The Case of Junk Bonds at Savings and Loan Associations
Much of the debate concerning the savings and loan (S&L) crisis has focused on questions regarding the various investments undertaken by S&Ls. The Financial Institutions Reform, Recovery, and Enforcement Act (FIRREA) of 1989, requires, among other things, that S&Ls' existing holdings of corporate debt securities not of investment grade ("junk" bonds) be divested by July 1, 1994.(1) Proponents of this restriction believe that S&Ls should return to their original purpose and concentrate on providing credit to potential and existing homeowners. They argue that junk bonds are inappropriate investments for institutions with federal deposit insurance. However, others contend that investing in junk bonds may improve the diversification of an S&L's assets and therefore lead to less risky, healthier institutions. Has allowing investment in junk bonds contributed to the severity of the S&L crisis by permitting increased risk-taking by some institutions? Or did holdings of junk bonds actually reduce S&L portfolio risk through the benefits of diversification? This is an important empirical question because the FIRREA restrictions have adversely affected the low-grade-bond market by eliminating a potential source of demand for these securities. It is also important because
many of the large losses of the S&L industry in the 1980s were borne by the taxpayer.
It may be argued, however, that debates about which assets thrifts should be allowed to hold are focusing on the wrong questions. There are many types of risky assets that thrifts are still permitted to hold in their portfolios even after the passage of FIRREA, for example, fixed-rate, thirty-year mortgage loans. If an institution wishes to increase its risk exposure, prohibiting junk bond investment will not prevent it from doing so. Thus, a more relevant policy question is what factors induce thrifts to take on additional risk. We believe that by studying the effects of junk bond investment on S&Ls, we can better understand the motivation for greater S&L risk-taking in general. In the case of junk bonds, we find empirical support for the view that the existence of deposit insurance created a moral hazard situation that gave poorly capitalized institutions a greater incentive to increase their risk exposure.
Several recent studies suggest that poorly capitalized institutions have actively sought to take additional risk. Benston and Koehn (1989) reported that increased emphasis on riskier nontraditional activities resulted in greater stock return volatility for poorly capitalized S&Ls and lower volatility for healthier institutions. Brewer (1991) found that shifts in asset composition toward nontraditional activities resulted in increases in the return on equity for distressed institutions but had no effect on healthy institutions. This suggests that the shareholders rewarded risk-shifting actions that raised the value of the insurance subsidy.
This paper differs from the previous studies in that we analyze the impact of S&L junk bond exposure on market risk. Using a sample of seventy-four S&Ls from September 1985 to the end of 1989, we report that institutions with larger shares of junk bonds (as a proportion of their market value of net worth) had greater stock return volatility. This suggests that these institutions did use junk bonds to increase rather than reduce their risk exposure.
Next, we examine whether S&Ls with larger shares of junk bonds in their portfolios paid higher interest rates to depositors. One explanation for the growth in junk bonds at some S&Ls is that it is a manifestation of a moral hazard problem that is endemic to a system of fixed rate deposit insurance pricing. Merton (1977) and Buser, Chen, and Kane (1981) have shown that providing deposit guarantees at less than their market value subsidizes S&Ls. By increasing asset risk through additional holdings of junk bonds, an S&L can enhance the subsidy associated with fixed rate deposit insurance. This argument, referred to as the Moral Hazard Hypothesis, implies that junk bonds increase the risk of deposits. If institutions holding junk bonds are perceived by depositors to have a higher probability of failure, then uninsured depositors would demand a higher interest rate. We find a significantly positive relationship between junk bond investments and deposit interest rates for the 1987-1989 period. Thus, we conclude not only that junk bonds increased S&L equity market risk but: also that institutions that held larger shares of junk bonds were perceived as more risky by uninsured depositors.
If holding junk bonds increases risk for S&ls, then (1) why would S&ls invest in these assets, and (2) why were almost all junk bond investments concentrated in a small number of institutions? An S&L should increase its investment in junk bonds (or any asset) if the expected marginal benefit of doing so is greater than its expected marginal cost. If the stock market is operating efficiently, then this should be reflected in stock returns. However, the existence of deposit insurance alters the risk-return trade-off for some institutions. If S&Ls with large junk bond holdings were also poorly capitalized and had a higher probability of failure than other S&Ls, the deposit insurance option becomes more valuable and the expected gain of larger junk bond investments would exceed the S&Ls' expected loss. Thus, the stock market should reward these S&Ls with higher returns for increasing their riskiness. However, a well-capitalized institution that increased its holdings of junk bonds should experience lower (or no) stock return gains, because the deposit insurance option is not as valuable. We test this hypothesis by dividing our sample of seventy-four S&Ls into high- and low-capital firms. We find a significantly positive relation for the low-capital S&Ls between changes in junk bond holdings and stock return and an insignificant relation for the highly capitalized firms. These results provide an empirical example of an asset that increases both the expected cost to the deposit insurance fund and stock return for low capital S&Ls, while having no effect on stock returns for highly capitalized S&Ls.
This paper is divided into five sections. Section 1 describes the regulations concerning S&L junk bond investment and presents descriptive data on the extent of S&L holdings over the sample period. Section 2 develops the method used to test the effects of junk bond holdings on stock market risk. Section 3 analyzes the effects of S&L junk bond holdings on the cost of deposit funds. Section 4 tests the impact of junk bond investments on S&L stock returns. Section 5 concludes.
