How Large Are the Coinsurance Benefits of Mergers?

Transcription

1 How Large Are the Coinsurance Benefits of Mergers? SAKYA SARKAR * November 2014 University of Southern California ABSTRACT Using a structural model, I estimate the value gain from coinsurance for a sample of mergers. For most mergers, the estimated gains from coinsurance are small smaller than the counterfactual gains if firms were to merge randomly suggesting that coinsurance is not the primary motivation for mergers. Coinsurance from diversifying mergers is also small, comparable to related mergers. In the sample of 1,884 mergers, only 174 produces coinsurance exceeding 1% of combined pre-merger target and acquirer firm value: these are mergers between small firms, or mergers between firms of similar size, or when highly levered targets merge with healthy acquirers. The cumulative abnormal return around merger announcement increases 0.89% for every 1% estimated value gain from coinsurance, suggesting that stockholders benefit from coinsurance. * University of Southern California, Marshall School of Business, 3670 Trousdale Parkway, Los Angeles, CA, Website: Comments are welcome, please sakyasar@usc.edu. I am grateful to my Dissertation chair, John Matsusaka, for his valuable guidance and constant encouragement. I thank Kenneth Ahern, Daniel Carvalho, Lee Cerling, Tom Chang, Cary Frydman, Gerard Hoberg, Derek Horstmeyer, Chris Jones, Scott Joslin, Maria Ogneva, Oguzhan Ozbas, Garrett Swanburg, Yongxiang Wang, the seminar participants at USC FBE Brownbag, and the USC Corporate Finance Study group for their valuable suggestions. I acknowledge financial support from USC. 1

2 Coinsurance is the idea that when two firms with imperfectly correlated cash flows merge, it reduces their probability of default, because when a negative shock hits one division of the merged firm, the other division rescues it by transferring cash (Lewellen (1971)). In theory this reduction in probability of default increases firm value through reduced deadweight costs of financial distress and increased tax benefits of debt. Coinsurance is regarded as an important motive for mergers, particularly diversifying mergers, as evidenced by its extensive coverage in academic papers, textbooks (Ross, Westerfield and Jaffe (2008), Brealey, Myers, and Allen (2006)), and review papers (Khanna and Yafeh (2007), Maksimovic and Phillips (2013)). 1 Other than mergers, coinsurance has been advanced as a rationale for business group affiliation (Khanna and Yafeh (2005), Fisman and Wang (2010), Jia, Shi and Wang (2013) and Luciano and Nicadano (2014)). Although coinsurance is as an important theory, no previous study has directly estimated the value gain from coinsurance in mergers. Previous studies have indirectly estimated coinsurance, using cash flow correlation as a proxy (Duchin (2010), Hann, Ogneva and Ozbas (2013)). But coinsurance, theory tells us, also depends nonlinearly on factors such as leverage, size, and volatility (Leland (2007)). Moreover, low cash flow correlation can increase firm value through channels other than avoidance of default by avoiding costly external financing of investment (Froot, Scharfstein and Stein (1993)), or through winner-picking (Stein (1997)). In contrast to these previous studies that rely solely on cash flow correlation, this paper uses a structural model that takes into account not only correlation, but also leverage, size, 1 Penas and Unal (2004), Billet, King and Mauer (2004), Leland (2007), Kuppuswamy and Vilalonga (2010), Banal- Estañol, Ottaviani and Winton (2013), and Hann, Ogneva and Ozbas (2013), Ahern (2012), Devos, Kadapakkam and Krishnamurthy (2009), Duchin (2010), Fulghieri and Sevilir (2011), Saretto and Tookes (2013). 2

3 and volatility to estimate coinsurance benefits. The structural model is a trade-off model of capital structure (Leland (1994)), customized to account for correlated cash flows. The model values two firms, when they are separate and when they merge, and the difference in firm values estimates coinsurance benefits in the form of lower expected cost of distress and higher tax benefits of debt. The parameters of the model are estimated using pre-merger stock price and accounting information, and post-merger firm value is simulated. Because the structural model uses only pre-merger information to simulate firm value, the coinsurance benefits are identified: their estimates are not contaminated with operational synergies, as would be the case if a part of the actual value change from the merger were to be ascribed to coinsurance. 2 Using the structural model, coinsurance is estimated for a sample of 1,884 mergers in the United States between 1981 and The main finding of the paper is that coinsurance benefits, for both related and diversifying merger, are small. The mean coinsurance benefit is 0.32% of the target and acquirer s combined pre-merger value; only 174 of the 1,884 mergers produced coinsurance benefits exceeding 1%; coinsurance, though small, is always positive. As a point of comparison, mean coinsurance benefits are about one-fifth the size of abnormal announcement returns. The estimated coinsurance benefits remain small, even with high deadweight cost parameters. Although coinsurance from most mergers are small, it could still be the case that coinsurance from diversifying mergers are large. Given that coinsurance has been traditionally 2 The value gain (loss) from the merger may come from various other sources: operational synergies, asset complementarities, agency problems. Without a model, it is difficult to decouple the value gain and attribute a part to coinsurance. Moreover, if coinsurance reduces default risk, then firms should increase their leverage after the merger, again increasing the default risk, rendering any comparison of default risks before and after the merger unfruitful. The structural model bypasses these problems by using only pre-merger information. 3

