Real estate collateral, debt financing, and product market outcomes

Transcription

1 Real estate collateral, debt financing, and product market outcomes Aziz Alimov * City University of Hong Kong May 15, 2014 Abstract How does debt financing affect product market outcomes? This paper exploits variations in the market value of firms real estate collateral caused by movements in local real estate prices to identify changes in firms ability to borrow ex ante. Using these changes, I show that debt financing significantly impacts competitive outcomes. A one standard deviation increase in the market value of a typical firm s collateral leads to a 6.5% increase in the firm s product market share. The results are robust to alternative estimation methods as well as instrumenting real estate price growth with local housing supply elasticity. Increases in collateral value lead to less competitive markets, as characterized by higher market concentration and fewer product rivals. Firms with appreciating collateral tend to gain market share at the expense of their competitors by spending more on capital expenditures, R&D, advertising, and acquisitions. Competitive effects of collateral are positively valued by the market. Overall, these results indicate that variations in firms collateral value and thus their ability to raise debt play an important role in shaping product market competition. Preliminary draft-comments welcome * I thank Jarrad Harford for helpful comments. aaalimov@cityu.edu.hk; telephone:

2 1. Introduction The role of a firm s mode of financing- in particular, its use of debt- in its product market performance has long been an important question to financial economists. A large theoretical literature formalizes the ways a firm s debt financing policy can influence its competitive conduct but reaches conflicting predictions as to whether debt motivates the firm to be a more or less aggressive competitor. For example, in a seminal paper, Brander and Lewis (1986) show that access to debt financing can create ex-ante incentives for a firm to compete more aggressively in the product market, for example, by altering its production to gain a competitive advantage over rival firms. In contrast, Chevalier and Scharfstein (1996) demonstrate that a reliance on debt financing commits the firm to compete less aggressively and thus leads to a less competitive outcome. Bolton and Scharfstein s (1990) theory of predation further suggest that firms with high reliance on debt financing are vulnerable to predation by the rival firms that have greater access to internal funds. Therefore, it is an on-going theoretical debate on the exact interplay between a firm s use of debt and its competitive performance. To study this issue empirically, researchers have examined the relation between a firm s chosen amount of debt and several different aspects of the firm s product market behavior including pricing, investment, advertising, and overall market share (e.g. Chevalier 1995; Phillips 1995; Zingales 1998; Grullon et al. 2006; Campello 2003 and 2006). Overall, this literature shows that higher levels of debt that a firm decided to take tend to adversely affect the firm s ability to compete in the product market. While these studies are carefully executed and generated many interesting findings, they tend to face empirical challenge of identifying the causal effect of debt financing owing to a variety of endogeneity problems (Parsons and Titman, 1

3 2007). One vexing issue, in particular, is the possibility that both a firm s debt financing policy and its competitive strategy may be jointly influenced by unobserved factors arising from the product market environment. Thus, if omitted variables (such as a firm or industry-specific change in demand) simultaneously cause the firm to choose a certain financing policy and change its competitive conduct, then a positive correlation between leverage and product market outcome is not reflective of a causal relation. This article addresses this identification challenge by exploiting variations in the market value of real estate holdings due to movements in local real estate prices to identify the causal effect of changes in the firm s ability to borrow ex ante on its product market behavior. The focus on corporate real estate holdings offers a promising research environment to study the competitive effects of firm debt financing policy for several reasons. Theoretical studies suggest that improvements in collateral values such as real estate assets can boost firms borrowing capacity (see, for example, Barro 1976; Stiglitz and Weiss 1981; Hart and Moore 1994). This collateral lending channel builds on the idea that financing frictions stemming from moral hazard and adverse selection problems between lenders and borrowers can be alleviated by pledging more valuable collateral. Rampini and Viswanathan (2013) develop a model that shows that collateral is a first-order determinant of firm debt capacity and financing choices. Consistent with the theory, Chaney, Sraer and Thesmar (2012) show that changes in the market value of firms real estate holdings positively affect their propensity to raise debt relative to equity and lead to higher capital expenditures. Gan (2007) and Benmelech and Bergman (2009)) also show that collateral value affects debt financing decisions and the cost of debt. Given the significant body of evidence on the important role of collateral in raising debt finance, the premise of this 2