In the May 1983 regulation implementing the Garn-St Germain Depository Institutions Act of 1982, the Federal Home Loan Bank Board (FHLBB) authorized federally chartered S&Ls to invest up to (i) 1 percent of their assets in commercial paper and corporate debt securities and (ii) 10 percent of their assets in commercial loans. The FHLBB classified junk bonds under the category of commercial loans. Many state governments have also enacted statutes that broadened asset powers for their state-chartered. S&Ls. State-chartered S&Ls were permitted by several states to invest almost unlimited amounts directly in junk bonds.(2) Recently, these junk bond. investments have been associated with some of the largest S&L failures.
Table 1 reports S&L holdings of junk bonds by different classifications from 1985 to 1989. Several points are worth noting. First, from the end of 1985 to the end of 1988, total S&L holdings of junk bonds grew from $5.59 billion to $14.64 billion, an increase of over 160 percent in three years. After 1988, however, S&Ls began to reduce and/or write down their holdings of junk bonds, so that by year-end 1989 the amount held had declined to $10.46 billion. Second, throughout the sample period, the top fifty holders had over 95 percent of all S&L junk bond holdings. Third, even though these holdings were concentrated in a small number of institutions, these investments still represented a substantial amount relative to tangible capital. For the publicly traded stock associations that are among the top fifty junk bondholders, the dollar value of junk bonds exceeded their tangible capital.
[TABULAR DATA 1 OMITTED]
Junk bond investments are frequently perceived as relatively risky assets in the sense that the distribution of returns associated with a single asset of this kind or even a group of such assets has a large variance: some institutions will earn generous returns on these investments while others will suffer low or negative returns. Recent studies of the junk bond market have verified that, other things equal, junk bonds are more risky than investment-grade bonds but less risky than equity. For example, Perry and Taggart (1990) found that the standard deviation of monthly junk bond returns was greater than that of investment-grade bonds but less than that of equities. Blume, Keim, and Patel (1991) found that, from 1977 to 1989, low-grade bonds exhibited more volatility than equivalent government bonds. They also report that there was no indication that junk bonds are either overpriced or underpriced, and this corroborates the findings of a 1988 General Accounting Office study. In general, junk bonds are less liquid than either Treasury or investment grade bonds and more liquid than consumer and commercial loans.
2. THE RELATION BETWEEN JUNK BONDS AND S&L MARKET RISK
A. Theoretical Considerations
Do changes in S&L holdings of junk bonds significantly influence S&L riskiness? We address this question by examining the relation between the volatility of S&L stock returns and holdings of junk bonds. The first step in the development of the model, following Black and Scholes (1973) and Galai and Masulis (1976), is to relate the volatility of the market return on S&L equity, [[sigma].sub.MV], to the volatility of the return on S&L asset, [[sigma].sub.A]:
(1) [Mathematical Expression Omitted] where ([differential]MV/[differential]A)I(A/MV) is the elasticity of market value of equity with res the value of the assets of a representative S&L. Equation (1) indicates that the volatility of S&L equity returns is a function of the volatility of the asset returns, [[sigma].sub.A] the change in market value capital with respect to the change in total assets, [th]V/[th]A; and the leverage ratio, A/MV
Because we cannot observe all the right-hand-side variables in equation (1), a simplified econometric specification of equation (1), following Christie (1982), can be written as:
(2) [[sigma].sub.i,t] [S.sub.O] + [S.sub.1][LEV.sub.i,t] + [[epsilon].sub.i,t] where [[sigma].sub.i,t] is the equity return volatility ([[sigma].sub.MV] of the ith S&L in period t total asset-to-market value capital ratio of the ith S&L in period t; [[epsilon].sub.i,t] is an erro term; and the coefficients [S.sub.O] and [S.sub.1] are parameters to be estimated. Since greater leverage increases S&L riskiness, we predict [S.sub.1] > 0.
Christie (1982) indicates that the volatility of equity returns is affected by other variables besides leverage. For example, if an S&L holds a portfolio of mortgage and nonmortgage assets of differing degrees of risk, then changes in asset mix can either increase or decrease the volatility of equity returns.(3) The precise behavior of [[sigma].sub.i,t], will depend on the variance/covariance structure of the S&L asset returns. Three potential categories of risky nonmortgage assets are real estate direct investments (DIRECT), nonmortgage loans (NONMORT), and junk bonds (JUNK). Changes in the relative investment in these different assets might affect the volatility of S&L equity returns.
An S&L's riskiness is also influenced by the composition of its mortgage loan portfolio. During the early 1980s, S&Ls were given broader powers to hold commercial mortgage loans. If S&Ls altered the composition of their mortgage portfolios (moving, for example, from residential mortgage loans to commercial mortgage loans), this might also affect S&L stock return volatility. Barth and Bradley (1989) find that, within the mortgage category, insolvent institutions have rapidly increased their commercial mortgage lending. Barth, Bartholomew, and Labich (1989) present evidence indicating that acquisition and development loans, which are loans to finance the purchase of land and the accomplishment of all improvements required to convert it to developed building lots, have a positive and statistically significant effect on resolution costs. In our empirical analysis, four mortgage loan categories are examined: residential mortgage loans (RMORT), commercial mortgage loans (CMORT), acquisition and development loans (ADL), and other mortgage assets (OMORT). We expect that returns on commercial mortgage loans and acquisition and development loans would be more volatile than the returns on residential mortgage loans.
