Asset Volatility and Financial Policy: Evidence from Corporate Mergers

Transcription

1 Asset Volatility and Financial Policy: Evidence from Corporate Mergers Oliver Levine University of Wisconsin-Madison Youchang Wu University of Wisconsin-Madison November 10, 2014 The presence of costly financial distress suggests that business risk and optimal leverage are inversely related, yet empirical studies have found this connection to be weak. We identify the effect of business risk on capital structure by exploiting the cross-section variation in the predictable change in asset volatility induced by corporate merger. Our estimates demonstrate a strong link, indicating a one standard deviation decline in asset volatility leads to a 10.2 percentage point, or 0.55 standard deviation, increase in a firm s leverage ratio. In addition, we find asset volatility strongly predicts cash holdings as well as the method of payment in merger. Our results suggest that asset volatility is an economically significant determinant of firms financial policies, and help to explain the persistent firm and industry-specific components of leverage identified in the literature. Levine: Patrick Thiele Fellow in Finance, Finance Department, Wisconsin School of Business, University of Wisconsin-Madison, 975 University Ave, Madison, WI 53706, olevine@bus.wisc.edu. Wu: Finance Department, Wisconsin School of Business, University of Wisconsin-Madison, 975 University Ave, Madison, WI 53706, ywu@bus.wisc.edu. Youchang Wu is currently a visiting financial economist at the U.S. Securities and Exchange Commission. The Commission, as a matter of policy, disclaims responsibility for any private publication or statement by any of its employees. The views expressed herein are those of the author and do not necessarily reflect the views of the Commission or of the author s colleagues on the staff of the Commission. We thank seminar participants at Virgina Tech, University of Oregon, Georgetown University, Securities and Exchange Commission, and University of Wisconsin-Madison for helpful comments.

2 When financial distress is costly, business risk is crucial to understanding a firm s capital structure choice. An increase in the riskiness of a firm s operations increases the likelihood of bankruptcy and the expected costs of financial distress, all else equal, lowering the firm s optimal leverage. Yet nearly every empirical study which explores the determinants of leverage find business risk, measured by either cash flow or asset volatility, to be of little economic significance, for example Titman and Wessels (1988), Welch (2004), Lemmon, Roberts, and Zender (2008), and Frank and Goyal (2009), or do not consider it at all, for example Rajan and Zingales (1995), Kayhan and Titman (2007), and Leary and Roberts (2014). 1 At the same time, previous studies have found large and persistent differences in capital structure across firms and industries that cannot be explained by traditional determinants of leverage. One challenge in assessing the effects of business risk on leverage is that business risk itself is inherently hard to measure, either because it relies on low-frequency accounting data, as in the case of cash flow volatility, or because it relies on measures of leverage itself, in the case of unlevered equity volatility. Compounding this problem is that business risk is intimately tied to other firm and industry-specific characteristics, making identification challenging. These difficulties have made understanding the role that business risk plays in firms financial policy decisions elusive. We use the unique opportunity provided by corporate merger to assess the importance of business risk, measured by asset volatility, on firms capital structure and cash holdings policies. Asset volatility represents the riskiness of the present value of all future cash flows generated by a firm for both its debt and equity holders. Corporate mergers provide a setting in which the asset volatility of a firm changes in a large and predictable way. If asset volatility is an important determinant of capital structure and cash holdings policies, then these variables should respond in predictable ways. Because the changes in asset volatility around mergers are predictable using pre-merger information, this setting allows us to avoid 1 One exception is Bradley, Jarrell, and Kim (1984), who find that long-run mean leverage, estimated over 20 years, is significantly negatively related to long-run cash flow volatility in the cross section. Another exception is Reindl, Stoughton, and Zechner (2013), who measure market-implied bankruptcy costs using a structural estimation approach. They find their structural estimates of asset volatility are significantly negatively related to observed leverage. 1

3 the measurement concerns normally associated with estimating asset volatility. In addition, our identification strategy exploits the cross-sectional variation across mergers, instead of the average post-merger leverage and cash ratio changes. Therefore, it does not rely on exogeneity of the merger event itself. Using a sample of public takeovers between 1980 and 2011, we find that anticipated changes in asset volatility have a large impact on both short and long-term changes in firm leverage. We find that firms respond to a decline in asset volatility by increasing leverage in an economically significant way: our coefficient estimates imply that a one standard deviation reduction in asset volatility corresponds to as much as a 10.2 percentage point, or 0.55 standard deviation, increase in leverage. These results supports the predictions of the tradeoff theory of capital structure, and suggest that asset volatility plays a much more important role in firms capital structure choice than previously documented. Similarly, we find that anticipated changes in asset volatility also predict sizable changes in cash holdings following acquisition, evidence of a precautionary savings motive consistent with Bates, Kahle, and Stulz (2009) and Duchin (2010). Performing the analysis on within-firm changes rather than levels helps to alleviate concerns about unobserved firm-specific heterogeneity driving the results. The results are robust to the inclusion of numerous controls for well-known determinants of financial policy, and cannot be attributed to leverage changes passively induced by the target s existing capital structure. In addition, we perform a placebo test using withdrawn merger deals to show that our results are not driven by an unobserved characteristic dictating jointly the choice of merger partner and leverage. After identifying significant effects of asset volatility on leverage and cash holding decisions, we test whether firms use the merger transaction itself as a tool to adjust capital structure. We run a multinomial logit to determine if the acquirer s method of payment choice cash only, stock only, or hybrid depends on the anticipated change in asset volatility. We find that for mergers which are more diversifying the anticipated change in asset volatility is lower the acquirer is most likely to make a cash-only offer, and least likely to offer pure stock. Our results suggest that post-merger capital structure and cash holdings 2