4 paper is that an increase in the value of a firm s real estate assets will increase the firm s ability to raise more debt ex-ante at lower cost. This paper confirms that I closely follow Chaney et al. (2012) and identify changes in the market value of real estate assets owned by firms caused by variations in real estate prices either at the state or Metropolitan Statistical Areas (MSA) level. My identification strategy exploits the cross-sectional and timeseries variation in the value of real estate assets across and within firms to isolate the effect of changes in collateral values (and thus debt financing policy) on firms product market performance. More specifically, I use a generalized version of a difference-in-differences regression model with firm and year fixed effects that compares performance across firms experiencing differential changes in their real estate values (where the treatment is continuous). A key advantage of this identification strategy is that it not only captures variation in exogenous shocks to debt financing capacity, but also solves the omitted variables concern by allowing for multiple shocks to different firms at different times and locations. I measure product market performance using relative-to-rivals market share growth and define a set of product market rivals according to the new Text-based Network Industry Classification (TNIC) developed by Hoberg and Phillips (2010 and 2014). TNIC industry defines peers for each firm based on the similarity in firms' product description in their annual 10Ks. Using a panel data of U.S. public firms from 1996 to 2008, I find that the fluctuations in firms collateral value have a significant impact on product market outcomes: The typical firm experiencing positive shocks to the market value of its collateral subsequently gains market share at the expense of their industry rivals. The economic magnitude of the estimated effect is substantial: A one-standard deviation increase in the average firm s real estate value leads to a 3

5 6.5% increase in the firm s market share, which corresponds to about 13% of the market share growth s sample standard deviation. The positive relation between real estate value and product market performance is robust to various estimation methodologies as well as to alternative definitions of product market peers. To address endogeneity concerns and to identify the causal effect of higher real estate prices, I instrument for the growth in real estate prices using the interaction of nationwide interest rate and MSA-level housing supply elasticity measure developed by Saiz (2010). This measure exploits geographic and regulatory constraints to local housing supply to differentiate MSAs where a decrease in nationwide interest rates (and thus increase in housing demand) translates into higher real estate prices or into higher construction volume. As such, this measure is exogenous to an individual firm s financing and product market choices. The I.V. regressions confirm that the typical firm located in areas with rising real estate prices (and thus increasing collateral value) expands its product market share. This competitive effect is also present in manufacturing sector, where the results are less likely to be driven by local demand. The paper then asks whether these competitive effects of real estate collateral-backed debt financing vary across firms and product markets in ways that are consistent with the theoretical models. For example, Brander and Lewis (1986) suggest that the competitive effects of debt financing should be greater in markets with higher firm concentration, where firms stand to gain more by implementing more aggressive strategies. The results are indeed more pronounced in more concentrated industries, especially for nonleader firms. I next study more specific strategies that firms might implement given shocks to the value of their collateral and find that firms appear to gain competitive advantage by spending more on capital expenditures and acquisitions than their rivals. Taken together, the results in this paper 4

6 consistently indicate that a typical U.S. firm uses the enhanced debt financing capacity caused by the appreciation of its real estate collateral to gain competitive advantage and expand market share at the expense of its product rivals. My evidence therefore is more consistent with the debt makes aggressive competitor theories, such as Brander and Lewis (1986). Given these finding, it is natural to ask whether changes in real estate value influence not only industry firms competitive performance but also the overall industry structure. I find that they do: both the concentration and the number of distinct rivals in a given industry decrease following positive shocks to the typical firm s collateral value. The final part of the analysis examines the impact of collateral value on two measures of firm value. The results show that firms whose real estate assets appreciated subsequently experience increases in their overall market value in comparison with their industry rivals, and more specifically, in market value around acquisition announcement dates. This set of results suggests that the competitive effects of real estate collateral are value enhancing (as perceived by the market). This paper contributes to two strands of empirical finance literature. It primarily contributes to the research on the interplay between firms debt financing decisions and their product market conduct (Chevalier (1995), Phillips (1995), Zingales (1998), Campello (2003, 2006)). The overall finding in these studies is that chosen levels of debt subsequently hurts firm product market performance. In contrast, this paper identifies plausibly exogenous shocks to a firm s exante ability to raise debt capital and finds that greater ability to use debt boosts the firm s performance in the product market. My findings on the product market effects of real estate collateral value also add to a growing body of evidence on the important role that collateral plays in determining firms debt 5

7 financing, investment and employment policies (Gan (2007), Chaney et al. (2012), Benmelech et.al. (2005), Benmelech and Bergman (2009, 2011)), and Adelino et al. (2014)). 2. Theory and evidence on debt financing-product markets interaction In general, a firm s decision to issue debt can affect product market competition because debt changes the ownership of cash flows and the firm s ability and incentives to invest in or protect market share. In response, rival firms may also change their product market conduct. In this section, I briefly review relevant theoretical studies that model how firms debt financing policies affect product market outcomes and the empirical studies that tested the implications of these models. For more detailed surveys of the theoretical and empirical literature, I refer the interested reader to Maksimovic (1995) and Parsons and Titman (2007), respectively. 2.1 Theory Brander and Lewis (1986) is the seminal study to explore the link between debt financing and competition in the product market. Their model shows that, in oligopolistic markets, issuing risky debt commits a firm to be more aggressive competitor, and this aggressive market conduct may confer a strategic advantage to the firm at the expense of its product market rivals. This advantageous effect of debt financing on product market strategy arises due to the limited liability enjoyed by shareholders and their option-like payoff structure: higher debt leads to a strategy that raises returns in good states and lowers returns in bad states. An alternative line of argument suggests that excessive dependence on debt financing can put a firm in a financially fragile position, thereby hindering the firm s ability to compete successfully in the product market. For example, Bolton and Scharfstein s (1990) deep pockets theory of predation suggests that a high reliance on debt financing can be perceived as a sign of 6