4 advocated as a rationale for corporate diversification (Lewellen (1971), Hahn, Ogneva and Ozbas (2013), Duchin (2010)). But, even from diversifying mergers, estimated coinsurance benefits are small, comparable to coinsurance from related mergers. In light of this finding, it seems unlikely that diversifying mergers are pursued for coinsurance. Though the structural model estimates coinsurance benefits to be small, it does not immediately follow that coinsurance is unimportant. It is possible that the structural model underestimates coinsurance benefits. 3 In particular, coinsurance from mergers could still be large relative to the counterfactual coinsurance if firms were to merge randomly. Hence, as a relative benchmark, I randomly draw firms from COMPUSTAT and pair them, drawing 1,884 random pairs over the same time period as the mergers sample. The coinsurance benefits from mergers are smaller than the counterfactual coinsurance benefits from these random pairings. Reinforcing this result, a logistic regression of merger incidence on coinsurance produces a negative slope: coinsurance does not make firms more likely to merge questioning its position as an influential theory of mergers. While coinsurance benefits from most mergers are small, for a minority of mergers they are substantial: 177 of the 1,884 mergers produce coinsurance benefits exceeding 1%. To 3 In order to assess whether the structural model used in this paper is reliable, I take the following steps. First, in Appendix C, I evaluate the performance of the structural model in predicting defaults out-of-sample. The model predicts default well, even after controlling financial indicators like size, leverage, volatility or correlation, demonstrating that the model captures important non-linearities that predict default, but would be missed by a reduced form approach. Second, as an alternative modeling choice, I estimate coinsurance benefits using Merton s (1974) model (unreported). Finally, I adopt an alternative accounting approach, using Altman Z -scores to estimate coinsurance benefits (Altman (1997)), which is presented in Appendix D. To the extent that these approaches differ in their assumptions, the estimated coinsurance benefits are different. Yet despite their differences, these approaches agree so far as the main findings about coinsurance are concerned. 4

5 identify these high-coinsurance mergers, I construct a coinsurance index of accounting variables that can be easily computed without having to simulate a structural model. This index is constructed by selecting the variables using LASSO, a popular statistical method for parsimonious model selection (Tibshirani (1996)). Its correlation with estimated coinsurance benefits is 67%, compared to 2% for cash flow correlation the traditional proxy for coinsurance. The coinsurance index loads positively on relative size: coinsurance is large when small targets are acquired. It also loads positively on target leverage (squared): coinsurance is large when a highly levered target, which would otherwise head towards distress, is acquired. The structural model estimates the total value gain from a merger. But, how are coinsurance benefits apportioned between the bondholders and equity holders? While significant empirical evidence suggests that bondholders benefit from coinsurance (Billet, King and Mauer (2004), Penas and Unal (2004)), there is little evidence that equity holders benefit from coinsurance. Instead, the literature asserts that equity holders suffer from coinsurance, because the reduction in asset volatility due to coinsurance transfers value from equity holders to bondholders (Galai and Masulis (1975), Higgins and Schall (1976), Mansi and Reeb (2002)). So ingrained is the notion that equity holders suffer from coinsurance, that MBA textbooks routinely label coinsurance as a deleterious reason to merge: This mutual guarantee, which is called the coinsurance effect, makes the debt less risky, and more valuable than before. There is no net benefit to the firm as a whole. The bondholders gain the coinsurance effect, and the stockholders lose the coinsurance effect. (Ross, Westerfield, and Jaffe (2008), Edition 8, page 825) 5

6 In contrast to this view, I find that, using the direct estimates of coinsurance benefits from the structural model, the equity holders benefit from coinsurance. For every 1% estimated coinsurance benefit, the cumulative abnormal announcement return increases by 0.89%. These regression results are statistically significant at the 1% level after including standard controls. The results are also robust to portioning the sample in various dimensions. This result suggests that when coinsurance benefits are large it can be a valid motive for mergers. My paper is closely related to Leland (2007), who also models the financial synergies from mergers, but using a two-period tradeoff cost framework. A theory paper, Leland calibrates the financial synergies for some typical values of parameters; unlike this paper, he does not estimate financial synergies for any actual sample of mergers. Leland calibrates financial synergies to be small for reasonable parameter values. In this, my paper agrees with Leland (2007), but mine goes a step further: it documents that coinsurance from mergers is smaller than coinsurance from random pairings. Another conclusion from Leland (2007) is that, contrary to Lewellen (1971), financial synergies can be negative for some parameter realizations: this he advances as a rationale for spinoffs. In contrast, I find that, empirically, coinsurance benefits are always positive even though there is nothing in my model that mechanically restricts them to be positive. How does this paper reconcile with the several previous papers that document the importance of coinsurance ((Billet, King and Mauer (2004), Penas and Unal (2004) Duchin (2010), Hann, Ogneva and Ozbas (2013))? In this paper, following the theoretical work on coinsurance (Lewellen (1971), Leland (2007), and Banal-Estañol, Ottaviani and Winton (2013)), the benefits from coinsurance stems solely from reduction of default probability and the 6