In order to account for each of the above asset mix variables, an expanded version of equation (2) is rewritten here as equation (3):(4)
(3) [Mathematical Expression Omitted] where all asset variables are divided by market value of capital, [W.sub.t] is a time dummy variable that is equal to one for quarter t (t = 2, ..., 7) and zero otherwise, and [Z.sub.i] is a cross-sectional dummy variable that is equal to one for the ith S&L (i = 2, ..., N) and zero otherwise. We included both cross-sectional and time dummy variables in the empirical specification to control for possible correlation either across institutions or across time.
B. Data Sources and Estimation Procedure
The data used in this paper are for seventy-four S&L organizations whose stocks were traded on the New York Stock Exchange, American Stock Exchange, or over the counter, and who filed FHLBB (now the Office of Thrift Supervision) Reports of Condition for each quarter from September 1985 to December 1989. A few of the seventy-four S&Ls were resolved by thrift regulators prior to the end of the sample period. These institutions are included in the sample period for the quarters before resolution, and are excluded from the sample for the time period after resolution. Stock market data are from Interactive Data Services, Inc. For multiple S&L holding companies, the assets of individual S&L subsidiaries are consolidated using reports of condition to construct the balance sheet variables discussed below.(5)
To obtain our measure of risk, we use daily stock market data. For each quarter in the sample period, estimates of the standard deviation of the daily returns on an S&L's equity were computed using data covering the three-month period ending with the last month of the quarter. The market value of equity is calculated by multiplying the number of shares outstanding at the end of each quarter by the price of the S&L's equity at the end of the quarter.
The asset-to-capital ratio (LEV) is calculated as the ratio of total book value assets to the market value of capital. The variables RMORT, CMORT, ADL, and JUNK represent values of residential mortgage loans, commercial mortgage loans, acquisition and development loans, and junk bonds, respectively. The other mortgage asset variable (OMORT) is the sum of multifamily mortgage loans and mortgage-backed securities. The real estate direct investment variable (DIRECT) is calculated by taking the sum of equity securities (except FHLB stock), real estate investments, and investments in service corporations or subsidiaries. The nonmortgage loan ratio (NONMORT) is the sum of total business and consumer loans. All asset variables were divided by the market value of capital.
Two versions of equation (3) are estimated for a pooled cross-section, time series sample of S&Ls over the period 1985:3 through 1989:4. One version of equation (3) uses only financial leverage and junk bonds as independent variables. The second version includes the remaining asset mix variables as independent variables. The models are also estimated over the March 1987-December 1989 period to conform with tests conducted later in the paper.
C. Empirical Results
Results from estimating the two versions of equation (3) using ordinary least squares are reported in Table 2. The estimated values of the parameters represent their cross-sectional average values.(6) The results from version one of equation (3) show a significant positive relationship between S&L return volatility and junk bond holdings. This supports the claim that S&Ls with larger proportions of junk bonds in their portfolios also exhibited higher volatility of stock returns. As expected, the coefficient on LEV is statistically significant and positively signed, verifying that higher stock return volatility is associated with higher financial leverage.
[TABULAR DATA 3 OMITTED]
The second column presents the results from estimating version two of equation (3). Again, the coefficient on junk bonds is significantly positive at the 1 percent level. One other asset category--acquisition and development loans (ADL)--has a positive and statistically significant coefficient, while the OMORT, NONMORT, and DIRECT variables have significantly negative coefficients. The result for ADL is consistent with previous studies. It is also worth noting that the coefficient on junk bonds is larger than any other asset coefficient except ADL. What this implies is that, holding market value and total assets constant, a portfolio shift from any asset except ADL into junk bonds would increase stock return volatility. Thus, we conclude that holdings of junk bonds increased the volatility of S&L stock returns for our sample of institutions over the 1985:3-1989:4 period. The results for the 1987:1-1989:4 sample period reported in the fourth and fifth columns of Table 2 are consistent with those over the entire sample period.
We also examined whether more liberal regulations on junk bond investment available to some state-chartered S&Ls may have resulted in their incurring greater risks than federally chartered S&Ls. These liberal guidelines have been blamed by federal regulators for some of the large losses of failed state-chartered S&Ls. Thus, we expected that changes in junk bond holdings should have a greater impact on the stock return volatility of state-chartered S&Ls than federally chartered firms. We tested this prediction by interacting a dummy variable that equals one for federally chartered S&Ls and zero otherwise with JUNK. The results are presented in columns three and six of Table 2. The negative coefficient on the multiplicative dummy variable indicates that junk bonds have less of an impact on the stock return volatility of federally chartered S&Ls than state-chartered firms.(7) Thus, we conclude that the greater range of junk bond authority available to state-chartered S&Ls allowed them to take greater risks than federally chartered associations.
