The Relation Between Reporting Quality and Financing and Investment: Evidence from Changes in Financing Capacity

3.2.1. Investments, Collateral Value, and Reporting Quality

To examine the effect of reporting quality and collateral values on investment, we build on specifications used in prior literature. Traditionally, previous research has predicted that investment is a function of growth opportunities (Tobin's q), cash flows, and other explanatory variables (e.g., Fazzari, Hubbard, and Petersen [1988], Hubbard [1998], Lamont [1997], Richardson [2006], Almeida and Campello [2007]). More recently, Chaney, Sraer, and Thesmar [2012] extend this literature by estimating a model that includes exogenous fluctuations in collateral prices to study the relation between collateral values and investment. Chaney, Sraer, and Thesmar [2012] estimate the following model:

(1)

where is a firm fixed effect, INV is capital expenditures scaled by lagged assets, and X is a vector of control variables (measured as of t – 1) that includes year fixed effects. The innovation in Chaney, Sraer, and Thesmar [2012] is to introduce a measure of exogenous fluctuations in the value of real estate assets by including RE_VALUE in equation 1. RE_VALUE is computed as follows:

(2)

Thus, RE_VALUE is the market value of real estate in 1993 (calculated as described in section 3.1), multiplied by an index of state-level real estate prices from 1993 to year t. Since firms had different amounts of real estate assets in 1993, RE_VALUE consists of cross-sectional variation in the initial value of real estate assets as well as time-series variation in the values of those assets due to variation in real estate prices. Our identification uses this variation to isolate the effect of changes in collateral values on investment, financing, and reporting quality. We normalize RE_VALUE by total assets in year t – 1, which is also the deflator for investment and financing.

In equation 1, STATE_INDEX_st is a control for the change in real estate prices in state s from 1993 to year t. This variable, in conjunction with year fixed effects, controls for macroeconomic changes (e.g., recessions, expansions, changes in interest rates) that affect the economy as a whole. The firm fixed effects give the coefficients a changes interpretation, that is, Chaney, Sraer, and Thesmar [2012] find a positive coefficient β₁, and interpret it as evidence that a change in real estate value is associated with a change in investment.

Note that the firm fixed effect in equation 1 removes the firm mean. It can also be estimated (with an appropriate adjustment to degrees of freedom) as:

(3)

where the “∼” above the variable name indicates that the firm mean has been subtracted from the variable, for example,

(4)

Our prediction in H1 is that changes in investment will be less sensitive to real estate prices when the level of reporting quality is higher (we examine changes in reporting quality in H2). In order to test this prediction, we include an interaction between reporting quality and the effect of interest. Thus, we modify equation 3 by introducing an interaction between and reporting quality. We also include a main effect for reporting quality.3 Specifically we estimate the following:

(5)

where FRQ, measured at year t – 1, is one of our six proxies for reporting quality (the information asymmetry component of the bid-ask spread (IAC_SPREAD), timely loss recognition (TLR_BASU), the length of MD&A (MDA_LENGTH), the number of 8-Ks filed by the company (COUNT_8K), the number of management forecasts (MGMT_FCST), and a composite measure of reporting quality (FRQ_INDEX1)). We describe these measures below.

In equation 5, we remove the means from the same variables as in equation 1, but not from FRQ. This specification allows us to interpret β₁ as the average sensitivity of the change in investment to changes in collateral prices. Chaney, Sraer, and Thesmar [2012] show that the sensitivity of changes in investment to changes in collateral prices (i.e., β₁) is positive, a finding that suggests that firms invest more (less) when real estate assets experience an increase (decrease) in value. Thus, if reporting quality attenuates this effect, then we predict that the estimated coefficient β₂ will be negative.