8 weakness in product markets and thereby encourage cash-rich ( deep pockets ) rivals to pursue predatory market strategies. Such predatory behavior is predicted to be stronger in concentrated industries, where predators stand to gain more market share when they drive a weaker competitor out of the business. The implication of the Bolton and Scharfstein s model is that firms should use less debt and hoard more cash so that they are better able to fend off potential predatory behavior. This prediction is the opposite of that in Brander and Lewis (1986). Finally, the models in Chevalier and Scharfstein (1996) and Dasgupta and Titman (1998) show that reliance on debt financing reduces a firm s incentive to invest in market share. This is because higher debt increases the prospect of default, thereby leading firms to discount the benefits of investing in future market shares, especially during economic downturns. Chevalier and Scharfstein predict that if most firms in an industry are externally financed, then they increase their prices (markups) more during recessions rather than expansions. To sum, using different assumptions about the nature of competitive interactions, theoretical models suggest that use of debt capital can make a firm either more or less aggressive competitor. Hence, the role of debt financing in product markets is an empirical question. 2.2 Empirical evidence There have been several empirical studies linking a firm's product market behavior and its debt financing policy. Most of these studies rely on some experiments involving adverse shocks to either a rival firm s financial structure or macroeconomic shocks. When interpreting evidence on this issue, it is worth noting that these studies examine how chosen debt amount and thus leverage ratios affect firm competitive strategy, especially against rivals with lower leverage. For example, Phillips (1995) investigates firms' competitive responses to leverage recapitalizations in four industries. Zingales (1998) examines the survival of trucking firms following the industry 7

9 deregulation. Chevalier (1995a, 1995b) studies stock market reaction for firms in the supermarket industry when a rival firm undertook an LBO. She and Chevalier and Scharfstein (1996) also analyze the prices at local supermarkets after leveraged buyouts, and Kanna and Tice (2000) examine how department retailers respond when Wal-Mart enters their market thereby changing their competitive environment. In general, all these studies find that high leverage leads to poor performance in the product market, such as relative-to-rivals sales and market share decline, lower product price, inability to aggressively respond to new entry. Overall, these conclude that a high amount of debt (leverage ratio) appears to constrain aggressive product market conduct. More recently, Campello (2003 and 2006) has studied the interactions between leverage and competitive outcome for a large panel of Compustat firms 1. In 2003 paper, he finds that leverage has a negative impact on market share growth during recessions in industries where rivals have low leverage. In 2006 paper, he finds that the relation between debt amount and product market performance is not uniform over the range of leverage: while high levels of debt is associated with market share losses, a moderate level of debt is associated with market share gains. While all these studies are carefully executed and use creative methodologies, nevertheless, as summarized by Parsons and Titman (2007) virtually all empirical studies that attempt to shed light on the connection between capital structure and a firm s corporate strategy potentially suffer from significant endogeneity problems. In particular, since firms tend to choose leverage ratio and product market strategies simultaneously, studies may assign a spurious causality to financial structure if both performance and debt financing decisions are influenced by underlying trends in market conditions, such as excess capacity or demand growth. 1 Frezard (2010) examines the effects of cash holdings on firm performance in product markets. 8

10 To solve this identification challenge, I exploit exogenous shocks to the market value of a firm's real estate assets that can boost the firm s ability to use debt financing. These shocks are caused by fluctuations in local or nationwide real estate prices and thus are exogenous to any individual firm and are separate from changes in industry-wide conditions. Accordingly, this setting allows me to plausibly identify the causal effects of firms increased ex-ante ability to use debt financing (collateralized borrowing) on product market outcomes Data and variable construction 3.1 Sample construction and definition of product market competitors The intimal sample includes all U.S.-headquartered firms covered by Compustat between 1996 and 2008 with non-missing book assets and total sales. Following Chaney, Sraer, and Thesmar (2012), I require that the firm was active in 1993 because this was the last year when firms were required to report data on accumulated depreciation on buildings, used to estimate the current value of real estate. I exclude firms in financial, real estate, and construction industries All financial ratios are winsorized at the 1 and 99 percentiles, to reduce the impact of outliers. Defining the relevant industry product markets and thus a set of likely competitors for the sample firms is central to this study. My main measure of product markets is based on the new Text-based Network Industry Classification (TNIC) developed by Hoberg and Phillips (2010, 2014) 3, which assigns a set of industry competitors to a given firm based on mutual product similarity. Specifically, TNIC is based on textual analysis of the product description sections in firms' annual 10-Ks filed with the Securities and Exchange Commission (SEC). Using these 2 This study shares its identification spirit with Hadlock and Sonti (2012), who study exogenous shocks to a firm s liability structure arising from asbestos litigation and find that exogenous increases (decreases) in asbestos liabilities are interpreted by as negative (positive) news for a firm s close competitors. 3 Available on 9

