In this section, we test our hypotheses regarding how the implications of provisions for loan losses and loan charge-off for current returns, future cash flows, and future returns depend on how banks’ income smoothing behavior varies across the bust vs. boom phases of the business cycle and homogeneous vs. heterogeneous loan types. Because we can observe loan charge offs and recoveries but not provisions for loan losses by loan type, we conduct this analysis in two distinct ways. Our primary analyses distinguish loan charge-offs and recoveries by loan type. We also conduct cruder analyses distinguishing provision for loan losses for homogenous and heterogeneous banks.
Current Returns Analysis
Distinguishing Net Loan Charge-offs by Loan Type
In this section, we test our hypothesis that net loan charge-offs for homogenous loans have positive implications for current returns during the 1990s boom. We regress market-adjusted returns from April 1 during the fiscal year to March 31 following the fiscal year end (hereafter current returns), R, on the change in earnings before the provision for loan losses, the change in the provision for loan losses, the change in net loan charge-offs for homogeneous and heterogeneous loans – NLCO(HOM) and NLCO(HET), respectively – and the change in non-performing assets during the year, with all explanatory variables deflated by price on April 1 of year t13
13 The main difference between equation (3) and those estimated in Liu and Ryan (1995) and Ahmed, Takeda, and Thomas (1999) is that the earnings before the provision for loan losses, provision for loan losses, and net loan charge-off variables are all in first differenced, not levels, form. We specify equation (3) in this way because, unlike during the 1980s, we find that first differenced variables have more explanatory power over returns in the 1990s, likely due to these variables’ greater persistence in the later period.
. )
( )
( t U t t t
T t t
t X PLL NLCO HOM NLCO HET NPA
R =α+β∆ +γ∆ +δ ∆ +δ ∆ +ς∆ +ε (3)
Studies prior to Ahmed, Takeda, and Thomas (1999) invariably find that net loan charge-offs have significantly negative implications for current returns. In contrast, Ahmed, Takeda, and Thomas find that net loan charge-offs have no implications for current returns in their later sample period from 1987-1995. As discussed above, we expect net loan charge-offs for homogeneous loans to have a positive association with returns.
Table 4 reports the estimation of equation (3). As expected, the coefficient δT on NLCO(HOM) is significantly positive, consistent with loan charge-offs for homogeneous loans being good news. In contrast, the coefficient δU on NLCO(HET) is insignificantly negative.
These results imply that it is important to distinguish loan charge-offs for homogeneous and heterogeneous loans in our sample period, and they suggest that Ahmed, Takeda, and Thomas’
(1999) finding that loan charge-offs have no implications for returns reflects a weighted-average of the valuation implications of charge-offs of homogeneous and heterogeneous loans.
Distinguishing Provisions for Loan Losses by Bank Type
In this section, we test our hypothesis that provisions for loan losses for both
homogenous and heterogeneous banks have negative implications for current returns during the 1990s boom. Similar to Liu and Ryan (1995), to assess whether the implications of provisions for loan losses for returns vary across banks based on the types of loans they hold, we estimate a modified version of equation (3) that allows the coefficient on the provision for loan losses to
differ for the homogeneous and heterogeneous bank sub-samples, with the indicator variable IHOM denoting the homogeneous bank sub-sample14
.
* HOM t t t
t HOM t
t
t X PLL PLL I NLCO NPA
R =α +β∆ +γ∆ +γ ∆ +δ∆ +ς∆ +ε (3’)
Liu and Ryan (1995) find that the coefficient γ on the provision for loan losses variable is significantly positive for their full sample over the 4Q:1983-2Q:1991, but that this result is attributable only to the heterogeneous bank sub-sample. In contrast, Ahmed, Takeda, and Thomas (1999) find that this coefficient is significantly negative over 1987-1995 for both sub-samples. Reflecting the timeliness of provisions for loan losses for heterogeneous banks in our sample period, we expect Ahmed, Takeda, and Thomas’ finding of negative coefficients for both sub-samples of banks to remain in our sample period.
Table 5 reports the estimation of equation (3’). Consistent with Ahmed, Takeda, and Thomas (1999), for the overall sample we find that the coefficient γ on the provision for loan losses is significantly negative for both samples and insignificantly different across the sub-samples. Thus, provisions for loan losses have negative implications for current returns in the 1990s boom. While this aspect of our results differs from those of Liu and Ryan (1995), they also find relatively timely provisions for loan losses to be bad news.
