The Effect of Stock Liquidity on the Firm’s Investment and Production

Abstract

We propose that stock market liquidity affects corporate investment and production. Illiquidity, which raises firms’ cost of capital, lowers investment in capital assets, R&D, and inventory. This effect holds after we control for endogeneity using exogenous liquidity events, the 2001 decimalization, and the 1997 Nasdaq reform and after employing instrumental variable estimation. Illiquidity affects investment regardless of firms’ financial constraints. Consequently, illiquidity induces firms to adopt less capital-intensive production processes. Illiquid firms have higher marginal productivity of capital, greater labor input increases for given increases in assets, and lower operating leverage, which means lower reliance on fixed costs.

“The Stock Market and Investment: Is the Market a Sideshow?”

Morck et al. (1990)

This paper proposes that one channel by which the stock market affects investment and production is through market liquidity. It is well known that illiquidity raises the expected return required by investors,¹ thus raising the cost of capital that managers use in evaluating investment projects. We show that illiquidity lowers corporate investment and leads firms to adopt production processes that are less capital intensive. This results in higher revenue productivity of capital, higher labor input relative to capital, and lower operating leverage and greater reliance on variable costs.

The main takeaway from our results is that secondary market liquidity affects the real economy through its effect on firms’ real investment and production processes. We show that reforms in the secondary market trading procedures that improve market liquidity, such as the 1997 Nasdaq market-making reform and the 2001 decimalization, induced higher corporate investment. The results in this paper on the beneficial effects of market liquidity on investment and production suggest a motivation for regulatory reforms and corporate policies that improve liquidity.²

Our evidence on the real-economy effects of trading in the secondary market is related to the analysis of Bond, Edmans, and Goldstein (2012). These authors distinguish between the real effects of primary financial markets, where firms raise capital, and the real effects of secondary financial markets, where securities are traded and prices are set but there is no active flow of capital to the firm. The real effect of the secondary markets results from a feedback effect proposed by Chen, Goldstein, and Jiang (2007) by which managers incorporate in their investment decisions private information that they learn from market prices. We offer another channel of feedback: from secondary market price information into real corporate decisions. Managers gather information from the secondary market on their firm’s stock liquidity and illiquidity premium and use it to set the investment hurdle rate, and select investment projects that are commensurate with investors’ required return.³ We find that the negative effect of illiquidity on investment holds regardless of capital raising needs, and after controlling for Chen, Goldstein, and Jiang’s (2007) feedback effect from market prices.

The negative effect of stock illiquidity on corporate investment holds after controlling for firm characteristics (including fixed effects); it holds consistently over time and across all five major industries considered. The negative investment-illiquidity relation remains significant after accounting for potential endogeneity by employing two exogenous liquidity-increasing events, the 2001 decimalization and the 1997 Nasdaq market-making reform. We find that investment increased in firms whose stock liquidity benefited more from decimalization and from the reform. The negative investment-illiquidity relation also holds when employing an instrumental variable (IV) analysis.

In addition to illiquidity raising the cost of capital, it may constrain firms in the raising of capital, thus inhibiting investment. This alternative channel by which the secondary market affects real activity matters mostly for financially constrained firms; see Morck et al. (1990) and Bond, Edmans, and Goldstein (2012).⁴ If illiquidity lowers investment only by exacerbating the firm’s financial constraint, its effect should be muted for unconstrained firms; however, if illiquidity also affects investment by raising the cost of capital, it should lower investment even for unconstrained firms. We find that the negative investment-illiquidity relation does not depend on financial constraints and it holds strongly for unconstrained firms. Using 10 common measures of financial constraint⁵ and following Fazzari, Hubbard, and Petersen (1988), we divide firms by financial constraints into three groups and estimate the investment-illiquidity relation for each group. We find that the negative illiquidity-investment sensitivity is similarly strong for all groups including that of unconstrained firms without any systematic pattern across the groups. Also, the significant negative effect of stock illiquidity on investment remains after controlling for corporate liquidity (cash and marketable assets). The fact that the negative investment-illiquidity relation holds even for firms that are not financially constrained suggests that the effect of illiquidity on investment is through the secondary, rather than the primary, market. The results suggest that when selecting projects, even managers that do not need to access the primary market to raise capital would consider the secondary market illiquidity because it reflects stockholders’ required return.⁶

We test the effect of lagged stock illiquidity on investment, measured by capital expenditures or changes in total assets, on investment in intangible assets that include research and development expenses (using Peters and Taylor’s [2017] measure), and on inventory investment (all scaled by lagged assets). Stock illiquidity is measured by Amihud’s (2002) ILLIQ or by the relative bid-ask spread (SPRD). The results are consistent across measures of investment and of illiquidity. The estimation models include control variables that significantly affect investment: current cash flow and lagged Tobin’s Q (following Fazzari, Hubbard, and Petersen 1988), as well as lagged total assets, return volatility, and cumulative 2-year stock return. The models also include firm (or industry) fixed effects and year fixed effects. The estimation is over 57 years, 1963 through 2019, and over two subperiods, 1963–1990 and 1991–2019. The negative and significant effect of illiquidity on investment is similar for the entire period and for both subperiods. The negative illiquidity effect is also similar in estimations using industry fixed effects and when using the Fama-MacBeth (1973) cross-section estimation method.

The negative and significant investment-ILLIQ relation robustly holds across all five major industries using a one-digit SIC code (excluding financials and utilities). As a robustness test, we estimate the model in changes rather than in levels and find that lagged changes in ILLIQ have a negative and significant effect on investment, controlling for lagged changes in all control variables and further controlling for lagged changes in investment. And, to mitigate a concern of simultaneity, we estimate the model with the illiquidity variable lagged by 2 or 3 years relative to investment. In all estimates, we find a negative and significant effect of illiquidity.

We test another channel by which illiquidity may affect investment: the information feedback effect. Chen, Goldstein, and Jiang (2007, p. 620) propose that “managers learn from the private information in stock price when they make corporate investment decisions” and find that the sensitivity of investment to stock price increases in the extent of private information incorporated into the price. Given that illiquidity increases in the extent of traders’ private information (see Glosten and Milgrom 1985; Kyle 1985), we follow Chen, Goldstein, and Jiang’s (2007) empirical specification and add to our model an interaction term of illiquidity and |$Q$|⁠. We find that both effects are in play. Consistent with Chen, Goldstein, and Jiang (2007), the investment sensitivity to the firm’s Tobin’s Q significantly increases in illiquidity while the direct negative effect of illiquidity on investment remains highly significant.

We deal with concerns of endogeneity or codetermination of illiquidity and investment by studying two exogenous liquidity-improving events. The first is the decimalization of quoted prices in the U.S. exchanges in 2001, which reduced the minimum tick (price increment) from 6.25 cents ($1/16) to 1 cent, thus enabling narrower quoted bid-ask spreads. Bessembinder (2003) documents a greater decline in illiquidity following decimalization for stocks with smaller bid-ask spreads. We find that, following decimalization, investment increased significantly more for firms whose stock liquidity benefitted most—those whose initial bid-ask spread was smaller. Also, for such stocks the effect of decimalization on investment should have been greater since, by Amihud and Mendelson (1986), required returns decline more for a given fall in the bid-ask spread in stocks with narrower spreads.

The second liquidity-improving event is the 1997 reform in Nasdaq trading. Barclay et al. (1999) document a significant reduction in the bid-ask spread for Nasdaq stocks with even-eighths quoted prices before the reform due to increased competition of traders with market makers who were allegedly colluding beforehand to raise the spread. We find that for such stocks there was a significant increase in investment after the reform.

We also account for endogeneity by employing an instrumental variable estimation. Finding a valid instrument is usually challenging. We instrument ILLIQ with the concentration of institutional holdings that is known to increase illiquidity; see theory and evidence in Bolton and von Thadden (1998) and Rubin (2007). Yet, the concentration of institutional holdings is not known to affect investment. We find that the instrumented ILLIQ has a negative and significant effect on investment.

Our investment model includes lagged |$Q$| as an explanatory variable, which may preclude the use of other explanatory variables, following Hayashi’s (1982) proposition that investment is sufficiently explained by marginal |$Q$|⁠, which equals average |$Q$| under some conditions. However, we find that, in addition to the effect of lagged |$Q$| on investment, there are significant effects of all explanatory variables: current cash flow and lagged values of illiquidity, total assets, return volatility, and cumulative 2-year stock returns. Our finding is consistent with those of Chen, Goldstein, and Jiang (2007) and Bai, Fairhurst, and Serfling (2020): that capital expenditures are significantly affected by a number of lagged firm characteristics in addition to lagged |$Q$|⁠. The reason may be that standard average |$Q$|⁠, which we use, is different from marginal |$Q$|⁠, which the model demands. Although the two are equal when financing is frictionless and profits are linear in capital (Hennessy, Levy, and Whited 2007), frictionless financing is inconsistent with stock illiquidity. Also, |$Q$| is measured with error,⁷ potentially because its calculation employs assets’ book value, which differs from replacement value and does not fully account for intangible capital.⁸

Inventory investment is also negatively affected by the cost of capital, but, unlike capital investment, it is not affected by Tobin’s Q. We thus expect that inventory investment declines in illiquidity, which raises the cost of capital, as is the case for capital investment. Our findings support this prediction. Lagged illiquidity has a significant negative effect on inventory investment after controlling for lagged |$Q$|⁠, which has an insignificant effect, and all the other control variables used in the capital-investment model as well as for the change in sales.

Next, we study the real effect of market liquidity on firm production. We propose that the negative effect of illiquidity on investment induces illiquid firms to reduce the capital intensity of their production processes. We find the following:

• (1) Illiquidity raises the value of the marginal productivity of capital so that it is commensurate with the higher firm’s cost of capital. Empirically, illiquidity raises the sensitivity of sales growth to asset growth and it raises the sales-to-assets ratio.

• (2) Illiquidity raises the labor-to-capital marginal rate of substitution reflected in a greater increase in labor for a given increase in capital. This result holds for all five main manufacturing industries (using a one-digit SIC code). We also find that illiquidity raises the labor/capital ratio. These results follow from our earlier finding that illiquidity inhibits investment, thus inducing firms to economize on their use of capital and switch to more labor-intensive production processes.