11 product descriptions, the authors compute a measure of product similarity for every pair of firms based on the number of common words that the two firms use to describe their products supplied to the market. Hoberg and Phillips then define each firm s industry product market (with the same granularity as the three-digit SIC codes) as a set of distinct competitors with whom the firm has the product similarity above a certain threshold. In other words, for each firm, the TNIC is defined as a set of peer firms that use similar words to describe their products and therefore are likely to compete in the same product market. Hoberg and Phillips argue that TNIC has three advantages over the standardized industry classification (SIC or NAICS) that the product market literature has traditionally used to define industry boundaries. First, unlike relatively static and rigid SIC and NAICS codes, a firm s TNIC can change over time when the firm modifies its products range or enters a new product market and thus faces a different set of competing firms. Second, TNICs are based on the products that a firm supplies to the market and not on the production processes which is the case for SIC. Third, unlike SIC and NAICS that impose transitivity between two firms that compete with the third one but may not necessarily compete against each other, TNICs defines a distinct set of industry peers relative to each firm. As such, TNIC are more likely to capture continuous measures of product market similarity and relatedness both within and across industries. Hoberg and Phillips (2014) provide evidence that TNIC-based metrics offer an economically significant improvement in explaining competitive environment and firm strategies to achieve product market differentiation than the SIC and NAICS industry-based metrics. In Section 4.2, I also employ, for robustness, alternative industry definitions that rely on traditional SIC-based industry classifications and obtain qualitatively similar results. 10

12 3.2 Measuring product market performance Following Campello (2003 and 2006), I measure a given firm s performance in the product market using a its product market share growth. More specifically, I use a firm s sales growth from year t-1 to year t in excess of that of its TNIC-industry rivals in a given year; this metric thus roughly gauges expansion or loss of a firm s product market share at the expense of competing firms. Campello notes that the advantage of this performance measure over other metrics is that it can be consistently estimated across many industries and years and it summarizes information from the combined effects of a number of strategies that firms use to influence their competitive position, such as decisions about spending on capital outlays, acquisitions, advertising and marketing, research and development, logistics, product pricing etc. The product market share growth variable is computed only for those firm-years that have a minimum of four distinct peer firms in its TNIC industry. This selection procedure results a final sample of 14,474 firm-year observations. As shown in the summary statistics reported by Table 1, the average unadjusted and industry-adjusted sales growth (market share growth) in my sample is about 12.8% and -15.1%, respectively. 3.3 Market value of real estate My empirical strategy is aimed at estimating the effect of the market value of a firm's real estate assets that can be pledged as collateral for debt financing on its product market conduct. Hence, the key variable in this study is the value of a firm's real estate collateral. I closely follow the procedure in Chaney et al. (2012) and calculate the market value of real estate assets for a firm i in location s in a year t (RE VALUE) as a function of the amount of real estate assets the firm owned in 1993 inflated by the state or metropolitan statistical area (MSA) -level real estate price growth from 1993 to year t, and scaled by lagged total assets 11

13 RE VALUE i,s,t = (RE ASSETS i,1993 PRICE INDEX s,1993 t )/Total Assets i,t-1 Where the value of real estate assets of a firm i in location j in 1993 is computed by inflating their purchase price (historical costs) with state or metropolitan statistical area (MSA) -level real estate price growth from estimated purchase year to 1993: RE ASSETS i,s,1993 = (Historical Real estate assets) i,1993 REprice s,1993 /REprice s,1993 Age i Historical real estate assets are defined as the sum buildings, land and improvement, and construction in progress, all of which are valued at historical cost on Compustat. To compute the average year of the acquisition of the real estate assets, I calculate the average age of all real estate assets by taking the ratio of the accumulated depreciation of buildings to the historic cost of buildings and multiplying by the assumed average depreciable life of these assets 40 years. Because COMPUSTAT provides only the geographic location of the firm headquarters (STATE, COUNTY and ZIP CODE), following Chaney et al., this paper measures the value of a firm s real assets using real estate prices in the state or MSA where the firm s headquarters is located. Figure 1 show the variation in MSA-level real estate price growth for selected five metropolitan areas-atlanta, Cincinnati, Dallas, Minneapolis and San Francisco- that my empirical analysis exploits. While we observe trending in real estate prices in all five cities, there is a distinct difference in the change in real estate prices across the cities. As in Chaney et al., this paper does not consider the value of any real estate acquisitions after 1993 and studies only the effect of variation in the value of real estate assets held in 1993 driven only by variation in real estate prices. The advantage of ignoring post-1993 real estate purchases is that it helps mitigate any potential endogeneity between changes in firms real estate holdings and their investment decisions. Following these authors, I perform several sensitivity 12