Our approach in equation 5 assumes that the reporting quality prior to the change is a predetermined variable that will affect the increase in information asymmetry due to the change in real estate prices. As discussed in the beginning of this section, reporting quality at t – 1 is arguably exogenous with respect to a subsequent (unanticipated) change in financing capacity (although it may be endogenous with respect to expected financing and investment). When we later test H2, we examine whether firms, subsequent to observing their financing capacity following a real estate change, then adjust their reporting quality.

Following prior literature (e.g., Richardson [2006], Almeida and Campello [2007], Biddle, Hilary, and Verdi [2009]), in equation 5 we control for contemporaneous cash flow (cash flow from operations divided by lagged assets),4 lagged Tobin's q (measured as the market value of assets divided by the book value of assets), the logarithm of lagged market value of equity, lagged age (measured as the logarithm of the number of years a firm has a record in Compustat), and lagged leverage (measured as the sum of short- and long-term debt divided by the book value of assets). In addition to these control variables, all of our specifications include year fixed effects.

Following Chaney, Sraer, and Thesmar [2012], we cluster standard errors using a two-dimensional cluster at the state and year levels, which addresses both cross-sectional and firm-specific dependence. By clustering in this way, our standard errors are conservative, given that the explanatory variables of interest are defined at the firm level (see Bertrand, Duflo, and Mullainathan [2004]).

Before we proceed, we discuss a potential source of endogeneity. Changes in real estate prices could proxy for something other than changes in collateral value and financing capacity (e.g., changes in growth opportunities). If the firm's choice to own real estate is correlated with its expected growth opportunities, the estimated sensitivity of investment to real estate prices can be biased upwards. For example, for a home builder operating in a given state, an increase in real estate prices indicates greater growth opportunities as well as greater collateral. To mitigate this concern, we exclude industries that are more likely to be affected by this issue (e.g., real estate and construction). Further, we control for proxies intended to capture growth opportunities, such as state-level real estate prices, year fixed-effects, Tobin's q, and cash flows.

In addition, although the direction of the bias is in favor of a positive sensitivity of investment to real estate (i.e., β₁ in equation 5), in terms of the interaction with reporting quality (i.e., β₂ in equation 5), there could be a bias against us finding our results. Specifically, Bushman, Piotroski, and Smith [2011] find that high–reporting quality firms have a higher sensitivity of investment to changes in growth opportunities because reporting quality lowers adverse selection costs and allows these firms to raise more capital and further tap into the growth opportunities. In contrast, we predict that high–reporting quality firms will have a lower sensitivity of investment (and financing) to changes in collateral values because reporting quality lowers the firm's need for collateral.

3.2.2. Financing, Collateral Value, and Reporting Quality

To test H1 with respect to financing, we modify equation 5 by replacing investment as the dependent variable with measures of external finance raised by the company. Specifically, we estimate the following:

(6)

where is one of the three measures of net finance raised by a firm—, , and As before, FRQ is one of our six proxies for reporting quality, as in equation 5, and the “∼” above the variable name indicates that the firm mean has been subtracted from the variable. We focus on debt, equity, and total financing, as prior research has shown that reporting quality can affect both debt and equity financing.

Chaney, Sraer, and Thesmar [2012] show that financing is sensitive to changes in collateral prices. Our hypothesis is that financing will be less sensitive to a change in real estate prices when reporting quality is higher. Thus, we predict that the estimated coefficient β₁ will be positive and the coefficient β₂ will be negative.5

Following Bradshaw, Richardson, and Sloan [2006], we measure net debt financing as the net cash received from (paid for) the issuance (reduction) of debt (Compustat variable DLTIS minus DLTR plus DLCCH, where we set DLCCH to zero if missing). DEBT_NET is the ratio of net debt financing in a given year scaled by the lagged value of assets (Leary and Roberts [2005], Lemmon and Roberts [2010]). We measure net equity financing as the net cash received from the sale (repurchase) of equity, less common dividends (Compustat variable SSTK minus PRSTKC minus DVC).6 EQUITY_NET is the ratio of net equity financing in a given year (Compustat variable SSTK) scaled by the lagged value of assets. FIN_NET is the sum of debt and equity financing in a given year scaled by the lagged value of assets. Finally, we control for contemporaneous cash flow (cash flow from operations divided by lagged assets), lagged Tobin's q (measured as the market value of assets divided by the book value of assets), and the logarithm of lagged market value of equity, lagged age (measured as the logarithm of the number of years a firm has a record in Compustat).