4 policy is an important consideration in method of payment choice. However, the effects of expected changes in asset volatility on leverage and cash holdings mentioned above are not purely driven by the choice of payment method, as we find similar effects in the subsample of stock-only mergers. The results from the merger sample suggest that asset volatility is an economically important determinant of leverage and cash holdings. Because these results are unlikely to be driven by measurement error or reverse causality, they serve as a natural benchmark. We then explore the ability of our measure of asset volatility to explain the leverage and cash holdings of the entire universe of public firms from In Table 12 (see Section 6), asset volatility alone is shown to be an important determinant of leverage, having an adjusted R-square of 14% in a univariate regression, compared to 20% for a multivariate regression which includes eight determinants commonly used in previous literature, as well as year fixed effects. The coefficient estimate implies a one standard deviation decline in asset volatility corresponds to a 9.3 percentage point, or 0.45 standard deviation, increase in leverage. Critically, this is very similar to the magnitude implied by our previous results which control for measurement error. While we cannot rule out measurement error in the full-sample results, the similarity of the estimates suggests that the effects of measurement error are likely to be small. Our results indicate that heterogeneity in firm asset volatility can help to explain the persistent firm-specific component of leverage identified in Lemmon, Roberts, and Zender (2008). Furthermore, we find asset volatility is able to explain about 38% of the industry effect on leverage, which has been noted as one of the strongest predictors of leverage. Our analysis for cash holdings in the full sample reveals a similar strong predictive power of asset volatility, consistent with the results from our merger sample. We estimate predicted changes in asset volatility in two ways. First, we use the model of Merton (1974) to estimate asset volatility and the predicted changes resulting from the merger. Using the anticipated merger-induced change in asset volatility is advantageous in that it does not rely on post-merger leverage information. Because we are interested in business risk, we measure asset volatility as the volatility of the firm s non-cash assets. 3

5 Second, we exploit the fact that the predictable change in asset volatility from a merger is a function of the correlation of the acquirer and target firms asset values. Because we are able to easily measure this correlation using equity returns, we can construct a measure of anticipated changes in asset volatility without ever measuring asset volatility itself. This eliminates concerns of measurement error and bias that arise from estimating asset volatility directly. There are several additional advantages to our empirical design. First, if there are fixed costs to adjusting capital structure, firm s are more likely to adjust capital structure to a target after a significant economic event creates a large deviation from their optimum. A merger event has the potential to trigger this adjustment. Second, our framework does not require the merger event itself to be exogenous. Instead, we condition on the merger event and use cross-sectional variation in asset diversification to identify the relationship between asset volatility and financial policy. Last, we perform the analysis using within-firm changes. This avoids concern that the estimated relationship between asset volatility and financial policy is driven by an omitted firm-specific characteristic. For comparison to traditional measures of business risk, we perform the analysis including realized cash flow volatility as a predictor of leverage and cash changes and find mixed results. The sign is consistent with the predictions of our measure of asset volatility, but the economic magnitude of the estimates are small, consistent with prior studies. A firm s asset value includes information about future cash flows, investment opportunities, and discount rates, making asset volatility a function of all of these forward-looking components. Realized cash flow volatility, on the other hand, is not only difficult to measure given low-frequency accounting data, but is backward-looking and therefore does not truly measure business risk or uncertainty. That our measures of asset volatility is a much stronger predictor than cash flow volatility suggests that business risk is better captured by the former. Our contribution is related to a large body of literature in capital structure, cash holdings, and mergers and acquisitions. A well-established view in the literature is that a large portion of the heterogeneity in capital structure is highly persistent and not easily explained by 4

