Investment–cash flow sensitivity measures investment thirst, but not financial constraint

2.1 Soft budget constraint

According to Kornai (1979, 1980, 1986), investment thirst is driven by soft budget constraint. A soft budget constraint can occur for three reasons: first, when investors lack the wherewithal to ensure an enterprise stays on budget; second, when the investment plans proposed by the manager are ex-ante efficient; and third, when the manager overinvests at his or her own volition. The issue therefore with soft budget constraint is that it may motivate managers to follow irresponsible incentives and thus risk enterprise finance without adequate surveillance (Dewatripont and Maskin, 1995; Qian and Roland, 1998). The soft budget-constraint problem is especially prominent in transitional economies. Qian and Roland (1998) model the relationship between soft budget constraint and federalism tailored for the China economy to show that soft budget constraint significantly characterises Chinese firms. This is also why we chose Chinese markets to test the relationship.

2.2 Investment thirst, free cash flow and overinvestment

Investment thirst, therefore, arises from two particular investment approaches: first, when firms characterised by soft budget-constraint overinvest, and second, from managers’ investment impulsion for short-term gains. The often-researched relationship between overinvestment and cash flows is derived from Jensen (1986) theory paper on free cash flows (FCFs). FCFs are considered cash flows beyond those required to fund all projects with positive net-present values. Lang and Litzenberger (1989) find a significantly positive relationship between the extent of managerial overinvestment and market reaction. They conclude that overinvestment denotes that the situation firms possessing high FCFs are investing in unprofitable projects.

Furthermore, Richardson (2006) creates a measure of overinvestment to provide direct empirical evidence of a significantly positive relationship between FCFs and overinvestment. Therefore, the positive relationship between FCF and overinvestment is valid both theoretically and empirically. Vogt (1994) investigates the over- and underinvestment effects of ICFS to find that, for large firms of low dividends, ICFS signifies overinvestment, while ICFS in small firms of low dividends, it indicates financial constraint. Such evidence suggests a positive relationship between ICFS and overinvestment.

Although both streams of theories indicate a positive relationship between ICFS and overinvestment, these theories vary in three ways: (i) the investment thirst theory introduced by Kornai (1979, 1980, 1986) stresses the necessity of soft budget constraint and expansion impulsion, while Jensen (1986) FCF theory does not. Instead, Jensen attributes the main reason to the agency problem induced by free cash flows. (ii) Soft budget constraint in a firm refer to the situation whereby the manager violates his or her promises to investors about using investment cash flows previously budgeted for (e.g. the original investment plan would have been considered efficient by investors, while the investment in violation ex post is not). This occurs because of a lack of surveillance received by managers, whereas FCF is linked to investments after the investment plan promised by the manager has been executed. (iii) While we predict that investment thirst is closely related to ICFS, this is not necessarily the case for FCFs. Investment thirst induces a firm to invest eagerly, resulting in a higher level of ICFS. In contrast, FCFs without monitoring relax the direction of investments and hence do not necessarily change the sensitivity of investment–cash flows. Therefore, the point is that overinvestment caused by investment thirst is closely related to ICFS, while overinvestment caused by FCF may not be.

Our study differs from the existing literature in three significant ways. First, almost all the existing finance studies explain the overinvestment problem within the framework of the FCF theory developed by Jensen (1986).5 However, we find the investment thirst theory from the macroeconomics work of Kornai (1979, 1980, 1986) better explains the overinvestment phenomenon in transitional economies such as China.

Second, we discover that ICFS validly measures investment thirst at the firm level. In existing macroeconomics studies (see Lee et al., 2012), the ratio of incremental capital to output is commonly used to measure investment thirst at the country level. However, no analogous index at the firm or industry level has been offered. We propose that investment–cash flow sensitivity (hereafter ‘ICFS’) is a good firm-level measure of investment thirst.

Third, we find that the ICFS index combines the effects of financial constraint and investment thirst on investment, whereas many of the existing studies do not discern these two effects.6 In this paper, our empirical tests are designed in a way such that the effects of investment thirst and financial constraint on ICFS can be separated, and therefore, the relationship between investment thirst and ICFS can be directly tested.

Correspondingly, our findings enrich the literature in three main ways. First, we show the connection between ICFS and investment thirst theoretically. For example, we propose that ICFS is more closely related to investment thirst than financial constraint. Second, we present evidence to demonstrate that in a transitional economy like China, in which firms face severe financial constraint and exhibit it concurrently with investment thirst, ICFS measures investment thirst after financial constraint is controlled for. Third, using the natural experiment setting we have chosen for this research offers two benefits: it makes our empirical results less prone to endogeneity problems;7 second, the empirical findings support our conjecture that ICFS is a good measure of investment thirst.

