The Long-Run Performance of Firms Following Loan Announcements

Transcription

1 The Long-Run Performance of Firms Following Loan Announcements by Matthew T. Billett a Henry B. Tippie College of Business, University of Iowa Mark J. Flannery b Warrington College of Business, University of Florida Jon A. Garfinkel a Henry B. Tippie College of Business, University of Iowa ABSTRACT Unlike equity offerings or public debt offerings, bank loan financing elicits a significantly positive announcement return. The lengthy literature on firm financing decisions relies (in part) on this finding to characterize bank loans as unique or special forms of external finance. We further explore the uniqueness of private lending agreements by examining the long-run equity performance of bank borrowers. In contrast to the positive announcement returns, the typical borrowing firm significantly underperforms relevant benchmarks over the subsequent three years. The bank borrowers abnormal returns are quite large, roughly comparable to the long-run returns following seasoned equity offers or public bond issues. This conclusion is robust to numerous variations in our measurement methods. At least over the longer-run, private loans do not reveal, or add, value for the borrowing firms. Rather, it seems that undertaking any major type of external finance portends poor future stock returns. We thank Jason Karceski, Anand Vijh, and especially Jay Ritter for helpful comments and suggestions, and Jennifer Marietta-Westberg for excellent research assistance. All remaining errors are our own. November 26, 2001

2 1. Introduction Myers and Majluf [1984] argue that a firm s decision to issue external securities instead of using internally generated funds may indicate that insiders consider the firm to be overvalued. Asymmetrically informed outside investors will then make valuation inferences based on how insiders choose to raise capital. The degree of inferred overvaluation increases in the sensitivity of the offered security's value to the asymmetric information. For example, an equity issue signals greater overvaluation than a bond issue, and convertible bonds should reflect more negative information than straight bonds. Short-term event studies largely support this lemons model of security valuation. The announcement of a seasoned equity offering (SEO) results in an average stock price decline of 2 3% (Asquith and Mullins [1986], Masulis and Korwar [1986], Bayless and Chaplinksy [1996]), while announcements of public bond issues generate zero or slightly negative equity returns (Eckbo [1986], Jung, Kim, and Stulz [1996], Howton, Howton, and Perfect [1998]). 1 A large body of theoretical work compares the benefits of private debt (e.g. bank loans) to armslength (public) borrowing. Institutional lenders are generally viewed as insiders, who may enhance a borrowing firm s value by reducing information asymmetries or by monitoring firm performance more efficiently (Bernanke [1983], Fama [1985], Berlin and Loeys [1988], Kwan and Carleton [1998]). In contrast to the announcement effects of public security issues, bank loan announcements generate significantly positive abnormal returns (Mikkelson and Partch [1986], James [1987], Lummer and McConnell [1989], and many others). The combination of private lending s theoretical benefits with bank loans positive announcement effects has made private loans appear special or unique among a firm s financing alternatives. 2 1 Ritter [forthcoming, Table 5] summarizes many other studies of the impact of financing decisions on firm equity value. 2 Boot [2000] and Ongena and Smith [2000] review the theoretical and empirical evidence about private lending relationships. The empirical loan announcement literature includes some relatively minor unresolved issues about the importance of lender identity (Preece and Mullineaux [1994], Billett, Flannery and Garfinkel [1995], Carey, Post and Sharpe [1998]) and about how a private lender gathers inside information (Lummer and McConnell [1989], Best and Zhang [1993], Wansley, Elayan and Collins [1992]). Overall, however, the literature clearly indicates that private loans generate positive announcement effects.

3 2 Although the short-run valuation effects of security issuances are consistent with the existing theory of asymmetrically informed outside investors, recent work on the long-run performance following security issuance has raised some important doubts. Numerous studies document substantial subsequent abnormal returns during the 3 5 years after a firm issues external securities. At face value, these results imply that market investors initially under-react to, or perhaps misunderstand, the implications of public security issuances. Moreover, the apparent under-reaction is substantial: negative abnormal returns in the range of 4 10% per year are not unusual. The long-run performance literature began with Ritter [1991] who showed that firms substantially underperform the market over the three years following their IPO. Extending this line of inquiry, Spiess and Affleck-Graves [1995] find that the mean five-year buy-and-hold abnormal return (BHAR) following a seasoned equity offering is 30.99%. 3 Loughran and Vijh [1997] document significant underperformance by firms that pay for acquisitions with shares instead of cash. Firm equity values also fall sharply in the wake of public debt issues (Lee and Loughran [1998], Spiess and Affleck-Graves [1999]). Datta, Iskandar-Datta, and Patel [2000] document similar underperformance following a firm s initial bond offerings. The estimated abnormal returns are again large: e.g. Spiess and Affleck-Graves [1999] find significant median five-year BHARs of 18.7% following issuance of straight public bonds and 19.8% for convertible bonds. In sum, prior studies suggest that both stock and bond issuances portend poor future performance. An issuing firm s long-run performance generally reinforces the implications of issuance announcement effects: long-run returns following equity issues are more negative than those following public debt issues, with convertible bond results in between. Private equity issuance deviates from this pattern. The initial announcement elicits a positive return (Hertzel [1993]), similar to announcements of private loan agreements. However, private equity s positive announcement effect is followed by substantial longerrun underperformance. Hertzel et al. [2001] find that the mean buy-and-hold return for firms issuing 3 Loughran and Ritter [1995] also find long-run underperformance following a seasoned equity offering.

