6676 E-mail: Yoon.Choi@bus.ucf.edu ABSTRACT I propose a duration gap measure, “duration ratio”, which measures a relative duration gap between assets and liabilities. The new measure may be a true risk measure in the sense that it captures not only the effect of interest rate changes, but also other exogenous shocks, such as market risk and exchange rate risk. This new approach provides us with a good proxy for the duration ratio based on interest income and expenses, and U.S. large banks are examined for their risk management based on the duration ratios. June 14, 2001 2 “Duration Ratio” as a New Risk Measure in Bank Risk Management Risk management has been a crucial element of success in the bank industry. However, with so many hedging instruments and strategies, it becomes so difficult to measure risk and assess the quality of these risk management techniques. This note proposes a new risk measure that would help investors and bank monitors and regulators in assessing risk in the bank industry. I also empirically examine risk management in the large banks in U.S. using this new approach. 1. Duration Gap and Immunization (Saunders, 2000) The overall interest rate risk faced by bank owners whose equity is the difference between assets and liabilities is normally defined as sensitivity of the value of equity with respect to interest rate changes. In notation, E / I = A / I - L / I, (1) where E = equity, I = interest rate, A = assets, and L = liabilities. The well-known duration model shows: A / A = - DA * I / (1 + I), (2) L / L = - DL * I / (1 + I), (3) where DA and DL are the durations of the bank’s asset and liability portfolio, respectively. From equations (2) and (3), we obtain A = - DA * A * I / (1 + I), (4) L = - DL * L * I / (1 + I). (5) Following the “Macauley” duration model which assumes that the interest rate and its shock are the same for both assets and liabilities, 3 E = - [DA * A - DL * L] * I / (1 + I). (6) Rearranging terms, we get E = - [DA - DL * k] * A * I / (1 + I), (7) where k = L/A is a measure of the bank’s leverage. According to the traditional immunization strategy based on equation (7), managers should structure the balance sheet so that DA = DL * k such that E = 0. 2. “Duration Ratio” as Elasticity between Assets and Liabilities Instead of calculating the difference, dividing (2) by (3) leads to the following: (A / A) / (L / L) = DA / DL (8) Now the duration ratio in equation (8) can be interpreted as the elasticity of the asset value with respect to liability values. I attempt to provide an economic meaning to the duration ratio in terms of the interaction between assets and liabilities. In general, bank risk management involves a way to offset asset risk by liabilities risk (e.g., maturity or duration matching), such that equity risk will be immunized from the source of risk. Suppose that both assets and liabilities are stochastic, and bank managers monitor assets and liabilities so as to generate riskless equities. Equity can be seen a portfolio of assets and liabilities. In a mathematical term, E = A – L. And VAR(E) = VAR(A) + VAR(L) – 2 * COV(A, L). (11) VAR and COV indicate variance and covariance, respectively. For simplicity, assume VAR(A) = VAR(L). Then, VAR(E) = 2 * VAR(A) * (1 - CORR(A, L)). (12) 4 Thus, equation (12) suggests that managers can hedge the total portfolio against any risk by achieving a perfect correlation (e.g., CORR = 1) between assets and liabilities. Just like elasticity, correlation is a statistical measure of a linear relationship between asset and liabilities. Therefore, the duration ratio is interpreted to measure overall risks rather than interest rate risk faced by banks. Another important implication of equation (8) is that we can obtain a simple proxy for the overall duration ratio: The ratio of the percentage change in the market value of assets to that of liabilities. Since these variables are in market value, I use two alternative variables in the balance sheet and income statement: book values of asset and liabilities and interest revenues and expenses. I use these variables of the U.S. large banks in order to determine its validity as a proxy for the duration ratio. 3. Other Issues: Convexity and Capital Ratio Requirement 3.1. Convexity and Duration It is well known that the duration model assumes a linear relationship between rate shocks and bond price changes. However, the actual bond price-yield relationship exhibits a property called convexity rather than linearity. Specifically, the increase in price due to a rate fall tends to be greater than as predicted by the duration model, while the price falls more for rate increases. We can incorporate the convexity factor in the duration model by adding the curvature parameter as follows: A / A = - DA * I / (1 + I) + ½ * C * (I )2, (9) The second term on the right hand side of the equation indicates the curvature term. 5 Given the convexity property, we need to adjust the duration ratio accordingly because we do not obtain the simple duration ratio as in equation (8), due to the convexity factor. Therefore, we suggest a modified duration ratio as follows: (A / A) / (L / L) = DA / DL + CV (10) where CV is a convexity factor with a property being that greater curvature is associated with higher CV. Of course the duration model implies CV = 0. 3.2. Capital Ratios and Immunization The result in equation (8) has a very interesting implication for risk management under capital ratio restriction. Regulators set minimum target ratios for a bank’s net worth, E. The simplest one is the ratio of bank capital to its assets, E/A. Suppose that a bank wants to immunize against falling below this target ratio; that is (E/A) = 0. Kaufman (1984) shows that in order to immunize the net worth to set (E/A) = 0, the manager needs to set DA = DL, instead of setting DA = k * DL such that E = 0. This implies, according to equation (8), that the percentage change in assets should be the same as the percentage change in liability. Furthermore, the risk management approach would be different depending on the bank’s objective of risk management, including the bank’s capital ratio requirement. 