5.1 The regression model

We build our regression model as:

(6)

Where I_t/K_t−1 represents the firm's capital expenditure I_t scaled by beginning-of-period total asset, K_t−1; Tobin's Q_t−1 is defined as the previous year's Tobin's Q. (S_t−S_t−1)/S_t−1 is the sales growth rate. CF_t/K_t−1 is the firm's internal cash flow (net operating cash flow), with CF_t deflated by its beginning-of-period total asset, K_t−1. We also introduce the interaction term Tobin's Q_t−1×CF_t/K_t−1 here according to Vogt (1994), of which the sign of the coefficient represents the characteristics of ICFS.13 In detail, if the sign is negative, we can deduce that ICFS rises while investment opportunities decline, and that ICFS should be correlated with soft budget constraint, firm risk-seeking behaviour and overinvestment.14 Hence, introducing the interaction term here can help identify the relationship between ICFS and investment thirst. We also include year- and industry-fixed effects.

5.2 Data and sample selection

To determine the relationship between ICFS and investment thirst, we select a sample of Chinese publicly listed firms for a number of reasons. First, according to Kornai (1979, 1980, 1986) and Qian and Roland (1998), the problem of investment thirst usually arises from transitional economies characterised by soft budget constraint. Indeed, China is the largest emerging and transitional economy in the world. Second, Chinese firms might be facing severe investment thirst and financial constraint problems at the same time, which could provide benefits in testing wherein ICFS can be a good proxy for investment thirst. For example, an International Monetary Fund (IMF) report entitled ‘Is China Over-Investing and Does it Matter?’ suggests that China has experienced prominent overinvestment problems over the last decade; this conclusion draws on the evidence of comparing the incremental capital–output ratio among a large numbers of countries (Lee et al., 2012). Meanwhile, according to a series of surveys conducted by the Development Research Center of State Council in China, 2007 to 2012, more than 90 percentage of managers in Chinese firms believe they are facing serious financial constraint.15 These two pieces of evidence indicate that financial constraint and investment thirst co-exist in Chinese firms.

Although the sample for our research was set to be derived over 2001 to 2012, the regression model was narrowed to a sample period of 11 years from 2002 to 2012, after we took into account that some variables are 1-year lagged. We also eliminated observations from the finance industries as well as those missing values in the variables used in the study. Our final sample hence consists of 15,576 firm-year observations from 2297 firms in 30 provinces of China. We obtained all accounting data from the China Stock Market and the Accounting Research (CSMAR) database, which is widely used in China-related research.

5.3 Summary statistics

We split the sample into four groups according to the quartile values of financial constraint in each year. In Panels A and B of Table 1, we present summary statistics of each subsample based on the WW index of Whited and Wu (2006) and the SA index developed by Hadlock and Pierce (2010) to denote the degree of financial constraint for each.16 The two studies suggest that the WW index is supposed to be positively related to the level of financial constraint, whereas the SA index has a reverse relation to financial constraint. In Panel A, while we observe Tobin's Q_t−1 increases with a rising value in WW index and financial constraint, Size_t and (S_t−S_t−1)/S_t−1 decrease with the augment of the WW index, I_t/K_t−1 and CF_t/K_t−1 do not show any obvious relationship with the WW index. In Panel B, we observe that the coefficient of Tobin's Q_t−1 is decreasing with the value of SA index goes up (and augment of financial constraint), and I_t/K_t−1, Size_t, CF_t/K_t−1, and (S_t−S_t−1)/S_t−1 increase when the value in SA index goes up (and decreasing financial constraint). The results on the SA index are more persistent than those of using the WW index, which indicate the SA index provides a better measure of financial constraint than the WW index in this study. Moreover, in Panel B of Table 1, the value of Tobin's Q_t−1 in groups with less financial constraint (Groups 3 and 4) is lower than 1, which coincides with the definition of ‘bad firms’ in Lang and Stulz (1994) and the definition of ‘overinvestment companies’ in Lang and Litzenberger (1989).

We hereby identify the relationship between ICFS and investment thirst in three steps: first, we need to disentangle the relationship between ICFS, financial constraint and overinvestment. By doing this, we can show that ICFS is more closely related to overinvestment than to financial constraint. Second, we offer direct evidence of the positive relationship between ICFS and a firm's risk-seeking preferences. Third, we present empirical results to show that the value of ICFS is related to soft budget -constraint and investor infringements. In particular, using the exogenous setting of a natural experiment of the ‘Four Trillion Stimulus’ policy in China, we provide valid evidence on the explanatory power of ICFS on investment thirst.

