A. Data

To test the above theories, I use a panel dataset of balance sheet and income statement variables of all 4,037 nonfinancial Japanese firms between 1995 and 2013 provided by Thomson Reuters Worldscope.⁷ Following Opler et al. (1999) and Bates et al. (2009), I consider the following explanatory variables for cash holdings, because their explanatory power can be rationalized in terms of the transaction cost and precautionary demand theories:

• Cash flow from operations: The propensity to save operating cash flow is a measure of financing constraints (Almeida et al, 2004). This variable therefore tests both the transaction cost and precautionary demand motives for holding cash, which are affected by the degree to which a firm is financing constrained.

• Cash flow volatility: A positive relationship between cash holdings and cash flow volatility would reflect that firms with riskier cash flow choose to hold more cash, which would provide evidence for the precautionary demand motive. On the other hand, a negative relationship would mean that firms experiencing riskier cash flow end up with less cash as a result. A negative relationship would therefore provide some evidence against the view of firms as active optimizing agents.

• Capital expenditure: Capital expenditure is an important control variable for measuring the effect of other variables on cash holdings, but its relationship with cash holdings could be positive or negative, and could reflect active or passive firm behavior. A negative relationship between cash holdings and capital expenditure could reflect the fact that investment improves collateral and borrowing capacity, reducing the cost of raising external finance when needed, which is consistent with both the transaction cost and precautionary demand motives. On the other hand, if capital expenditure measures the desired level of investment, then a positive relationship between cash holdings and capital expenditure would reflect financing constraints causing firms to build up cash reserves to finance their desired investment. Positive or negative associations between cash holdings and capital expenditure are also consistent with the view of cash as the unplanned outcome of shocks to the firm. In the short run, changes in capital expenditure must be financed to some extent by changes in the cash stock when averaging over many firms and points in time, giving rise to a negative relationship. In the long run however, a consistently high level of capital expenditure requires a high cash stock to sustain it due to timing differences between income and expenditure, which would reflect a positive relationship.

• Total assets: A negative relationship between cash holdings and total assets would be consistent with economies of scale in the costs of raising finance or selling assets when needed, which would support both the transaction cost and precautionary demand theories.

• International assets: Firms with international operations may be liable for repatriation tax on foreign profits. In addition, these firms could use a higher level of cash to insure themselves against exchange rate movements. By holding cash in multiple jurisdictions, a firm can avoid transferring cash between them when exchange rates are temporarily unfavorable. A positive relationship between the cash holdings and the proportion of international assets in total assets would reflect such a motivation, which is consistent with the precautionary demand theory. A Japanese manufacturer with productive facilities in low-cost developing countries where financing constraints are more severe would also lower its transaction costs by holding more cash to finance its overseas running expenses. A negative relationship between a firm’s cash holdings and international assets would reflect that a firm purchases international assets out of cash holdings.

• Market-to-book ratio: A positive relationship between cash holdings and the market-to-book ratio would indicate that firms with more investment opportunities find cash shortfalls more costly, supporting both the transaction cost and precautionary demand motives. A negative relationship would indicate that firms with better corporate governance, and therefore higher valuations, are more successful at curbing managers’ incentives to exceed the shareholder wealth-maximizing level of cash holdings, which indicates an agency cost motive.

• Leverage: A negative relationship between cash holdings and leverage indicates there is an opportunity cost of holding cash rather than paying it out to debtholders.⁸ However, the level of correlation between investment needs and operating cash flow determines the preference for more cash relative to less debt (Acharya et al., 2007). Firms with a high correlation between investment needs and operating cash flow require less cash and hence less debt, while firms with low correlation between investment needs and operating cash flow require a cash reserve to protect themselves against times where they need to invest while operating cash flows are not available. Hence if all firms are equally financing constrained and vary only by this level of correlation, then we expect a positive association between cash and leverage, because firms are either high correlation (low cash and debt) or low correlation (high cash and debt). This preference is a form of precautionary demand motive for holding cash.