4 3 private equity fall far short (20 45%) of the non-issuing peer firms return over the subsequent five years. In this paper, we evaluate whether the short-run valuation effects of inside loan announcements persist. 4 Do investors initially under-react to the benefits associated with bank loans? Or do loans resemble other sources of external funding enough so that the longer-run effects are negative? This is an important research issue because loans have a prominent place in the capital-raising process. In the aggregate, bank loans account for approximately 30% of all U.S. nonfinancial corporations credit, and this proportion is doubtless higher abroad. 5 Furthermore, bank loan financing is typically a major financing event. The mean (median) ratio of loan size to borrower s market value of equity is 86.9% (51.1%) in our sample, which comes from the years By contrast, Spiess and Affleck-Graves [1999] report that the mean (median) ratio of bond issue size to the market value of equity is 53.64% (28.86%) during their sample period. We find that firms announcing private lending agreements substantially underperform over the long run, just as equity and public debt issuers do. In the three years following a loan announcement, the mean (median) annual BHAR for the sample of loan announcing firms is 10.0% ( 9.5%). Because the appropriate evaluation of long-run security returns is controversial (Lyon, Barber and Tsai [1999], Brav et al. [2000], Mitchell and Stafford [2000], Eckbo et al. [2000], Loughran and Ritter [2000], and others) it is important that our conclusions be robust to alternative methodologies. In particular, some authors have challenged the robustness of prior researchers abnormal return statistics because security-issuance is endogenous to firm conditions. If security-issuance events tend to cluster in calendar time or in specific industries, the issuing firms subsequent returns may be correlated, invalidating the classical test statistics 4 Dichev and Piotroski [1999] have previously examined long-run performance following public and private debt issuance, inferring new debt issues from year-to-year changes in a firm s balance sheet. They find very weak evidence that firm performance improves following a private debt issue. Their methods differ so greatly from ours (and that of the other papers we cite here) that comparing results is quite difficult. 5 Data is from Federal Reserve Bulletin, table 1.59 and is for the year 1996 (the first year bank loan data is reported). For comparison, corporate bonds accounted for 47% of corporate credit in Note that nonbank loans are not included in these figures so private lending agreements account for far more than 30% of all U.S. corporate credit.

5 4 used to establish significance. We address these potential econometric problems with a variety of measurement techniques. In addition, Section 2 demonstrates that our sample of loan announcements is likely to be better balanced than prior studies samples of public securities issuance. Our basic results are confirmed when we select peer firms on the basis of size and book-to-market ratio, size alone, size and industry, or size and momentum. In addition, estimates from the Fama-French three-factor model imply that loan announcers suffer significantly negative abnormal returns in the range of % annually over the ensuing three years. By contrast, the (non-borrowing) peer firms exhibit slightly positive abnormal returns. Finally, we show that Vijh s [1999] method of computing calendartime abnormal returns (CTARs) implies borrower underperformance on the order of 4.6% 10.8% annually for the three years following a loan announcement. Our empirical results have two primary implications. First, they add to the growing literature suggesting that investors systematically misunderstand the implications of certain financing events. Second, the special nature of private ( bank ) lending appears to be a transitory phenomenon. From the perspective of long-term shareholders, private loans resemble the issuance of public securities. Our results can therefore motivate a new discourse on the value of financial intermediation. The remainder of this paper is organized as follows. Section 2 describes our data sources. Section 3 discusses the methodological issues associated with long-run performance measurement. Our results are presented in Section 4, and the final Section concludes. 2. Data In collecting data for an earlier paper (Billett, Flannery and Garfinkel [BFG, 1995]), we used a keyword search of news stories to identify 1,468 announced loan agreements between nonfinancial borrowers and banks or nonbank lenders during the calendar years 1980 through That paper s focus on loan announcement effects led us to concentrate on a clean sub-sample of 626 announcements with no confounding events around the announcement date (e.g. merger discussions or new investment