6 traditional determinants of leverage. Notable papers supporting this view of persistence and heterogeneity include Rajan and Zingales (1995) and Lemmon, Roberts, and Zender (2008). 2 Our results provide evidence that asset volatility, which has received little attention in the empirical literature, is an important determinant of leverage. In the trade-off theory of capital structure, it is common to specify business risk, defined as either asset or cash flow volatility, as an exogenous primitive. These models obtain a negative relationship between business risk and optimal leverage; see, for example, Fischer, Heinkel, and Zechner (1989), Leland (1994), Goldstein, Ju, and Leland (2001), and DeAngelo, DeAngelo, and Whited (2011). Our results provide support for these models of capital structure. 3 As in this study, Harford, Klasa, and Walcott (2009) use the merger event to study capital structure choice. They look at leverage changes around acquisitions to determine if firms have leverage targets, and find strong evidence that they do. Consistent with our results that firms use method of payment choice to effect capital structure changes, they also find that deviations from target leverage can predict method of payment. Uysal (2011) finds that firms with higher than target debt ratios are less likely to make acquisitions and less likely to use cash in a takeover. These papers, however, do not consider asset volatility in capital structure or method of payment, which is the focus of our study. Our empirical design is motivated by the diversifying impact of asset combination resulting from merger. Coinsurance motives for merger were first highlighted in Lewellen (1971), and later studied empirically in Asquith and Kim (1982) and Ghosh and Jain (2000). Leland (2007) was the first to show that while coinsurance from merger does have the benefit of decreasing expected distress costs, it also has a cost for the same reason: the target and acquirer cannot choose capital structure independently and therefore cannot maximize the option value of default for the individual firms by setting individual leverage policies. 2 One exception is DeAngelo and Roll (2014) who argue that firm-level capital structure is not very stable. 3 The implication of our findings for the perking order theory is less clear. On the one hand, riskier firms are more likely to face cash flow shortfalls. So according to the perking order theory, they should use leverage more conservatively to lower the chance of costly outside financing (Fama and French (2002)). On the other hand, it is possible that riskier firms suffer more from adverse selection. If so, then the pecking order theory would predict that they are less likely to issue equity, and therefore should have higher leverage (Frank and Goyal (2009)). 5

7 The merger model of Morellec and Zhdanov (2008) predicts that the bidder with the lowest leverage will win the takeover bid in equilibrium, and that acquirers increase leverage after acquisition. Mansi and Reeb (2002) shows that diversifying mergers benefit debtholders through the coinsurance effect, at the expense of shareholders. There are several studies which explore precautionary savings by firms, including Acharya, Davydenko, and Strebulaev (2012), Almeida, Campello, and Weisbach (2011), Riddick and Whited (2009), and Han and Qiu (2007). Duchin (2010) finds a strong negative relationship between diversification across business segments and cash holdings, and finds that correlations in industry investment opportunities provide a good predictor of future cash holdings. Finally, while many studies have focused on the agency conflicts and signaling motives of method of payment choices, our method of payment results are consistent with Heron and Lie (2002) who finds that post-acquisition operating performance is not impacted by method of payment, suggesting the choice is based on financial, rather than operational, motives. A robust empirical feature, shown in Travlos (1987), is that announcement returns to cash bids are significantly higher than stock bids, which he finds supports a signaling theory of payment choice. However, this also is consistent with our results in which shareholders gain from the increased debt tax shield created from a cash bid; Ghosh and Jain (2000) provides evidence of this shareholder tax benefit. The remainder of the paper is organized as follows. In Section 1 we detail the empirical approach we use to identify the relationship between asset volatility and financial policy, and state our hypotheses. Section 2 describes the data, and Section 3 describes the changes in the relevant variables around the merger event. Our main empirical results for the merger exercise are shown in Section 4, followed by exploring method of payment choice in Section 5. We extend the analysis to the entire universe of firms in Section 6 before concluding in Section 7. 6

8 1 Empirical Design and Hypotheses A significant body of empirical capital structure literature has focused on finding which firm characteristics determine leverage choice for example, Rajan and Zingales (1995), Kayhan and Titman (2007), Lemmon, Roberts, and Zender (2008), and Frank and Goyal (2009). If financial distress is an important cost associated with the use of debt, e.g. in the tradeoff theory of capital structure, the riskiness of a firm s business will impact leverage choice. Despite this link, past studies have found little evidence that business risk is economically significant in understanding leverage choice, or do not consider it as a determinant. One potential reason why a characteristic as fundamental as business risk has received little empirical support or attention is that it is difficult to measure. Most studies proxy for business risk using realized cash flow volatility from accounting data. Not only does this require a long horizon to measure accurately, but also does not capture uncertainty about future cash flows and asset value. 4 An alternative approach to measuring business risk is to estimate asset volatility by multiplying the firm s equity volatility by the equity-to-value ratio, a simple unlevering. 5 This approach incorporates uncertainty about future cash flows, and allows measurement over a shorter horizon. However, this approach suffers from potential measurement error that could generate a spurious correlation between estimated asset volatility and leverage. This results from the fact that simple unlevering of equity volatility does not account for the volatility of debt, which generates measurement error in asset volatility that is negatively correlated with leverage. 6 Thus any estimated relationship 4 Recognizing the difficulty in measuring cash flow volatility, Fama and French (2002) proxy it by firm size. However, firm size is related to many other factors potentially relevant for financial policy, such as the ease of access to capital markets. This makes the interpretation of its effect difficult. 5 This simple unlevering of equity volatility was used by Welch (2004) as a control in leverage regressions exploring whether firms adjust debt to changes in equity values, as well as by Frank and Goyal (2009). 6 Specifically, for a firm composed of debt and equity, the variance of a firm s asset value, σ 2, can be decomposed as σ 2 = ( ) 2 ( ) 2 E D σe 2 + σd E D Cov(E, D) (1) D + E D + E (D + E) 2 where E and D are equity and debt values, and σ E and σ D are their associated volatilities. The simple unlevering approach assumes σ D 0, making insignificant the final two terms of (1). In practice, the final two terms become larger for firms with riskier debt, thus resulting in a negative bias in estimates of asset volatility, a bias which is increasing in leverage. Thus this simple approximation of asset volatility and 7