• Dividends: Dividends are a sign of maturity, stability and access to capital markets. High-dividend firms are less financing constrained and therefore experience a lower cost of selling assets / raising finance when needed and they experience less cash flow uncertainty, so they hold less cash. However, a relationship between cash holdings and dividend payments is also consistent with the interpretation of cash as an unplanned outcome, rather than a conscious firm choice. In the short run, dividends are paid out of cash holdings, so increases/decreases in dividends should be reflected in decreases/increases in cash holdings. In the long run, a high level of dividends must be sustained by a high level of liquidity and therefore a high cash stock.

In addition I consider whether the demand outlook can explain the level of cash holdings. The demand outlook data comes from the Business Outlook Survey conducted by the Cabinet Office and the Ministry of Finance. The reported survey measures the proportion of firms reporting a favorable domestic demand two quarters ahead less the proportion of firms reporting an unfavorable domestic demand two quarters ahead.⁹ I use published aggregates indexed jointly by size, sector and calendar year, which I match to my firm—year Worldscope data using total assets, ICB sector and firm financial year.¹⁰ To the extent that firm financial years end early in the calendar year, the two-quarter-ahead demand outlook would actually measure a demand outlook over anything up to six quarters ahead. A firm’s demand outlook at horizons of between two and six quarters could be expected to be strongly correlated.

To test the agency theory of cash holdings, I obtain the following variables from Bloomberg, which I match to Worldscope data by ISIN and year:¹¹

• Audit committee meetings, measured as the number of meetings of the board of directors’ audit committee during the reporting period. If more audit committee meetings reflect more care exercised by management on behalf of shareholders, then the number of audit committee meetings would measure the degree of alignment of management and shareholder interests.

• Board average age, measured as the average age of members of the board of directors. To the extent that older boards of directors are more risk averse, they might be inclined to hold more cash.

• CEO duality, a dummy variable indicating whether the firm chief executive officer is also chairman of the board of directors. A dual CEO working closely with the board of directors could indicate closer alignment between management and shareholder interests, but a dual CEO exerting influence on the board of directors could limit the accountability of managers and allow them to pursue their own objectives at the expense of shareholders.

• Ethics policy, a dummy variable indicating whether the firm has established ethical guidelines and/or a compliance policy for its nonmanagement employees in the conduct of the firm’s business. An established ethics policy could be a measure of accountability between employees and management and/or a sign of management transparency. On the one hand, accountability and transparency indicate closer alignment of manager and shareholder interests, better corporate governance and lower cash. On the other hand, better governance is associated with lower shareholder risk, financing costs, and opportunity costs of holding cash.

• Disclosure score, a proprietary Bloomberg score between 0 and 100 measuring the proportion of data items that Bloomberg is able to obtain about the firm from the firm’s public disclosures. Items are weighted according to their importance in the firm’s industry. The disclosure score is a measure of transparency. Like the existence of a firm ethics policy, the disclosure score could be associated with lower cash if it indicates better alignment of manager and shareholder interests, or higher cash if it indicates lower opportunity costs of holding cash rather than paying that cash out to financiers.

• Independent directors, measured as the proportion (in percent) of members of the board of directors that are external to the firm. Independent directors could be expected to act more impartially than directors who have business dealings with the firm. This impartiality could be an advantage for shareholders if it better aligns manager and shareholder interests, but it could be a disadvantage to shareholders if it is accompanied by complacency that concentrates power with the remaining nonindependent directors.

• Nonexecutive directors, measured as the proportion (in percent) of members of the board of directors that are not part of the firm’s executive management team. Nonexecutive directors differ from independent directors in that the former may hold an ownership stake in the firm.

Better corporate governance measures like transparency and accountability could reduce the risks of insolvency, fraud, fines and legal action. Such risk reductions also reduce the cash flow uncertainty of the firm and therefore reduce the need for firms to self-insure through holding cash.