3. THE RELATIONSHIP BETWEEN DEPOSIT INTEREST RATES AND JUNK BOND INVESTMENTS
Since junk bonds raise stock market volatility, one would expect, ceteris paribus, a positive relation between the deposit insurance liability and an S&L's holdings of junk bonds. While the deposit insurance liability is not directly observable, one can use the risk premium on an S&L's uninsured (or partially insured) deposits as a proxy. Assuming uninsured depositors behave as if they are not implicitly fully insured, the Moral Hazard Hypothesis predicts a positive relation between the interest rate on uninsured CDs and the volume of junk bonds. Following Baer and Brewer (1986), we test this hypothesis by estimating the following empirical model:
(4) [Mathematical Expression Omitted] where [RCDI.sub.i,t] represents the interest rate paid by the ith S&L in period t on large, partially insured certificates of deposit (deposits in excess of $ 1 00,000) with a maturity of six to twelve months and was obtained from the Quarterly Report of Condition. S&Ls were riot required to submit deposit interest rate data to regulators prior to 1987; hence, our sample period in this section is from the beginning of 1987 to the third quarter of 1989. RTB, is the riskless interest rate and is computed by taking
the average of the yields on 182-day Treasury bills and 364-day Treasury bills.(8) The [DEFAULT.sub.i,t] variable measures the default or credit risk of the S&l's asset portfolio and is captured by two variables. The first measure, [RISK.sub.i,t] is a market-based measure of the riskiness of an S&L's asset returns and is obtained by multiplying the variance in stock returns in a quarter by the square of the market value of equity to total assets.(9) The second proxy for credit risk is an accounting-based measure and is captured by an S&L's holdings of junk bonds ([JUNK.sub.i,t]) and acquisition and development loans ([ADL.sub.i,t]) as defined earlier. The variable [CAP.sub.i,t] is the ratio of the market value of common stock to total assets of the ith S&L at the end of quarter t; [SIZE.sub.i,t] represents the natural logarithm of total assets; [AGROWTH.sub.i,t] is the percentage change in total assets over quarter t for the ith S&L; and [v.sub.i,t] is an error term.
Since CDs and Treasury bills are close but not perfect substitutes, we expect the coefficient on RTB to be positive but less than one. S&Ls do not adjust their CD rates as rapidly as market interest rates change. We predict the coefficient on CAP should be negative: because a higher capital-asset ratio implies a lower probability that depositors would suffer a loss. 10 The coefficients on the credit risk variables should be positive because an increase in asset risk implies that there is a greater chance that the value of an S&L's assets will fall below the level needed to repay all depositors. We include an asset size measure as an additional explanatory variable to account for the possibilities that purchasers might view the CDs of larger S&Ls as being more liquid. The last variable chosen was asset growth (AGROWTH), included because rapid asset growth was linked by the now-defunct FHLBB both to high likelihoods of failure and to high costs to the deposit insurance fund to resolve those failures. Asset growth is expected to be positively related to the deposit rate because rapidly growing institutions often must bid more aggressively for funds both within and outside their normal deposit market.
Three versions of equation (4) are estimated. One version uses [JUNK.sub.i,t] and [ADL.sub.i,t] as the measure of credit risk. The second version uses the adjusted variance (which should capture not only the impact of JUNK and ADL on an S&L's asset risk but other factors as well). The third version of equation (4) includes both sets of credit risk measures. The results reported in Table 3 indicate that all the coefficients are significantly different from zero. As expected, the CD rate is, for all three versions of the model, positively related to the Treasury bill rate and negatively related to both the capital-asset ratio and asset size. Moreover, the coefficients on the JUNK and ADL variables are positive and statistically significant. This result suggests that CD rates reflect the credit risk of the S&L. In the second and third columns of Table 3 the coefficient on the RISK variable is significantly positive, indicating that depositors received higher interest rates for bearing additional risk. Finally, the results show a significantly positive relationship between S&L CD rates and asset growth, supporting the comcerns of many that institutions growing rapidly are paying higher rates to increase their deposits. These results are consistent with previous studies that found a risk premium in interest rates paid on large CDs [see, for example, Baer and Brewer (1986) and Hannan and Hanweck (1988)]. Thus, we conclude that institutions that had larger shares of junk bonds in their portfolio were perceived as more risky by uninsured depositors.