• (3) Illiquidity lowers the firm’s operating leverage, which is the extent of its reliance on fixed costs in production. We find that illiquidity raises the response of operating expenses to change in sales, implying greater reliance on variable cost.

Our findings that firms’ decisions on investment and production deviate from those that would be selected if their stock were liquid contribute to explaining the negative effect of illiquidity on firm value (Fang, Noe, and Tice 2009).

Our analysis is in line with studies on the real effects of financial markets, presented in Bond, Edmans, and Goldstein (2012),⁹ and with the results of Goldstein, Yang, and Zuo (2020). Employing a quasi-natural experiment—the implementation of the Electronic Data Gathering, Analysis, and Retrieval (EDGAR) system in the United States in the years 1993–1996—they find that it significantly improved stock liquidity and raised corporate investment. Studies on the effects on investment of deletion/additions of stocks from/to the S&P 500 and the FTSE 100 indexes find conflicting results. Whereas Becker-Blease and Paul (2006) find that firms whose stock was added to the S&P 500 index enjoyed an increase in stock liquidity and they increased their investment, Gregoriou and Nguyen’s (2010) study of firms whose stock was deleted from the FTSE 100 index concludes that “deletion from a major stock index does not influence corporate investment decisions” (p. 267).¹⁰ Notably, additions and deletions of firms from indexes are often not random, and may affect firms’ information environment, and are therefore a confounded setting for testing how liquidity shocks affect investment.

1. The Effect of Illiquidity on Corporate Investment

We focus on the effect of lagged illiquidity, measured by Amihud’s (2002) ILLIQ, and hypothesize that |$b1 < 0$|⁠; that is, investment is a declining function of the firm’s stock illiquidity. ILLIQ|$_{j,t}$| is the (logarithm of the) average ratio of the daily absolute return to dollar volume for stock |$j$| in year |$t$|⁠. ILLIQ is highly correlated with Kyle’s (1985) theoretical measure of illiquidity, |$\lambda $|⁠, the price impact of trades, and with the fixed cost of trading resulting from the bid-ask spread.¹¹ In calculating the annual average ILLIQ, we exclude trading days with volume of less than 100 shares and require that a stock has at least 150 trading days for the year. To avoid outliers, we delete 1|$\%$| of the daily observations with the highest values of ILLIQ in each stock-year.¹² As a robustness test, we replace ILLIQ|$_{j,t}$| with SPRD|$_{j,t}$|⁠, the bid-ask spread (in logarithm) calculated as the dollar quoted spread divided by the quote’s midpoint, averaged over the days of year |$t$|⁠. Data for SPRD are available from CRSP¹³ since the end of 1992; thus we use the 1993 average spread for cross-sectional analysis that begins in 1994. Consequently, the sample size is smaller. The estimation results for SPRD|$_{j,t-1}$| are presented in the appendix.

The control variables include contemporaneous cash flow, CF, following Myers and Majluf (1984) and Fazzari, Hubbard, and Petersen (1988), who propose that firms invest first from available internal resources. CF is net income (before extraordinary items) plus depreciation and amortization, scaled by lagged total assets. All other explanatory variables are lagged by 1 year. |$Q$| (an estimate of Tobin’s Q) is the market value of assets at the end of the calendar year divided by the book value of assets, where the market value of assets is defined as the market value of equity plus book value of assets minus book value of equity and balance sheet deferred taxes (following the definition in Fang, Noe, and Tice 2009). This variable reflects growth opportunities and positively affects investment. TA is total assets (in logarithm), a measure of firm total size. VOL is volatility or risk, the standard deviation of weekly stock returns calculated over the year. Volatility positively affects the cost of capital, and thus should negatively affect investment, as does ILLIQ. RET2 is the 2-year cumulative stock return, which captures recent change in market expectations about the firm’s opportunities, and thus its coefficient is expected to be positive. The inclusion of the control variables TA, VOL, and RET2 is necessary because, in addition to their direct effect on investment, they are correlated with illiquidity. TA and RET2 are negatively correlated with ILLIQ, and VOL is positively correlated with it. Omitting these variables may lead to the “missing variable problem” by which their effect on investment is erroneously attributed to ILLIQ, with which they are correlated. RET2 also controls for the effect of market sentiment on investment, following Morck et al.’s (1990, p. 167) proposition that a rise in the firm’s stock price improves its access to cheaper financing through the stock market.¹⁴

Our sample includes firms whose stock traded on the New York Stock Exchange (NYSE) and the American Stock Exchange (AMEX) during 1963–2019. We conduct a separate analysis for firms whose stock traded on Nasdaq during a shorter period, 1998–2019, which follows the Nasdaq reform that enabled direct trading between buyers and sellers, similar to the trading regime on the NYSE and AMEX. Before that, trading volume (used to calculate ILLIQ) was usually double counted, reflecting buying and selling through market makers. We exclude firms in the financial industry (SIC code 6000–6999) and utilities (code 4900–4999), and we exclude REITs and firms with ADR whose stock is traded on a foreign market. We also exclude firm years if the assets or sales more than doubled or halved in that year. We require that firms have total assets of at least $10 million and share price of at least $1 at the beginning of the year, and we winsorize all variables at the 1|$\%$| level on both tails of their distribution.

Table 1 presents statistics for our data. (The table includes some variables whose construction is detailed below when employed in estimations.) Our sample includes 64,798 firm-years except for the data on SPRD|$_{j,t-1,}$| which includes 28,971 firm years.

The estimation results of Model (1), presented in Table 2, strongly support our hypothesis that corporate investment is negatively related to lagged stock illiquidity. Panel A presents results for the entire sample period for NYSE-AMEX stocks. The coefficient of ILLIQ|$_{j,t-1}$| is negative and significant, |$-$|0.009 (⁠|$t=-13.91$|⁠) and |$-$|0.009 (⁠|$t=-13.46$|⁠) for |${\it INV}_{j,t} = {\it CEx}_{j,t}$| and CExRD|$_{j,t}$|⁠, respectively. The economic significance of the estimated effects of illiquidity is illustrated as follows. A one-standard-deviation increase in ILLIQ (controlling for firm fixed effects)¹⁵ lowers subsequent investment relative to assets, measured as either CEx or CExRD, by 0.015. This effect is sizable given that the mean of CEx is 0.076 and the mean of CExRD is 0.094, implying a decline of about 1/5 to 1/6 in investment in response to an increase of one standard deviation in ILLIQ. In panel E, we present estimation results of Model (1) by the Fama-MacBeth (1973) method, replacing the firm fixed effects with industry fixed effects. In this estimation, the coefficients of ILLIQ|$_{j,t-1}$| are |$-$|0.004 (⁠|$t=-12.54$|⁠) and |$-$|0.008 (⁠|$t=-14.61$|⁠) for |${\it INV}_{j,t} = {\it CEx}_{j,t}$| and CExRD|$_{j,t}$|⁠, respectively. By these estimates, a one-standard-deviation increase in ILLIQ across firms lowers investment, measured as CEx, by 0.010 and lowers CExRD by 0.020.

The results are similar for illiquidity measured by SPRD (logarithm of the average bid-ask spread): the coefficient of SPRD is negative and significant in all regressions; see Appendix Table A.1. For example, the coefficient |$b1$| of SPRD|$_{j,t-1}$| is |$-$|0.008 with |$t=-6.84$| for |${\it INV}_{j,t} = {\it CEx}_{j,t}$|⁠. As noted, the data for SPRD, which begins in 1994, includes less than a half of the years and the firm-years used for tests with ILLIQ as a measure of illiquidity. As for the economic effect of the estimates, a one-standard-deviation¹⁶ increase in SPRD lowers CEx by 0.010, which is close in magnitude to the effect of ILLIQ.

The control variables’ coefficients have the predicted signs. CF|$_{j,t}$| (cash flow) and |$Q_{j,t-1}$| have positive and significant effects on investment, as found by Fazzari, Hubbard, and Petersen (1988). The effect of VOL|$_{j,t-1}$|—return volatility—is negative and significant, similar to the negative effect of ILLIQ|$_{j,t-1}$|⁠. Both illiquidity and risk raise the firm’s cost of capital and thus have negative effects on investment.¹⁷ Investment is lower for larger firms (with higher TA|$_{j,t-1})$| and higher for firms with better past stock performance (higher RET2|$_{j,t-1})$|⁠, which captures expectations about future growth. The significant effects on investment of the control variables, which are correlated with ILLIQ|$_{j,t-1}$|⁠, highlight the importance of not omitting them from a model that is intended to estimate the effect of ILLIQ|$_{j,t-1}$|⁠.

In addition to the illiquidity’s direct negative effect on investment, it negatively affects |$Q$| (Fang, Noe, and Tice 2009), which in turn positively affects investment. In our estimation, the effect of ILLIQ|$_{j,t-1}$| is already conditional on the contemporaneous |$Q_{j,t-1}$|⁠. Here, we estimate separately this indirect channel of the effect on CEx of ILLIQ through its prediction of |$Q$|⁠. First, we estimate Model (1) with |$Q_{j,t}$| as the dependent variable, omitting |$Q_{j,t-1}$| from the right-hand side. The coefficient of ILLIQ|$_{j,t-1}$| is |$-$|0.149 with |$t=-13.96$|⁠. Since the coefficient |$b3$| (the effect of |$Q_{j,t-1}$| on CEx|$_{j,t})$| is 0.005, the indirect effect of ILLIQ on CEx through its negative effect on |$Q$| is |$-$|0.0007 (⁠|$= 0$|⁠.005|$*-$|0.149), which is substantially smaller than the direct effect of ILLIQ on CEx.

We replicate our analysis measuring investment by dTA, the annual change in total assets scaled by lagged total assets, as do Chen, Goldstein, and Jiang (2007), which reflects changes that may be voluntary or involuntary. The variable dTA is positive due to capital expenditures, investment in current assets and acquisitions, or it can be negative following a spinoff or asset sale when the proceeds are distributed to shareholders or used to redeem debt. An asset sale decision can possibly be affected by a rise in illiquidity, which raises the hurdle rate and leads the firm to divest. But dTA may be positive or negative for other reasons. It may be positive following an asset sale at a gain over book value, or it may be negative because of involuntary impairment of value. Not all these changes in asset values are driven by considerations related to stock illiquidity. We find that |${\it dTA} < 0$| for 27|$\%$| of the firm years in our sample.