6 5 programs), and for which the borrower s equity traded on the announcement day. Our longer-term focus in the present paper is less sensitive to concurrent events, and we therefore require only that the CRSP master file report the announcing firm s equity market value at the preceding calendar yearend. Our main sample thus includes 1,385 loan announcements from , although we also report results for a variety of sub-samples in Section 4.4. The sample s summary statistics in Table 1 reveal several noteworthy points. First, loan agreements are significant external financing events: the average loan or commitment size is 86.9% of the firm s market value of equity. Second, loan announcers tend to be small firms: the median market value of equity for our sample of firms is $58.1 million compared to the NYSE listed firm median of $197 million over the period. Viewed another way, the mean (median) size decile (based on NYSE cuts) for our sample of firms is 3.09 (2) with more than two-fifths our sample belonging to decile 1. 6 Third, our average firm closely resembles the average Compustat firm (which tends to be small) in terms of average growth potential: the median sample firm s market to book assets ratio is 1.13, while the corresponding Compustat universe s median value is 1.12 over The mean market-to-book asset ratios are a bit farther apart, with our sample mean equal to 1.53 and the Compustat mean equal to Overall, univariate statistics are consistent with the traditional view that bank borrowers tend to be smaller firms with relatively valuable growth opportunities. 7 An obvious concern with this type of sample is that the expected returns to small, high-growth firms may be poorly described by available asset pricing models (Fama and French [1993], Fama [1998]). This would potentially bias long-run performance assessments. We address this possibility by utilizing a variety of return measurements and several distinct criteria for identifying control firms. Event clustering (in time or within industries) may render the sample firms abnormal returns cross-sectionally correlated. This would bias upward the computed t-statistics for long-run returns (Lyon, 6 Decile 1 is the smallest and decile 10 the largest. See also Panel C of Table 2. 7 The minimum stock price of $0.06 in our sample raises the question whether low-priced firms drive our results. Although not reported in the tables, our results are unchanged when we omit firms with stock prices less than $5.

7 6 Barber and Tsai [1999], Mitchell and Stafford [2000]). However, Table 2 indicates that our sample exhibits no strong clustering of loan announcements, either in terms of calendar time or borrowers industries. Panel A of Table 2 suggests that our sample is quite evenly distributed over the 1980s, unlike equity issues which cluster significantly in particular years (Mitchell and Stafford [2000]). Panel B of Table 2 describes the distribution of loan announcements by industry, defined by the first digit of the borrower s SIC code. We formally test whether events cluster within certain industries using a Kolmogorov Smirnov two-sample test to compare the distribution of our sample firms industries against the population s industry distribution. Specifically, we compare the cumulative distributions of one-digit SIC codes for our sample and the Compustat universe across the entire sample period (1980 through 1989), and on a year-by-year basis. The full sample period s test statistic does not reject the null of similar distributions of SIC codes for loan announcing firms and all Compustat firms over the 1980s (p=0.246). When we conduct our test on individual years, the null hypothesis is rejected (p-value = 0.078) for only one year (1989). (The next closest p-value to the in 1989 is the p-value in 1985.) Given ten separate tests (one for each year of the 1980s), it is not surprising that one of the tests would reject the null at the 10% level. We conclude that our test statistics are unlikely to be substantially compromised by the effects of calendar or industry clustering. 3. Methodology Simulation experiments by Lyon, Barber, and Tsai [1999] lead them to conclude that misspecification in nonrandom samples is pervasive. Thus analysis of long-run abnormal returns is treacherous (emphasis added). Mitchell and Stafford [2000] echo this sentiment and provide evidence that the abnormal returns documented by prior studies of acquisitions, share repurchases and seasoned equity offerings are less significant than standard parametric tests indicate. Therefore, we examine the long-run performance using all the latest techniques, and we find that the resulting underperformance of bank borrowers is robust to both variations in sample construction and method of analysis.

8 Prior Methodological Evaluations Early studies of long-run performance simply extend event study techniques to a longer horizon. These analyses compare the subsequent equity performance of each individual firm associated with the event under study (e.g. stock issuance) to that of a reference portfolio. Some researchers make specific adjustments for the security s beta or other factor loadings. However, Kothari and Warner s [1997] simulation evidence suggests that the size and power of these parametric tests are both over-stated. In particular, the abnormal returns computed by subtracting benchmark portfolio returns from an individual security s returns tend to be substantially skewed. Barber and Lyon [1997] reiterate the importance of simple abnormal returns skewness, and describe additional potential biases that may arise from new listings and market portfolio re-balancings. All of these problems can be ameliorated by using peeradjusted, buy-and-hold abnormal returns (BHARs) to measure long-run performance effects, as in Ritter [1991]. For each sample firm, Barber and Lyon [1997] suggest choosing a peer firm based on market capitalization and equity s book-to-market ratio. The difference between the sample firm s and its peer firm s holding period returns then indicates the impact of the studied event on subsequent performance. Barber and Lyon conclude that this BHAR technique performs well in randomized samples when appropriate peer firms are chosen. Lyon, Barber and Tsai [1999] point out that events often cluster in calendar times (e.g. hot issue IPO markets), or affect similar firms at the same time (e.g., firms in the same industry, firms experiencing a rapid stock price run-up, value vs. growth firms). With either type of clustering, the firms post-event returns are likely to be positively correlated with one another. 8 Specifically, test statistics based on peer-adjusted returns are biased in the presence of non-random samples or clustering 8 Similar calendar time event occurrence can be driven by either of two factors. First, different firms may all tend to experience the event around the same time (for example firms like to issue seasoned equity following a long price run-up). Alternatively, the same firm may have multiple events in close time proximity. The second occurrence therefore falls during the long-run return calculation window that followed the first occurrence. According to Lyon, Barber and Tsai [1999], the first situation causes little trouble for peer-adjusted returns, but multiple firm events can have more serious consequences.