7 consequent avoidance of deadweight costs and loss of tax shield associated with default. But, even though avoidance of deadweight costs of distress is perceived as the primary benefit from coinsurance, as Hann, Ogneva and Ozbas (2013) argue, coinsurance benefits may also stem from adverse selection and transaction costs of external finance and resulting investment distortions, forgone business opportunities due to defections by important stakeholders such as suppliers, customers, or employees, and so on. What my paper does then is to show that, if at all coinsurance benefits are substantial, as is argued by the previous literature, then, perhaps, its benefits stem not from default avoidance, but from those secondary channels. The rest of the paper proceeds as follows: Section I presents a structural model of coinsurance; Section II describes the data; Section III discusses the first part of the results: Estimated Coinsurance Benefits; Section IV narrates the second part of the results: The Stock Price Response to Coinsurance; Section V concludes. I. A Structural Model for Estimating Coinsurance Benefits Consider two firms, labelled 1 and 2, that merge at time tt. The value of firm 1, the acquirer, is VV 1 (tt); the value of firm 2, the target, is VV 2 (tt); and the value of the merged firm is VV 12 (tt). The value gain from the merger is then given by the accounting identity: = VV 12 (tt) VV 1 (tt) VV 2 (tt). (1) In order to estimate the value gain,, we need a model that values the pre-merger standalone firms, VV 1 (tt), VV 2 (tt), as well as the merged firm, VV 12 (tt). To value the pre-merger firms, I use a simple yet widely cited trade-off model: the Leland (1994) model. But Leland s model, in itself, cannot value the merged firm. So I extend Leland s model imposing more structure and customizing it to value the combined firm when two firms with correlated cash flows merge. 7

8 A. Valuing the Standalone (Pre-Merger) firms Following Leland (1994), each pre-merger firm ii has assets in place, the value of which is XX ii (tt); the asset value is unaffected by any financial decisions. The asset value of each firm evolves following two correlated geometric Brownian motions: ddxx ii (tt) XX ii (tt) = (μμ ii δδ)dddd + σσ ii ddww ii(tt), ii = 1,2 σσ ii > 0, δδ 0. (2) The correlation coefficient of the two processes is constant: ρρ [ 1,1]. Also constant are the expected growth rate of assets, μμ ii, its volatility, σσ ii, and the payout rate to equity, δδ. The firm issues debt: a claim that pays a constant coupon, CC ii, per instant of time, as long as the firm is solvent. The firm is solvent until the value of its assets falls below a default boundary, KK ii. Denote the time at which the firm ii becomes insolvent by TT ii = MMMMMM{tt 0; XX ii (tt) = KK ii }. When the firm becomes insolvent, it defaults on the debt: on default, a fixed proportion, αα, of the value of assets is lost due to the deadweight costs; the debt holders get the remaining (1 αα)kk ii. This is the cost of debt. 4 The benefit to using debt stems from its tax deductibility. A tax benefit of ττcc ii accrue to the firm per instant of time, ττ being the tax rate. The tax benefit is lost when the firm defaults. The net benefit to leverage is thus the tax benefit of debt minus the expected cost of distress. The levered firm s value, VV(XX ii (tt)), is the sum of all future expected cash flows, discounted at the risk free rate, rr, with expectations computed under the risk neutral probability measure. 5 4 The concept of default in Leland (1994) differs somewhat from Lewellen (1971). While Leland s model is in continuous time, Lewellen s is a one period model. In Leland s model, default occurs if the market value of assets falls below the default barrier, whereas, in Lewellen s model, default occurs if cash flow falls short of the current portion of debt payable. 5 Following established notation, the superscript denotes expectations computed under the risk neutral measure. 8

9 TT ii ττ=tt VV(XX ii (tt)) = XX ii (tt) + EE ττττ ii ee rrrr dddd EE [ee rrtt ii ααkk ii ]. (3) Equation (3) demonstrates that the levered firm s value, VV(XX ii (tt)), is the sum of three components: the value of the firm s assets, XX ii (tt), which is unaffected by the leverage; the tax shield, EE TT ii ττ=tt ττττ ii ee rrrr dddd, which is the expected value of all future tax deductions, discounted to the current time; and, the cost of debt, EE [ee rrtt ii ααkk ii ], which is the expected deadweight loss on default, discounted to the current time. Computing these expectations, the value of the firm simplifies: VV(XX ii (tt)) = XX ii (tt) + ττ CC ii 1 rr XX ii(tt) γγ ii ααkkii XX ii(tt) γγ ii, KK (4) ii where, γγ ii = rr δδ 1 σσ 2 ii 2 + rr δδ 1 σσ 2 ii σσ 2 2 ii rr 2 /σσii. KK ii 1 The default barrier, KK ii, is endogenous: equity holders choose when to default, so that equity value is maximized. The endogenous default barrier is given by KK ii = (1 ττ)cc ii. This value rr+0.5σσ 2 ii of KK ii can be substituted in equation (4), to obtain the firm value. Equation (4) presents the value of the firm as a function of the coupon rate CC ii, where CC ii depends on the firms leverage. If we further assume that firms choose their leverage optimally, then we can choose the value of CC ii that maximizes firm value. Empirically, though, the actual leverage for most firms depart significantly from the optimal leverage predicted by Leland s model, or for that matter the optimal leverage from most tradeoff models (Fama and French (2002), Leary and Roberts (2005)). As an extreme case, in my sample of 1,884 mergers, as many as 242 targets and 118 acquirers have zero debt significantly different from the optimal leverage predicted by any tradeoff model (Strebulaev and Yang (2013)). Moreover, observed 9