9 between this simple estimate of asset volatility and leverage must be viewed with caution. In this paper, we circumvent these measurement concerns by exploiting a unique setting provided by corporate merger events. When two firms merge, their future cash flows are pooled, providing a coinsurance of the individual firms debt liabilities. As long as their cash flows aren t perfectly correlated, this combination creates a diversification effect. Specifically, if the acquirer and target have asset volatilities σ acq and σ tar and asset values V acq and V tar prior to the merger, ignoring synergy effects, the expected asset volatility of the combined firm will be E [σ acq+tar ] = ω 2 σ 2 acq + (1 ω) 2 σ 2 tar + 2ω(1 ω)ρσ acq σ tar (2) where ω V acq /(V acq + V tar ), and ρ is the correlation of the target and acquirer asset values. 7 Anticipated post-merger changes in asset volatility are increasing in ρ, a result of the diversification effect. The merger event provides an abrupt and predictable change in a firm s asset volatility, allowing us to identify the firm s financial policy response to these asset volatility changes. First, we use the merging firms asset correlation ρ as a proxy for predicted changes in asset volatility. If asset volatility is an important determinant of capital structure, ρ should predict, inversely, post-merger leverage changes. Using ρ as a proxy for predicted changes in asset volatility eliminates concerns that measurement error may be driving the results because this approach does not use any direct measure of asset volatility or leverage. 8 Second, we extend our analysis by estimating the anticipated change in asset volatility around the merger transaction. We estimate acquirer and target asset volatilities prior to the merger to measure directly the post-merger asset volatility, E [σ acq+tar ], given by leverage will be negatively correlated due to measurement error. 7 To be clear, our goal is not to uncover the best forecast of post-merger asset volatility. Instead, we investigate whether the predictable component of the merger-induced change in asset volatility based on a simple model can explain the financial policies around the merger. To the extent that financial activities, such as the financing of the merger deal itself, may need to be planned far ahead of time, they may well be based on the predicted rather than the realized asset volatility. 8 We estimate ρ as the correlation between acquirer and target equity returns. Precisely, ρ is the correlation in the unobserved asset returns, not equity returns. However, since equity is a call option on the underlying asset value, instantaneously equity value and asset value are perfectly correlated. Therefore, when returns are measured over short time intervals, equity return correlation is a good approximation to asset return correlation. 8

10 (2). The pre-merger predicted change in the acquirer s asset volatility, E[ Asset Vol] (E [σ acq+tar ] σ acq ), is then used to predict post-merger changes in leverage. Because the anticipated change in asset volatility depends only on pre-merger information, and critically does not depend on post-merger leverage, this avoids concerns that measurement error could lead to a mechanical correlation between our predictive variable and post-merger leverage changes. To further mitigate concerns of measurement error, we estimate asset volatility using an approach motivated by Merton (1974), which treats equity as a call option on the firm s assets with a strike price equal to the face value of debt. 9 This model accounts for the increase in debt volatility that occurs as leverage increases. Given data on equity volatility, the risk-free rate, the market value of equity, and the face value of debt, it is straightforward to use this option pricing model to estimate the implied asset volatility. In essence, this is a structural approach to unlever the firm s equity volatility that avoids the bias that comes from ignoring the volatility of the firm s debt. Technical details on the estimation are provided in Appendix A. Furthermore, we remove the effects of cash from our measure of asset volatility, as we do not want cash holdings, a financial policy potentially driven by business risk itself, to affect our measure of business risk. Thus our measure of asset volatility is the volatility of the value of all non-cash assets of the firm. Removing the effects of cash holdings also eliminates the mechanical negative correlation between the asset volatility estimate and cash holdings.also, we control for the direct impact of the target s capital structure and cash holdings on the merged firm. In addition to the advantages of our approach regarding measurement error described above, our empirical design has three other advantages. First, because we focus on withinfirm changes, we remove unobserved firm-specific effects that could otherwise drive the results. This avoids concerns of an omitted variable bias resulting from an excluded firm characteristic. Second, we use only pre-merger information about asset volatility to predict 9 The approach is similar to that described in Crosbie and Bohn (2003) and Vassalou and Xing (2004). 9