[TABULAR DATA OMITTED]
4. THE IMPACT OF JUNK BOND INVESTMENTS ON S&L STOCK RETURNS
A. Theoretical Considerations
In this section., we examine the effects of junk bond holdings on S&L stock returns. We have already shown that S&Ls with higher proportions of junk bonds had more volatile stock returns, even after controlling for the effects of financial leverage and asset mix. S&Ls with large holdings of junk bonds were also less capitalized than those with small holdings. Figure 1 compares the aggregate tangible accounting principles (TAP) capital-to-asset ratios for the twenty-one S&Ls in the sample classified as "high" junk bondholders with those for the fifty-three S&Ls classified as "low" junk bondholders.(11) To be considered a "high" junk bondholder, the S&Ls in our sample must have ranked among the top fifty junk bondholders at the end of 1988, the date when industry junk bond holdings were greatest. The remaining S&Ls were classified as "low" junk bondholders. For every quarter between 1986 and 1989, the capital-asset ratio for the "high" junk bond group was lower than the "low" junk bond group.(12) Other things held constant, lower capital-asset ratios at high junk bond S&Ls indicate a greater exposure to the risk of failure. However, previous research has associated high stock market elasticity with low capital-asset ratios (Ang et al. 1985). Thus, it is possible that the positive correlation that we found between stock return volatility and junk bond holdings may be the result of S&Ls with greater junk bond holdings having higher stock market elasticities rather than higher volatility of asset returns. To examine this issue, we estimated the following model for our sample of S&Ls over the March 1987-December 1989 period:(13) (5) [RET.sub.i,t] = [alpha] + [[beta].sub.1][MRET.sub.t] + [[beta].sub.1,1]([MRET.sub.t])(DUM) + [[o where [RET.sub.i,t] is the holding-period return for the ith S&L stock in time period t; [MRET.sub.t] is the holding-period return on the market portfolio in time period t; DUM is a binary variable equalling one for a high junk bond S&L and zero otherwise; and [[omega].sub.i,t] is an error term. The stock market portfolio used to compute MRET in this study is the value-weighted portfolio (NYSE and AMEX) obtained from the Center for Research in Security Prices (CRSP) data base. The coefficient on the multiplicative dummy variable, ([MRET.sub.t])(DUM) measures the impact of the return on the market portfolio on the stock return of high junk bond S&Ls relative to low junk bond S&Ls. The results of estimating equation (5) are provided below: [RET.sub.i,t] = -0.0930 + 1.2205 [MRET.sub.t] - 0.3576 ([MRET.sub.t])(DUM)
(- 11.838) (14.523) (-2.386) Adj. R-Sq = 0.2189 F-Statistic = 119.421
These results indicate that S&Ls with higher proportions of junk bonds had lower elasticities of equity.(14) The behavior of S&L elasticity of equity does not appear to explain the positive correlation between the volatility of stock returns and junk bond holdings; thus we can conclude that the volatility of asset returns increased with junk bond holdings. The CD rate analysis also supports this conclusion. The evidence presented on uninsured CD rates is consistent with the Moral Hazard Hypothesis that increased junk bond holdings (and acquisition and development loans) increases overall S&L asset risk.
The impact of junk bond investments on S&L stock returns may differ across firms because underpriced deposit insurance may be more valuable to poorly capitalized S&Ls than to others. To examine this issue, we first relate the rate of return on S&L stock to the rate of return on a stock market index and the market return on junk bonds, [RHY13OND.sub.t]:
(6) [RET.sub.i,t] = [[beta.sub.0] + [[beta].sub.1[MRET.sub.t] + [[beta].sub.2][RHYBOND.sub.t] + [[
This equation is then expanded to take into consideration the value of access to underpriced fixed-premium deposit insurance. When the government offers deposit insurance, it assumes a liability: if the institution fails, the government must redeem the claims of the insured depositors. Kane (1985) indicates that, given access to underpriced fixect-premium deposit insurance, an S&L consists of two types of assets, its portfolio, holdings and the value of its deposit insurance. As a result, the common stock returns of S&Ls reflect the return on both types of assets. The effect of underpriced fixed-premium deposit insurance on S&L stock returns can be illustrated using the option pricing framework. Merton (1977) valued deposit insurance as a (European) put option using the Black and Scholes formula. For simplicity, following Mertoli, we consider a federally insured S&L that has a capital structure consisting of deposits, D, and equity only. The S&L currently holds total assets with value of [[alpha].sub.0].
Assume that the value of the S&L's total assets, A, follows a lognormal diffusion process with constant variance, [[sigma].sub.A], and that the other standard assumptions of option pricing hold.(15) At a time period T, [[alpha].sub.T] has a lognormal distribution, and the value of the deposits, with guaranteed principal and interest, is [e.sub.rT]D, where r is the risk-free rate of interest. If, at the time of the deposit insurer's examination date T, the value of the S&L's assets is greater than the value of deposits, the deposit insurer pays nothing. The value of shareholder's equity is max(0, [[alpha].sub.T] - [e.sup.rT]D). However, i on the examination date the value of the S&L's assets is less than its deposits, the deposit insurer must make up the difference. Merton has shown that the deposit insurance liability, L, is min(0, [[alpha].sub.T] - [e.sub.rTD]D).
The deposit insurance liability is a negative number because it creates a potential cash inflow to the S&L. Following Esty and Baldwin (1991), the deposit insurance liability can be written as the asset value of the S&L, [alpha].sub.0], minus the value of deposits, D, and the option value of equity, S:
(7) L = [[alpha].sub.0] - D - S. Dividing both sides of equation (7) by Ao results in
(8) L/[[alpha].sub.0] = [1 - D/[[alpha].sub.0]] - S/[[alpha].sub.0].
The deposit insurance liability per dollar of assets equals the S&L's capital-to-asset ratio (CAP) minus the option value of equity (divided by assets). But the option value of equity is a function of the capital-to-asset ratio, volatility of asset returns, and other variables. Thus:
(9) L/[[alpha].sub.0] = CAP - S(CAP, [[sigms].sub.A], [tau], r).