9 8 on the basis of non-matched firm characteristics (such as industry or pre-event returns). On the other hand, calendar time portfolio approaches perform well in the presence of calendar time clustering of events but perform poorly in the presence of firm characteristic clustering by event firms. Lyon, Barber and Tsai suggest that alternative peer choices can reduce the likelihood of clustering on the basis of nonmatched characteristics. Mitchell and Stafford [2000] suggest several potential solutions to the calendar time event-clustering problem. First, one can compute peer-adjusted (or benchmark portfolio adjusted) returns, and adjust their standard errors to reflect positive cross-correlations of abnormal returns. A substantial difficulty here is that the researcher must make an assumption about the strength of this correlation. Mitchell and Stafford s other two suggestions involve creating a sequence of calendar time portfolios. Each month, the researcher forms a portfolio containing all the firms that announced the event within the last (say) three years. These calendar time portfolio returns may then be evaluated either of two ways. First, the researcher can regress the portfolio s time series of excess returns (over the riskfree rate) on Fama and French s three factors: the market risk premium, the difference in returns between small and large stocks, and the difference in returns between high and low book-to-market stocks. The intercept from this regression then measures abnormal performance. Alternatively, Mitchell and Stafford note that one may subtract monthly returns on portfolios of firms with similar size and book-to-market characteristics from the calendar time portfolio of sample firms returns Our Methodological Design We provide empirical results for several types of statistical tests, based on several definitions of peer firms. The first method we describe here is the peer-adjusted buy-and-hold abnormal returns (BHARs) approach, which permits easy comparison with earlier analyses of the long-run wealth effects following other financing events. In addition, BHARs measure an investor s experience if s/he were to try to profit from expected underperformance (Barber and Lyon [1997]). For each loan-announcing firm, 9 Loughran and Ritter [2000] argue that these latter two approaches fail to pick up time variation in abnormal performance that may be correlated with the frequency of events.

10 9 we select a peer firm that resembles the sample firm except for the announcement of loan financing. We then compute each firm s subsequent holding period return as: HPR i = = T i t 1 ( 1+ R ) 1 x 100% it where R it is the i th firm s stock return on the t th day, and T i is the number of trading days in the 3-year period following the loan announcement. After calculating HPR for each sample firm and for its matching firm, we evaluate the mean and median holding period return differences (BHARs) to determine if loan-announcing firms exhibit distinctive performance. The definition of peer firms is crucial in this analysis. We always select matching firms from the same trading venue (NYSE/AMEX or NASDAQ) as the sample firm. Subject to this constraint, we employ four different sets of matching criteria: size and book-to-market ratio, size alone, size and 2-digit SIC code, and size and momentum. In order to save space, we describe the matching process in detail only for one criterion: style based matches. Matching sample and peer firms on the basis of their style involves three dimensions: the trading venue (NYSE/AMEX vs. NASDAQ), size, and book-to-market ratio. We first discarded any sample firm for which Compustat reports non-positive book equity at the fiscal yearend preceding the loan announcement. For the remaining loan announcers, we followed Spiess and Affleck-Graves [1999] in identifying all firms that traded on the same exchange whose equity market value lay within 10% of the sample firm s at the prior calendar yearend. 10 The chosen peer has the smallest sum of the absolute percentage differences in size and book-to-market equity value, using data from the preceding fiscal yearend. A cursory analysis of the summed absolute percentage differences reveals the precision of our 10 The restrictions that firm book value be positive and that the peer be within 10% of the sample firm reduced our sample to 1,169 loan announcements. The 10% size proximity criterion addresses Barber and Lyon s [1997] finding that matched firm (peer) adjusted returns are misspecified when the event firm is very large. They attribute this to allowing peer firms to be within [70% to 130%] of the event firm s market cap. Very large event firms may have significantly smaller peer firms unless the size match criterion is tightened.

11 10 matching algorithm. The mean (median) absolute difference in firm size is 2.9% (2.4%), and the corresponding measures for book/market are 7.1% (2.9%). The mean summed difference is 9.99% and the median is 5.94%. More than three-quarters of our matches have summed differences below 10%. A peer-selection methodology issue must handle delistings of sample firms and peer firms that occur before the end of the performance measurement window. In about 20% of our style-matched pairs (231 out of 1,169), the peer firm is delisted before three years have passed. On these delisting dates, we switch the peer return series to that of the sample firm s second-closest matching firm. If an announcing firm s second peer was delisted, we continue the computations using its third-closest peer from the announcement day, and so forth. 11 Conversely, a sample firm was delisted within three years of the loan announcement 241 times out of 1,169 observations (20.6%). In these cases, we terminate the computations for both firms in the pair. Examining the reasons for either sample firm or peer firm delisting, we see that mergers were associated with 60.2% and 48.5% of delistings for peer and sample firms respectively. Also, 26.4% and 44.0% of peer and sample firm delistings were due to an exchange (NYSE/AMEX/NASDAQ) dropping the issue. Liquidations were rather infrequent, representing 2.6% and 1.66% of the peer and sample firm delistings. The remaining delistings (10.8% of peer firms and 5.8% of sample firms) involved exchanges for a different issue trading on NYSE, AMEX, or NASDAQ. In addition to computing BHARs relative to peer firms, we measure abnormal long-run performance using the Fama-French [1993] calendar time portfolio regression method. Each month, we form a portfolio containing all firms that announced a loan agreement within the past 36 months. We then regress a time series of these monthly portfolio returns, net of the risk free rate, on the three Fama- French factors: (R pt R ft ) = α + β( R mt R ft ) +ssmb t + hhml t + ε t (1) where R pt is the return on the portfolio of sample firms in month t; 11 Among these 231 peer firms who required replacement, 176 had two peers (one delisted and one replacement) during the subsequent five years, 40 had three peers, and 15 had four peers.