Estimating Model (1) with |${\it INV}_{j,t} = {\it dTA}_{j,t,}$| we find again that illiquidity significantly lowers investment. The coefficient of ILLIQ|$_{j,t-1}$| is |$-$|0.025 with |$t=-12.12$|⁠. When measuring illiquidity by SPRD|$_{j,t-1}$|⁠, its coefficient is |$-$|0.038 with |$t=-6.79$|⁠. We also replicate the analysis where dTA|$_{j,t}$| includes only firm-years in which dTA|$_{j,t} \ge 0$|⁠, which is more likely to reflect voluntary action by firms. This constraint reduces the sample to 47,113 firm-years. We find that the coefficient of ILLIQ|$_{j,t-1}$| is |$-$|0.019 with |$t=-9.12$|⁠, again highly significant.

1.1 Robustness tests

1.1.1 Tests across industries

In panel B of Table 2, we test whether the negative investment-illiquidity relation holds across industries by estimating Model (1) separately for each of the five major one-digit SIC code industries. While the firm fixed effects subsume the industry effects in terms of the level of investment, the slope coefficients may vary across industries. We use SIC codes 1 through 5, which are, respectively, mining and construction; two types of manufacturing; transportation, communication, electric, gas, and sanitary services; and retail and wholesale trade. Results are presented for |${\it INV} = {\it CEx}$|⁠. We find that all five coefficients of ILLIQ are negative and significant, varying between |$-$|0.006 (⁠|$t=-8.80$|⁠) and |$-0.019 (t=-5.64)$|⁠. We conclude that our result on the negative investment-illiquidity relation applies to all five industry groups.

1.1.2 Consistency over time

In panel C, we examine the consistency over time of the negative INV-ILLIQ relation by splitting the sample into two subperiods and estimating the model separately for each subperiod. Also, given that firm fixed effects are assumed to control for time-invariant firm characteristics,¹⁸ a shorter estimation period makes this assumption more reasonable because some firm characteristics may change over time. We find that the coefficient of ILLIQ is negative and significant in both subperiods. For |${\it INV} = {\it CEx}$|⁠, the coefficients of ILLIQ are |$-$|0.011 (⁠|$t=-12.02$|⁠) and |$-0.007$| (⁠|$t=-9.78$|⁠) for the first and second subperiod, respectively. Similarly, for |${\it INV} = {\it dTA}$| the coefficients of ILLIQ|$_{j,t-1}$| for the first and second subperiods are |$-$|0.029 (⁠|$t=- 11.20$|⁠) and |$-$|0.027 (⁠|$t=-7.79$|⁠), respectively. We conclude that the negative effect of illiquidity is consistent over time.

1.1.3 The effect on Nasdaq firms

In panel D, we estimate the model for Nasdaq stocks. The data begin in 1998, after the Nasdaq market-making reform that enabled direct trading between stockholders in a way similar to that done on the NYSE and AMEX, which makes the estimated ILLIQ for Nasdaq similar to that for NYSE and AMEX. Notably, Nasdaq accommodates relatively younger firms in developing industries (such as high tech) that are different from those firms listed on NYSE|$\backslash $|AMEX. Thus, a separate estimation of the investment-illiquidity relation for Nasdaq firms tests our hypothesis for a different type of firm.

Panel D shows similar results for Nasdaq and NYSE|$\backslash $|AMEX firms. The coefficient of ILLIQ is negative and highly significant. We also estimate the model for |${\it INV} = {\it dTA}$| and find that the effect of ILLIQ is negative and highly significant. In Appendix Table A.1, panel B, we present the results for Nasdaq firms with SPRD replacing ILLIQ. There, too, we find that the INV-SPRD relation is negative and significant.

1.1.4 Using industry fixed effects instead of firm fixed effects

We estimate Model (1) by replacing firm fixed effects with industry fixed effects that employ Fama and French’s 49-industry classification.¹⁹ The estimation is done by two methods: a panel regression and annual cross-section Fama-MacBeth regressions. Roberts and Whited (2012) propose performing model estimations with and without firm fixed effects and checking whether the estimated coefficients change in a meaningful way between the two models, which can indicate whether a low-frequency, unobserved omitted variable affects the results. The authors also suggest that estimation without firm fixed effects may provide a better understanding of the cross-sectional effects of the variables because a model with firm fixed effects estimates the within-firm variation rather than the cross-sectional variation that is of interest. Panel E presents the estimations of Model (1), which include industry fixed effects instead of firm fixed effects and employ both panel regressions and annual cross-section Fama-MacBeth (1973) regressions where the annual cross-section coefficients are averaged over the 57 sample years.²⁰

We find that the coefficients of ILLIQ in these estimations are similar to those in panel A in both magnitude and statistical significance. This finding suggests that the model is unlikely to omit an unobservable variable that affects the results.²¹ For example, for |${\it INV} = {\it CExRD}$|⁠, the coefficient of ILLIQ in the panel regression here is |$-$|0.009 (⁠|$t=-10.91$|⁠) with industry fixed effects, compared with |$-$|0.009 (⁠|$t=-13.46$|⁠) in a model with firm fixed effects (panel A). Results with illiquidity measured by SPRD are presented in Appendix Table A.1, panel C. There again, the effect of illiquidity is negative and significant and the magnitudes of the coefficients are similar to those in panel A, where we use firm fixed effects. Employing the Fama-MacBeth estimation procedure, we again find that the coefficients of ILLIQ and SPRD are negative and highly significant with magnitudes that are close to those presented in panel A for panel regressions with firm fixed effects. We also estimate the model with |${\it INV}_{j,t} = {\it dTA}_{j,t}$| and find similar results. For the Fama-MacBeth cross-section regressions procedure, with industry fixed effects, the coefficient of ILLIQ|$_{j,t-1}$| is |$-$|0.012 with |$t=-10.14$|⁠.

1.1.5 Estimating the model with changes in all variables

In this robustness test we convert all variables in Model (1) from levels to changes (first difference) and estimate the model without firm fixed effects.²² We add to the model lagged changes in investment following Eberly, Rebelo, and Vincent’s (2012, p. 370) suggestion that the “lagged-investment effect is empirically more important than the cash-flow and |$Q$| effects combined.” Panel F of Table 2 presents the coefficients of dILLIQ|$_{j,t-1}$| and those of the lagged investment changes; the coefficients of all the variables are presented in Appendix Table A.2 with the prefix |$d$| indicating the first difference of the variable. The models include only year fixed effects.

We find that in the model with |${\it dINV}_{j,t} = {\it dCEx}_{j,t}$|⁠, the coefficient of dILLIQ|$_{j,t-1}$| is negative and highly significant, |$-$|0.008 with |$t=-11.58$|⁠. The magnitudes of these estimates are close to those in panel A for the levels of the variables in a model that includes firm fixed effects. When we add the lagged dependent variable dINV|$_{j,t-1}$| to the model, the coefficient of dILLIQ|$_{j,t-1}$| is |$-$|0.010 with |$t=-13.45$|⁠. The coefficient of the lagged dependent variable dINV|$_{j,t-1}$| is significantly negative, indicating partial reversals of investment, which is often bulky and changes intermittently in large increments. The coefficients of all other variables retain their sign and significance, as in panel A.

1.1.6 Controlling for corporate liquidity: The level of cash and changes in cash

Corporate liquidity—cash and marketable securities—may affect investment and may also be correlated with stock liquidity. We, thus, reestimate Model (1), augmenting it by including Cash|$_{j,t-1}$|⁠, the level of cash and marketable securities, or dCash|$_{j,t-1}$|⁠, the annual change in Cash|$_{j,t-1}$|⁠, both scaled by lagged total assets.

We find in panel G that the effect of ILLIQ on investment remains negative and highly significant. With Cash|$_{j,t-1}$| as a control variable, the coefficient of ILLIQ|$_{j,t-1}$| is |$-0.009$| with |$t=-13.95$|⁠, and it is negative and significant for each of the two subperiods. The coefficient of Cash|$_{j,t-1}$| is |$-$|0.006 with |$t=-1.60$|⁠. Similarly, when we add dCash|$_{j,t-1}$| to Model (1), the coefficient of ILLIQ|$_{j,t-1}$| remains negative and significant for the entire period and for the two subperiods, whereas the coefficient of dCash|$_{j,t-1}$| is negative and significant for the entire period and for the first subperiod, and it is insignificant for the second subperiod; see panel G.

Our results thus show that the effect of stock illiquidity on investment remains negative and highly significant after controlling for corporate liquidity. Below, we present findings that the negative investment-illiquidity sensitivity is similar when estimating the model separately for firms with high or low corporate liquidity.

1.1.7 Inventory investment

In standard inventory models, optimal inventory declines in its carrying cost, which includes the cost of capital. We thus expect that inventory investment declines in illiquidity, which raises the cost of capital. Notably, Tobin’s Q is not expected to affect inventory investment, therefore, it is not included in the inventory investment models of Carpenter, Fazzari, and Petersen (1994) and Jones and Tuzel (2013). Jones and Tuzel (2013) find that inventory growth is a declining function of the firm’s implied cost of capital that is derived from projected future firm earnings based on analysts’ expectations or earnings forecasting models.²³ They attribute their result to the effect of risk premium on the cost of capital. We test the effect of illiquidity and risk on inventory investment and find that both have negative and significant effects.

We estimate Model (1) with |${\it INV}_{j,t} = {\it dINVTRY}_{j,t}$|⁠, the change in inventory in year |$t$|⁠, scaled by lagged total assets. The model is augmented by dSales|$_{j,t}$|⁠, the annual change in sales, scaled by lagged total assets, following Carpenter, Fazzari, and Petersen’s (1994, p. 76) suggestion that “inventory investment is positively correlated with contemporaneous sales shocks.”