10 leverage that is very different from the optimal leverage predicted by the model tends to be stable over time, often stable over twenty years (Lemmon, Roberts and Zender (2006)). Hence, instead of assuming that the firms are optimally levered, I treat the debt as exogenous, obtaining the actual debt before the merger from COMPUSTAT. The product of the observed debt and the risk free rate gives the coupon rate CC ii (Elkamhi, Ericsson and Parsons (2012)). Now, to value the pre-merger firm using equation (4), we only need the structural parameters: σσ 1,σσ 2, and ρρ. These structural parameters are estimated from the past stock price information, following the procedure described in sub-section D. C. Valuing the Merged firm To value the merged firm, I extend the Leland (1994) model in the following ways. First, I assume that, when the two firms merge, the value of assets of the merged firm, XX 12 (tt), is the sum of the pre-merger asset values: XX 12 (tt) = XX 1 (tt) + XX 2 (tt). This additively abstracts away all operational synergies (Lewellen (1971), Leland (2007)). 6 Applying Ito s Lemma, the dynamics of the merged firm s assets, XX 12 (tt), is given by ddxx 12 (tt) where, ss(tt) = XX 12 (tt) = [(μμ 1 δδ)ss(tt) + (μμ 2 δδ) 1 ss(tt) ]dddd + σσ 1 ddww 1(tt) + σσ 2 ddww 2 (tt), (5) XX 1 (tt) XX 1 (tt)+xx 2 (tt) is itself stochastic. Equation (5) demonstrates that, unlike the premerger standalone firms, the dynamics of asset value is not a geometric Brownian motion, and 6 Are the coinsurance benefits correlated with operational synergies? When operational synergies are high, that is, XX 12 (tt) > XX 1 (tt) + XX 2 (tt), then default risk is lowered not because of cash flow pooling but because combined cash flows are greater and coinsurance is consequently higher. Abstracting away operational synergies can thus underestimate coinsurance. However, simulations show that the effect is second order. For reasonable operational synergies, coinsurance estimates don t shift much. 10

11 thus obtaining a closed form solution for firm value, as in the Leland (1994) model, is no longer feasible. 7 Second, following Lewellen (1971), I assume that the debt of the merged firm is given by the sum of the pre-merger debts. In the context of the model, this means that coupon rates are additive: the merged firm s coupon rate, CC 12 = CC 1 + CC 2. An alternative to the additivity assumption is to compute the optimal leverage for the merged firm: trade off the cost of debt against its tax benefit (Leland (2007)). But in subsection A, I obtained the pre-merger leverage from COMPUSTAT, instead of assuming that the pre-merger firm is optimally levered. So, assuming the post-merger firm to be optimally levered, while the pre-merger firms are far away from their optimal leverage, will be comparing apples with oranges. 8 Third, I assume that the post-merger default boundary, KK 12, is sum of the pre-merger default boundaries: KK 12 = KK 1 + KK 2. Unlike pre-merger firms, the equity holders are not choosing the default boundary optimally; instead, they face a default barrier before the merger. This assumption is because the pre-merger bonds may have positive cash flow covenants, which may be difficult to supersede after the merger. This assumption, again, makes the estimated coinsurance benefit a conservative estimate. In robustness checks, I relax this assumption, and estimate coinsurance benefits for the endogenous default case; those estimates are not materially different. 7 The reader may ask: If the pre-merger firm s which itself may have many divisions and engage in diverse activities asset value be approximated by a geometric Brownian motion, why not for the post-merger firm? Because, mathematically, the sum of two geometric Brownian motions is not a geometric Brownian motion. 8 In theory firms should increase their leverage following the merger, taking advantage of the cheaper cost of debt due to coinsurance. Assuming post-merger leverage to be the sum of pre-merger leverages, in general, makes the estimated coinsurance benefit a conservative estimate of the true coinsurance. 11

12 Finally, I assume that the proportion of firm value lost in distress, αα; the tax rate, ττ; and, the payout rate, δδ; are the same for both firms, and they are unchanged by the merger. 9 The value of the merged firm, VV(XX 12 (tt)), is then the sum of all expected future cash flows, discounted to the present at the risk free rate, with expectations computed under the risk neutral measure. VV(XX 12 (tt)) = XX 12 (tt) + EE TT 12 ττ=tt where TT 12 is time to default for the merged firm. ττττ 12 ee rrrr dddd EE [ee rrtt ii ααkk 12 ]. (6) Once the values of the firms before and after the merger are estimated, I estimate the coinsurance benefits,, by inserting these values from equations (3) and (6) into equation (1): The term, ( ττττ 1 rr = ( ττττ 1 rr ααkk 1)EE (ee rrtt 1 ee rrtt 12) + ( ττττ 2 rr ααkk 2)EE (ee rrtt 1 ee rrtt 12). (7) ααkk 1)EE (ee rrtt 1 ee rrtt 12), represents the change in net benefit from leverage due to the merger for the first firm. The second term has a similar interpretation. D. Parameter Estimation Estimating the coinsurance benefits through equation (7) requires the structural parameters: the volatilities, σσ 1, σσ 2, and the correlation between the two Brownian motions, ρρ. To estimate the volatility, I exploit the closed-form expression of equity price for the premerger firms. From Leland (1994), the equity price, EE ii (tt), for the pre-merger firms is given by 9 In general, the cost of distress, the tax rate, or the payout rate is neither homogenous, nor unchanged by the merger. Glover (2013) demonstrates that there is significant cross-sectional variation in cost of distress. Zhu and Singhal (2011) document that the proportion of value lost in bankruptcy is higher for diversified firms. In reality, the tax rates too differ across firms. While it is straight-forward to obtain pre-merger tax rates, it is less clear what the tax rate will be for the merged firm. Similarly, there is significant cross-sectional variation in payout rates. But, when two firms with different payout rates merge, it is not clear ex-ante what the post-merger payout rate will be. 12