12 11 R ft is the 3-month T-bill yield in month t; R mt is the return on the value-weighted index of NYSE, Amex, and NASDAQ stocks in month t; SMB t is the return on small firms minus the return on large firms in month t; and HML t is the return on high book-to-market stocks minus the return on low book-to-market stocks in month t. A significant intercept term in (1) implies that abnormal returns are associated with the event analyzed. We estimate this Fama-French regression using two alternate portfolio-weighting schemes: valueweighting the announcing firms returns and equal-weighting them. Previous writers have observed that variation in the number of firms included in different months portfolios may cause heteroskedastic residuals in (1). We address this concern by reporting both OLS and weighted least squares (WLS) coefficient estimates, where the WLS weights equal the square root of the number of announcing firms in the portfolio for that month. Despite these precautions, it remains possible that any abnormal returns detected by this methodology reflect only the asset pricing model s inability to explain the returns on the type of firm represented in the sample. Fama [1998, page 299] and Mitchell and Stafford [2000, pages 324-5] argue that the Fama-French three-factor model performs especially poorly for small firms and high book-tomarket firms. Loughran and Ritter [1995] note that if the abnormal returns indicated by the Fama-French model reflect a bad model, significant intercept terms should also occur for the style-matched peer firms. We therefore estimate Fama-French regressions for the portfolio of peer firm returns, and compare the intercepts (α s) to those from regressions using sample firms. We more formally compare the sample firms performance to their peers via a peer-adjusted Fama-French test of long-run performance. We subtract monthly peer portfolio returns from the monthly sample firm returns and regress this difference on the usual factors. This approach also allows us to assess whether sample and peer firms returns differ

13 12 in their sensitivity to size and book-to-market factors, in addition to reaffirming differences in the intercepts. Another concern with the Fama-French regression approach is its assumption of parameter stability over the entire estimation window (Mitchell and Stafford [2000]). We address this possible problem by implementing Vijh s [1999] methodology to compute calendar time abnormal returns (CTARs). Specifically, we first calculate the annual returns on the portfolio of loan-announcing firms by compounding monthly returns starting in January and ending in December of each year (for firms who had announced the loan within the last three years). We then subtract the annual returns on a similar portfolio of peers to obtain annual excess returns. We calculate a t-statistic for the average of these annual excess returns using the time series standard deviation of annual excess returns over Results 4.1. BHAR Results Table 3 compares loan-announcers buy-and-hold abnormal returns to those of their peers. Panels A through D are based on style-matched (i.e. size and book-to-market matched) peers. In Panel A, both the mean (-27.2%) and median (-25.8%) three-year BHARs are significantly negative, with 99% confidence. Despite bank loans well-known positive announcement effect, they are associated with significant long-run underperformance. Indeed, loans seem to generate a more negative impact than public debt offerings: Spiess and Affleck-Graves [1999] report mean (median) abnormal returns of % (-18.7%) during the five years after a straight bond is issued. (Their mean return does not differ significantly from zero.) The fact that bank-loan borrowers fare even worse than bond-issuers calls into question much of the extant banking theory, which presumes that private debt (uniquely) enhances firm value. 12 We also perform the same analysis using monthly portfolio returns, with similar conclusions.

14 13 Panel B of Table 3 reports the peer-adjusted performance separately for each year in the threeyear period after borrowing. Mean and median underperformance is surprisingly similar in all three years, averaging between 10 and 11 percent per year. Underperformance is not concentrated in a small time window. Panel C reports mean and median BHARs for various announcing firm size deciles. While the smallest eight size deciles exhibit double-digit mean and median underperformance, significance is limited to groups of firms below the NYSE median size deciles one through five. This result foreshadows our subsequent discovery that equally-weighted abnormal returns are more prominent than value-weighted returns. Even so, the average loan announcement is surely not good news for smaller firms long term shareholders. Panel D investigates whether significant underperformance by loan announcers is concentrated in a particular time period. It appears not. Either mean or median performance is significantly negative in every year except 1980, with most years indicating underperformance for both measures. Panels E through G of Table 3 demonstrate that our general conclusion about loan announcers holds for a variety of peer firm definitions. Given the evidence in panel C that borrower size is an important determinant of long-run returns, we choose another set of peer firms on the basis of firm size alone. These matches are very close: the mean and median absolute valued percentage size difference between sample and peer firms is 0.2% and 0.1% respectively, and the largest absolute percentage difference is 6.37%. Panel E shows that borrowing firms significantly underperform size-matched peers. The mean 3-year BHAR is 35.5% and the median is 32.6%. Panel F reports the results when we choose peer firms to match borrower industry and size. For 1,055 announcing firms, we were able to find a peer firm traded on the same exchange, with the same (two-digit) SIC code, and with equity market value within 10% of the sample firm s. The borrowers mean (median) 3-year underperformance is 15.4% (18.5%), with each statistic differing from zero at the 99% confidence level. Finally, Panel G of Table 3 controls for the firms momentum, which has been identified as an important effect for many