In Table 3, we find that ILLIQ|$_{j,t-1}$| has a negative effect on inventory, which is similar to its effect in the investment model of Table 2. The coefficient of ILLIQ|$_{j,t-1}$| is |$-$|0.005 with |$t=-10.75$| for the entire period and |$-$|0.006 (⁠|$t=-9.53$|⁠) and |$-$|0.003 (⁠|$t=-7.24$|⁠) for the first and second subperiods, respectively. Notably, the effect of |$Q_{j,t- 1}$| is not robust, and its coefficient is altogether negative, as opposed to being positive and significant in the capital-investment model. The negative coefficient of |$Q_{j,t-1}$| is significant for the entire period but insignificant for each of the two subperiods. The effect of past stock returns, RET2, which may predict better future prospects for the firm, is positive and significant. The effect of volatility, VOL, is negative and significant, which is consistent with Jones and Tuzel (2013)’s results. The effects of dSales|$_{j,t}$| and Cash Flows|$_{j,t}$| are positive and significant, consistent with Carpenter, Fazzari, and Petersen (1994). When we add dSales|$_{j,t-1}$| to the model, its coefficient is 0.003 with |$t = 1.34$|⁠, which is insignificant.

We reestimate the model measuring illiquidity by SPRD|$_{j,t-1}$| instead of ILLIQ|$_{j,t-1}$| and find that the coefficient of SPRD|$_{j,t-1}$| is |$-$|0.005 with |$t=-6.56$|⁠.

We conclude that illiquidity has a negative and significant effect on inventory investment, similar to its effect on capital investment. In both cases, the negative effect of illiquidity can be explained by its effect on the firm’s cost of capital, which negatively affects investment.

1.1.8 Using illiquidity lagged by 2 and 3 years

We estimate the effect on current investment of ILLIQ lagged by |$2$| years or |$3$| years relative to investment, which mitigates concerns about a simultaneous codetermination of investment and illiquidity. We estimate Model (1) replacing ILLIQ|$_{j,t-1}$| with its value lagged by 2 or 3 years, ILLIQ|$_{j,t-2}$| or ILLIQ|$_{j,t-3}$|⁠, while leaving all other variables at a 1-year lag as before; CF|$_{j,t}$| is contemporaneous. We find that the effect on investment of lagged ILLIQ, which is persistent, remains negative and significant when it is lagged further back by 2 or 3 years. For |${\it INV}_{j,t} = {\it CEx}_{j,t}$|⁠, we find:

• (i) Two-year lag: The coefficient of ILLIQ|$_{j,t-2}$| is |$-$|0.006 with |$t=-9.29$|⁠.

• (ii) Three-year lag: The coefficient of ILLIQ|$_{j,t-3}$| is |$-$|0.003 with |$t=-4.41$|⁠.

The results are similar for |${\it INV}_{j,t} = {\it CExRD}_{j,t}$|⁠. We replicate this estimation for inventory investment, dINVTRY|$_{j,t}$|⁠, using Model (1) augmented by dSales|$_{j,t}$| and find that the negative effect of 2- and 3-year-lagged illiquidity remains negative and significant:

• (i) Two-year lag: The coefficient of ILLIQ|$_{j,t-2}$| is |$-$|0.003 with |$t=- 7.53$|⁠.

• (ii) Three-year lag: The coefficient of ILLIQ|$_{j,t-3}$| is |$-$|0.002 with |$t=- 4.28$|⁠.

1.1.9 Investment in intangible assets

We now use as a dependent variable InvInt|$_{j,t}$|⁠, the investment in intangible assets, in knowledge capital, and in organizational capital. Following Peters and Taylor (2017), InvInt|$_{j,t}$| is the sum of R&D spending plus a fraction of the firm’s selling, general, and administrative (SG&A) expenditures,²⁴ scaled by |${\it TA}{^{{tot}}}_{j,t-1}$|⁠, the sum of the physical assets and intangible assets at replacement cost (the cumulative past investment in intangible assets depreciated at an industry-specific rate) and the book value of property, plant, and equipment (Compustat PPEGT item). We also estimate the determinants of RDex|$_{j,t}$|⁠, the spending on R&D alone scaled by TA|${^{tot}}_{j,t-1}$|⁠. The explanatory variables include TA|${^{tot}}_{j,t-1}$|⁠, |$Q^{{tot}}_{j,t-1}$|⁠, Peters and Taylor’s (2017) Tobin’s Q, which accounts for intangible assets in the denominator, and cash flow CF|${^{tot}}_{j,t,}$|⁠, defined as net income (before extraordinary items) plus depreciation and amortization scaled by TA|${^{tot}}_{j,t-1}$|⁠. Data for |$Q{^{tot}}_{j,t}$| and the firm’s intangible assets (at replacement cost) are obtained from Wharton Research Data Services (WRDS). The sample includes 46,151 firm-years between 1976 and 2017 after applying the sample selection criteria suggested by Peters and Taylor (2017).

The estimation results, presented in Appendix Table A.3, show that the coefficient of ILLIQ|$_{j,t-1}$| remains negative and highly significant. For InvInt|$_{j,t}$| and RDex|$_{j,t}$| as dependent variables, the coefficients of ILLIQ|$_{j,t-1}$| for NYSE-AMEX stocks are, respectively, |$-$|0.006 (⁠|$t=- 11.64$|⁠) and |$-$|0.001 (⁠|$t=-7.38$|⁠). The results are similar for Nasdaq stocks.

1.2 The information feedback effect

Another channel by which stock illiquidity may affect investment follows from the feedback effect proposed by Chen, Goldstein, and Jiang (2007). Managers glean from their firm’s stock price the information provided by the trading of informed traders and use it in their investment decision. “[A] positive relation between the investment sensitivity to stock price and the amount of private information incorporated into the price by speculators would imply that the private information in price is new to managers and that managers look at the price to learn this information and use it in their investment decisions” (p. 620). Chen, Goldstein, and Jiang (2007) test whether corporate investment increases in INFO|$_{j,t-1}$|*|$Q_{j,t-1}$|⁠, where INFO|$_{j,t-1}$| is a measure of the private information in the stock price. They find a positive and significant effect of this interaction term, meaning that the investment-|$Q$| sensitivity increases in the extent of private information in prices.

For INFO, Chen, Goldstein, and Jiang (2007) use two measures that are linked to stock illiquidity and largely reflect adverse selection cost due to information asymmetry between informed traders—those possessing private information—and liquidity traders, who are uninformed. Glosten and Milgrom (1985) and Kyle (1985) propose that illiquidity, measured by the bid-ask spread or the price impact of trading, results from asymmetric information between informed and uninformed traders. Chen, Goldstein, and Jiang’s (2007) first measure is the ratio of return idiosyncratic variance to the total variance. Benston and Hagerman (1974) find that the bid-ask spread increases in the idiosyncratic risk while not being affected by the systematic risk, and Ho and Stoll (1981) show theoretically that the bid-ask spread rises in idiosyncratic risk because of market makers’ risk aversion. Evidence shows a positive correlation between idiosyncratic return variance and illiquidity measures including ILLIQ; see Spiegel and Wang (2005) and Han and Lesmond (2011).

The second measure of INFO used by Chen, Goldstein, and Jiang (2007) is PIN, the probability of informed trading, derived from the asymmetry in the arrival rate of orders from informed and uninformed traders (Easley, Kiefer, and O’Hara 1996; Easley et al. 1996). Easley, Hvidkjaer, and O’Hara (2002) find that PIN is positively and significantly correlated with the bid-ask spread and Duarte and Young (2009) find that ILLIQ subsumes the effect of PIN, with which it is positively correlated, in explaining expected return. Ferreira, Ferreira, and Raposo (2011) find that both PIN and ILLIQ have similar explanatory power as measures of stock price informativeness, and Balakrishnan et al. (2014) find that PIN, ILLIQ, and SPRD are similarly affected by corporate disclosure. Bakke and Whited (2010) propose that PIN captures liquidity. In summary, PIN is positively related to illiquidity both theoretically and empirically.²⁵

We augment Model (1) by INFO|$_{j,t-1}$|*|$Q_{j,t}$|⁠, where INFO is either ILLIQ or SPRD. By Chen, Goldstein, and Jiang (2007), the coefficient of INFO|$_{j,t-1}$|*|$Q_{j,t- 1}$| is positive. Our estimation supports this hypothesis. For INV|$_{j,t}$| = CEx|$_{j,t}$|⁠, the coefficient of ILLIQ|$_{j,t-1}$|*|$Q_{j,t-1}$| is 0.001 with |$t = 4.12$| and the coefficient of SPRD|$_{j,t-1}$|*|$Q_{j,t-1}$| is 0.002 with |$t = 3.78$|⁠. For |${\it INV}_{j,t} = {\it CExRD}_{j,t}$|⁠, the coefficient of ILLIQ|$_{j,t-1}$|*|$Q_{j,t-1}$| is 0.001 with |$t = 2.62$|⁠, and the coefficient of SPRD|$_{j,t-1}$|*|$Q_{j,t-1}$| is 0.002 with |$t = 3.45$|⁠. Throughout, the coefficients of ILLIQ|$_{j,t-1}$| and SPRD|$_{j,t-1}$| remain negative and highly significant. For example, for INV|$_{j,t}=$|CEx|$_{j,t}$|⁠, the coefficient of ILLIQ|$_{j,t-1}$| is |$-$|0.010 with |$t=-13.39$|⁠. Thus, the indirect positive feedback effect of ILLIQ|$_{j,t-1}$| on CEx|$_{j,t}$| offsets about 1/10 of the direct negative effect of ILLIQ|$_{j,t-1}$| for |$Q =1.0$|⁠. For INV|$_{j,t}=$|dTA|$_{j,t}$|⁠, as used by Chen, Goldstein, and Jiang (2007), we find that the coefficient of ILLIQ|$_{j,t-1}$|*|$Q_{j,t-1}$| is 0.004 with |$t = 3.53$| and the coefficient of ILLIQ|$_{j,t-1}$| is |$-$|0.029 with |$t=-12.32$|⁠. Here, the indirect positive feedback effect of ILLIQ|$_{j,t-1}$| on dTA|$_{j,t}$| offsets about 1/7 of the direct negative effect of ILLIQ|$_{j,t-1}$| for |$Q = 1.0$|⁠.