13 EE ii (tt) = XX ii (tt) + (1 ττ) CC ii rr (1 ττ) CC ii rr KK ii XX ii(tt) KK ii γγ ii, (8) where γγ ii = rr δδ 1 σσ 2 ii 2 + rr δδ 1 σσ 2 ii σσ 2 2 ii rr 2 /σσii, and, KK ii = (1 ττ) rr+ 1 2 σσ 2. ii In equation (8), the risk free rate, rr, is known; and, the stock price, EE ii (tt), can be easily obtained, at least for most firms with traded common stock. Following Leland (1994), the tax rate, ττ, is assumed to be 35%, and the payout rate, δδ, is assumed to be 1%. The only unknowns in this equation are then the asset value, XX ii (tt); and, the volatility of asset returns, σσ ii. As an initial estimate of the volatility, σσ ii, I use the standard deviation of the daily stock returns, computed over a 252 day window that ends 42 days before the merger announcement. The data ends 42 days before the announcement because previous literature has found a runup in target prices prior to a merger announcement, starting roughly 42 days before merger announcement (Schwert (1996)).The initial estimate of volatility, σσ ii 1, is inserted into equation (8), (the superscript 1 denoting that this is the first iteration). Then, equation (8) is solved numerically to estimate the asset value XX ii 1 (tt), for each of these 252 days. From these asset values, XX ii 1 (tt), asset returns are computed: RR ii 1 (tt) = 1 XX ii 1 (tt) XX ii 1 (tt 1), 1. The standard deviation of these asset returns over the 252 day window gives an updated estimate for the volatility, σσ ii 2. The process is repeated till the absolute difference between the estimates of volatility, from two successive iterations, is less than That is σσ ii NN σσ ii NN 1 < Usually, convergence occurs within five iterations. When convergence occurs, the estimate from the last iteration, σσ ii NN, is taken as the final estimate for volatility: σσ ii = σσ ii NN. In the course of estimating the volatility, I also estimated a series of asset values, XX ii NN (tt). These estimated asset values, from the final iteration step, are used to generate the asset CC ii 13

14 returns for each firm. The sample correlation between the asset returns of firm 1 and 2 gives the estimate of the correlation coefficient between the two Brownian motions, ρρ. Once the structural parameters are estimated, the coinsurance benefits are estimated using equation (7), by evaluating the expectations through Monte-Carlo simulations. The simulation procedure is described in Appendix B. II. Data A. Sample Construction I compute coinsurance benefits for two different samples. One comprises mergers that took place in United States between 1981 and The other is a sample of randomly paired firms, also corresponding to the same period. The sample of mergers is collected from the Securities Data Corporation s (SDC) U.S. Mergers and Acquisitions Database. The sample consists of mergers in the United States provided it satisfied the following criteria: (1) the announcement date was between January 1, 1981 and December 31, 2013; (2) deal size was above $10 million (2013 dollars); (3) the transaction was completed; (4) the acquirer did not have more than a 5% stake in the target before the merger; (5) after the merger, the acquirer owns more than 99% of the target; (6) at least one of the target and the acquirer have some debt outstanding both are not all equity firms (7) neither the acquirer nor the target is a financial firm, as indicated by their primary SIC code (8) the target and the acquirer are both public firms, with daily stock market return data available on Center for Research in Security Prices (CRSP) and fundamentals data available from Compustat North America. 14

15 The sample contains 1,884 mergers. The sample size of 1,884 mergers, between 1981 and 2013, looks reasonable. In comparison, Akbulut and Matsusaka (2010) study 4,764 mergers from 1950 to 2006, including mergers between financial firms. The sample of randomly paired firms is constructed as following. First, I take all firm years in the Fundamentals Annual file from the Compustat North America Database, covering the same time period as the merger sample: 1981 to For comparability with the merger sample, I retain only those observations with market capitalization above $10 million (2013 Dollars) and drop all financial firms, as indicated by their primary SIC codes. Let us call this the population of comparable firms. From this population, I randomly draw 1,884 firm years without replacement. This constitutes the sample of hypothetical acquirers. For each observation, the year gives the year of the hypothetical merger. Next, for each hypothetical acquirer, I randomly draw without replacement one firm from the population of comparable firms, such that the firm belongs to the same year cohort as the acquirer. This firm is the hypothetical target. Finally, from that year, I draw a random date the hypothetical announcement date. This completes the sample of randomly paired firms. For both these samples the mergers as well as the randomly paired firms I collect data on stock price, return, and shares outstanding for both target and acquirer from the CRSP Daily Stock File. Data are collected for each day, over a 252 day window, ending 42 days prior to the merger announcement. The data ends 42 days before the announcement because the previous literature has found a run-up in target prices prior to a merger announcement, starting roughly 42 days before the announcement date (Schwert (1996)). When data is not available for all the 15

16 252 days, I retain only those firms with at least 90 days of data. I also collect accounting data on short term and long term debt from the COMPUSTAT North America Database. The fiscal year for these data corresponds to a year before the merger announcement. B. Summary Statistics I compute the size, leverage and cumulative abnormal return on merger announcement, for both the samples. Detailed variable descriptions are available in Appendix A. For both the target and acquirer, size is computed as the market capitalization 42 days before merger announcement. Market leverage is computed as the ratio of book value of debt to the sum of book value of debt and market capitalization. Cumulative abnormal returns (CAR) are computed using the Fama-French three-factor model, over a three day (-1, 1) window around the announcement date. The betas for the model are computed using daily returns, over a 252 day window, which ends 42 days before merger announcement. Data on Fama-French factors are collected from Kenneth French s website. 10 The CARs of the target and acquirer are weighted by the respective pre-merger market capitalizations, and their average is taken. This gives the combined CAR. Table I presents the summary statistics. Panel A pertains to the mergers; Panel B to the randomly paired firms. For the sample of mergers, the median size of the acquirers is $ 1.2 billion, whereas the median target size is only $100 million. So the median acquirer is 12 times larger than the median target. By contrast, for the random pairings, the acquirer and target are similar in size: the median acquirer is worth $160 million only, while the median target is worth $140 million