Equation (9) shows that the deposit insurance fund's exposure is affected by the actions of individual S&Ls. The higher the capital-to-asset ratio, the lower will be the value of the access to deposit insurance since it is less likely that an S&L will fail and exercise the option and impose losses on the insurance system. The lower the asset risk, [[sigms].sub.A], the lower will be the value of the access to deposit insurance. The primary factors affecting the return on access to underpriced, fixed-premium deposit insurance are changes in total assets relative to capital, and changes in asset volatility due to changes in asset composition (see Brewer 1991). Equation (6) can be expanded to account for each of these factors:
(10) [Mathematical Expression Omitted] where DFINLEV is the change in total assets relative to capital; [delta][[alpha].sub.i,t]([kappa]) i change in the holdings of the kth asset at time t of the ith S&L; and [mu][V.sub.i,t-1] is the market value of capital of the ith S&L in period t - 1. With underpriced, fixed-premium deposit insurance, value-maximizing S&Ls have incentives to shift risk to the deposit insurance fund in an attempt to expropriate wealth. From equation (10), two sets of factors that S&Ls can potentially manipulate for risk-shifting purposes are examined: change in total assets relative to capital and asset mix variables. S&Ls' attempts to shift risk are successful if the net effect of their manipulations of financial leverage and/or asset mix results in an increase in the value of underpriced, fixed-premium deposit insurance. The approximate benefits to shareholders of such manipulations are captured by the last two terms in equation (10). To control for the possible impact of other S&L-specific factors on stock returns, individual S&L dummy variables, [[zeta].sub.i], multiplied by the returns on the market and bond portfolios are included to allow the market betas and interest rate coefficients to vary cross-sectionally.
The DFINLEV variable should capture the impact of changes in financial leverage on the value of access to deposit insurance and shareholders' common stock returns The k individual assets studied include: direct real estate investments (DDIRECT); nonmortgage loans (DNONMORT); other nonmortgage assets (DOASSET); junk bonds (DJUNK); residential mortgage loans (DRMORT); commercial mortgage loans (DCMORT); acquisition and development loans (DADL); and other mortgage assets (DOMORT).(16)
Much of the concern about deregulation of S&L nonmortgage activities has to do with low-capital S&Ls gambling some of the institutions' deposits on investments with large but unlikely payoffs. In order to examine this issue, the S&Ls in this study are ordered according to their TAP capital-to-asset ratio and divided into high-capital and low-capital associations. The high-capital category includes those firms with TAP capital in excess of 3.0 percent.(170 The low-capital category is comprised of the remaining S&Ls in the sample.
B. Data Sources
The data sources for the stock prices, market values, and balance sheet items are described in section 2. The common stock returns over a quarter are calculated by compounding weekly common stock returns within a quarter. The stock market portfolio used to compute MRET in this study is the value-weighted portfolio (NYSE and ANLEX) obtained from the CRSP data base. We also compute MRET using the thrift index obtained from SNL Securities to check the robustness of our results.(18) A measure of the total returns on junk bonds is constructed from a high-yield bond index obtained from Merrill-Lynch.(19)
C. Empirical Results
The results of estimating the common stock return equations are shown in Table 4. The first two columns of Table 4 present the results using the value-weighted stock market portfolio as a measure of MRET The last two columns report the results using the thrift industry index as a measure of MRET For low-capital S&Ls, the results in column one indicate that common stock returns increase significantly with growth in junk bonds (DJUNK). However, growth in junk bonds does not have a statistically significant impact on the common stock returns of high-capital S&Ls. With the exception of real estate direct investments, all of the other asset mix variables have positive coefficients and most have a significant impact on the common stock returns of low-capital S&Ls. Surprisingly, the variable measuring the change in total assets relative to capital (DFINLEV) is significantly related to common stock returns for high-capital S&Ls.(20) When the industry index is used, the results are similar to those reported for the value-weighted market index. In particular, for low-capital S&Ls, common stock returns increase significantly when funds are shifted from residential mortgage loans into junk bonds and acquisition and development loans. There is no evidence of an impact of balance sheet shifts on the common stock returns of high-capital S&Ls.
[TABULAR DATA OMITTED]
In this paper, we first examine whether the financial markets view S&Ls with relatively large exposure to junk bonds as more risky than S&Ls with smaller exposure to junk bonds. We test this hypothesis using data on S&L stock returns and interest rates paid on large CDs. We find that equity return volatility appears to be positively related to the proportion of junk bonds held in an S&L's portfolio. In addition, we find evidence that CD holders demand higher rates when junk bond holdings increase relative to market value of equity.
We attempt to explain why junk bond holdings are concentrated among a small number of institutions and why these holdings grew so rapidly in the 1986-1988 period. Because of the low capital-asset ratios of the large junk bondholders, we test the Moral Hazard Hypothesis by dividing the sample of institutions into two groups based on their TAP capital ratios, and examine the relation between their stock returns and changes in their asset mix. We find that the stock returns of S&Ls who have relatively low capital are positively related to shifts in funds from residential mortgage loans into junk bonds (and acquisition and development loans). The stock returns of other S&Ls, however, are not related to changes in junk bond holdings relative to their capital. These results support the notion that the stock returns of S&Ls on the "edge" respond to volatility increases as if deposit insurance is a valuable subsidy. Access to deposit insurance is not as valuable for other types of S&Ls (that is, those with relatively more capital).