We add a falsification test to verify the learning channel. Cash flows are known to be positively correlated with Tobin’s Q (since both are correlated with the firm’s investment opportunity set) but do not incorporate informed traders’ private information. Thus, showing that the results on the interaction effects of INFO (measured by illiquidity) with Tobin’s Q do not replicate with cash flows would help support the learning channel. To that end, we reestimate the model in Table 4, which is Model (1) augmented by INFO|$_{j,t-1}$|*|$Q_{j,t-1}$|⁠, adding the term INFO|$_{j,t-1}$|*CF|$_{j,t-1}$|⁠. The results, presented in Appendix Table A.4, support the learning channel since all the coefficients of INFO|$_{j,t-1}$|*CF|$_{j,t-1}$| are insignificantly different from zero. Using |${\it INV}_{j,t} = {\it CEx}_{j,t}$|⁠, these coefficients for INFO |$=$|ILLIQ and SPRD are, respectively, 0.002 (⁠|$t = 0.87$|⁠) and 0.002 (⁠|$t = 0.38$|⁠). Using INV|$_{j,t}=$|CExRD|$_{j,t}$|⁠, the respective coefficients are 0.001 (⁠|$t = 0.36$|⁠) and |$- $|0.001 (⁠|$t=-0.06$|⁠). At the same time, the coefficients of ILLIQ|$_{j,t- 1}*Q_{j,t-1}$| remain positive and significant for both measures of INV|$_{j,t}$| and both measures of INFO|$_{j,t}$|⁠. Also, the coefficients of ILLIQ|$_{j,t-1}$| are negative and significant, as before.

Our findings thus support two channels by which illiquidity affects investment: investment declines in the cost of capital, which rises in illiquidity, and investment is positively related to the extent of private information incorporated into stock prices, which is positively related to illiquidity.²⁶

1.3 Financial constraint and the effect of illiquidity

We have proposed that illiquidity lowers investment because it raises the hurdle rate on investment projects. Another explanation for the negative effect of stock illiquidity on investment is that illiquidity indicates financial constraint. Morck et al. (1990) and Bond, Edmans, and Goldstein (2012) propose that the stock market affects firms’ investment behavior through its effect on the issuance of new securities to finance investment. The question is whether it follows that illiquidity more strongly affects investment in financially constrained firms.

Following the methodology of Fazzari, Hubbard, and Petersen (1988), we divide firms in each year into three groups by lagged measures of financial constraint and estimate Model (1) separately for each group. First, we sort firms by Hoberg and Maksimovic’s (2015) measure Delaycon|$_{j,t-1}$|⁠, derived from textual analysis in the section “Managerial Discussion and Analysis” in the 10-K report. It indicates that “firms with higher values are more similar to a set of firms known to be at risk of delaying their investments due to issues with liquidity.” The data are available from the authors’ website,²⁷ and include 15,917 firm-years for our sample between 1998 and 2016. We also sort firms by either stock illiquidity, ILLIQ, or by |$-$|1*Equity Capitalization, which is positively correlated with ILLIQ. Higher values of ILLIQ and |$-$|1*Equity Capitalization may indicate higher financial constraint. We estimate Model (1) using INV|$_{j,t}=$|CEx|$_{j,t}$|⁠.

We propose that if the negative effect of ILLIQ|$_{j,t-1}$| on investment is due to illiquidity reflecting financial constraint, its coefficient |$b1$| should be more negative for firms with greater financial constraint and negligibly small for firms that are unconstrained. However, if the channel by which illiquidity lowers investment is by raising the corporate cost of capital, |$b1$| should be negative and significant for both constrained and unconstrained firms.

In Table 5.1, we find that the negative and significant investment-ILLIQ relation holds for all three groups sorted on Delaycon, ILLIQ, or |$-$|1*Equity Capitalization. To save space, we present only the coefficient of ILLIQ|$_{j,t-1}$| and CF|$_{j,t}$|⁠. For the constraint measures that span the entire sample period 1963–2019, we also present results for the two subperiods, 1963–1990 and 1991–2019, that examine whether our results are consistent over time. In Appendix Table A.5, we present the complete results with coefficients for all the variables.

We find that for all classifications of firms by measures of constraint, the coefficient |$b1$| of ILLIQ|$_{j,t-1}$| is negative and significant for all three financial-constraint groups, including the group of the least constrained firms. For the first classification by Hoberg and Maksimovic’s (2015) Delaycon|$_{j,t-1}$| measure of the need to delay investment, |$b1$| is more negative for more constrained firms: moving from low-constraint to high-constraint firms, |$b1$| is |$-$|0.005 with |$t=-4.09$| and |$-$|0.009 and |$t=-4.40$|⁠, respectively; the difference between the coefficients is insignificant.²⁸ However, for the classification by ILLIQ|$_{j,t-1}$|⁠, the coefficient |$b1$| has an opposite pattern: Moving from the lowest-ILLIQ to the highest-ILLIQ group, |$b1$| is |$-$|0.015 (⁠|$t=-9.69$|⁠), |$-$|0.010 (⁠|$t=-8.92$|⁠), and |$-$|0.007 (⁠|$t=-8.80$|⁠), respectively, with the difference between the coefficients being statistically significant. The same pattern holds in each of the two subperiods, and it also holds when moving from big to small firms, that is, from the least constrained to the most constrained firms.

The more negative investment-ILLIQ sensitivity for financially constrained firms, as observed for the Delaycon measure, as well as for some measures of constraint in Table 5.2 below, is consistent with the secondary market liquidity facilitating capital raising Edmans, Goldstein, and Jiang (2012). The greater investment-ILLIQ sensitivity for firms with the more liquid stocks is consistent with Amihud and Mendelson’s (1986) theoretical prediction and empirical evidence on the greater positive effect of illiquidity costs on expected return for more liquid stocks,²⁹ which leads to a more negative effect on corporate investment for firms with liquid stocks.

In Table 5.2, we use seven other measures of financial constraint to split our sample and estimate Model (1) for each group. Three of these measures indicate the availability of corporate liquidity: Cash Distribution equals dividends plus stock purchases divided by market value of equity, following Fazzari, Hubbard, and Petersen (1988); Cash Flow is income before extraordinary items plus depreciation divided by lagged total assets; and Cash Balance is cash and cash equivalents divided by total assets. For all these measures, a higher value means a lesser need for external financing and a lower constraint. Next, following Hovakimian and Titman (2006), we use Age, the number of years from IPO, and Firm Size, measured by total assets, because younger and smaller firms are more financially constrained.³⁰Leverage is the sum of long-term and short-term debt minus cash, divided by total assets; higher debt overhang may indicate constraint. Finally, Whited-Wu is a weighted average of the firm’s characteristics using Whited and Wu’s (2006) model and estimated weights. We multiply the last two measures by |$-$|1 so that lower value implies greater financial constraint, thus making the presentation consistent with that of the other measures.

In Table 5.2, we find that |$b1$|⁠, the coefficient of ILLIQ|$_{j,t-1}$|⁠, is negative and highly significant for all three groups of financial constraint. In particular, ILLIQ negatively affects investment even for the least-constrained firms. The pattern of the coefficient |$b1$| is not consistent across the three terciles for all measures of constraint. For some measures, |$b1$| decreases with financial constraints; for others, it increases; and for some, there is no monotonic pattern across the constraint terciles. In particular, |$b1$| is significantly more negative and less negative for constraint measured by Firm Size and by Leverage, respectively.

For the dINVTRY-ILLIQ sensitivity, results are presented in Table 5.3 using Model (1) augmented by dSales|$_{j,t}$| as we do in the model estimated in Table 3. For the sake of parsimony, we present results for five measures of financial constraint: Delaycon, Cash Distribution, Age, Firm Size, and Whited-Wu. Delaycon and Whited-Wu are multiplied by |$-$|1 to make their ranking consistent with that of the other variables. We find that the coefficient |$b1$|— the dINVTRY-ILLIQ relation—is negative and significant for both financially constrained and financially unconstrained firms. For Delaycon, |$b1$| is more negative for the least constrained firms, while for Firm Size and Whited-Wu, |$b1$| is significantly more negative for more constrained firms. For the other measures, there is no clear pattern across the constraint terciles.

In summary, even for unconstrained firms that can raise funds more easily or that have available internal resources, illiquidity has a significant negative effect on investment in capital assets and in inventory. This finding supports our view that the channel by which illiquidity affects investment is through its effect in raising the expected return required by investors and the firm’s cost of capital.

2. Controlling for Endogeneity

We account for potential endogeneity of illiquidity in two ways. We study the differential effect across firms of two exogenous liquidity-improving events—the decimalization of quotes in 2001 and the Nasdaq reform of 1997—and we employ an instrumental variables method to conduct a 2SLS estimation of Model (1), testing whether the instrumented ILLIQ has a negative effect on investment.

2.1 The effect of decimalization

The 2001 decimalization of quoted stock prices enabled trading at minimum price increments of $0.01 compared to $1/16, or $0.0625, beforehand. Consequently, the minimum bid-ask spread could decline from $0.0625 to $0.01, leading to lower trading costs. Decimalization started on January 29, 2001 for NYSE-AMEX stocks and on April 9, 2001 for Nasdaq stocks.

Bessembinder (2003) finds that the decline in quoted and effective dollar bid-ask spreads after decimalization was greater for stocks that initially had narrower spreads for which decimalization relaxed the lowest bound constraint of $0.0625 on the dollar spread. We thus expect that stocks whose initial spread was more often quoted at the minimum tick experienced a greater decline in their expected return or in their cost of capital. Also, the decline in the cost of capital should be stronger for firms with a narrower spread based on Amihud and Mendelson’s (1986) theory and evidence that the expected return-illiquidity sensitivity is greater for more liquid stocks. By the investors’ clientele effect, stocks with a lower bid-ask spread are held by frequently trading stockholders, who value liquidity more, and thus more liquid stock should realize a greater decline in required return for any given decline in the bid-ask spread.

Post is a dummy variable that equals one for the 2 postdecimalization years 2002–2003. SPD|$_{j}$| captures the “treatment” variable, using either of the following two variables that are based on the predecimalization quoted dollar bid-ask spread of the firm’s stock:

• (i) |$P625_{j}$| is the proportion of quoted dollar spread at 6.25 cents, or $1/16, during the period. Spreads of stocks with a higher value of |$P625_{j}$| were more often constrained before decimalization by the minimum tick of $1/16.

• (ii) LoSP|$_{j}$| is a dummy variable that equals one if the average quoted dollar spread of stock |$j$| is in the bottom quartile, and zero otherwise. Stocks with LoSP|$_{j}$| = 1 are viewed as “treated” stocks whose liquidity was more likely to improve after decimalization.