4. Data and Empirical Results 4.1. Data I collected the data on assets and liabilities for the top 100 largest bank holding companies as of December 31, 1997. The data were from the National Information 6 Center of the Federal Financial Institutions Examination Council. Only 89 of the top 100 bank holding companies had the variables we use in the empirical investigation. Table 1 shows the summary statistics of our sample. The average asset size (AA) over the sample period ranges from 6.19 billion to 281 billion dollars, with a mean of about 33 billion dollars. On average, the interest revenue doubled interest expenses, indicating a profitable performance in 1997. Tier I capital ratio is about 7.5% on average, which is well above the 5% benchmark rate for “well-capitalization” according to the category defined by the 1991 FDIC Improvement Act. In fact, the minimum Tier I leverage ratio in our sample was 5.03%. 4.2. Duration ratio in changes in assets and liabilities in book value Although we use the book value in calculating duration ratios using assets and liabilities (DR1), the ratio may capture the current market value changes since we use the recent percentage changes in assets and liabilities instead of the levels. The average duration ratio based on assets and liabilities was 0.919 close to 1, which implies successful hedging strategies used in our sample banks. However, if we examine the results carefully, the duration ratio may not be reliable for some individual banks. I find a few negative sign, sometimes with a very high number (e.g., -6.2), and this is not acceptable because the duration ratio is supposed to be positive. 4.3. Duration ratio in interest revenues and expenses (DR2) Since the price changes in assets and liabilities in banks are accurately reflected in interest revenues and expenses, we may obtain the duration ratio using interest revenues 7 and expense changes instead. Table 2 shows that the duration ratio is again very close to 1.0. DR2 seems to measure the duration ratio more accurately, since the standard deviation (0.06) of the measure is much smaller than DR1. Furthermore, all duration ratios calculated in this sample are positive, which is another indication of its reliability for the proxy for the duration ratio. 4.4. Two Alternative Risk Management: DA = DL vs. DA = k * DL The previous section discussed two different risk management objectives: DA = DL vs. DA = k * DL . If management wants to hedge the level of equity against risks (e.g., E = 0), then the manager needs to set DA = k * DL (Strategy I). In this case, the duration ratio is close to the leverage ratio. If the manager wants to hedge the capital ratio instead (e.g., (E/A) = 0), he needs to set DA = DL (Strategy II). Here, the duration ratio is close to one. Table II shows summary statistics for the two different sets of banks, based on these strategies. There are 17 banks whose immunization strategy is to DA = k * DL so as to keep the level of equity stable. These banks seem to be larger in size and produce less interest revenues than those banks using Strategy II. Finally, the capital leverage ratio for the banks using Strategy I seems to be lower than the banks using Strategy II. 5. Summary and Conclusion We examine two proxies for the “duration ratio”. One is based on assets and liabilities; the other is based on interest revenues and expenses. Given the calculated duration ratios of the well-capitalized large U.S. banks, the second measure based on 8 interest revenues and expenses seems to be a more reliable proxy. The underlying assumption is that these U.S. banks in our sample are engaged in successful risk management. We find that an interesting pattern exists among our sample banks in terms of immunization strategies. Those banks using Strategy I seem to be larger in size, have lower capital leverage ratios, and produce less interest revenues than those banks using Strategy II. Several extensions are in order. 1. We can extend our analysis with much longer time-series data. We may observe how duration ratio evolves over time. 2. We may examine smaller banks of different capital ratios to see whether different duration ratios are associated with these banks in a systematic way. References Kaufman, G. “Measuring and Managing Interest Rate Risk: A Primer.” Federal Reserve of Chicago, Economic Perspectives, 1984, p. 16-29. Saunders, A. Financial Institution Management: A Modern Perspective, McGraw-Hill Co. 2000. 9 Table I Summary Statistics of the Sample This table contains the summary statistics of the major variables considered for 1997. INTREV is the interest income per average assets, INTEXP is the interest expense per average assets, AA is the average asset size in millions of dollars. CAP is the ratio of Tier I capital ratio. DURAT_1 ((A / A) / (L / L)) is the first proxy for duration ratio. DURAT_2 is the second proxy, defined as (int rev / int rev) / (int exp / int exp). ____________________________________________________________________ Variables Mean Minimum Maximum Std Dev. ____________________________________________________________________ INTREV (%) 7.5% 3.55% 14.15% 1.74% INTEXP (%) 3.55% 1.83% 5.32% 0.66% AA (Million) 32,907 6,190 281,678 52,346 CAP (%) 7.53% 5.03% 15.92% 1.9% 2.215 0.8 DURAT_1 0.9196 -6.169 DURAT_2 1.0011 0.8846 1.291 0.06 ____________________________________________________________________ 10 Table II Immunization Strategies This table contains the summary statistics of the major variables considered for 1997 for the banks with different immunization strategies. INTREV is the interest income per average assets, INTEXP is the interest expense per average assets, AA is the average asset size in millions of dollars. CAP is the ratio of Tier I capital ratio. L/A is the leverage ratio. DA / DL is the duration ratio. Strategy I is to set DA = k * DL, while Strategy II is to set DA = DL . ____________________________________________________________________ Strategy I Strategy II Variables Mean Median Mean Median ____________________________________________________________________ INTREV (%) 6.89% 6.97% 7.64% 7.43% INTEXP (%) 3.34% 3.35% 3.59% 3.51% AA (Million) 46,302 15,532 29,743 13,340 CAP (%) 6.5% 6.36% 7.77% 7.24% DA / DL 93.4% 94.2% 101.7% 99.9% L/A 93.5% 93.6% 92.2% 92.7% ____________________________________________________________________