14 tests to assess the effect of relaxing this requirement (untabulated), such as including firms that acquired some or disposed all real estate assets after 1993 and reached very similar conclusions. Table 1 presents summary statistics for the market value of the firm s real estate holdings and other variables used in study. The average RE value calculated using state-level real estate price index is 29.7 percent of lagged assets and the median is 13.7 percent. The industry-year mean adjusted RE value has a mean and median very close to zero. 3.4 Empirical strategy The empirical specification exploits the cross-sectional and time-series variation in the value of real estate assets across and within firms to isolate the effect of changes in collateral values (and thus debt financing capacity) on product market performance. The formal specification is: ΔMarket Share i,t = α+ β 1 RE Value i,s,t-1 + β 2 Price Index s,t-1 +XControls i,t-1 +Firm and Year Fixed Effects +ε i,t where i indexes firms, t indexes years, and s indexes the state or MSA of the firm s headquarter. The dependent variable is a firm s sales growth from year t-1 to year t minus its TNIC industry-year average, so that this variable measures a firm s sales growth relative to that of its competitors in its product space. RE Value i,s,t-1 is the market value of the firm s real estate assets in year t-1 based on the state or MSA-level price index, is the key variable of interest. Price Index s,t is a control for the real estate price index at the state or MSA level from 1993 to year t-1. This variable, in conjunction with year fixed effects, controls for changes that affect the nationwide and local state or MSAlevel economy as a whole. The vector X includes a set of firm-specific one-year lagged control variables that may influence product market performance, as used by Campello (2006) and Frezard (2010). These 13

15 firm-specific factors are: Size (log of firm book assets), z-leverage (ratio of total debt to total assets, divided by industry-year standard deviation), z-cash (cash-to-assets scaled by industryyear standard deviation), profitability (EBITDA-to-assets), market-to-book ratio, and past market share growth. All independent variables are industry adjusted by removing their industry-year means. 4 Recognizing the selection bias noted above when computing real estate values in 1993, I calculate the industry means using all firms in the TNIC and COMPUSTAT databases. Finally, all specifications control for firm and year fixed effects to mitigate the concern that unobservable variables omitted from Eq. (1) that affect the value of a firm s collateral value might be correlated with the firm s market share growth. The inclusion of firm fixed effects removes the firm-specific means, which allows us to interpret β 1 as the average sensitivity of the change in market share to changes in collateral value. All of the standard errors of the estimated coefficients from all regressions are clustered at the firm level. 4. Results 4.1 Does real estate collateral affect debt financing? At the basis of my empirical strategy is the idea that a firm's collateral value is a central determinant of its debt financing policy (e.g. Rampini and Viswanathan 2013). Therefore, before proceeding to main tests, it is necessary to verify the validity of this argument. Table 2 reports the results of the regressions relating fluctuations in a firm s real estate collateral value on its relative-to-industry rivals debt and equity financing policy and leverage ratios. The regression 4 As noted by Campello (2006), purging industry-year specific effects from the firm-specific characteristics minimizes the concern of spurious correlation driven by unobservable industry-related factors. 14

16 structure is identical to Equation, except that debt and equity issuance and leverage ratios replace market share growth as the dependent variable. The results in Column 1 show that the typical firm indeed issues more debt than its industry peers when the value of its real estate assets appreciates. The coefficient on RE Value of is positive with a p-value below This implies that an increase in real estate value by one standard deviation (0.454) raises relative-to-rivals debt issuance (as a percentage of total assets) by 1.8 percentage points. Given that the mean industry-adjusted debt issuance is -0.2% and its standard deviation is 9.3%, the effect of the change in real estate collateral value on debt financing is economically significant. Column 2 of the table studies the effect of real estate collateral value on equity issuance. In contrast to debt issuance, the typical firm does not raise more equity capital when its real estate assets appreciate in value. Interestingly, however, the results in Columns 3 and 4 show that increased borrowing following a positive shock to real estate value does not translate into higher industry-adjusted book or market leverage ratio on average. In fact, I find that market leverage ratio, on average, decreases following positive shocks to firms real estate value. Potential explanation for this finding is provided by the findings in Section 4.6, which shows that the new debt proceeds (and potentially other sources of funds) appear to be used to fund new capital investment and acquisitions that increase book assets size, thereby potentially leaving the book leverage ratio unchanged. In addition, the results in 5.1 indicate that positive shocks to the average firm s real estate value substantially raise the market value of its assets, which in turn may lower its market leverage ratio. Finally, to complete the picture, Column 5 studies the effect of the change in a firm s real estate value on the cost of debt financing. Specifically, I examining the effect on the bank loan 15