The variables are calculated for each stock over the period January–July³¹ of 2000 using the TAQ database.³² The variables |$P625_{j}$| and average spread are calculated daily using data between 9:30 am and 4:00 pm, and then we average the variables across the trading days from January 1 to July 31 of 2000. The mean of |$P625_{j}$| is 0.248, the interquartile range is 0.129–0.335 with a standard deviation of 0.16, and the lowest quartile of the average bid-ask spread is $0.132. The sample includes 1,348 NYSE-AMEX and Nasdaq firms that satisfy data requirements and have data for the 2 predecimalization years, 1999 and 2000, and for at least 1 year in the postdecimalization period, 2002–2003. There are 5,252 firm-years in the sample.

We expect that |$b1 > 0$|⁠. This means that following decimalization, firms whose stocks were quoted with a narrower bid-ask spread increased investment by more because their bid-ask spread declined by more, which induced a greater decline in their cost of capital.

The results are presented in Table 6. In panel A, which shows estimates for Model (2), the coefficient |$b1$| is positive and significant for both measures of SPD|$_{j}$|⁠. For |${\it INV}_{j,t} = {\it CEx}_{j,t}$| (columns (1)–(2)), the estimated |$b1$| for |$P625_{j}$| and LoSP|$_{j}$| is, respectively, 0.024 (⁠|$t = 2.78$|⁠) and 0.006 (⁠|$t = 2.01$|⁠).³³ For INV|$_{j,t}=$|CExRD|$_{j,t}$|⁠, the respective values of |$b1$| are 0.034 (⁠|$t = 3.32$|⁠) and 0.010 (⁠|$t = 2.81$|⁠). By the results in column (2), the firms with the more liquid stocks—those in the lowest quartile of bid-ask spread—whose liquidity improved more by the decimalization increased their investment by 0.006 compared with the more illiquid stocks. This is economically meaningful, given that the average CEx|$_{j,t}$| over the 2 predecimalization years is 0.079. For INV|$_{j,t}=$|CExRD|$_{j,t}$|⁠, the increase in investment for firms in the lowest quartile of the bid-ask spread was 0.010, which is meaningful relative to 0.122, its predecimalization average level.

We conduct a falsification test by estimating Model (2) over two sets of time periods that follow the 2001 decimalization event. In Set 1, the “pre” and “post” 2-year periods are 2002–2003 and 2005–2006 (skipping 2004), respectively, and in Set 2, the respective “pre” and “post” periods are 2003–2004 and 2006–2007 (skipping 2005). SPD|$_{j}$|⁠, which indicates a lower dollar spread is either mSP|$_{j}$|⁠, which is minus the average spread, or LoSP|$_{j}$|⁠, which equals one for stocks whose average spread is in the lowest quartile, as before. The dollar spread is estimated over January–July of 2003 for Set 1 and over January–July of 2004 for Set 2. In the decimalization test, a lower dollar spread—higher SPD—in the “pre” period predicts a higher investment in the “post” period, captured by |$b1 > 0$|⁠. The results for the falsification test are different. In Appendix Table A.6, we find that the coefficient |$b1$| is mostly insignificantly different from zero and in one case it is negative and significant. Thus, in the falsification test, a lower dollar spread in the “pre” period does not predict a higher investment in the “post” period.

Yr00, Yr02, and Yr03 are dummy variables that equal one in 2000 (the year before decimalization), 2002, and 2003, respectively. Model (2.1)³⁴ estimates the investment response compared to the base year, 1999 (2 years before decimalization), for the stocks that benefited more from decimalization. We expect that both |$b{\it 12}$| and |$b{\it 13}$| are positive, meaning that investment in the two postdecimalization years is higher than in 1999 for treated firms.

The estimation results of Model (2.1) are presented in Figure 1. Both |$b12$| and |$b13$| are positive and significantly greater than zero, indicating an increase in investment (relative to that in 1999) for treated firms after decimalization, while |$b11$|⁠, which accounts for the predecimalization year, is insignificant (the tests use a 0.10 confidence interval). These results suggest a sustainable and significant rise in investment following decimalization for firms whose stock was quoted at a narrower spread before decimalization, which made them benefit more from the decline in the lower bound on the spread’s quoted value.

For inventory investment, the effect of decimalization is similar. We estimate Model (2) with INV|$_{j}=$|dINVTRY|$_{j}$|⁠, with the control variables as in Table 3. For SPD|$_{j}=P625_{j}$| and LoSP|$_{j}$|⁠, the coefficient |$b1$| is, respectively, 0.039 (⁠|$t= 4.74$|⁠) and 0.012 (⁠|$t = 4.59$|⁠), suggesting that firms that benefited more from the exogenous improvement in liquidity increased more of their inventory investment.

2.2. The effect of the Nasdaq reform

We test the effect of the 1997 Nasdaq reform on investment of Nasdaq firms, following evidence that the reform lowered the bid-ask spreads in this market. Christie and Schultz (1994, 1999) find that Nasdaq dealers avoided quoting odd-eighths prices, which resulted in the inside bid-ask spreads being at least $0.25 (or multiples of $0.25) for some Nasdaq stocks, suggesting a tacit collusion to inflate the spreads. They also find that when dealers in a stock started using odd-eighth quotes routinely, there was a decline in inside spread of that stock. The Nasdaq reform enabled all investors to display their limit orders that competed with the dealers’ quotes, leading to narrower bid-ask spreads. Barclay et al. (1999, Table II) find that, following the reform, inside bid-ask spreads declined for stocks whose bid and ask prices were both quoted in even-eighths prices beforehand. They conclude (p. 14): “... the decline in the avoidance of odd-eighth quotes has a significant and dramatic impact on the width of average inside spreads.” Therefore, we expect a greater improvement in liquidity—a greater decline in the bid-ask spread—for stocks that were more frequently quoted in even-eighths before the reform.

Even8|$_{j}$| is the proportion of quotes in which both bid and ask prices are even-eighths prices. Even8|$_{j}$| is estimated from June to December of 1996, just before the reform. We consider all quoted prices between 9:30 am and 4:00 pm on each trading day using the TAQ database and define as Even8|$_{j}$| the average of the daily proportion of quotes in which both bid and ask prices are at even-eighths between June 1 and December 31 of 1996.³⁵ For f(Even8|$_{j}), $|we use either LEven8|$_{j} =$| ln(1 |$+$|Even8|$_{j})$| or HiEven8|$_{j}$|⁠, which equals one if Even8|$_{j}$| is above the median and zero otherwise. A higher value of f(Even8|$_{j})$| implies a greater likelihood of a decline in the bid-ask spread after the reform. Post is a dummy variable that equals one in the 2 postreform years 1998 and 1999 and zero in the 2 prereform years 1995 and 1996, skipping the reform year of 1997. The sample consists of 599 Nasdaq stocks and 2,286 firm-years. The much smaller number of firm-years in this analysis is expected to reduce statistical significance. The estimation results are presented in Table 7.

We find higher investment for firms with higher f(Even8|$_{j})$|⁠, that is, for firms whose stock’s inside bid-ask spread was more likely to decline after the Nasdaq reform. Investment increased by 0.013 with |$t = 2.16$|⁠, which is significant,³⁶ for stocks with HiEven8|$_{j}$|⁠. This increase is economically meaningful given that the average CEx|$_{j}$| was 0.095 during the 2 prereform years. For CExRD|$_{j}$|⁠, the coefficient of HiEven8|$_{j}$| is 0.017 with |$t = 2.17$|⁠, which is nearly 1/10 the average CExRD|$_{j}$| during the 2 prereform years, which is 0.155. The coefficients for LEven8|$_{j}$| are similarly positive and significant, 0.031 (⁠|$t = 1.87$|⁠) and 0.049 (⁠|$t = 2.48$|⁠) for CEx|$_{j}$| and CExRD|$_{j}$|⁠, respectively.

We conduct a falsification test by estimating Model (3) over two sets of time periods that follow the 1997 Nasdaq reform. In Set 1, the “pre” and “post” 2-year periods are, respectively, 1998–1999 and 2000–2001 (skipping 2000), and in Set 2, the respective “pre” and “post” periods are 1999–2000 and 2002–2003 (skipping 2001). As before, f(Even8|$_{j})$| is either LEven8|$_{j}$| or HiEven8|$_{j}$|⁠. The values of Even8|$_{j}$| are estimated over June–December of 1999 for Set 1 and June–December of 2000 for Set 2.³⁷ In the Nasdaq reform test, the coefficient |$b1$| of Post*f(Even8|$_{j})$| is positive and significant, implying a higher postreform investment for firms whose stock’s bid and ask prices were both quoted more often at even-eighths before the reform. This is not the case for the falsification tests; see the results in Appendix Table A.7. In none of these tests is the coefficient |$b1$| significantly different from zero.

Yr96, Yr98, and Yr99 are dummy variables that equal one in 1996 (the year before the reform), 1998, and 1999, respectively. We expect that both |$b{\it 12}$| and |$b{\it 13}$| are positive, meaning that investment in the 2 postreform years is higher than in 1995 for treated firms.

The estimation results, presented in Figure 2, support our hypothesis. Both |$b{\it 12}$| and |$b{\it 13}$| are positive and significantly greater than zero, indicating a higher investment that was sustainable over the 2 postreform years by firms whose stock liquidity improved more by the Nasdaq reform. The coefficient |$b{\it 11}$| for the prereform year is insignificantly different from zero.

Finally, we test the effect on inventory investment of the liquidity improvement after the Nasdaq reform by estimating Model (3) with INV|$_{j} =$|dINVTRY|$_{j}$| with dSales|$_{j,t-1}$| added to the explanatory variables. We find that the coefficient of LEven8|$_{j}$| is 0.049 with |$t = 3.41$|⁠, and the coefficient of HiEven8|$_{j}$| is 0.019 with |$t = 3.37$|⁠, suggesting that firms whose stock liquidity benefited more from the reform had a greater increase in inventory investment.