Future Cash Flows Analysis
Following Wahlen (1994), Liu, Ryan, and Wahlen (1997), and Ahmed, Takeda, and Thomas (1999), as an alternative to the current returns analysis, in this section we assess the
14 We do not include intercept dummies to capture mean effects when slope dummies are included in regressions in this paper; such intercept dummies are invariably insignificant and have no noticeable effect on any aspect of the estimations.
association of provisions for loan losses and loan charge-offs with future cash flows. Since the provision for loan losses is the main accrual estimate for banks, as in these prior studies we measure cash flows as earnings before the provision for loan losses.
Distinguishing Net Loan Charge-offs by Loan Type
In this section, we test our hypothesis that net loan charge-offs for homogenous loans have positive implications for future cash flows. To assess the implications of net loan charge-offs for future cash flows, we regress each of the next three years’ earnings before the provision for loan losses on the current year’s earnings before the provision for loan losses, the provision for loan losses, net loan charge-offs for homogeneous and heterogeneous loans, and the change in non-performing assets, with all variables deflated by market value at the beginning of the current year15
While Wahlen (1994) and Liu, Ryan, and Wahlen (1997) find that net loan charge-offs have significantly negative implications for future cash flows, Ahmed, Takeda, and Thomas (1999) find the opposite in their later sample period 1987-1995, consistent with our hypothesis that loan charge-offs become good news in the early 1990s. Reflecting our prior discussion and results for the current returns analysis, we expect net loan charge-offs for homogeneous loans to have a positive association with future cash flows in our sample period.
15 Equation (4) is similar to one estimated by Liu, Ryan, and Wahlen (1997), but differs somewhat from those in Wahlen (1997) and Ahmed, Takeda, and Thomas (1998), who use changes in earnings before the provision for loan losses variables and measures of the unexpected provision for loan losses, net loan charge-offs, and change in non-performing loans. As discussed below, we find similar results regardless of which equation we estimate, and so we
Table 6 reports the estimation of equation (4). As expected, the coefficient dT on NLCO(HOM) is significantly positive at all horizons s from 1 to 3, consistent with loan charge-offs for homogeneous loans being good news. In contrast, the coefficient dU on NLCO(HET) is insignificant at all three horizons. Again, these results imply it is important to distinguish loan charge-offs for homogeneous and heterogeneous loans in our sample period, and they suggest that Ahmed, Takeda, and Thomas’ (1999) finding that loan charge-offs have positive
implications for future cash flows is driven by charge-offs of homogeneous loans.
Distinguishing Provisions for Loan Losses by Bank Type
In this section, we test our hypothesis that provisions for loan losses for both
homogenous and heterogeneous banks have negative implications for future cash flows during the 1990s boom. Again similar to Liu and Ryan (1995), to assess whether the implications of provisions for loan losses for future cash flows vary across banks based on the types of loans they hold, we estimate a modified version of equation (4) that allows the coefficient on the provision for loan losses to differ for the homogeneous and heterogeneous bank sub-samples.
.
Wahlen (1994) for the 1977-1988 period and Liu, Ryan, and Wahlen (1997) for the 1Q:1984-4Q:1989 period find that the coefficient c on the provision for loan losses is significantly positive, while Ahmed, Takeda, and Thomas (1999) find that this coefficient is significantly negative in the 1987-1995 period. We again expect Ahmed, Takeda, and Thomas’ finding of negative coefficients for both sub-samples of banks to remain in our sample period.
Table 7 reports the estimation of equation (4’). Inconsistent with our hypothesis and the results of Ahmed, Takeda, and Thomas (1999) and our prior returns analysis, the coefficient c on
the provision for loan losses is insignificant at all horizons s from 1 to 3 for both sub-samples of banks, implying that provisions for loan losses have no implications for future cash flows during the 1990s.
We conjecture that the main explanation for this difference in the results in the current returns and future cash flow analyses is that during the 1990s provisions for loan losses conveyed (or possibly just were thought by the market to convey) some relatively timely information about loan losses that was reflected in returns, but that these provisions also contained considerable noise about the cash flows in any given future period because of excessive reserving for loan losses. This conjecture also explains why we find the coefficient f on the change in non-performing assets in equations (4) and (4’) to be much more negative and significant than do Ahmed, Takeda, and Thomas (1999), consistent with the change in
non-performing assets containing less noise about future cash flows than the provision for loan losses.