17 interest rate, which is measured as the natural logarithm of the All In Drawn loan spread that the borrower pays in basis points over LIBOR (London Interbank Offered Rate) or its equivalent. In addition to firm characteristics, the regression also controls for several loan characteristics that might be correlated with the price of debt (see, for example, Alimov 2012). The specification includes the natural logarithms of loan maturity and the dollar amount of a loan, and a set of binary indicators for whether the loan is secured, loan type, and loan purpose. The bank loan data comes from DealScan. The point estimate on RE Value in Column 5 is negative with a p-value of This indicates that the typical borrower experiencing an increase in the market value of its real estate assets subsequently is able to obtain bank loans at relatively lower price. In summary, the results confirm that exogenous increases in the market value of real estate collateral make it easier for firms to raise more debt capital (at lower cost) than their product market rivals. I thus conclude that shocks to real estate value represent a promising research setting to study causal effects of debt financing on product market outcomes. 4.2 Baseline results Table 3 reports the coefficients from estimating different variants of Eq. (1). Column 1 reports the results of the regression that relates the changes in firms real estate collateral value calculated using the state real estate price index (RE Value State ) and the state real estate price index (RE Price State ) to firms market share growth, without any of controls. The coefficient on RE Value is positive and statistically significant at the 1% level, suggesting that firms experiencing positive real estate shocks, on average, expand their product market share relative to industry peers. In terms of economic magnitude, all else equal, a one-standard deviation (0.454 from Table 1) increase in the typical firm s real estate value in relation to rivals in year t-1 16

18 leads to a 6.5% gain in market share between years t-1 and t. This effect corresponds to about 13% of a market share growth s standard deviation (0.505) 5. Therefore, the effect of real estate price shocks on product market outcomes (as measured by the growth in sales relative to industry peers) is both statistically and economically significant. The regressions in Columns 2 and 3 of Table 3 gradually include firm-level characteristics that could influence changes in the average firm s market share growth. These variables include market share growth in the prior fiscal year, beginning-of-year log of assets, leverage and cash ratio, as well as the market-to-book ratio and profitability. The results in both columns show that after controlling for firm characteristics, the effect of RE Value on market share growth is virtually unchanged and it continues to be positive and highly significant. Among the controls for firm-characteristics, we can observe that a firm s leverage ratio (zscore) has no effect on its product market share growth 6. This result appears to be inconsistent with prior work which concludes that leverage tends to adversely affects competitive position. However, as noted before, the leverage ratio is endogenous as firms are likely to choose the amount of debt and their corporate strategy simultaneously. Similar to Frezard (2012), I find that having more cash than industry rivals positively associates with higher future market share growth and that prior change in market share is negatively associated with subsequent market share growth. Finally, the results also indicate that, on average, firms with more growth opportunities (as characterized by higher market-to-book ratio) and less profitable firms tend to expand their market shares on average. 5 Because all regressions in this paper include firm fixed effects, all coefficients should be interpreted as the change from each firm s sample average. 6 Using a simple leverage (not z-score) or lagging it two years does not change the results. 17

19 One interesting observation is that the addition of all these other controls for firmcharacteristics does not affect the R-squared of regressions much. The R-squared in Column 1, which includes only the two real estate measures as the independent variables, is and it increases only slightly to in Column 2 which includes the z-scores of leverage and cash holdings. This result potentially implies that the contribution of the market value of real estate collateral to the variation in market share growth is significantly larger than those of other variables. Some readers may wonder whether, given that all other variables are measured relative to their industry-year means, it is necessary to industry-year adjust the main independent variable of this study as well. A related concern could be that the change in real estate prices in some areas induces a local demand shock that especially affects firms in certain product markets that are concentrated in those areas (such as computer programing industry that is concentrated in the San Francisco metropolitan area). Note, however, that industry-year adjusting RE Value may introduce sample (survivorship) bias and additional measurement noise due to the data requirements for the calculation of the value of real estate holdings as of Nevertheless, in Model 4 of Table 3 I address this concern by replacing RE Value and RE Price Index measures with their industry-year adjusted values. The conclusions are unchanged: the coefficient on the adjusted RE Value variable in Column 4 is positive with a p-value of In terms of economic magnitude, all else equal, a one-standard deviation (0.422) increase in the average firm s real estate value in relation to its rivals in year t-1 leads to a 3.7% gain in market share between years t-1 and t. Finally, in Column 5, I measure the RE Value using the MSA-level real estate price index. Chaney et al. (2012) argue that using MSA-level instead of state-level real estate prices 18

20 potentially offers a more accurate measure of variation in the market value of a given firm s real estate holdings. The use of MSA-level prices is consistent with the instrumental variable strategy used in the next section, where it identifies the causal effect of real estate prices on by instrumenting real estate price growth with the measure of housing supply elasticity of Saiz (2010), which varies at the MSA level. However, the use of MSA-level prices requires an assumption of all of firm s the real estate assets are located in that metro area and thus may increase measurement error. As we can see in Column 5, the results show the value of a firm s real estate assets calculated using MSA prices is positively and significantly related to the firm s market share growth: the coefficient on RE Value is positive and has a p-value of less than Identification issues and Robustness Tests The evidence in the prior section shows a strong positive effect of the value of a firm s real estate collateral on its product market performance. In this section, I address remaining concerns regarding endogeneity and the definition of relevant industry product markets Endogeneity issues As in Chaney et al. (2012), two sources of endogeneity might affect the results in this paper: 1) the decision to own real estate is not necessarily random and could be correlated with the firm s growth and financing opportunities, and 2) real estate price shocks are picking up some other unobserved local economic shocks, and that real estate-holding firms are simply more sensitive to those shocks. The first source of endogeneity implies that if firms that decided to own more real estate are also the firms whose financing and product market strategies are more sensitive to state or MSA- 19