3.2.3. Reporting Responses to Changes in Collateral Value

In the previous section, we used reporting quality prior to the change as a conditioning variable that mitigates the effect of real estate price changes on investment. That specification assumes that reporting quality in the prior period is a predetermined variable (with respect to an unanticipated change in financing capacity). However, as discussed in section 2., a complementary prediction is that firms adjust their reporting choices based on cost–benefit tradeoffs, and that a change in collateral values affects future reporting choices (H2).

We test this hypothesis by estimating a regression of reporting quality on changes in collateral. The model is similar to the one in equation 1 but uses reporting quality as the dependent variable (as opposed to investment):

(7)

where FRQ_t is one of our proxies for reporting quality (MDA_LENGTH, COUNT_8K, MGMT_FCST, IAC_SPREAD, and a composite measure, FRQ_INDEX2, described below) and is a firm fixed effect.7 Note that we estimate equation 7 directly with firm fixed effects, whereas we estimate the investment and financing models above by first demeaning the variables and estimating equations 5 and 6 above. This is because our first hypothesis pertains to the level of reporting quality, whereas H2 relates to the change in reporting quality.

Our second hypothesis is that firms will increase (decrease) reporting quality in response to a decrease (increase) in real estate prices. Thus, we predict that the coefficient β₁ will be negative. Again, the presence of the firm fixed effect allows the coefficients to be interpreted as changes. Thus, if we find a negative coefficient β₁, we will interpret it as a decrease in the value of real estate assets being associated with an increase in reporting quality.

In equation 7, consistent with the prior literature (e.g., Healy, Hutton, Palepu [1999], Lang and Lundholm [2000], Leuz and Verrecchia [2000]), we control for ROA (measured as operating income before depreciation and amortization divided by total assets) and the logarithm of the market value of equity (MVE). We also control for Q, the logarithm of LEVERAGE, and the logarithm of AGE. Finally, in addition to these control variables, we include year indicator variables.

3.2.4. Measures of Reporting Quality

In our tests, we use five proxies for reporting quality and two composite indexes.

IAC_SPREAD measures the extent to which unexpected order flow affects prices and is increasing in information asymmetry. This variable measures the effect of information asymmetry on the stock price (i.e., the price impact or adverse selection that results from information asymmetry among investors). It is intended as an ex post proxy for (the inverse of) reporting quality based on several prior studies that document a negative relation between reporting quality and the bid-ask spread (e.g., Verrecchia [2001]). A caveat with this measure, however, is that high levels of information asymmetry might lead firms to increase disclosure in response to it, which is not the relation we want to capture. Thus, the use of this measure assumes an average negative relation between reporting quality and information asymmetry.

We estimate IAC_SPREAD following Madhavan, Richardson, and Roomans [1997] (as described in Armstrong et al. [2011]). The resulting estimate expresses the information asymmetry component of the bid-ask spread as a percentage of price (IAC_SPREAD). Because running the algorithm is very time consuming, we measure IAC_SPREAD for each firm once a year at its fiscal year-end, using all intraday data for that month. Again, to give the variable the interpretation of increasing reporting quality, we multiply it by negative one. Further, given the noise inherent in the process of estimating IAC_SPREAD, we include this measure as a ranked variable (converted into deciles) in all our tests. We then rescale it to range from zero to one.