11 post-merger leverage changes. This lessens concerns about reverse causality. By exploring leverage changes over various horizons, such concerns are further mitigated. Last, because merger events represent a significant and abrupt change in asset volatility and other characteristics, firms may have strong incentive to reset their leverage to their new optimum. If leverage deviations from target are small, we may not expect firms to be willing to pay the fixed costs associated with adjusting capital structure, as shown by Fischer, Heinkel, and Zechner (1989) and Leary and Roberts (2005). Leverage changes around merger events, which generate potentially large deviations from target leverage, are therefore more likely to be informative about firms financial decisions. We summarize our hypothesis regarding capital structure and asset volatility with the following: Hypothesis 1. An acquiring firm s post-merger choice of leverage responds negatively to (1) its pre-merger asset value correlation with the target, and (2) the predicted post-merger change in its asset volatility. This unique situation provided by the merger transaction also allows us to explore the impact of asset volatility on corporate cash holdings. In contrast to agency-conflict explanations, one explanation for the presence of cash holdings within firms is that cash serves as precautionary savings to protect against future cash flow shocks. Similar to leverage, we test whether firms respond to changes in asset volatility by adjusting their cash holdings. If asset volatility is expected to decline due to the merger, firms have less incentive to hold cash as insurance. In contrast, higher volatility will induce higher cash holdings. This leads to our second hypothesis: Hypothesis 2. An acquiring firm s post-merger choice of cash holdings responds positively to (1) its pre-merger asset value correlation with the target, and (2) the predicted post-merger change in its asset volatility. In addition, this framework allows us to explore the firm s choice of method of payment in the merger transaction. Choosing between cash and stock has a strong impact on the 10

12 capital structure of the combined firm. While there are many possible explanations for the method of payment decision including tax incentive, agency motives, signaling and asymmetric information the above approach gives us an opportunity to look specifically at the capital structure motives of the method of payment. If firms use method of payment to effect capital structure changes, then we should expect predicted asset volatility changes to predict method of payment. This leads to our third hypothesis: Hypothesis 3. The likelihood a firm chooses cash as the method of payment in acquisition decreases in (1) its pre-merger asset value correlation with the target, and (2) the predicted post-merger change in its asset volatility. Because payment method has a direct impact on the post-merger leverage and cash ratio, it is an interesting question whether the effects of expected asset volatility changes on the post-merger leverage and cash come only through the endogenous choice of payment method. Furthermore, since our model may not fully capture the factors determining the choice of payment method, there is a potential concern that this choice itself may lead to a false inference about the firm s response to anticipated changes in asset volatility. For example, suppose that diversifying mergers are more likely to be made with cash, for some reason unrelated to asset volatility and for which we do not control. This could induce a positive correlation between changes in cash and asset volatility, and a negative correlation between changes in leverage and asset volatility. To address these questions, we repeat the empirical tests of Hypotheses 1 and 2 conditioning on method of payment. Specifically, we look only at the subset of acquisitions that were all-stock deals. For this subset, the method of payment itself does not impact the post-merger leverage and cash holdings of the combined firm. 2 Sample, Variables and Summary Statistics Starting with the SDC Mergers and Acquisitions Database from 1980 to 2011, we apply the following filters to construct our sample, based on the objective and data requirements of our analysis: (1) both the acquirer and the target are nonfinancial public firms and can be 11

13 matched to the CRSP monthly stock database using the historical CUSIP code; (2) both the acquirer and the target can be matched to the Compustat quarterly database at the time of merger announcement; (3) the deals must be completed within 12 months after announcement, with the two firms fully merged; (4) the shares acquired during the merger must be more than 50% of target s share outstanding; (5) the transaction values must be at least 5% of the acquirer s market capitalization; 10 and (6) by the last fiscal quarter end prior to the merger announcement, there are at least 100 overlapping daily stock return observations available for estimating the correlation between the acquirer and the target. Our final sample consists of 1778 merger events that satisfy these criteria. Panel (A) and (B) of Table 1 present the summary statistics for our sample of corporate mergers. Details of variable definitions are provided in Appendix B. Panel (A) shows the characteristics of the merger deals. About one-third (595) of them are stock deals, i.e., paid for by a stock exchange. The numbers of cash-only (505) and hybrid deals (477) are slightly lower. Hybrid deals are substantially larger in terms of transaction value, reflecting the need for multiple payment methods to finance bigger mergers. The mean and median ratios of the transaction value to the acquirer s market capitalization are 62.5% and 35.3%, respectively, suggesting significant post-merger changes in acquirers asset compositions, which potentially generate needs for adjustments in corporate financial policies. The ownership acquired during the merger is 99% on average, suggesting that the toehold is on average very small prior to the merger. Panel (B) shows the characteristics of the acquirer and the target before the merger announcement. Both the book leverage (BL), market leverage (ML) and cash ratios (Cash) of the acquirers and the targets are similar, and are on average above the industry medians (see variables indicated with (firm ind) ). The average (non-cash) asset volatilities (Asset Vol), estimated from daily stock returns over the 12-month period through the last fiscal quarter before the merger announcement, are 52.2% for the acquirer, and 63.9% for the target. The cash flow volatility of the acquirer is also lower than that of the target. Acquirers are bigger 10 If the transaction value is too small, the merger will be unlikely to have a significant impact on the acquirer s asset volatility and financial policies. 12