21 level demand shocks, then one would find a spurious correlation between changes in real estate prices and firms market share growth. This means that the analysis needs to ensure that any changes in firm product market decisions in response to increases in the value of its real estate assets come only through the collateral channel. To this end, I follow Chaney et al. (2012) and control for the observable determinants in the real estate ownership decision- firm age, book assets size, and profitability- all interacted with contemporaneous real estate prices. Channey et al. suggest that if those firm characteristics are related to both firm decision to own real estate in 1993 and firm sensitivity to subsequent fluctuations in real estate prices, then controlling for the interaction between those firm characteristics and the real estate prices allows one to more accurately identify the effect of collateral. Column 1 of Table 4 shows the results of the regression model described by Eq. (1), adding the interactions of initial (pre-1993) characteristics of firms with contemporaneous state-level real estate prices that control for a firm s decision to own real estate and its sensitivity to fluctuations in real estate prices. The results are unchanged: estimated coefficient on RE Value is positive with a p-value of less than The second source of endogeneity implies that if an unobserved positive economic shock impacts both local real estate prices and firm financing and growth opportunities, this would bias the estimates because this unobservable shock might increase both real estate prices and firm market expansion opportunities. To address this source of endogeneity, I follow Channey et al. (2012) and instrument the MSA-level real estate prices using the interaction of local housing elasticity provided by Saiz (2010), interacted with the aggregate real interest rate as in Himmelberg et al. (2005). 20

22 The rationale for this instrumental variables (I.V.) strategy is that shifts in the interest rate affect real estate prices differently for locations with different land supply elasticities. In general, when interest rates and thus mortgage rates decrease, the demand for housing and other real estate increases. If the supply of land in a given MSA location is relatively plentiful and this is very elastic, then the increased demand will translate mostly into more construction of real estate rather than higher real estate prices. In contrast, if the supply of land is limited and thus very inelastic, then the increased demand will translate mostly into higher real estate prices rather than more construction. Thus, the change in the aggregate interest rate should have a larger impact on the real estate prices in the MSA locations with lower land supply elasticity. For example, the San Francisco metro area has a very low elasticity of 0.66 and, as Figure 1 shows, it experienced a drastic boom in the real estate prices in the 1990s. In contrast, Indianapolis has a high elasticity of 4.0 and it experienced only moderate real estate price appreciation. Overall, the use of this instrument should thus provide variation in real estate prices that is not correlated with omitted local economic shocks. In the first-stage of the I.V. regression, I predict the MSA-level real estate prices using the interaction of local housing elasticity provided by Saiz (2010), interacted with the aggregate real interest rate as in Himmelberg et al. (2005). RE Price msa,t =α +ρelasticity msa Interest Rate t + γinterestrate t +MSA and Time FE +ε i,t (2) The elasticity measure is available for 95 MSAs (with population over 500,000 in the 2000 Census) which reduces the sample to 9600 firm-year observations. The results of the first-stage regression (not reported) are consistent with expectation and show that the interaction of MSA-level housing supply elasticity and interest rate has a positive and significant effect on local real estate price level. Column 2 of Table 4 report the estimation results of the second- 21

23 stage regression, where RE Value is based on the predicted RE Price from the first stage. When instrumented, the effect of the real estate collateral value on firm market share growth not only remains positive and significant but increases in magnitude: the coefficient is with a p- value of less than This estimate indicates that a one standard deviation increase in instrumented RE Value (0.45 in Table 1) is associated with a 8.2 percentage point increase in market share. This finding also implies that the non-instrumented results in Table 3, which are economically and statistically significant, are likely understating the true effect of real estate value on product market outcomes. In Column 3 of Table 4 I further tighten the identification by limiting the sample to firms in the manufacturing industries (SIC ). As noted by Adelino et al. (2014 ), the manufacturing industries the least likely to be affected by local demand conditions. In addition, firms in these industries typically have significant amount of other types of fixed assets, which makes it harder to find the effect of shocks to their real estate collateral. The coefficient of interest drops only slightly to and remains significant at the 1 percent level. In unreported results, I also go one step further and remove any unobserved economic shocks in the firm's location by using MSA or state-year fixed effects instead of only year fixed effects. The results are very similar to the specifications that use firm and year fixed effects Alternative definitions of industry peers As discussed above, the new text-based network industry classification (TNIC) developed by Hoberg and Phillips (2010) offers several advantages over relatively static SIC and NAICSbased industry classification that has been in the prior product market studies (e.g. Campello 2003 and 2006; Frezard 2010; Girourd and Mueller 2010). Nevertheless, to show robustness of my estimates to alternative product market definition, I replicate the regression in Table 3 by 22