2.3 Instrumental variable estimation

We employ an IV estimation as another way to deal with endogeneity. We instrument ILLIQ by the concentration of institutional investors’ holdings of the firm’s stock, denoted CIH. While CIH strongly explains ILLIQ, there is no theory or evidence suggesting that CIH affects investment. This satisfies the exclusion restriction, as proposed by Edmans, Goldstein, and Jiang (2012).³⁸

Theory and evidence suggest that CIH positively affects ILLIQ. Bolton and von Thadden (1998) propose that CIH is associated with a lower level of market-making activity and with a greater information asymmetry, which both lead to lower liquidity. Rubin (2007, p. 221) proposes that “ownership concentration proxies for information asymmetry as ownership concentration measures quantify the incentives of few shareholders to obtain, analyze, and trade on information,” and finds that stock illiquidity is rising in CIH. Blume and Keim (2012, p. 17) propose that an “increase in the number of institutional holders of a stock decreases the average number of shares of the stock held by individual institutions and, thereby, reduces the potential size of a trade and its accompanying liquidity-induced impact.”³⁹

We measure CIH by two variables: (i) HHI, the Herfindahl-Hirschman index of institutional investors’ holdings, calculated as the sum of the squared values of the fraction of each institutional holding of the total share outstanding, and (ii) Breadth, following Chen, Hong, and Stein (2002), the ratio of the number of institutions holding the stock to the number of all institutional investors in the sample at that time. Goldstein, Yang, and Zuo (2020) propose that a broader investor base attracts liquidity to the secondary market. We obtain from Thomson Reuters Institutional Holdings data on institutional ownership reported by the 13F filing for the last quarter of each year. If the number of shares outstanding is missing on Thomson, we use the shares outstanding reported on CRSP for December of that year. We exclude cases in which the total number of shares owned by institutional investors exceeds the number of shares outstanding. Both HHI and Breadth are in logarithm, and we use -Breadth so that the effects of both HHI and -Breadth on ILLIQ are in the same direction. Valid values of HHI|$_{j,t}$| and Breadth|$_{j,t}$| are available for 31,448 firm-years for the period 1979–2019. The two variables are naturally related. In a regression of HHI|$_{j,t}$| on |$-$|Breadth|$_{j,t}$|⁠, in a model that includes firm and year fixed effects, the coefficient of |$-$|Breadth|$_{j,t}$| is 0.700 with |$t = 61.57$|⁠, and the partial |$R^{2}$| is 0.126.

In the first stage of the 2SLS analysis, we estimate Model (1) with ILLIQ|$_{j,t-1}$| as the dependent variable (instead of it being among the explanatory variables), adding to the explanatory variables CIH|$_{j,t-1,}$| which is either HHI|$_{j,t-1}$| or –Breadth|$_{j,t-1}$|⁠. The results are presented in columns (1) and (4), respectively, of Table 8. The coefficients of HHI|$_{j,t-1}$| and –Breadth|$_{j,t-1}$| are both positive and significant, 0.499 with |$t = 13.29$| and 0.538 with |$t = 11.01$|⁠, respectively, which is consistent with both theory and evidence.⁴⁰

In the second stage, we estimate Model (1) replacing ILLIQ|$_{j,t-1}$| with FILLIQ|$_{j,t-1}$|⁠, the fitted value of ILLIQ|$_{j,t-1}$| from the first-stage regression. The results in Table 8 show that FILLIQ|$_{j,t-1}$|⁠, the instrumented value of ILLIQ|$_{j,t-1}$|⁠, has a negative and significant effect on investment. For CEx|$_{j,t-1}$|⁠, we find that for CIH|$_{j,t-1}=$|HHI|$_{j,t-1}$| (column 2), the coefficient of FILLIQ|$_{j,t-1}$| is |$-$|0.013 with |$t=-6.26$| and for CIH|$_{j,t-1}=-$|Breadth|$_{j,t-1}$| (column 5), the coefficients of FILLIQ|$_{j,t-1}$| is |$-$|0.013 with |$t=-6.02$|⁠. The results are similar for CExRD|$_{j,t}$|⁠.

For inventory investment, dINVTRY|$_{j,t}$|⁠, we follow the same procedure using Model (1) that is augmented by dSales|$_{j,t} $|and the instrument CIH|$_{j,t-1}$|⁠. We find that the coefficient FILLIQ|$_{j,t-1}$| for CIH|$_{j,t-1}$|⁠, HHI|$_{j,t-1}$| and |$- $|Breadth|$_{j,t-1}$|⁠, is, respectively, |$-$|0.006 with |$t=-4.12$| and |$-$|0.006 with |$t=-3.37$|⁠. The results thus show that instrumented ILLIQ has a negative and significant effect on investment in capital assets and in inventory.

3 The Effect of Illiquidity on the Firm’s Production Process

We present a novel link between the stock market and the firm’s production process. We propose that market illiquidity makes firms select production processes that are less capital intensive because illiquidity lowers capital investment. Specifically, greater stock illiquidity induces the following outcomes:

• 1) Higher marginal revenue productivity of capital.

• 2) Higher labor/capital marginal rate of substitution and higher labor/capital ratio.

• 3) Lower operating leverage or the extent of fixed costs in production, which includes fixed costs due to all types of investment.

Test 1 : Marginal revenue productivity of capital

The estimation results of Model (4) in Table 9, panel A, support our hypothesis. We find that |$b{\it 3}$| is 0.046 with |$t = 9.01$| and that it is positive and significant in each of the two subperiods. For robustness, we estimate the model without firm fixed effects and find that |$b{\it 3} = 0.048$| with |$t = 9.03$|⁠. The results are similar when replacing the firm fixed effects with industry fixed effects. These results support our proposition that higher illiquidity, which raises the cost of capital, makes the firm select a production process with a higher marginal revenue productivity of capital.

As an alternative measure for MRPK, we use Hsieh and Klenow’s (2009) model that allows for friction in capital markets, which raises the cost of capital. They assume a Cobb-Douglas production function with constant returns to scale, Y |$=$| AK|$^{\alpha} $|L|$^{(1-\alpha )}$|⁠, where Y, L, and K are output, labor, and capital, respectively, and A is a constant. The firm’s profit is |$\Pi =$| PY – wL – (⁠|$1+c_{K})$|*rK, where P is output price, w is market labor cost, r is the market cost of capital, and |$c_{K}$| is an assumed cost of friction in the capital market, which raises the firm’s cost of capital. In our case, |$c_{K}$| indicates the firm-specific illiquidity premium required by capital providers—the holders of its capital claims—that increases in illiquidity. The firm optimally sets its output so that the marginal revenue productivity of capital MRPK is |$\alpha $|*PY/K |$= (1+c_{K})$|*r. Thus, for a given production function, the firm’s MRPK, which is proportional to the revenue/capital ratio PY/K, increases in |$c_{K}$|⁠. Intuitively, a higher cost of capital due to illiquidity induces producing more output per unit of capital.

We test whether MRPK is an increasing function of ILLIQ by estimating Model (1) with the dependent variable Sales|$_{j,t}$|/TA|$_{j,t}$|⁠, which is proportional to MRPK|$_{j,t}$|⁠.⁴¹ We hypothesize that |$b{\it 1}>0$|⁠, meaning that Sales|$_{j,t}$|/TA|$_{j,t}$|⁠, the output value per unit of capital, increases in ILLIQ|$_{j,t-1}$|⁠. The estimation results in panel B of Table 9 support our hypothesis. The coefficient |$b1$| of ILLIQ|$_{j,t-1}$| is 0.023 with |$t = 5.45$| for the entire sample period, and it is positive and significant for each of the two subperiods, indicating consistency over time.

We employ the decimalization event of 2001 to test the effect of an exogenous change in stock liquidity on the firm’s production process. We estimate Model (2) with Sales|$_{j,t}$|/TA|$_{j,t}$| as the dependent variable. For SPD|$_{j}=P625_{j}$|⁠, the coefficient of Post*SPD|$_{j}$| is |$-$|0.114 (⁠|$t=-2.71$|⁠), indicating that firms whose liquidity improved by decimalization increased the capital intensity of production, which lowered output per unit of capital. For SPD|$_{j}=$|LoSP|$_{j}$|⁠, the coefficient of Post*SPD|$_{j}$| is negative, |$- $|0.021, but insignificant, |$t=-1.26$|⁠.

Testing the effect of the Nasdaq reform on the firm’s production process, we estimate Model (3) with Sales|$_{j,t}$|/TA|$_{j,t}$| as the dependent variable. We again expect that the coefficient of f(Even8|$_{j}) $|is negative. We find for f(Even8|$_{j})=$|LEven8|$_{j}$| and HiEven8|$_{j}$| that the coefficient of Post*f(Even8|$_{j})$| is, respectively, |$-$|0.065 with |$t=-0.78$| and |$-$|0.051 with |$t=-1.79$| (significant at the 0.10 level).

In both cases, the results suggest that firms whose liquidity improved by decimalization or by the Nasdaq reform increased the capital intensity of production, reflected in lower output per unit of capital.

Test 2 : The change in labor input relative to change in assets

We propose that firms with higher stock illiquidity substitute labor for capital because of their higher cost of capital. Employing again Hsieh and Klenow’s (2009) model, the firm’s optimal labor input is given by L |$= [(1-\alpha )$|r(⁠|$1+c_{K})$|/|$\alpha $|w]*K. Thus, dL/dK |$= (1-\alpha )$|r/|$\alpha $|w |$+c_{K}(1-\alpha )$|r/|$\alpha $|w, which increases in |$c_{K}$|⁠. Therefore, the marginal rate of substitution between labor and capital in production—the increase in labor input for a given increase in capital—is positively related to the illiquidity premium in the cost of capital, which increases in |$c_{K}$|⁠. And, the labor/capital ratio is increasing in |$c_{K}$|⁠.

We test whether dL/dK increases in illiquidity by estimating Model (4) with the dependent variable dLabor|$_{j}$|⁠, the annual change in the number of employees scaled by lagged total assets. Then, dL/dK |$= b2 + {\it b3*ILLIQ}_{j,t-1}$|⁠, where ILLIQ is positively related to |$c_{K}$|⁠. Naturally, |$b_{2} > 0$| since the marginal rate of substitution between labor and capital is |$\alpha $|r/|$(1-\alpha )$|w |$> 0$|⁠. We test whether |$b3 > 0$|⁠, that is, whether firms with a higher illiquidity premium realize a greater increase in labor input for a given increase in capital because the higher cost of capital induces firms to select a production process that is less capital intensive.