Given that macro-economic conditions were remarkably and unexpectedly robust during the 1990s, it is possible that the market interpreted provisions for loan losses to be worse news than they turned out to be, however, so that the cash flow analysis better reflects the actual valuation implications of provisions for loan losses. This possibility is diminished by a specification test discussed below, in which we observe no drift in returns in the year subsequent year to
provisions for loan losses.
Future Returns Analysis
In this section, we test our hypothesis that net loan charge-offs for homogeneous loans are associated with a positive drift in future returns. We also investigate whether these charge-offs’ association with future earnings and recoveries explains any drift we observe. To determine
whether a drift exists, we first estimate a regression of one-year-ahead market-adjusted returns on the change in earnings before the provision for loan losses, the change in the provision for loan losses, the change in net loan charge-offs for homogeneous and heterogeneous loans, the change in non-performing assets during the year. To explain the drift that we observe, we then estimate expanded regression equations that include the one-year-ahead change in earnings before the provision for loan losses and/or the one-year ahead recoveries for homogeneous and heterogeneous loans, denoted REC(HOM) and REC(HET), respectively. All explanatory variables are deflated by price on April 1 of year t. The most expanded equation is
.
A positive coefficient δT on NLCO(HOM) in the restricted equation implies a positive return drift subsequent to loan charge-offs for homogeneous loans. If δT becomes less positive (zero) as future net income and recoveries are included in the expanded equations, then the drift is partly (fully) explained by the association of these variables with NLCO(HOM).
Table 8 reports the estimation of equation (5). In the most restricted equation, the
coefficient δT on NLCO(HOM) is significantly positive, consistent with a return drift subsequent to net loan charge-offs for homogeneous loans. The coefficient β on ∆Xt is also significantly positive, perhaps because banks’ profitability was unexpectedly persistent throughout the 1990s boom, and because repeated profitability has positive valuation implications, as documented by Barth, Elliott, and Finn (1999). No other slope coefficients in the restricted model are significant.
As next year’s change in recoveries and especially change in net income are included in the equation the coefficient δT on NLCO(HOM) becomes less positive, and it is insignificant in
the most expanded equation. Thus the return drift associated with net loan charge-offs for homogeneous loans is explained fully by their association with future net income and recoveries.
Interestingly, the return drift subsequent to the change in earnings is strengthened significantly by the inclusion of next year’s change in net income, perhaps because inclusion of this variable captures the effect of repeated profitability.
We emphasize that these tests of future returns should not be viewed as contributing to the return anomalies literature, because of the selected time period and industry investigated. In particular, we do not expect positive return drift subsequent to loan charge-offs in the current credit riskier, less well reserved banking environment.
Summary and Specification Tests
The results in this section imply that the valuation implications of provisions for loan losses and loan charge-offs differ across the bust and boom phases of the banking industry business cycle and across homogeneous and heterogeneous loan types, reflecting the effect of these factors on banks’ income smoothing. In particular, healthy banks’ acceleration of loan offs for homogeneous loans during the 1990s led to the unusual situation of these charge-offs being positively associated with both current returns and future cash flows. Moreover, investors incompletely appreciated this manipulation, so that these charge-offs are associated with a drift in future returns that is explained by their association with future net income and recoveries.
In addition, we find that the provision for loan losses conveys either bad or no news about loan losses during the 1990s. Thus, our results are more consistent with those of Ahmed, Takeda, and Thomas (1999) than with research prior to that study. In particular, unlike Wahlen
(1994), Liu and Ryan (1995), and Liu, Ryan, and Wahlen (1997), in no analysis or sub-sample do we observe that provisions for loan losses convey good news.
We conducted five specification tests, none of which change the basic character of our results regarding the valuation implications of provisions for loan losses and loan charge-offs.
First, we specified the explanatory variables in equations (3)-(5) in both levels and first differences forms and also using Wahlen’s (1994) expectations models; deviations in the specifications from those in the reported results generally weaken the results in the returns analysis slightly and have minimal effect on the results in the future cash flow analysis. We also measured returns unadjusted, industry adjusted, and size adjusted, with little effect on our results.