14 than targets in terms of sales revenue, which may partly explain their lower volatilities. Acquirers are also more profitable. Other firm characteristics that are potentially relevant for leverage and cash policies, including asset tangibility, market-to-book ratio of firm value, R&D intensity, are similar for the acquirer and the target. Summary statistics for the full CRSP/Compustat annual database from 1965 to 2012, which will be used in Section 6, is presented in Panel Panel (C). 3 Merger-Induced Changes: Expected and Realized We first investigate the expected changes in asset volatility, leverage and cash ratios arising from a pro forma combination of the acquirer and the target, based on the information available at the last quarter before the merger announcement. The results are summarized in Panel (A) of Table 2. A key variable driving the expected asset volatility changes is the asset value correlation between the acquirer and the target, which we measure using daily stock returns over the 12-month period through the last fiscal quarter before the announcement. The average correlation, denoted by Rho, is 18.5%. Due to the low correlation, the average merger generates a sizable reduction in the acquirer s asset volatility in expectation (E[ Asset Vol]), despite the higher asset volatility of the target. The mean expected reduction in asset volatility is 7.6%. The dependence of the expected change in volatility on asset value correlation is apparent, as shown in Figure 3. We group mergers into 10 deciles with increasing before-announcement return correlation, and plot the average predicted asset volatility change for each decile. The figure shows the predicted change in asset volatility increases nearly monotonically with the correlation. In contrast to the sharp expected decline in asset volatility, a pro forma combination of the balance sheets of the acquirer and the target have very little effects on leverage and ratios: BL(pf acq), ML(pf acq), and Cash(pf acq) are close to zero. This is not surprising, as the pre-announcement leverage and cash ratios of acquirers and targets are similar. 13

15 We then investigate the realized changes in firm characteristics after the merger, shown in Panel (B) of Table 2. The leverage ratio increases, on average by 4.6 percentage points, while the cash ratio decreases, on average by 5.8 percentage points, from the last pre-announcement fiscal quarter to the fourth quarter following merger completion. These results are consistent with the findings of Ghosh and Jain (2000), Harford, Klasa, and Walcott (2009), and Duchin (2010), suggesting that mergers increase debt capacity, and reduce the marginal value of cash holdings. Since the leverage and cash ratios of the acquirer and the target are similar before the merger, such leverage and cash ratio changes cannot be explained by the passive effects of a pro forma combination of balance sheets. Panels (A) and (B) of Figure 2 show the mean and median of the leverage and cash ratios of the acquirer over the entire 32-quarter event window, from 12 quarters before the announcement to 20 quarters after merger completion. These figures show that abrupt changes occur mainly around the merger completion date. The realized asset volatility declines on average by 4.7 percentage points during the first post-merger year, confirming a substantial diversification effect. Panel (C) of Figure 2 shows that it declines even further in the last three years of the event window. Despite the lower asset volatility, the average equity volatility increases in the first five quarters after the merger, and gradually decreases afterwards, as shown in Panel (D) of Figure 2. This is because the increased leverage effect arising from higher debt and lower cash outweighs the asset diversification effect on equity risk in the short term. The realized change in asset volatility one year after merger completion is somewhat smaller than predicted based on the pre-announcement information (4.7 vs. 7.6 percentage points). This is not surprising as uncertainty about the complementarities and the integration process may increase asset volatility in the short term. By the fifth year after completion of the merger, the average forecast error becomes tiny: the gap between the realized and predicted change in asset volatility is less than 0.1 percentage points. In addition, the correlations between the predicted change in asset volatility and the one-year and five-year realized changes in asset volatility are 0.50 and 0.53, respectively. These results suggest that 14

16 our simple correlation-based forecasting model does quite well to predict realized changes in asset volatility. Panel (B) of Table 2 show that the realized changes in other firms characteristics relevant for leverage and cash holdings, such as firm size and asset tangibility, are statistically significant, suggesting a need for control for these changes to isolate the effect of asset volatility. One exception is the change in cash flow volatility, CF Vol, which is measured by the difference of cash flow volatility between the five post-merger years and the three pre-announcement years, each with a minimum of 8 quarters of data. While being negative as expected, CF Vol is small in magnitude, and is statistically insignificant, in contrast to the sharp decline in asset volatility. This highlights the difficulty in measuring business risk using low-frequency data, and the inadequacy of changes in cash flow volatility as a measure of the diversification effect. These univariate summaries of asset volatility and financial policy changes around the merger are consistent with our hypotheses that firms increase leverage and reduce cash in response to a lower asset volatility. However, part of the changes in leverage and cash ratio may be due to the endogeneity of the merger event itself. For example, acquirers on average may be under-levered and cash-rich before the merger announcement in anticipation of the need to finance the deal. This would lead to a future increase in leverage and decline in cash, on average, even in the absence of asset volatility changes. For this reason, our empirical approach does not rely on average post-merger changes and instead uses cross-sectional variation to identify the effects of asset volatility on leverage and cash. We examine the cross-section in the next section. 4 Asset Volatility and Changes in Leverage and Cash We now examine how changes in leverage and cash ratio are related to the predicted asset volatility change. To get a first impression, we plot in Figure 3 the post-merger increases in book leverage (Panels (A) and (B)) and cash-to-asset ratio (Panels (C) and (D)) for mergers in different deciles of equity return correlations and predicted asset volatility changes. 15