We also use timely loss recognition (TLR_BASU) computed at the industry level (e.g., Ball, Bushman and Vasvari [2008], Wittenberg-Moerman [2008]). To the extent that debtholders (common suppliers of collateralized capital) are more sensitive to borrower's losses and firms with more conservative financial reporting are less likely to report losses ex post, conservative accounting can facilitate access to credit markets. Consistent with this argument, Wittenberg-Moerman [2008] shows that firms with higher timely loss recognition (her proxy for conservative reporting) have lower debt financing costs.

Following Wittenberg-Moerman [2008], we employ an industry-level estimation of timely loss recognition using the Basu [1997] methodology at the three-digit SIC level. An advantage of this measure is that, because it is measured at the industry level, there is less of a concern about endogeneity.8 We measure TLR_BASU as follows. For each three-digit SIC industry-year in Compustat, we estimate the following piecewise linear regression of annual price-deflated earnings (NI_it) on annual stock returns (R_it, ending three months after the end of the fiscal year) and stock returns interacted with a dummy variable for negative returns (DR_it):

(8)

TLR_BASU is measured as the sum of b2 and b3. This industry-level measure of timely loss recognition is then assigned to each firm in a given industry.

We also use the length of the MD&A, the issuances of 8-Ks, and the issuance of management forecasts as proxies for reporting quality. The sample for these proxies begins in 1995 because it is the first year with available SEC filings on EDGAR's Web site to collect MD&A and 8-K information and because First Call coverage improves significantly post-1994 (e.g., Rogers and Stocken [2005], Ng, Tuna, and Verdi [2013]).

We use the length of the MD&A section of the 10-K reports filed by firms as a proxy for reporting quality. We retrieve firms’ 10-K filings from SEC EDGAR and then employ text-mining programs to extract the MD&A section. The MDA_LENGTH is the number of words in the MD&A section of the 10-K statement. Recent studies provide evidence that the forward-looking statements in a firm's MD&A are positively correlated with its future earnings and have explanatory power over other variables that can predict future performance (e.g., Li [2010], Feldman et al. [2010]). This suggests that the MD&A is a reasonable proxy for reporting quality. Using similar logic, Leuz and Schrand [2009] employ MD&A length as their primary measure of disclosure.

We also employ the number of 8-K forms (COUNT_8K) filed by the firm. This variable is also used by Leuz and Schrand [2009] when investigating disclosure responses to the Enron event. The SEC requires that firms use the 8-K form to disclose material information and to update any information provided in previous SEC filings. Although the SEC lays out specific reportable events, the guidelines are generic enough that firms have discretion in filing an 8-K for other information.

Our final proxy is the issuance of earnings forecasts by the management (MGMT_FCST). Several papers have used management forecasts as a proxy for voluntary disclosure activity. Overall, this literature shows that management forecasts provide information to the market and that they are associated with market returns and analyst forecast revisions. Prior research that examines earnings forecasts suggests that managers who wish to enhance transparency issue more frequent, specific, and accurate forecasts (Skinner [1994], Kasznik and Lev [1995]). In addition, investors and analysts react to these forecasts, which suggests that they have information content (e.g., Ajinkya and Gift [1984], Waymire [1984], Jennings [1987], Williams [1996]). We acknowledge that all of these disclosure measures are noisy because managers may disclose for reasons other than to improve the information environment; for example, they may disclose in response to abstain-or-disclose rules (Li, Wasley, and Zimmerman [2011]).

We note that, although we use several proxies for reporting quality, none of these proxies are perfect and they vary in terms of relative strength and measurement error. Thus, we also create two composite indexes in which we aggregate the individual proxies for reporting quality with the intent of reducing measurement error in the individual proxies. For the tests of H1, we create FRQ_INDEX1 as the average of standardized values of IAC_SPREAD, TLR_BASU, MDA_LENGTH, COUNT_8K, and MGMT_FCST (we require at least two of the individual proxies to be available to compute the index). Since these variables are used as explanatory variables in equations 5 and 6, all variables are measured as of year t – 1. For the tests of H2, we create FRQ_INDEX2 as the average of standardized values of IAC_SPREAD, MDA_LENGTH, COUNT_8K, and MGMT_FCST (i.e., TLR_BASU is not included since it is not used in the tests for H2). Again, we require at least two of the individual proxies to be available to compute the index. In this case, since these variables are used as dependent variables in equation 7, all variables are measured as of year t.