Second, we included firm and year (industry) effects; firm effects have little effect on either the returns or future cash flow analyses, but year effects typically reduce the coefficient (t statistic) by about half (one-third) in the returns analysis, consistent with the banking industry exhibiting correlated discretionary behavior in given years. Our reported results do not include year effects, because we are interested in discretionary reporting behavior at both the banking industry and individual bank levels.
Third, banks merged with considerable frequency over this period, and so we deleted observations for which a bank’s loans grew more than 10% in any quarter of a given year (about 5% of our observations), which slightly weakens but does not change the character of our results.
Fourth, we replicated all our analyses distinguishing the valuation implications of the change in net loan charge-offs and recoveries for banks with above- and below-normal
allowances for loan losses as reflected in Ihighall indicator variable described earlier. We find that banks with above-normal allowances for loan losses drive the positive implications of loan charge-offs for homogeneous loans for future cash flows and future returns, but not current
returns, consistent with over-reserved banks accelerating loan charge-offs and this discretionary behavior not being appreciated fully by investors.
Finally, as an indication of the efficiency of the market reaction to provisions for loan losses, we used one-year ahead market-adjusted returns as the dependent variable in equation (3’) and obtained an insignificant coefficient γ on the provision for loan losses for both sub-samples of banks, consistent with the market reacting in an unbiased fashion to provisions for loan losses.
VI. CONCLUSION
In this paper, we illustrate the complex and contextual nature of income smoothing in the setting of banks’ provisions for loan losses and loan charge-offs. Specifically, we show that while banks manage provisions for loan losses to smooth income, their exercise of discretion is constrained by whether they hold homogeneous or heterogeneous loans, and that these
constraints bind differently in the bust and boom phases of the banking industry business cycle.
During boom periods, banks have more ability to accelerate provisions and charge-offs for homogeneous loans, while during bust periods they have more ability to delay provisions and charge-offs for heterogeneous loans. By comparing and reconciling the results for our sample period of 1991-2000 to those of prior research, we show that banks’ exercise of discretion given these constraints has substantial effects on the valuation implications of provisions for loan losses and loan charge-offs.
We also find that the market does not react fully to banks’ income smoothing behavior during the 1990s boom, documenting a return drift subsequent to charge-offs of homogeneous loans that is explained by the association of these charge-offs with future net income and recoveries. We hypothesize that this drift occurs because of the perceived non-discretionary
nature of charge-offs based on number-of-days-past-due rules. As discussed by Dechow and Skinner (2000), earnings management is inherently more problematic when it is opaque.
These results bear on the recent policy debate about rules versus principles-based accounting standards. While rules-based standards are often thought to yield less discretionary accounting, our results suggest that a rules-based accounting policy that appears particularly non-discretionary – the use of number-of-days-past-due rules to determine charge-offs of
homogenous loans – leads to more discretion in certain contexts.
Relatedly, our results suggest that Barton and Simko’s (2002) arguments and evidence that income management can be constrained by the magnitude of balance sheet needs to be probed in specific contexts by future research. While the balance sheet surely influences the conduct and evaluation of income management, whether the balance sheet imposes real or perceived constraints appears to depend on the context. In particular, we provide an example of a context in which a perceived constraint facilitates income smoothing.
TABLE 1 Descriptive Statistics
Full Sample 1991-2000 Panel A: Means and Quartiles of Variables
mean minimum 25% median 75% maximum #obs
Raw R .153 -.520 -.117 .138 .387 .939 1734
ALL/NLCO 8.068 -62.210 3.038 5.453 10.436 77.702 1751
CAP .115 .041 .085 .110 .137 .307 1736
Panel B: Medians of Variables in Odd Years in Sample Period
1991 1993 1995 1997 1999
Raw R .396 -.005 .349 .573 -.187
ALL/NLCO 2.525 5.386 6.251 5.862 6.191
CAP .073 .114 .112 .110 .105
max #obs 122 141 197 252 329
TABLE 1 (continued)
Panel C: Pearson Correlations of Variables
(2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
TABLE 1 (continued)
The full sample includes all observations of any variable included in any subsequent analysis.
Observations in the outside .5 percent of either tail of the distribution of any variable in a given analysis or for which total assets or loans outstanding on the regulatory Y9-C reports differs from the value on Bank Compustat by more than 10% are deleted.
Variable definitions:
Raw R= raw returns over the 12-month period from April 1 during the reporting year
Raw R= raw returns over the 12-month period from April 1 during the reporting year