The results of this study should not be construed as support for the decision by Congress to force S&Ls to exit the junk bond market by 1994. Rather, we argue that some S&Ls had an incentive to take on excessive risk in many ways, including the purchase of junk bonds. Underpriced, fixed-premium deposit insurance, in effect, provides an incentive for S&Ls to take additional risks, since it induces a positive correlation between stock market returns and changes in holdings of risky assets. Closing the junk. bond market to S&Ls will not prevent S&Ls from taking more risk because there are many ways for depository institutions to acquire assets which are at least as risky as junk bonds. Legislative action that attacks excessive risk-taking by prohibiting institutions from acquiring particular classes of risky assets is attacking the symptoms of the disease instead of its causes and is doomed to fail. If the incentives to increase risk are there, then value-maximizing institutions will find a way to circumvent regulations and increase risk. The solution is to adopt policies that eliminate incentives for institutions with low capital to increase their risk exposure.
The authors thank Herbert Baer, George Kaufman, James Lindley, Jacky So, Steven Strongin, Vefa Tarhan, two anonymous referees, and the participants of finance seminars at DePaul University and th University of Notre Dame for valuable comments and suggestions. The research assistance of Loretta Ardaugh, George Rodriguez, and Gary Sutkin is greatly appreciated. All views expressed here are thos of the authors and are not necessarily those of the Federal Reserve Bank of Chicago or the Federal R System. (1.) Noninvestment grade securities may be transferred to a holding company affiliate or (for mutual to a separately capitalized subsidiary. (2.) California, Connecticut, Florida, Louisiana, Ohio, Texas, and Utah were the states with more le guidelines for state-chartered S&Ls. (3.) It is worth noting that our specification for the asset mix variables is equivalent to dividing variable by total assets and multiplying by leverage. Thus, leverage is implicitly interacted with t composition variables in our model. We thank a referee for clarifying this point. (4.) One might expect that stock return volatility may be positively related to growth since many S& suffering large losses also were growing rapidly during this period. We did not include this variabl we believe that growth is a consequence rather than a cause of S&L risk taking, since faster deposit growth enables an institution to acquire more risky assets. In this section, we choose to focus on t relationship between stock return volatility and asset choice. (5.) For each of the holding companies included, the S&Ls were the major activity of the holding com in terms of assets. The mean ratio of S&L assets to total holding company assets was 96 percent over the sample period. Other holding company activity included real property management, housing development, brokerage services, insurance products, data processing services, corporate debt and eq services, and real estate appraisal services. Assets for the holding companies were obtained from Moody's Banking and Finance Manual, various years. (6.) We performed two additional tests. We could not reject the null hypothesis of homoskedasticity using White's test. We also estimated the equations using Fuller's and Battese's (1974) error compon model. The results were qualitatively similar to those reported. (7.) The simple correlation coefficient between the junk bond-market value ratio and the charter dum variable was -0.09, which was significantly different from zero at the 1 percent level. (8.) Because the CD rate is for a deposit with a maturity between six and twelve months, it was nece to develop a riskless rate of interest that covers these two maturity dates. (9.) The calculation in the paper assumes that the variance of the return on debt is zero. This adju has been used by several other researchers including Marcus and Shaked (1984). (10.) For empirical evidence supporting this hypothesis see Pozdena (1991) and Gendreau (1991). (11.) TAP capital is computed by subtracting the value of goodwill and other intangible assets from GAAP capital. For a discussion of the use of goodwill by the FHLBB in the 1980s to augment regulator though not economic, net worth, see Black (1990) and Barth (1991). (12.) Similar results are obtained when GAAP capital-to-asset ratios are compared. (13.) This approach has been used, for example, by Brickley and James (1986). (14.) From option pricing theory, the following relation exists between the return on equity, [r.sub return on the assets of the firm, [r.sub.v]: (F1) [r.sub.s] = [[eta].sub.s] [r.sub.v], where [[eta].sub.s] is the elasticity of equity with respect to the value of the assets of the firm. text provides an estimate of the elasticity of equity and any change in the elasticity between high bond S&Ls and low junk bond S&Ls, holding the intercept constant across junk bond groups. To allow for cross-sectional differences in the intercept, the following equation was estimated over the 1987 period:
(F2) [Mathematical Expression Omitted] where [[zeta].sub.i] is a cross-sectional dummy variable that is equal to one for the ith S&L (i=2,. otherwise. The parameter [[beta].sub.1] is equal to 1.12 with a t-statistic of 13.47. The coefficien dummy variable is again negative but it has a t-statistic of -0.20. This result indicates no signifi differences in the elasticity of equity between high and low junk bond S&Ls, after allowing for cros differences in the intercept. The finding that the elasticity of equity for high junk bond S&Ls is not significantly different from that of low junk bond S&Ls is still consistent with the notion that junk bond holdings increases S&L asset risk. (15.) For example, frictionless markets and no arbitrage opportunities arc assumed. (16.) The model as specified in equation (10) cannot be estimated directly because assets are relate each other through the balance sheet identity. Perfect multicollinearity and singular cross-product will occur in the estimation unless one asset is deleted. DRMORT is omitted in the estimation. Using balance sheet identity, it can be shown that the coefficients of the included asset categories measu impact on stock returns of shifts from residential mortgage loans to the included asset categories. (17.) Alternatively, the sample of S&Ls was divided into two groups using TAP capital ratios of 2 an percent. The result; were qualitatively similar to those reported. (18.) The SNL securities' thrift index excludes dividend payments. Thus, the standard errors of the coefficient estimate s will likely be higher because of this errors-in-variable problem. (19.) The junk bond index started October 31, 1986. (20.) A partial explanation for these results is that when a firm's economic capital is positive as many of the high-capital S&Ls, an increase in assets relative to market capitalization means greater Greater leverage will lead to a rise in the value of access to deposit insurance. And as a result, this should generate higher stock returns. For economically insolvent S&Ls, however, Flannery (1985) has shown that an increase in assets relative to market capitalization initially generates higher ca ratios (less negative ratios), leading to lower stock returns.