4.5.1. Over- Versus Underinvestment

We conclude from our analysis thus far that reporting quality mitigates financing constraints. Implicit in this conclusion is an assumption that our sample firms, on average, have constrained financing capacity and that an increase in collateral value mitigates underinvestment. However, it is possible that some of our sample firms do not face financing constraints. In this case, when collateral values increase, any extra investment might be excessive instead of efficient investment.

In order to examine the sensitivity of our results to potential overinvestment, we follow Biddle, Hilary, and Verdi [2009] and categorize firms into two subsamples based on their propensity to under- or overinvest. Specifically, we first rank firms into deciles based on their cash balance and their leverage (we multiply leverage by minus one before ranking so that, as for cash, it is increasing with the likelihood of overinvestment) and rescale them to range between zero and one. We create a composite score measure as the average of the ranked values of the two partitioning variables and compare firms in the extreme terciles of the distribution. We classify firm-years in the highest tercile as those that are likely to overinvest; those in the bottom tercile are firms that are likely to underinvest.

We then examine the effect of changes in collateral values on investments, financing, and reporting quality in both these subsamples. Firms in the bottom tercile have low cash and/or high leverage and are likely to be financially constrained. These are the firms to which our above hypotheses are most likely to apply. For the firms in the highest tercile (classified as more likely to overinvest), however, the predictions are less clear. On one hand, evidence in Biddle, Hilary, and Verdi [2009] and others suggests that reporting quality mitigates overinvestment. This result suggests that firms more likely to overinvest with high-reporting quality also could be less likely to overinvest after an increase in collateral. On the other hand, firms more likely to overinvest have rich internal resources, so that it is unclear whether a change in collateral value would affect their investing, financing, and reporting choices.

Table 5 compares the results for the over- and underinvestment samples for investment, financing, and reporting quality, respectively.12 Columns 1 and 2 compare investment for the two samples. For the overinvestment sample (column 1), the main effect on RE_VALUE for investment is insignificant, but the interaction is significantly negative. In contrast, for the underinvestment sample (column 2), the main effect on RE_VALUE for investment is positive, and the interaction is significantly negative. The difference between the coefficients on RE_VALUE suggests that firms prone to underinvestment are more sensitive to changes in collateral value than firms prone to overinvestment (t-statistics for the difference of 3.17). In addition, this result holds for firms with higher reporting quality as proxied by the joint effect of RE_VALUE and its interaction with reporting quality (t-statistics for the difference of 3.19). The differences in financing shown in columns 3 and 4 are similar. Overinvesting firms do not react to the collateral change, while underinvesting firms show a significant change in financing. As for reporting quality (H2), we find evidence that changes in collateral are associated with changes in reporting quality for the overinvestment sample (column 5), but the underinvestment sample (column 6) reacts slightly more strongly (t-statistics for the difference of 1.60). Thus, our findings in table 5 are consistent with the effects we document arising from firms suffering from potential underinvestment.13

4.5.2. Identification Tests

Our tests implicitly assume that a firm's real estate assets (or at least a large portion of them) are located in the same state as the firm's headquarters. As a way of assessing the importance of this assumption, we rerun our tests using a U.S. price index instead of the state price indices. Column 2 of table 6 shows these results (first for investment, then for total financing, and finally for reporting quality as the dependent variable), which are very similar to the results shown in column 1, which shows the main results from tables 2 through 4. Given that state real price changes have a substantial systematic (U.S.) component, this result suggests that the idiosyncratic statewide component from state prices is not providing identification.