6.1 ICFS, financial constraint and overinvestment

We first divide the sample into four groups according to their degree of financial constraint (the WW index and the SA index, see Table 1). Group 1 consists of firms with the least financial constraint, while Group 4 consists of firms with the most financial constraint. In Panel A of Table 2, the sample is sorted by the value of the WW index using the regression model without the interaction term Tobin's Q_t−1 × CF_t/K_t−1 and the WW index (results displayed in the left-hand side of Panel A of Table 2).17 No definite pattern is observed in the relationship between the financial constraint and the coefficient of CF_t/K_t−1. Nevertheless, after introducing the interaction term Tobin's Q_t−1 × CF_t/K_t−1 and the WW index term, we observe that the coefficients of CF_t/K_t−1 in Groups 1 and 2 are larger than those in Groups 3 and 4 (results displayed in the right-hand side of Panel A of Table 2) showing that all the coefficient estimates are significant at a 1 percentage level.

Panel B of Table 2 shows the regression results of the sample sorted on the value of the SA index. In the regression model without the interaction term Tobin's Q_t−1 × CF_t/K_t−1 and the SA index term, the coefficients of CF_t/K_t−1 in Groups 1 and 2 are larger than those in Groups 3 and 4. After introducing the interaction term Tobin's Q_t−1 × CF_t/K_t−1 and the SA index term, the results only change slightly. Notably, in Panels A and B of Table 2, a significantly positive coefficient of CF_t/K_t−1 is accompanied by a negative coefficient of Tobin's Q_t−1 × CF_t/K_t−1,18 which implies that, in China, ICFS is significantly correlated with soft budget constraint and investor infringements.

Panel A of Table 2 shows that most of the coefficients of Tobin's Q_t−1 are negative and that the pattern does not change even after introducing the WW index term. On the contrary, Panel B of Table 2 shows that, after introducing the SA index term, the coefficients of Tobin's Q_t−1 become larger and most become positive. Thus, we hereby argue that the ability of the SA index in discerning the effect of financial constraint and the effect of investment opportunities (e.g. Tobin's Q) is greater than that of the WW index in our sample.

Following Richardson (2006), we next sort our firms based on the quartile values of overinvestment each year.19 Each sample firm falls into one of the four exclusive groups: Groups 1 and 2 represent firms of underinvestment, while Groups 3 and 4 represent firms of overinvestment. Table 3 shows ICFS levels in each of these four groups. In Panel A of Table 3, with the sample sorted on the value of the WW index using the regression model without the interaction term Tobin's Q_t−1 × CF_t/K_t−1 and the WW index, we observe a significantly positive relationship between the level of investment and CF_t/K_t−1. Furthermore, after introducing the interaction term Tobin's Q_t−1 × CF_t/K_t−1, the significant positive relationship between the level of investment and CF_t/K_t−1 is unchanged. This consistency implies that ICFS is closely related to overinvestment. Furthermore, the largest coefficient of CF_t/K_t−1 in Group 4 is accompanied by a significantly negative coefficient of CF_t/K_t−1.

Panel B of Table 3 shows more evidence of the relationship between ICFS and overinvestment after adding the interaction term Tobin's Q_t−1 × CF_t/K_t−1 and the financial constraint index term. When using the WW index as the proxy for financial constraint, the regression results are almost the same as for the results in the right-hand side of Panel A of Table 3. The coefficient of Tobin's Q_t−1 remains negative, and most coefficients of the WW index are positive, both of which illustrate that, using the WW index, it is hard to separate the effect of financial constraint and that of Tobin's Q in our sample. However, when the SA index instead of the WW index is used as the proxy for financial constraint, the regression results are different to those in Panel A of Table 3. Most of the coefficients of Tobin's Q_t−1 become positive, and the coefficients of the SA index are significantly positive. Therefore, this again demonstrates that the capability of the SA index to discern both the effect of financial constraint and the effect of Tobin's Q on investment is better than the WW index for our sample.