15 14 stocks. We selected a set of peer firms based on trading venue, equity market value and the firms cumulative raw return over the year preceding the loan announcement. The results in Panel G again indicate substantial and significant negative excess returns. The mean 3-year BHAR is 22.4%, while the median is 23.7%. In sum, it appears that the choice of benchmark has no qualitative effect on our conclusion that loan announcers underperform over the subsequent three years. This is particularly important given Fama s [1998] warning that results based on BHARs can vary dramatically depending on whether peer firms are chosen based on size and book-to-market versus based solely on size. In addition, the robustness of our results to industry and momentum based matching alleviates concerns raised by Lyon, Barber and Tsai [1999] regarding clustering on the basis of both items. The results using size-only matched peers begin to address concerns about small firm effects in our sample. We return to this issue below Calendar-Time Portfolio Approaches We address the possible effects of calendar time event clustering more directly in Table 4, which reports the results of regressing calendar time portfolios excess returns on the three Fama-French factors. We report OLS estimation results for both value-weighted and equal-weighted portfolio returns. Since the number of firms in monthly portfolios varies over time, we also control for potentially heteroskedastic residuals by undertaking weighted least squares (WLS) estimations using the square root of the number of announcing firms to weight each month s return. 13 Recall that the intercepts (α) from this regression measure abnormal returns. Panel A reports that our sample borrowing firms estimated intercepts are all significantly negative. Value-weighting the borrowing firms subsequent returns yields an estimated monthly abnormal return of 0.49% using OLS or 0.36% using WLS estimation. That is, loan announcers 13 Because we only have loan announcement data from January 1980 through December 1989, the monthly portfolio returns calculated early and late in the time period will probably have fewer firms returns in them than monthly returns calculated in the middle of the sample.

16 15 underperform by an average of % annually over the following three years. The intercepts t- statistics ( 2.6 and 3.3) indicate that these abnormal returns differ from zero with 99% confidence. It is not obvious that one should value-weight the portfolio returns. Loughran and Ritter [2000] suggest that value-weighting will reduce the extent of measured misvaluations, which are likely to be more prevalent among smaller firms. We therefore present results for equal-weighted portfolio returns in the last two rows of Panel A. As predicted by Loughran and Ritter [2000], the equal-weighted intercepts ( 0.84% and 0.95% per month) are somewhat larger than under value weighting, now implying annual abnormal returns of 9.6% ( 10.8%) with t-statistics of 3.3 ( 5.6) for the OLS (WLS) estimators. If the negative intercept terms in Table 4 Panel A result from the three-factor model s inability to fit the type of firms in our borrowing sample, we should also find negative intercepts for the stylematched peer firms. Panel B presents the result of estimating the Fama-French regressions for calendar time portfolios of the style-matched peer firms. The intercept terms are always positive (though not always significant), consistent with the hypothesis that loan agreements themselves are associated with the poor performance manifested in Panel A. We examine the robustness of this conclusion in Panels C through E, by estimating α s for alternative peer groupings. When we match only on firm size in Panel C, the intercepts are always positive, though statistically indistinguishable from zero. The next two Panels of Table 4 report Fama-French regression estimates for the peer firms chosen on the basis of industry and size, and of size and momentum. Once again, these firms exhibit small and insignificant intercept terms. It appears that the Fama-French model accounts well for the systematic components of stock returns for firms similar to our bank borrowers. A cursory examination of the factor loadings on sample and peer firms across the panels A through E suggests different sensitivities to these factors between peer and loan announcing firms. We formally test the significance of these differences in Table 5, by subtracting the monthly portfolio return on peers from the contemporaneous sample firms portfolio return, and regressing these differences on the three Fama-French factors. Once again, we present results for four distinct sets of peer firms. The