Ang, James, Pamela Peterson, and David Peterson. "Investigations into the Determinants of Risk: A New Look." Quarterly Journal of Business and Economics 24 (Winter 1985), 3-20. Baer, Herbert, and Elijah Brewer. "Uninsured Deposits as a Source of Market Discipline: Some New Evidence. " Economic Perspectives, Federal Reserve Bank of Chicago (September/October 14)86), 23-31. Barth, James R. The Great Savings and Loan Debacle. Washington, D.C.: The AEI Press, 1991. Barth, James R., Philip F. Bartholomew, and Carol Labich. "Moral Hazard and the Thrift Crisis: An Analysis of 1988 Resolutions." Proceedings of a Conference on Bank Structure and Competition, Federal Reserve Bank of Chicago, 1989. Barth, James R., and Michael G. Bradley. "Thrift Deregulation and Federal Deposit Insurance." Journal of Financial Services Research 2 (September 1989), 231-59. Benston, George: J., and Michael F. Koehn. "Capital Dissipation, Deregulation, and the Insolvency of Thrifts." Unpublished paper, Emory University (December 1989). Black, Fischer, and Myron Scholes. "The Pricing of Options and Corporate Liabilities." Journal of Political Economy 81 (May/June 1973), 637-59. Black, William K. "Ending Our Forbearers' Forbearances: FIRREA and Supervisory Goodwill." Stanford Law and Policy Review 2 (Spring 1990), 102-16. Blume, Marshall E., Donald B. Keim, and Sandeep A. Patel. "Returns and Volatility of Low-Grade Bonds," Journal of Finance 46 (March 1991), 49-74. Brewer, Elijah III. "The Impact of Deposit Insurance on S&L Shareholders' Risk/Return Trade-offs. " Unpublished paper, Federal Reserve Barik of Chicago (September 199 1). Brickley, James A., and Christopher M. James. "Access to Deposit Insurance, insolvency Rules, and the Stock Returns of Financial Institutions." Journal of Financial Economics 16 (July 1986), 345-71. Buser, Stephen A., Andrew H. Chen, and Edward J. Kane. (Federal Deposit Insurance, Regulatory Policy and Optimal Bank Capital." Journal of Finance 36 (March 1981), 51-60. Christie, A. A. "The Stochastic Behavior of Common Stock Variance." Journal of Financial Economics 10 (December 1982), 407-32. Esty, Benjamin C., and Carliss Y. Baldwin. "Market Discipline in the Thrift Industry." Working Paper #92-045. Harvard Business School (November 1991). Flannery, Mark J. "Recapitalizing the Thrift Industry. " Proceedings of the Eleventh Annual Conference, Federal Home Loan Bank of San Francisco (December 1985), 91-113. Fuller, W. A., and G E. Battese. "Estimation of Linear Models with Crossed-Error Structure." Journal of Econometrics 2 (May 1974), 67-78. Galai, D., and R. W. Masulis, "The Option Pricing Model and the Risk Factor of Stock." Journal of Financial Economics 3 (January/March 1976), 53-81. Gendreau, Brian. "U S. Deposit Insurance Reform." In World Financial Markets. Morgan Guaranty Trust Co. (January 25, 1991). General Accounting Office. "High Yield Bonds: Nature of the Market and Effect on Federally Insured Institutions." Washington: Government Printing Office (May 1988). Hannan, Timothy H. and Gerald A. Hanweck. "Bank Insolvency Risk and the Market for Large Certificates of Deposit." Journal of Money, Credit, and Banking 20 (May 1988), 203-11. Kane, Edward J. The Gathering Crisis in Deposit Insurance. Cambridge, Mass.: MIT Press, 1985. Kane, Edward J. The S&L Insurance Mess. Washington, D.C.: Urban Institute Press, 1989. Marcus, Alan, and Israel Shaked. "The Relationship between Accounting Measures and Prospective Probabilities of Insolvency: An Application to the Banking Industry." Financial Review 19 (March 1984), 67-83. Merton, Robert C. "Analytical Derivation of the Cost of Deposit Insurance and Loan Guarantees: An Application of Modem Option Pricing Theory." Journal of Banking and Finance 1 (June 1977), 3-11 Perry, Kevin J., and Fobert A. Taggart, Jr. "Development of the Junk Market and Its Role in Portfolio Management and Corporate Finance." In The High-Yield Debt Market: Investment Performance and Economic Impact, edited by Edward I. Altman, pp. 185-205. Homewood, Ill.: Dow Jones-Irwin, 1990. Pozdena, Randall. "Recapitalizing the Banking System." FRBSF Weekly Letter, Federal Reserve Bank of San Francisco (March 8, 1991).…