17 Decile one consists of mergers with the lowest pre-merger correlation, or mergers expected to generate the largest reduction in asset volatility, while decile ten consists of mergers with the opposite features. The increases in leverage and cash ratios are measured over different horizons: from the pre-announcement quarter to the end of the first, third and fifth years after merger completion. Consistent with our predictions, there appears to be a strong negative univariate relationship between leverage changes and predicted changes in asset volatility, as well as equity return correlations, and a positive relationship between cash changes and these two variables. We now test these relationships using multivariate regressions, which allows us to control for alternative effects. 4.1 Changes in Leverage To explore the effect of a change in asset volatility on a firm s leverage choice, we first examine the relationship between the post-merger leverage change and the equity return correlation, an easily measured proxy for the expected change in asset volatility. Table 3 shows the regression results for the acquirer s leverage increase in the first year after the merger completion, from the last quarter before the merger announcement. In model (1), we regress the book leverage increase ( BL(1)) on only the pre-announcement return correlation (Rho) and year fixed effects, representing the year of merger announcement. Because the correlation Rho is positively related to predicted changes in asset volatility, we expect the coefficient on Rho to be negative. The result provides strong support for this prediction. The coefficient is , and is significant at the 1% level. This coefficient implies that for mergers with a pre-announcement correlation one standard deviation (0.186) above the mean, the post-merger change in leverage is on average 1.6 percentage points lower. Because the average total increase in leverage is 4.6 percentage points, our estimates indicate correlation alone can explain a large fraction of the leverage change. In model (2) of Table 3, we control for the changes in other firm characteristics that the existing literature has found to be important for leverage, including firm size (measured by sales), market-to-book ratio, asset tangibility, profitability and R&D intensity. Consistent 16

18 with the previous findings, the leverage increase is positively related to the increase in firm size, but negatively related to the increase in profitability. Changes in other variables do not seem to have a significant effect. Importantly, the effect of pre-merger correlation remains largely unchanged. In model (3), we control for the acquirer s industry-adjusted book leverage ratio before the merger announcement (BL(acq ind)) to account for potential mean reversion in leverage. The industry is defined using the Fama-French 49-industry classification. To account for the passive effect of the target s capital structure on the post-merger leverage, we also control for the gap between the leverage arising from a pro forma combination of balance sheets and the acquirer s leverage before the merger announcement, BL(pf acq). The pro forma leverage ratio can be viewed as a status quo for the post-merger leverage. If the acquirer inherits the debt of the target and is slow in making adjustments, we should expect BL(pf acq) to be positively related to the post-merger leverage increase. The results are consistent with both mean reversion and slow adjustment from the status quo, as a higher industry-adjusted leverage before announcement leads to lower leverage increase after the merger, while a larger gap between the pro forma leverage and the acquirer s initial leverage is associated with a larger leverage increase. At the same time, the effect of pre-announcement return correlation becomes even stronger, both economically and statistically. In model (4), we add an additional control variable, i.e., the realized change in cash flow volatility( CF Vol). Theoretically, the importance of cash flow volatility for capital structure is well-established, although empirical evidence of this link has not been strong. If the asset volatility and cash flow volatility capture the same thing, we should expect the statistical significance of Rho to become weaker after CF Vol is included in the regression. The result in column (4) rejects this conjecture. While the coefficient of CF Vol is significantly negative, as expected, the significance of Rho is unchanged, and its magnitude increases slightly. This is because cash flow volatility, which itself is hard to measure precisely given the low frequency of data, is only one variable that affects asset volatility. Other economic fundamentals, such as volatility of investment opportunities and discount rates, may mani- 17