17 For the merger sample, the median target leverage is 21.5%, whereas the median acquirer leverage is 17.2%. The targets thus have slightly more debt than acquirers. For the randomly paired firms, the median target leverage is 31.5%, and that for acquirers is 32.4%. Thus, both the target and the acquirer for the random pairings have significantly more debt than merger participants. For the merger sample, the cumulative abnormal return ranges from -38.5% to 57.8%; the mean is 1.5%. There is thus a lot of dispersion. In contrast, the CAR s are closer to zero for the random pairings: the mean is -0.0%, and the median is -0.1%. This reflects that the CAR s are purely noise for the random pairings, unlike in mergers where the CARs correspond to the announcement of the merger. The summary statistics for my sample of mergers are comparable to the previous literature (for e.g. Akbulut and Matsusaka (2010), Andrade, Mitchell and Stafford (2001)). C. Inputs to the Structural Model A crucial input to the structural model is the correlation of asset returns, ρρ. Row 1 of the table (Panel A) shows that the correlation ranges from to 0.99, with mean 0.23, and median A second input is the volatility of asset return, σσ ii. As rows 2 and 3 show, the targets (median 47.0%) are generally more volatile than acquirers (median 34.7%). This is expected, given that targets are usually smaller. Some targets, and to a lesser extent some acquirers, are very volatile. Third, the coupon rates, CC ii, which depend on the leverage, are presented in rows 4 and XX ii (0) 5. The median annualized coupon rate is 1.1% of the firm value for the targets, and 0.8% for the 17

18 DD acquirers. Finally, the default barrier, ii, is presented in rows 6 and 7. The mean default XX ii (0) barrier is 6.2% of the firm value for the targets, and 5.9% for the acquirers. In comparison, as panel B demonstrates, the input parameters for the randomly paired firms are quite different. They are more volatile: the median for acquirers is 35.8%, and that for targets is 35.9%. They also pay higher coupon: the median is 1.6% for acquirers, and 1.5% for targets. This is because the randomly paired firms are smaller and more levered. III. Estimated Coinsurance Benefits A. Coinsurance Benefits for the Sample of Mergers This sub-section describes the distribution of the coinsurance benefits for the sample of mergers. Table III presents the results. Row 1 presents the results for the base case, which assumes the proportion of firm value lost in distress, αα, to be 16.5% the midpoint of Andrade and Kaplan s (1998) estimates. The mean gain from coinsurance is 0.32% of the combined premerger firm value. The median is 0.09%. The 90 th percentile of coinsurance benefits is 0.94%, while the 99 th percentile is 2.83%. The 1 st percentile is 0, and the minimum value is also 0: coinsurance benefits thus turn out to be always positive. 11 For most mergers, these numbers indicate, the coinsurance benefits are small. To better understand the economic magnitude of the coinsurance benefits, we can benchmark it against the announcement period cumulative abnormal return (CAR), which is the markets perception 11 Lewellen (1971) argues coinsurance is always positive. However, Leland (2007) demonstrates theoretically that financial synergies can be negative, if we relax the assumption that cash flows are always positive. Theoretically, in this paper, the coinsurance benefits may be negative, depending on parameters values, yet, empirically, the estimated coinsurance benefits turn out to be positive. 18

19 of the total equity value created through the merger. Row 5 presents the CAR (-1, 1) for the sample of mergers. The mean CAR is 1.25%; the median is 0.69%; the 75 th percentile is 4.17%; and, the 25 th percentile is -1.87%. In comparison to the CARs, the coinsurance benefits appear moderate. For a minority of mergers, particularly those in the top decile, coinsurance benefits appear substantial: Corresponding to the 90 th percentile of coinsurance benefits, a merger creates $94 million through coinsurance, assuming a pre-merger combined value of $10 billion. While $94 million sounds significant economically, compare it to the fees paid to investment banks for M&A advisory, which is on average 1.22% of the transaction value (Hunter and Jagtiani (2003)). Even at the 90 th percentile, the coinsurance benefits are not enough to offset the mean fees. One reason why the estimated coinsurance benefits are small may be because I did not consider fixed cost of distress. So, next, I consider a case where there are also fixed costs incurred at default. Following Elkamhi, Ericsson and Parsons (2012), I assume fixed costs φφ = 1.32 million, over and above the usual 16.5% proportional costs. Row 2 presents the results for fixed costs φφ = 1.32 million. The mean coinsurance benefit is 0.35%, slightly higher than the case with only proportional costs, but still quite small. Likewise, the median increases to 0.10% from 0.09%. The 90 th percentile is 1.01%, against 0.94% previously. Introducing fixed costs increases the estimated coinsurance benefits marginally. Perhaps, estimated benefits are small because I chose a small proportional cost parameter, αα. So instead of the midpoint of Andrade and Kaplan s (1998) estimates, I choose their upper limit: αα = 23%, while retaining the 1.32 million fixed cost of distress. Row 3 demonstrates that the mean coinsurance benefit is 0.36% and the median is 0.10%. The coinsurance benefits 19