6.2 ICFS and risk-seeking preferences

In the second step, we discuss the relationship between ICFS and risk-seeking preferences. We sort our firms based on the quartile values of relative idiosyncratic risk in each year20 and clarify all observations into four groups from the lowest risk (Group 1) to the highest risk (Group 4).21

However, it is difficult to determine the effect of risk-seeking preference on ICFS by computing ICFS directly in each group, because of its close association with soft budget constraint and investor-infringement effects. To account for the possible confounding effects, we have to manage our sample more carefully. Therefore, we construct a sample consisting of observations with low soft budget constraint only. We examine our results in Table 3 and find that the coefficient of Tobin's Q_t−1 × CF_t/K_t−1 in Group 2 is the largest; this suggests that firms in this group could, on average, afford the lowest soft budget constraint. However, even when we include firms in Group 2 of Table 3 in the sample, we cannot rule out the effect that firms with high soft budget constraint tend to have lower levels of soft budget constraint together, which can also lead to a high coefficient of Tobin's Q_t−1 × CF_t/K_t−1. To mitigate such effects as much as possible, we further examine the results in Table 2 and identify that the coefficient of Tobin's Q_t−1 × CF_t/K_t−1 in Group 4 is the largest. Consequently, we select the observations existing in both Group 4 of Table 2 and Group 2 of Table 3 simultaneously and divide them into four groups by their quartile values of relative idiosyncratic risk.

Table 4 presents the results of the relationship between ICFS and risk. We find the coefficients of CF_t/K_t−1 in Groups 3 and 4 are commonly larger than those in Groups 1 and 2. However, as this is an obviously exceptional case, the largest coefficient of CF_t/K_t−1 emerges in Column 1 of Panel B, although, at the same time, the only significant negative coefficients of CF_t/K_t−1 are also shown here. Therefore, the abnormal coefficient of CF_t/K_t−1 in Column 1 of Panel B not only represents the effect of risk-seeking preference, but also includes the effect of soft budget constraint and investor infringements. In short, our evidence shows a positive relationship between ICFS and risk-seeking preferences, although it is difficult to remove the effect of soft budget constraint from that of risk seeking.

Similar to Table 3, when using the WW index to denote financial constraint in Table 4, we still observe the coefficient estimate of Tobin's Q_t−1 being negative, whereas most of the coefficient estimates of the WW index are positive. This means that, with the WW index, separating the effect of financial constraint and that of Tobin's Q in our sample is again difficult. Additionally, when we use the SA index as the proxy for financial constraint, the sign of Tobin's Q_t−1 in most cases becomes positive, and the coefficients of the SA index are significantly positive. Again, we find the ability of the SA index in discerning the financially constrained effect and Tobin's Q effect on investment is better than the WW index in our sample.

6.3 ICFS, soft budget constraint and Investor infringements

Table 5 presents the regression results of ICFS in each year from 2002 to 2012. Panel A shows both the results for ICFS from the regression model without the interaction term of Tobin's Q_t−1 × CF_t/K_t−1 and the financial constraint proxy. We do not find any obvious pattern in the value of ICFS coefficient estimates. After introducing the interaction term Tobin's Q_t−1 × CF_t/K_t−1, we find that the coefficients of CF_t/K_t−1 are significant and increasing in the three periods of 2002–2004, 2007–2008 and 2009–2012. We note that they are insignificant in 2005 and 2006 based on the results shown in Panel B of Table 5. Furthermore, most of the significant coefficient estimates on CF_t/K_t−1 are accompanied by negative coefficient estimates on Tobin's Q_t−1 × CF_t/K_t−1, which implies ICFS, soft budgets and investor infringements are closely correlated.

To separate the effect of financial constraint on ICFS from that of investment thirst, Panels C and D introduce the WW index and SA index, respectively. Similar to Tables 3 and 4, in which we used the WW index to denote financial constraint, in Panel C of Table 5, we still observe that the signs of coefficient estimates on Tobin's Q_t−1 are negative in most cases. Nevertheless, these signs are positive in the WW index coefficient estimates in most cases, both indicating that, for the WW index, it is hard to separate the effect of financial constraint and that of Tobin's Q.

Next, we use the SA index (instead of the WW index) as the proxy for financial constraint, which shows that most of the coefficients of Tobin's Q_t−1 become positive, and a number of the coefficients of the SA index are significantly positive. The result of this comparison again shows the ability of the SA index to discern the effect of financial constraint and the effect of Tobin's Q on investment, as well as its relative efficiency compared to the WW index. Consequently, the results in Panel C are almost identical to those in Panel B, while the results in Panel D differ somewhat. In particular, the coefficient estimates of CF_t/K_t−1 are significant in that they increase in the three periods of 2002 to 2003, 2007 to 2008 and 2009 to 2012, while being insignificant between 2004 and 2006.

6.4 ICFS and incremental capital–output ratio

In this section, we discuss the third step used in our empirical work and test the explanatory power of ICFS on investment thirst. Existing studies in macroeconomics typically employ the incremental capital-to-output ratio to detect investment thirst and overinvestment in China (e.g. Lee et al., 2012).22 If ICFS is a good measure of investment thirst as we suggest, it should coincide with the incremental capital-to-output ratio.