17 16 intercept term in these regressions measures a peer-adjusted monthly abnormal return to bank borrowers, controlling for the effects of systematic risk, size, book-to-market, and calendar clustering of events. Style matched peers are used in Panel A, which contains several interesting facts. The intercepts are all negative with 99% confidence, implying average annual underperformance on the order of 7% 11%. However, the coefficients on the size and book-to-market factors are all positive and nearly always significant. Despite the fact that the peer firms are matched on the basis of size and book-to-market, the sample firms sensitivities to SMB and HML exceed those of the peer firms. The significant positive β coefficients for EW portfolios also indicate that sample firms may be more sensitive to the market. The underperformance of bank loan announcers relative to their peers does not appear to be driven by differences in the factor sensitivities. We evaluate the robustness of this conclusion for alternative sets of peer firms. Panel B matches on size (market capitalization) alone. The peer-adjusted returns still exhibit sensitivity to SMB, although (surprisingly) HML sensitivities are substantially reduced. Despite these differences, the regression α s remain significantly negative, with roughly the same magnitude as in Panel A. The remainder of Table 5 replicates the analysis for peers matched on SIC and Size (Panel C), or Size and Momentum (Panel D). None of these peer definitions produces a portfolio with zero factor loadings. However, in each case the results suggest that bank loan announcers underperform their peer firms even after controlling for differences in their sensitivities to the Fama- French factors. The CTAR approach advocated by Vijh [1999] provides an alternative way to control for calendar time clustering concerns, and it ameliorates concerns about the Fama-French regressions assumed 13-year parameter constancy. Table 6 presents the estimated CTARs constructed as the difference between sample firms portfolio returns and (a variety of) peers portfolio returns. In addition to OLS results, we again present WLS estimates that use the square root of the number of firms in the monthly portfolio to weight each month s returns.

18 17 The Table s first two lines present information about the CTARs based on style-matched peer firms. If we value-weight portfolio firms returns, borrowing firms significantly underperform peers by 4.63% or 7.19% annually, depending on whether we weight by the number of firms per month in the portfolio (to control for heteroskedasticity). Both measures are significant with at least 95% confidence. Under the more traditional equal weighting of portfolio firms returns, annual underperformance of loan announcers averages 10.81%, significant with 99% confidence, for both OLS and WLS estimates. All these estimates closely resemble the implied underperformance from Panel A of Table 5. The remainder of Table 6 presents corresponding estimates based on alternative peer definitions: Size alone, Industry and Size, or Size and Momentum. They all indicate similar levels of underperformance by loan announcers, most of which are statistically significant with at least 95% confidence. In sum, Tables 3 6 indicate that firms announcing loan agreements underperform in the three years following their announcement. Although the BHAR point estimates in Table 3 indicate greater underperformance than the Fama-French intercepts (Tables 4 and 5) or the CTARs in Table 6, this may simply reflect the different nature of the two calculations. The estimates in Table 3 are designed to measure investor experience associated with holding the event firm s stock, while the results in the other tables measure abnormal returns to a time-series portfolio of firms announcing bank loans. Given the fact that underperformance is implied by both BHAR and calendar time portfolio methodologies, we continue using the BHAR approach to investigate other possible explanations for borrowers underperformance Interactions with Equity Offerings We assess the robustness of our results by controlling for the effects of equity offerings. In particular, if the loan announcement occurs within three years of a previous IPO or SEO, the post loan announcement underperformance may merely reflect the negative long-run returns associated with an equity issuance. Likewise, a subsequent equity issuance may elicit negative returns that we mistakenly 14 Using BHARs simplifies our procedure to control for the effects of seasoned equity offerings in Section 4.3.

19 18 attribute to the loan announcement. In order to distinguish post-loan performance from the effects of equity issuance, we make several adjustments to our methodology. To address post-ipo underperformance concerns, we divide the sample into two groups, according to whether the borrower had been listed on CRSP for at least three years prior to the loan announcement (using CRSP beginning of data dates). We then re-examine BHARs for the two subsamples. Panel A of Table 7 reports the mean and median BHARs for borrowers based on whether a firm has been listed on CRSP for at least three years. Since both the young and old sample firms underperform significantly, IPO effects do not appear to drive our results. Removing the impact of SEOs on our loan announcement sample requires us to identify SEOs that occurred within 3-year windows before and after the loan announcement. We identify SEOs from Securities Data Corporation data and analyze the data in two ways. First, we re-examine the peeradjusted BHARs in Table 3 for announcements that were (not) preceded by an SEO. Second, we truncate our post-loan performance period at the occurrence of a subsequent SEO. Panel B of Table 7 describes the borrowers mean and median BHARs relative to their style-matched peers, from the loan announcement date to the earlier of 36 months later or the date of a subsequent SEO by the borrowing firm. We find that the mean and median peer-adjusted BHARs are statistically indistinguishable between the group with a prior SEO and the group without. Panel C repeats the panel B calculation but assumes a full three-year post-loan holding period, regardless of whether the borrower subsequently announces an SEO. Again, the prior-seo distinction has no significant effect on the mean or median peer-adjusted returns. We conclude that underperformance following loan announcements is not driven by the welldocumented underperformance associated with equity offerings Robustness Across Different Sub-samples An obvious concern with any empirical study is sampling methodology. To evaluate possibly spurious effects of our main sample selection technique, we present results from alternative samples in Table 8. Our conclusions are unaffected by these alternatives.