20 increase marginally. Further, following Glover (2013), I choose αα = 45%. Row 4 shows that the mean is 0.40% and the median is 0.11%. Again, the increase in coinsurance benefits is marginal. To sum up, Table III illustrates that even if firms lose almost half their value to deadweight costs in default, the avoidance of such hefty costs through mergers create but small value gains at least for most mergers. The coinsurance benefits for most mergers are small, regardless of the parameterization of the cost of distress. B. Counterfactual Coinsurance Benefits for the Randomly Paired Firms Table III suggests that coinsurance benefits are economically small. Relatively, though, the coinsurance benefits from mergers may still be large, relative to the coinsurance that would be produced if firms were to merge randomly. So as a benchmark, I estimate the counterfactual coinsurance benefits for the sample of randomly paired firms. There is another rationale for randomly pairing firms. Lewellen (1971) and the subsequent literature on coinsurance suggest that firms merge so that they can avoid distress through coinsurance. If coinsurance is the main motivation for firms to merge, then we should expect coinsurance to predict merger incidence. Ceteris paribus, mergers that can produce greater coinsurance benefits should be more likely to take place than mergers that can produce lesser coinsurance; conversely, the coinsurance benefits from mergers should be greater than the counterfactual coinsurance benefits for randomly paired firms. This approach of randomly pairing firms follows Hoberg and Phillips (2010) Hoberg and Phillips (2010) find that when firms are randomly paired asset complementarities from such random pairings are less than that from mergers. They conclude that high asset complementarities motivate mergers. 20

21 Table IV presents the counterfactual coinsurance benefits for the sample of randomly paired firms. The first row presents the results for the base case: the proportion of firm value lost in distress, αα = 16.5%, and there are no fixed costs. The mean estimated counterfactual value gain from coinsurance is 0.70%, more than twice that for the sample of mergers (0.32%). The median is 0.23%, again more than twice that for the sample of mergers (0.09%). Similarly, the 90 th percentile is 1.89%, compared to 0.94% for the mergers; the 99 th percentile is 6.48%. To determine if the coinsurance benefits from the two samples are significantly different, I perform a two-sample t-test of means. The t-statistic for the difference in means is , rejecting the null hypothesis of equality of means at the 1% level of statistical significance. The t-test says only compares the means. To instead compare the two distributions, I perform the Kolmogorov-Smirnov (KS) test. The null hypothesis for the KS Test is that the coinsurance benefits for both the samples are drawn from the same underlying distribution. The KS test produces a statistic of 0.077, which corresponds to a P-value of less than The null hypothesis of same distribution is thus rejected at the 1% level of statistical significance. Next I perform a logistic regression for the combined sample of 1,884 mergers and 1,884 randomly paired firms. The dependent variable is a dummy for merger incidence: the dummy is one for a merger and zero otherwise. The main explanatory variable is the estimated coinsurance benefit. I use the following controls: target size, relative size, target leverage, acquirer leverage, acquirer market-to-book, target market-to-book, correlation of stock return, and a dummy that is one when the three digit primary SIC codes are different. The results are presented in Table V. 21

22 Column (1) presets the results for the univariate regression. The coefficient is -0.52, statistically different from zero at the 1% level. Column (2) presets the results for the multivariate regression. The coefficient is -0.43, different from zero statistically at the 1% level. Likelihood of two firms merging does not increase, but decreases, with the potential for coinsurance from that merger. These results suggest that most mergers are not motivated by the desire to reduce distress risk through coinsurance. But why do randomly pairing firms produce greater coinsurance than mergers? Perhaps, the answer lies in Table I, where the characteristics of both populations of firms are compared. Randomly paired firms are usually smaller than merger participants smaller than the typical target, and much smaller than the typical acquirer even though only firms above $10 million are sampled; hence they have much higher scope for risk reduction. Moreover, randomly paired firms have higher debt than merger participants and are thus at higher risk of default, potentially benefitting more from coinsurance. Finally, most mergers are within the same industry, whereas randomly paired firms are more likely to be diversifying, with greater scope of coinsurance. C. Coinsurance and Corporate Diversification Although coinsurance benefits for most mergers are small, as many as 174 mergers produced coinsurance benefits exceeding 1%. Are these mergers that produce (relatively) higher coinsurance benefits diversifying mergers? It would appear so, given that, the literature generally associates coinsurance with diversification. Since Lewellen (1971) proposed coinsurance to explain diversifying mergers, several papers have suggested that coinsurance is a major benefit from diversification. Hahn, Ogneva and Ozbas (2013) report that coinsurance 22