24 assigning firms to industries based on their (historical) three and four-digit SIC codes. The results reported in Columns 4 and 5 of Table 4 are similar to those in Table 3. In both regression specifications, RE Value enters with a positive and highly significant coefficient. In sum, the results in Tables 3 and 4 offer consistent and robust evidence of plausibly causal effect of shocks to firms real estate collateral value on product market outcomes. 4.4 The Effect of Industry Characteristics In this section, I explore the specific industry characteristics that could determine the circumstances under which an increased access to collateralized debt financing matters more. For example, Brander and Lewis (1986) and Bolton and Scharfstein (1990) indicate that the strategic benefits of debt financing apply essentially to industries with higher concentration, where the gains associated with removing a weaker competitor are the greatest. In contrast, such market share effects might be practically non-existent in highly competitive industries where all firms tend to be price-takers and predation risk is low. Kovenock and Phillips (1997) and Zingales (1998) empirically show that strategic interactions are indeed more prevalent in highly concentrated industries. In addition, Bolton and Scharfstein suggests that increased access to debt financing would bring greater product market benefits to firms that have incentives to prey on weaker rivals, such firms with large market share ( market leaders ). In this section, I ask whether the competitive effects of real estate value shocks varies with the degree of market concentration and between leader and nonleader firms. I classify as market leaders those firm-years whose sales are in the top quartile of their industry-year. I classify firms into concentrated and nonconcentrated markets based on their concentration ratio (Herfindahl Index or HHI) in their TNIC in 1996, the first year TNIC HHI is available. This concentration classification is not allowed to vary over time since, as shown in 23

25 Section 4.6, changes in industry concentration are endogenous. Firms in the highest HHI tercile are assigned to concentrated, and those in the bottom tercile to nonconcentrated markets. Column 1 of Table 5 examines the differences in the competitive effects of real estate collateral between market leaders and other firms. The regression specification modifies Eq.(1) by including Market Leader dummy variable and two additional interaction terms between Market Leader and RE Value and RE Prices variables. The results show that the coefficient on RE Value remains positive and highly significant, indicating that a typical non-market leader firm expands its product market share when the market value of its real estate assets appreciates. The coefficient estimate on the interactive variable Market Leader x RE Value is also positive and significant at better than 10%, indicating that the positive relation between increases in real estate value and market share growth is more pronounced among market leaders. Columns 2 through 4 of the table contrast the competitive effects of real estate collateral between market leaders and nonleaders across industry concentration terciles. The main finding here is that shocks to real estate value are more important for nonleader firms in highly concentrated industries, both statistically and economically, than for nonleader firms in nonconcentrated industries. At the same time, market leadership does not appear to significantly alter the relation between collateral value and market performance in any of the concentration terciles. I also estimated all the regressions reported in Table 5 using instrumented real estate prices (using the interaction of interest rates and local constraints on land supply as described in Section 4.3.1). The results are essentially the same and, in the interest of conserving space, not reported. In summary, the results in this section provide some support for the proposition in theoretical models that industry concentration and to a lesser degree market leadership are the 24

26 important determinants of when increased debt financing leads to competitive (or aggressive) behavior in the product markets. 4.5 Product market strategies Next, I attempt to pin down the effect of collateral on product market performance by identifying some of the competitive strategies that firms might implement given shocks to their own real estate assets value and therefore ability to raise debt. While it is beyond the scope of the paper to try to fully identify all possible strategies, in this section I look at an important competitive strategy that a firm can fund using collateralized debt financing: discretionary spending on various forms of investment. More specifically, I look at a firm s spending on capital expenditure, research and development (R&D), advertisement, and acquisitions. I estimate the following augmented version of the standard investment-q specification: Investment CashHolding 5 i,t a + REValue i,t i, t 1 Leverage + PriceIndex i,t-1 2 Investment 7 state, t 1 i,t-1 Q 3 i,t-1 CashFlow Firm & Year FE + 4 e i,t-1 ij,t (2) where i indexes the firm and t is a year index. The dependent variable is capital expenditures, R&D, advertisement, and acquisitions, all scaled by total assets in year t - 1. Similar to all other variables in this study, each firm s changes in investment spending are measured relative to its industry competitors by removing the means of these variables of its TNIC peers. I use the Compustat data to compute capital expenditures, R&D and advertisement expenditures and the merger and acquisitions data from Thomson Financial SDC to compute changes in acquisition spending. Spending on R&D, advertisement and acquisition is set to zero if its missing. To account for the lagged investment effect (Eberly et al. 2012), I include the lagged values of each of the industry-adjusted investment variables in their respective regressions. 25