20 Sample Firms with Multiple Loan Announcements Lyon, Barber and Tsai [1999] raise serious statistical questions about samples in which some firms post-event return calculation windows encompass a repeat event. To investigate the impact of such observations on our results, we examine three sub-samples of the loan announcements. First, we examine firms with only one loan announcement during the loan announcement sample period ( ). For these 569 announcements, Panel A of Table 8 presents the mean and median three-year returns. The mean firm with only one loan announcement underperforms by 22.2% and the median firm underperforms by 19.7%, both significant with 99% confidence. Panel B of Table 8 examines the threeyear peer-adjusted returns following loan announcements by firms that do not announce another loan within three years of the first one and during our sample period. These 867 loan announcers suffer threeyear mean (median) underperformance of 22.0% (22.3%), both significant with 99% confidence. Neither of these calculations addresses the possibility that a loan is announced shortly after the end of our sample window (yearend 1989). We address this concern by focusing on loans prior to 1987, for which we are certain that there was no loan announcement by the same firm within three years of the original event. For these 594 events, the mean (median) three-year peer-adjusted BHAR is 17.4% ( 16.4%), both significant with 99% confidence. We also examine the BHARs of these 594 firms using our three alternative peer groupings, the size only matched firms, industry and size matched firms, and the size and momentum matched firms. The peer-adjusted buy and hold returns (BHARs) continue to indicate significant underperformance at better than the 95% confidence level. (Not reported in a table.) Another way to investigate the impact of repeated events in the sample is to conduct Fama- French tests and calculate CTARs for the sample firms with no subsequent loan announced within three years of the first one. This is the sub-sample employed in Panel B of Table 8 (N=867). The trade-off we face in using this sub-sample as opposed to the one from Panel C is that use of the latter eliminates loan announcements associated with clearly significant long-run underperformance (see Table 3, Panel D). However, use of the former opens up the possibility that some of our loan announcers in 1987 through

21 announce loan agreements again within three years of their last announcement in the 1980s. Given the evidence from Lyon, Barber and Tsai that calendar time portfolio approaches effectively control for overlapping event return periods, we feel more comfortable including events from 1987 through The results from tests mimicking those in Tables 5 and 6, now using the sub-sample of 867 announcements, closely resemble our earlier reported ones. Peer-adjusted Fama-French intercepts are all significantly negative with at least 95% confidence. So too are the CTAR estimates based on the methodology of Vijh. In general, our results seem robust to concerns about overlapping measurement windows Bank versus Nonbank Loans Fama [1985], James [1987], and many other authors claim that bank loans are not equivalent to private loans from nonbank lenders (such as finance companies). However, subsequent researchers have found no significant difference among the announcement effects associated with different kinds of private lenders (Preece and Mullineaux [1994], Billett, Flannery and Garfinkel [1995]). Table 9 compares longrun returns following announcements of loans from different types of lenders: only a bank(s), only a nonbank, some type of mixed lending group, and loans for which the lender was not identified in the news story. The bank loans are followed by significantly negative borrower performance, and the non-banks median return is reliably different from zero. The other two groups exhibit reasonably large negative returns, but the small number of observations apparently makes it impossible to distinguish these returns reliably from zero. Overall, Chi-Square tests fail to reject the hypothesis that all private loans generate similar 3-year BHARs. We conclude that the data are too noisy to reliably indicate lender effects in longrun performance The Clean Announcement Sample In their study of (one day) loan announcement returns, Billett, Flannery and Garfinkel [1995] focus on a sample of 626 clean loan announcements that have returns based on true transaction prices (not

22 21 bid-ask averages), and are uncontaminated by other valuation effects (e.g., merger activity, financial distress, dividends, earnings announcements). While we are less concerned with the effects of such additional information given our focus on long-run performance, it could be argued that the underperformance is an artifact of these events, unless we specifically control for them. Table 10 presents underperformance estimates for Billett, Flannery and Garfinkel s clean loan announcement sample. These results (and their implications) closely resemble those presented in Tables 3, 5, and 6 for the full sample. Panel A indicates that the mean and median BHARs associated with this clean sub-sample are significantly negative, with 99% confidence. Panel B reports the Fama-French intercepts for the clean sample calendar time portfolio regressions. The only substantial change from Table 5, panel A is a loss of significance in the intercept from the WLS equal-weighted regression (t = 1.56). We argue that this is of little concern for two reasons. First, the clean sample is not necessarily most appropriate for longer-run studies, in which the choice of a peer is designed to control for other effects. Second, as noted next, our other calendar time portfolio approach generates significant results across all weighting schemes. Panel C of Table 10 also supports our hypothesis that calendar time effects do not lead to improper conclusions about long-run underperformance of loan announcers. CTARs for the clean sample using the style matched peers are significant with 99% confidence in all cases except one (in which case the t-statistic indicates significance with 95% confidence). In summary, our results appear to be invariant to sample selection issues. Given prior criticisms of long-run return studies we view this robustness as critical to our conclusions regarding financing related underperformance in general and lack of bank loan uniqueness in particular Cross-Sectional Effects For the full sample of 1,385 loan announcements, we investigate whether long-run peer-adjusted performance is correlated with ex ante firm loan characteristics, such as the relative loan size, whether the loan is new or a renewal, the lender s credit quality, borrower systematic and total risk, borrower leverage and market-to-book (assets) ratio. Cross-sectional regression models reveal no significant correlations.