23 reduces cost of capital for diversified firms. Duchin (2010) finds that diversified firms hold less cash; he argues coinsurance enables diversified firms to manage with less precautionary cash holdings. In contrast, the literature is largely silent about coinsurance in the context of related mergers. There is, however, no strong theoretical reason except that related mergers have more correlated cash flows for limiting coinsurance to diversifying mergers. In view of this literature, I test whether coinsurance benefits are larger for diversifying mergers. I classify a merger as diversifying when none of the divisions of target or acquirer have any three digit SIC code in common. 13 Otherwise, it is classified as a related merger (Kaplan and Weisbach (1992), and Akbulut and Matsusaka (2010)). Of the 1,884 mergers, 265 are classified as diversifying, and the remaining 1,669 as related. In Table V, I present the coinsurance benefits for the related and diversifying mergers, assuming the proportion of firm value lost in distress, αα = 16.5%, and no fixed costs. The mean coinsurance benefit is 0.33% for both diversifying mergers and related mergers. Also similar are the medians: 0.11% for diversifying and 0.09% for related mergers. The 75 th percentile is 0.36% for diversifying mergers, and 0.37% for related mergers. The 90 th percentile for diversifying mergers is 1.51%, slightly higher than related mergers, 1.43%. The maximum is 5.23% for related mergers and 4.82% for diversifying mergers. Of the 174 mergers that produce coinsurance exceeding 1%, 151 are related and only 23 are diversifying. I plot the distribution of coinsurance benefits for the related and diversifying mergers in Figure 1. The distributions again appear very similar. The t-statistic for the null hypothesis that 13 The previous literature uses several other approaches for classification: they include defining industries at the two digit level (Matsusaka (1993); or, using text-based measures instead of SIC codes (Hoberg and Philliips (2010)). 23

24 the two means are equal is 0.16, under the assumption of equal variance. If instead, I assume different variances, then the t-statistic is 0.15, suggesting that the null hypothesis of equality cannot be rejected at the 10% level of significance. Next, to determine whether the distribution of coinsurance benefits is the same for both related and diversifying mergers, I perform the Kolmogorov-Smirnov (KS) test. The KS test produces a statistic of 0.053, corresponding to a P- value of less than The null hypothesis of same distribution thus cannot be rejected, even at the 10% level of significance. The finding that coinsurance benefits are no larger for diversifying than related mergers is hard to reconcile with the prevalent notion that pigeon-holes coinsurance into corporate diversification. Coinsurance is not exclusive to diversifying mergers. Rather, related mergers are often associated with significant coinsurance benefits, a finding not documented by the previous literature. This finding, however, does not contradict the claim that coinsurance is an important benefit from diversification. It may well be the case that firms undertaking diversifying mergers are already sufficiently diversified, and the incremental coinsurance benefit from another diversifying merger is small. Alternatively, coinsurance benefit from diversification may stem from other channels that Duchin (2010) discusses: avoidance of transaction costs of external finance, rather than avoidance of deadweight costs of distress. D. Proxies for Coinsurance Following Lewellen (1971), academics routinely use cash flow correlation, or its variant, as a proxy for coinsurance. For example, Duchin (2010) uses cross-divisional correlation in cash flow, and Hann, Ogneva and Ozbas (2013) use correlations of industry level cash flows based on 24

25 single-segment firms. But is cash flow correlation a good proxy? Since this paper is the first to provide structural estimates of coinsurance and to the extent that these estimates of coinsurance are accurate it enables me to test how good are the proxies for coinsurance. To test if cash flow correlation measured as the correlation between quarterly cash flows of the target and the acquirer, dating back up to eight quarters prior to the merger announcement is a good proxy, I compute its correlation with the estimated coinsurance benefits. Table VII reports that the correlation with cash flow correlation to be 0.02, not statistically different from zero even at the 10% level of significance. This contradicts the popular notion that cash flow correlation is a good proxy. Even return correlation, which is an input to the structural model, has a correlation of only Instead, target leverage (-0.54), acquirer leverage (-0.37), and acquirer size (-0.30), predict coinsurance better. Since cash flow correlation is not a good proxy, it may be worthwhile to construct an alternative proxy for coinsurance: a proxy constructed using easy-to-compute accounting variables, which researchers may easily use, without having to actually estimate a structural model. But, as we can see in Table VII, there are several variables that are correlated with coinsurance. Moreover, since the structural model is nonlinear, the quadratic as well as interaction terms of these variables are probably correlated with coinsurance. Using so many variables to predict coinsurance will no doubt result in good in-sample fit, but over fitting will render the predictions susceptible to high standard errors, and out of sample validity of such a model will be questionable. To avoid this over-fitting problem, I use LASSO to select the optimal model (Tibshirani (1996)). LASSO, or least absolute shrinkage and selection operator, is a popular statistical 25

26 method for parsimonious model selection. The LASSO produces an optimal five variable index, given by log( SSSSSSSS TT SSSSSSSS AA ) LLLLLL AA LLLLLL TT log ( SSSSSSSS AA ) LLLLLL TT log( SSSSSSSS TT SSSSSSSS AA ), (10) where coinsurance benefits are expressed in percentage; leverage in decimals; and, size is market capitalization in million dollars. The correlation between this coinsurance index and estimated coinsurance benefit is IV. The Stock Price Response to Coinsurance A. Coinsurance Benefits and Announcement Returns The paper to this point considers the total gain in firm value from coinsurance. But do the benefits from coinsurance reach the stockholders? Although several papers document that coinsurance benefits bond holders, there is little evidence that coinsurance benefits stock holders ((Billet, King and Mauer (2004), Penas and Unal (2004)). The dominant view in the literature is that coinsurance benefits bondholders at the expense of stockholders. The literature argues that coinsurance reduces volatility, consequently reducing the option value of equity, and transferring value from equity holders to bondholders (Galai and Masulis (1975), Higgins and Schall (1976), Mansi and Reeb (2002)). If the stockholders lose from coinsurance, as is argued by the literature, then when a merger that is high in coinsurance is announced, the stock price should decrease and not increase. This argument motivates the following test. The cumulative abnormal return (CAR) around the merger announcement date is regressed on the estimated coinsurance benefits. I add standard controls following the literature on announcement returns (Akbulut and 26