To further explore this result, we conduct randomization tests. We start by randomizing the state that a given firm is assigned. Specifically, instead of using the real estate price for the state in which the firm's headquarter is assigned, we assign a firm to a random state other than the state of the headquarters. Column 3 of table 6 presents these findings. We find that randomizing the state yields identical inference, again suggesting that the state location is not crucial for our identification. As discussed in the prior paragraph, a primary explanation for this is that the cumulative state indexes are highly correlated over time (average state pairwise correlation of 0.84), and, therefore, randomizing leads to little difference.

We then perform two additional randomization tests to determine whether the identification is coming from cross-sectional or time-series variation in the value of real estate assets. Specifically, to explore the time-series variation, we randomize over time holding the firm constant; this is akin to assigning a given firm a placebo yearly change in the value of its real estate assets over its time-series. Second, to explore the cross-sectional variation, we randomize across firms holding the year constant; in this case, the test is akin to assigning a given firm the change in the value of the real estate assets of a placebo firm in the cross-section.

The last two columns of table 6 present these results. In both cases, the main effect for RE_VALUE and the interaction between RE_VALUE and FRQ are no longer significant. This suggests that the identification in our paper crucially depends on the initial value of real estate assets; this initial value induces cross-sectional and time-series variation in the value of real estate assets among our firms. Finally, as noted above, about 42% of our sample has no real estate in 1993. To ensure that our randomization tests are not simply capturing the choice of ownership, we drop nonowners, and repeat the randomization tests. The results for this subsample (untabulated) are very similar to the results in table 6. This suggests that the identification in our paper depends not only on the choice of ownership in 1993, but also on the magnitude of real estate owned by a firm.

4.5.3. Measurement Error in Real Estate Values

We use the methodology in Chaney, Sraer, and Thesmar [2012] to identify exogenous variation in the value of a firm's real estate assets. The main advantage of this approach is that it mitigates endogeneity concerns (e.g., that variations in real estate assets proxy for future growth opportunities), but it requires assumptions that lead to potential measurement errors in the data. In this section, we discuss these assumptions and provide sensitivity tests.

An important measurement concern is that we value a firm's real estate assets in 1993, and then assume that the value evolves over time based on the changes in the real estate prices. This procedure ignores the firm's real estate purchases and sales after 1993. To investigate whether this source of measurement error affects our results, we consider two extreme cases. First, we identify firms that initially had real estate in 1993, but later sold all of their real estate (disposers). In our sample, there are approximately 1,688 firm-year observations that occur after a complete sale of real estate. Second, we identify firms that initially had no real estate in 1993, but later acquired real estate (acquirers). In our sample, there are approximately 3,382 firm-year observations that occur after an initial real estate purchase. We then reestimate the main regressions in the paper (equations 5–7) for these two subsamples.14

These results are presented in columns 2–4 of table 7, panel A (column 1 presents the results for the main sample to facilitate comparison). In column 2, we first present the results for the firms that own real estate in 1993 (1,610 firms as discussed in section 3.1) but without the firm-years after the “disposal.” The resulting sample is 11,959 firm-years. This provides a way to assess the measurement error driven by inclusion of firms (and firm-years) without real estate and also serves as a benchmark to compare the results for the “disposers” sample. The results in column 2 yield inferences similar to the results with the full sample, although the coefficients of interest tend to be larger in absolute magnitude. For instance, the coefficient on RE_VALUE increases from 2.20 to 2.98 for investment, increases from 2.17 to 7.09 for financing, and decreases from –0.15 to –0.21 for disclosure. As for the interaction with reporting quality, the magnitude of the coefficients also increases.