19 fest themselves in asset volatility without being fully reflected in cash flow fluctuations. A simple calculation suggests that for mergers with a cash flow volatility change one standard deviation above the average, the change in leverage is 0.75 percentage points lower. This effect is economically significant, but is less than half of the impact of one standard deviation increase in return correlation. This result suggests empirical models controlling only for cash flow volatility without considering asset volatility may not fully capture the importance of business risk in explaining leverage. The results for market leverage are largely similar. For brevity, we only report the results for the full specification, which appear in column (5) of Table 3. We control for industryadjusted market leverage of the acquirer (ML(acq ind)) and the deviation of the pro forma market leverage from the acquirer s leverage (ML(pf acq)) in the regression. The results are very similar to those for book leverage shown in column (4). In Table 4 we explore the effect of return correlation on changes in the acquirer s book leverage from the pre-announcement quarter to the end of the second through the fifth post-merger years. We control for realized changes in firm size, market-to-book ratio, asset tangibility, profitability, and R&D intensity over the same time horizons, as well as the pre-announcement industry-adjusted leverage, and the effect of a pro forma combination of balance sheets. The results are very similar to those in Table 4, except that the coefficient on cash flow volatility becomes insignificant. These results suggest that asset volatility s effect on leverage is much more important and robust, either because it is more precisely measured, due to the higher frequency of data, or because it captures more economically relevant factors. In Table 5, we estimate directly the effect of predicted changes in asset volatility on leverage changes. We keep the baseline specification of model (4) in Table 3, and replace Rho by the predicted change in asset volatility, (E[ Asset Vol]). Recall the effects of cash holdings have been removed from our measure of asset volatility. Similar to the previous results, we find strong support that leverage responds negatively to changes in asset volatility. The coefficient of E[ Asset Vol] is significantly negative for years two through five, suggesting 18

20 that a larger decline in predicted asset volatility is associated with a larger increase in firm leverage. At the same time, the cash flow volatility effect is insignificant. An exception is the year one leverage change in the full model (column (2)), where the cash flow volatility effect is marginally significant, while the asset volatility effect is insignificant. A further analysis suggests that the insignificance is likely due to the positive correlation of E[ Asset Vol] with BL(acq ind) and BL(pf acq). When the acquirer s asset volatility is low while the target s asset volatility is high, E[ Asset Vol] is large and positive, provided that the relative size of the target is significant. At the same time, BL(acq ind) and BL(pf acq) are also likely to be high in this case, as the acquierer s leverage is likely to be high due to its low asset volatility. Such correlations among the explanatory variables make the regression results less stable, unlike the results based on return correlation. If we exclude BL(acq ind) and BL(pf acq) from the regression, the coefficient of E[ Asset Vol] becomes significantly negative at the 5% level (see column (1) of Table 5). The results suggest that the firm s leverage response to changes in asset volatility is economically large. Given a standard deviation of asset volatility of acquiring firms of 0.43, the coefficient estimate in column (4) implies a one standard deviation decline in asset volatility leads to a 10.2 percentage point, or 0.55 standard deviation, increase in leverage. Because we perform the analysis using pre-merger predicted change in asset volatility that do not rely on post-merger information about leverage, this avoids concerns these results are driven by measurement error in asset volatility. In addition, the fact that the three-year effect is stronger than the one-year effect further suggests that our results are not driven by measurement error, method of payment of the merger deal, or debt passively inherited from the target. 4.2 Changes in Cash Ratio We explore in this section the impact of predicted changes in asset volatility on cash holdings. As in the last section, we first use the pre-announcement return correlation as a proxy for the predicted change in asset volatility. The model specifications mimic those in the leverage 19

21 regressions. We start with the simplest specification, in which the cash ratio change from the preannouncement quarter to the end of the fourth post-merger quarter is regressed on the return correlation Rho and year fixed effects. The results are in column (1) of Table 6, which exhibit a strong positive effect of return correlation on cash changes. The coefficient is 0.073, significant at the 1% level. This coefficient estimate suggests that for mergers with return correlation one standard deviation above the average, the post-merger change in cash ratio is 1.36 percentage points higher. This is strong evidence that firms adjust cash reserves in response to the anticipated changes in post-merger asset volatility. In model (2), we control for a variety of firm characteristics that may change during the merger event window. The effect of correlation becomes slightly stronger. The effects of the control variables are largely as expected. For example, an increase in sales negatively affects cash holdings, consistent with larger firms, which face less financial constraints than small firms, holding proportionally less cash. The market-to-book effect is positive, consistent with the idea that firms with more growth opportunities hold more cash to avoid financial constraints when investment opportunities arise. Asset tangibility reduces cash holdings, suggesting firms with more tangible assets require less precautionary savings because they are better able to obtain outside financing when needed. In model (3), we add two additional control variables, the pre-announcement industryadjusted cash ratio of the acquirer, and the deviation of the pro forma cash ratio of the merged firm from that of the acquirer. Again, the idea here is to control for potential mean reversion in cash holdings, and slow adjustment away from the cash ratio inherited from the balance sheets of the acquirer and the target. Although these two control variables are highly significant as expected, they do not significantly affect the coefficient on return correlation. In fact, the correlation effect becomes slightly stronger. In model (4), we add the change in cash flow volatility as a control variable. Somewhat surprisingly, this variable has virtually no effect, most likely because it contains large measurement errors, or because its variation has been captured by the change in asset volatility 20