In a typical transitional environment such as China, where investment thirst tends to connect with government policy (Kornai, 1980; Stiglitz, 1996), we test the effect of ICFS on investment thirst using a natural experiment of a typical government policy that occurred during the sample period. The Four Trillion Stimulus investment policy, announced in December 2008, is regarded as the most important policy adopted by the Chinese government. Thus, we use this event to analyse the relationship between ICFS and investment thirst as follows.

Figure 2a,b compare the changing trends of ICFS with that of incremental capital-to-output ratios from 2002 to 2012. The values of ICFS in Figure 2a are sourced from Panel D of Table 5, and the values of incremental capital-to-output ratios in Figure 2b are calculated using the data from investment on fixed assets and gross domestic product (GDP) during 2000 to 2013.23 We observe that the changes in ICFS (firm-level measure of investment thirst) accurately mirror the changes in incremental capital-to-output ratios (country-level measure of investment thirst) changes over the sample period.

7.1 Subperiod regressions

We find similar trends in both ICFS over time and the incremental capital-to-output ratios, which further confirms the significant relationship between ICFS and investment thirst. Moreover, the ICFS captures the effect of the Four Trillion Stimulus investment policy of 2008 accurately, whereas the incremental capital-to-output ratios exhibit a 1-year lag. Indeed, while the policy was announced in 2008, most of the investment activities occurred in 2009, which explains the 1-year-lagged macrodata. This difference also suggests that ICFS is more efficient than incremental capital-to-output ratios in detecting investment thirst.

We further split the sample period into four subperiods according to the trends of incremental capital-to-output ratios in Figure 2b, so as to present the ICFS of each subperiod displayed in Table 6. The results of ICFS are 0.1112, 0.0529, 0.1186 and 0.1061 in the using the data from 2002 to 2006 (considering 1-year lag, we are left with 4 years panel), respectively. A similar pattern is observed in the changes of the incremental capital-to-output ratios. Again, we show that the changes in firm-level investment thirst measure (ICFS) are consistent with the county-level measure results.

7.2 GMM results

To address the concern that the measurement errors in Tobin's Q might contaminate the OLS regression results in Section 5, we use high-order moment estimators of Erickson and Whited (2000). Based on the overidentification J-test results, we use the third up to the six order moments in our regression.24 Following EW, we first partial out the perfectly measured regressors (all other regressors except Tobin's q, it is assumed that Tobin's q is the only mismeasured variable and EW's approach is specifically used to address the measurement problem of Tobin's q.)

Our model is specified as follows,

In which, z_i denotes all other regressors except Tobin's q, and these regressors are assumed perfectly measured. X_i denotes the mismeasured regressor: Tobin's q. u_i is the regression disturbance; and ε_i is the measurement error. α, β, γ are coefficients. In EW's work, they partial out the perfectly measured regressors to obtain:

where

Therefore, the six order moment (GMM6) equations are shown as follows.

Table 7 presents the test results implementing the EW's GMM method. Given that our previous tests showed that the SA index is found to be better than the WW index in measuring financial constraint, we therefore use the SA index to proxy for financial constraint. The same concern of measurement errors can apply to the interaction term used in the models, as exemplified by the interaction terms containing Tobin's Q. To address this problem, we introduce a dummy variable to replace Tobin's Q and create a new interaction term, I-Dummy_t × CF_t/K_t−1, to detect investor infringement. The dummy variable I-Dummy is defined in the condition: if the sustainable growth rate is larger than the real growth rate, (S_t−S_t−1)/S_t−1, then I-Dummy equals to 1, otherwise 0. In general, if the sustainable growth rate is larger than the real growth rate, managers would invest more, causing an increase in ICFS. If the sustainable growth rate is below the real growth rate, investment detriments can be found and a negative sign of the coefficient of I-Dummy_t × CF_t/K_t−1 is expected.

Table 7 reports the GMM estimation results showing that the trend in the ICFS GMM estimators is similar to its OLS estimators in Table 5. We conclude that ICFS is a good measure of investment thirst and thus that our results in Section 5 are robust.

7.3 Relative idiosyncratic risk calculated through CAPM

Table 8 presents some robust tests for the results of the relationship between ICFS and risk. Here, we use the CAPM model instead of Fama and French (1992) method to compute idiosyncratic risk. We find that the coefficients of CF_t/K_t−1 in Groups 3 and 4 are commonly larger than those in Groups 1 and 2, which is similar to the results in Table 4. Therefore, our evidence in Table 8 also implies a positive relationship between ICFS and risk-seeking preferences.