In column 3, we find that disposers, after the disposal, have a lower sensitivity to real estate prices and an interaction with reporting quality that is smaller in magnitude. For example, the coefficient on RE_VALUE, when compared to the firms with real estate (column 2), decreases by 48% (from 2.98 to 1.56) in the investment model. This suggests that real estate disposers, after they sell real estate, have a diminished sensitivity to real estate prices. In column 4, we find that real estate purchasers, after they acquire real estate, have a positive sensitivity to real estate prices and a negative, but insignificant, interaction with reporting quality. Overall, these results suggest that (1) real estate disposers, after they sell real estate, create measurement error that diminishes the sensitivity to real estate prices; and that (2) real estate purchasers, after they acquire real estate, have a significant sensitivity to real estate prices.15

We perform two additional tests to deal with measurement error in the value of real estate. Specifically, we assess the sensitivity of our findings using ranked values and an alternative winsorization procedure. Following Chaney, Sraer, and Thesmar [2012], in our main specification we measure RE_VALUE as a continuous variable, but winsorize all variables at the 5% and 95% levels. As additional analyses, we convert RE_VALUE to deciles as well as use it continuously but winsorize variables at the 1% and 99% levels. The results for all these tests are presented in columns 5 and 6 of table 7, panel A. In general, our findings are similar to those reported in tables 2 through 4 shown in column 1. The only exception is that the investment regression yields weaker results when using ranked RE_VALUE or when winsorizing the data at the 1% and 99%, suggesting that measurement error is indeed an issue in this model (the results with financing and reporting quality are robust to these tests).

Finally, we examine alternative measures of our dependent variables and our reporting quality index. First, our measures of investment and financing exclude operating leases (which are excluded from financial statements under U.S. accounting rules). It is possible, however, that, when faced with a reduction in collateralizable assets, firms might switch away from on-balance-sheet to off-balance-sheet investment and financing. Following Beatty, Liao, Weber [2010], we estimate the asset/liability associated with operating leases,16 and add the change in asset/liability associated with operating leases to our measure of investment and financing. Second, we measure our dependent variables (investment, financing, and reporting quality) in year t + 1 (instead of year t) because it is possible that firms might take some time to react to the changes in real estate assets. Last, in the construction of FRQ_INDEX1 in our tests above, we require that only two of the individual proxies be available in order to maximize sample size. To test the sensitivity of this assumption, we reestimate our models for the subsample of firms with data for all the individual FRQ proxies.

The results for all these tests are presented in columns 7–9 of table 7, panel A. Again, in general, our findings are similar to those reported in tables 2 through 4 shown in column 1. An exception is that our results with investment also become slightly weaker when investment is measured in year t + 1, suggesting that investment actions manifest in the same year of the changes in collateral values. The results with financing, however, become stronger when the dependent variable is moved to year t + 1. One possible interpretation of this result is that financing activities are more lumpy (i.e., have higher adjustment costs) and take more time for firms to adjust their financing activities subsequent to changes in financing capacity.17

4.5.4. Type of News

Another concern is that our results may be capturing a “bad news” effect. In particular, years in which real estate values decline may be eventful (e.g., greater possibility of restructuring) and, hence, associated with higher levels of reporting quality and transparency in general. To address this issue, we partition our sample into firm-years in which the yearly change in the real estate index was above 4%, which is the median change in our sample. We then reestimate our regressions separately for years with above (below) the median change in the state price index—our proxy for good (bad) news.18 In addition, we reestimate our models after excluding the period of the financial crisis (years 2008 and 2009). This mitigates a concern that our results could be capturing other factors beyond variations in real estate prices (e.g., supply of lending). Finally, we examine the effect of large changes in real estate values. Firms are less likely to change their investment, financing, and reporting activities in response to small variations in financing capacity. To test this argument, we keep only firms in the top and bottom tercile of yearly changes in state real estate price index (i.e., drop firm-years in the middle tercile). The results of these tests are presented in table 7, panel B (first for investment, then for total financing, and finally for reporting quality as the dependent variable). With the exception of an insignificant coefficient on the interaction for positive news firm, the results are very similar to those reported in tables 2 through 4, suggesting our findings are not specific to a particular type of news.