The estimation results are presented in Table 10, panel A. Naturally, |$b2 > 0$|⁠, meaning that labor input rises to match an increase in capital. Importantly, |$b3 > 0$| as we hypothesize, which means that illiquidity raises the increase in labor input for a given increase in capital. We find that |$b3 = 1.201$| with |$t = 7.60$|⁠. The results are consistent across the two subperiods, |$b3 = 1.478$| with |$t= 7.82$| and |$b3 = 0.371$| with |$t = 9.23$| for the first and second subperiods, respectively.⁴²

As a robustness test, we estimate the model for each of the five one-digit SIC code industries because production functions differ across industries. We find that |$b3$| is positive and significant for all five industries; see panel B of Table 10.

Panel C of Table 10 presents a second test of the effect of illiquidity on the labor/capital ratio. We estimate Model (1) with LaborTA|$_{j,t}$|⁠, the ratio of the number of employees to total assets in year |$t$|⁠, in logarithm, as the dependent variable. We expect a positive coefficient of ILLIQ|$_{j,t-1,}$| meaning that as illiquidity raises the cost of capital, the firm selects a higher labor/capital ratio, thus reducing the capital intensity in production. The results support our hypothesis. The coefficient of ILLIQ|$_{j,t-1}$| is 0.010 with |$t = 2.14$| for the entire period, and it is 0.014 (⁠|$t = 3.06$|⁠) and 0.009 (⁠|$t = 1.98$|⁠) for the first and second subperiods, respectively.

We test the effect of the 2001 decimalization on the labor/capital ratio, estimating Model (2) with the dependent variable LaborTA|$_{j,t}$|⁠. We expect the coefficient of Post*SPD|$_{j}$| to be negative (⁠|$b1 < 0$|⁠), implying a reduction in the labor/capital ratio after decimalization for firms with a greater improvement in liquidity (given that beforehand their spreads were quoted more often at the minimum tick), that is, such firms adopted a production process that was less labor-intensive. We find that the coefficients for SPD|$_{j}$| that equal |$P625_{j}$| and LoSP|$_{j}$| are, respectively, |$-$|0.148 with |$t=- 3.71$| and |$-$|0.015 with |$t=-1.02$|⁠.

Testing the effect of the 1997 Nasdaq reform on the labor/capital ratio, we estimate Model (3) with the dependent variable LaborTA|$_{j,t}$|⁠. We find that the coefficient of Post*f(Even8|$_{j}$|⁠) for f(Even8|$_{j})=$|LEven8|$_{j}$| and HiEven8|$_{j}$| is, respectively, |$-$|0.028 (⁠|$t=-1.00$|⁠) and |$-$|0.028 (⁠|$t=-0.35$|⁠), both negative as expected, though insignificant.

Test 3 : Operating leverage

We test the effect of illiquidity on the firm’s operating leverage, OL, the extent of fixed costs in the total cost of production. This test supplements our analysis of the effect of illiquidity on investment and accounts for alternatives to investing in assets such as leasing them.⁴³ If switching from buying to leasing assets is a costly endeavor driven by higher illiquidity that would not have been done otherwise, it means that illiquidity imposes a cost that may inhibit the use of capital assets. Reducing the use of capital induces firms to adopt production processes that rely less on fixed costs and more on variable costs. We test this hypothesis below.⁴⁴

Cost|$_{j,t}$| and Sales|$_{j,t}$| are contemporaneous, whereas ILLIQ|$_{j,t-1}$| is lagged. Cost|$_{j,t}$|⁠, defined as Sales|$_{j,t}$| – EBIT|$_{j,t}$| (earnings before interest and taxes), includes all costs, both variable and fixed.⁴⁵ Its main components are cost of goods sold; selling, general, and administrative costs; R&D costs; and depreciation. Sales|$_{j,t}$| and Cost|$_{j,t}$| are scaled by lagged assets.

Operating leverage is defined as |${\it OL} = 1-b2$|⁠, following Lev (1974). Here, operating leverage is |${\it OL} = 1-b2-$|b3*ILLIQ. We hypothesize that |$b3 > 0$|⁠, that is, higher illiquidity reduces operating leverage, inducing greater reliance on variable costs in production.

The estimation results of Model (5), presented in Table 11, support our hypothesis: |$b3 = 0.001$| with |$t = 2.61$|⁠. The estimated coefficient |$b3$| is consistent in the two subperiods in both magnitude and in statistical significance: for the first and second period, |$b3$| is 0.002 (⁠|$t = 2.20$|⁠) and 0.002 (⁠|$t = 2.28$|⁠), respectively.

We expect that |$b3 < 0$|⁠, meaning that stocks with a narrower bid-ask spread, for which liquidity improved the most, increased their operating leverage and relied less on variable costs that normally rise with sales. For SPD|$_{j} = P{\it 625}_{j}$| and LoSP|$_{j}$|⁠, we find that the coefficient |$b3$| is negative and significant, |$-$|0.029 (⁠|$t =-3.02$|⁠) and |$-$|0.012 (⁠|$t=-1.95$|⁠), respectively. The negative |$b3$| implies that |${\it OL} = 1-b2-b3 > 1-b2$|⁠. That is, operating leverage has increased for the lower-spread firms whose liquidity improved more by decimalization.

We test the effect of the Nasdaq reform by estimating Model (5.1) with f(Even8|$_{j})$| replacing SPD|$_{j}$|⁠. We again expect |$b3 < 0$|⁠, meaning that firms whose bid-ask spread was likely to decline by more increased their reliance on fixed costs in production. We find for f(Even8|$_{j})$|⁠, which equals LEven8|$_{j}$| and HiEven8|$_{j,}$| the coefficient |$b3$| is, respectively, |$-$|0.024 with |$t=-1.73$| and |$-$|0.007 with |$t=-1.00$|⁠.⁴⁶

4. Concluding Remarks

This paper presents a channel by which Wall Street affects Main Street: the secondary market liquidity affects corporate investment and production decisions. Because illiquidity raises the expected return required by stockholders, it raises the firm’s opportunity cost of capital and, consequently, lowers corporate investment. The negative investment-illiquidity relation holds even for firms that are not financially constrained or have no need to raise capital, suggesting that the effect of illiquidity on investment is through the secondary, rather than the primary, market. Because illiquidity inhibits capital investment, it induces firms to economize on the use of capital in production. Firms with higher illiquidity have higher output per unit of capital and rely relatively more on labor input in production. Generally, their costs consist less of fixed costs and more of variable costs.

Our results provide a broad interpretation of Chen, Goldstein, and Jiang’s (2007) proposed feedback effect from the secondary market to the real economy by which firms’ investment decisions employ information learned from secondary market prices. In our context, managers learn about their firm’s stock liquidity and cost of capital from market prices and trading activity and use this information in their investment decisions. Just as most managers use their firm’s |$\beta $| risk and risk premium derived from market price data to calculate the cost of capital by the CAPM (Graham and Harvey 2001), managers learn from the secondary market about their firm’s stock illiquidity and its premium and use this information to estimate the expected return required by investors, which is used to determine the investment hurdle rate. Managers can employ asset pricing models to estimate the illiquidity premium (see Amihud, Mendelson, and Pedersen 2013) or learn about the illiquidity effect from the price/earnings multiple, which declines in illiquidity (Loderer and Roth, 2005). This analysis is consistent with Bond, Edmans, and Goldstein’s (2012) proposed feedback from secondary market price information to the real economy.

Our results suggest that firms may benefit from spending resources on improving their stock liquidity, which would in turn reduce their cost of capital. Liquidity can improve by improved disclosure of corporate information, by a security design that makes the corporate claims more liquid, and by increasing the company float (see Amihud and Mendelson 1988). Balakrishnan et al. (2014) find that stock liquidity increased in firms that increased voluntary disclosure by providing timely and informative earnings guidance. Amihud, Mendelson, and Uno (1999) find a rise in stock liquidity and stock price after firms reduce their stock’s minimum trading unit, and allow smaller investors to invest in their stock. And Amihud, Lauterbach, and Mendelson (2003) find that stock liquidity rises after firms eliminate fragmented trading in their equity securities by consolidating them.

Our results suggest that reforms in the secondary market trading procedures that improve market liquidity, such as the 1997 Nasdaq market-making reform and the 2001 decimalization, benefit the real economy by affecting investment and production. More generally, government policies and regulations that increase the information available to traders, reduce asymmetric information, and increase liquidity are expected to have a similar effect. For example, mandating periodic disclosure of accounting information and setting a uniform and comprehensive format of disclosure improves liquidity and lowers the cost of capital. Leuz and Verrecchia (2000) find that liquidity improved for firms in Germany that switched to an international reporting regime (IAS or U.S. GAAP). Similarly, transparency of trading in the market improves liquidity, as seen for example when corporate bond trading started to be reported on the TRACE system (see Bessembinder, Maxwell, and Venkataraman 2006). Henry (2000) finds that broad stock market liberalizations, which allowed foreigners to purchase shares in the country’s stock market, and increased net capital inflows lead to private investment booms. Among the explanations for this finding, Henry (2000) proposes that increased capital inflow increases market liquidity, which, following Amihud and Mendelson (1986), lowers the equity premium. Our analysis focuses on liquidity-related market reforms, showing how they raise investment.

Macroeconomic analysis by Naes, Skjeltorp, and Odegaard (2011) finds that aggregate stock market liquidity positively affects growth in aggregate private real investment. Our analysis provides microeconomic firm-level evidence that is consistent with the positive macroeconomic relation between stock market liquidity and investment.

Appendix

Acknowledgement

We thank two anonymous referees and particularly the editor, Itay Goldstein, for comments and suggestions that helped improve the paper. We also thank Alex Edmans, Eti Einhorn, Vivian Fang, Avner Kalay, Haim Mendelson, Roni Michaely, Thomas Philippon, Alexi Savov, Dan Weiss, Ivo Welch, Avi Wohl, and David Zvilichovsky and seminar participants at WHU, NYU-Stern, City University of Hong Kong, NYU-Shanghai, Tsinghua University, Tel Aviv University, Ben-Gurion University, and IDC Arison School for helpful comments. Amihud is the Ira Leon Rennert professor of Entrepreneurial Finance. Levi’s research was supported by the Henry Crown Institute of Business Research and the Jeremy Coller Foundation.

Footnotes

References