B. Method

Fixed effect and system GMM regressions for cash holdings and investment

Let the latent target cash-to-asset ratio for firm in period be given by

where x_it is a K × 1 vector of determinants for firm in period t, ϕ_t is a deterministic component describing the contribution of firm-invariant time-specific variables like aggregate interest rates or economic growth to the firm’s target cash-to-asset ratio, and ε_it is a noise component with zero mean. Equation (1) could arise as the first order condition in an optimization problem for the level of cash holdings of firm i in period t, or it could arise in some heuristic manner if firm i in period t does not necessarily optimize. Equation (1) is a static relationship for firm in period in the sense that it does not depend on lagged values of the cash-to-asset ratio. Suppose that firm i in period t rebalances its actual cash-to-asset ratio y_it from one period to the next according to

where 0 ≤ λ ≤ 1 is the speed of adjustment. When λ = 1 the firm immediately adjusts its cash-to-asset ratio to the target from one period to the next, and when λ = 0 the firm does not adjust its cash-to-asset ratio. For an optimizing firm when λ < 1, the quantity 1 − λ > 0 could therefore be interpreted as the cost of adjusting cash-to-asset ratios from one period to the next, which rationally induces the firm to slow its adjustment of cash-to-asset ratios relative to target. Such costs could include the fixed costs associated with selling illiquid assets or raising external finance to raise cash. For optimizing firms, the cost of adjustment should be increasing in the sampling frequency, so that λ should be higher if periods are measured in years than if periods are measured in quarters, for example. For a nonoptimizing firm, the speed of adjustment parameter λ can be interpreted as an outcome of the firm’s heuristic behavioral rule.

Combining (1) and (2) and re-parametrizing α = 1 − λ, λδ = β, θ_t = λɸ_t and λε_it = c_i + u_it gives the data generating process for the cash-to-asset ratio for firm in period t

where c_i is a random component that controls for unobserved time-invariant firm-specific variables like the type of firm, the firm’s objectives and mission, or the time-invariant component of the firm-specific level of management risk aversion or corporate governance structure.¹², ¹³ When λ = 1, we have α = 0 and model (3) is equivalent to model (1) with the target cash-to-asset ratio equal to the actual cash-to-asset ratio.

The parameters in model (3) cannot be consistently estimated by fixed effects regression because the lagged dependent variable y_it-1 is correlated with the firm-specific effect c_i. When α = 0, these parameters can be consistently estimated by fixed effects, but this consistency relies on the exogeneity assumption $E\left(u_{is}\right|x_{i1}\dots x_{iT})=0$ for all i and s, which requires that the correlation between the idiosyncratic error u_it and the explanatory variables X_it be uncorrelated at all leads and lags. In particular, we may be concerned that some of the explanatory variables like capital expenditure might be correlated with past idiosyncratic shocks to the cash-to-asset ratio. The system generalized method of moments (GMM) estimator of Arellano and Bond (1991), Arellano and Bover (1995) and Blundell and Bond (1998) weakens this exogeneity condition to

E(x_isΔu_it) = 0 for all i and for all i and for all s < t

while strengthening it by adding the exogeneity condition

E(Δx_it(c_i + u_iT)) = 0 for all i and for all t

so that we only require that lags of the explanatory variables be uncorrelated with the idiosyncratic error. In other words, we place no restriction on feedback from past shocks to the cash-to-asset ratio on current and future explanatory variables like capital expenditure. The second exogeneity condition requires that the steady state level of the cash-to-asset ratio is not correlated with lagged changes in the explanatory variables, which is plausible in our context.

Tobit and linear probability model regressions for dividends

I test both whether cash holdings affect the magnitude of a firm’s dividend declarations and the likelihood of a firm declaring dividends by estimating two types of regression. To test for the effect of cash on the magnitude of dividends, I estimate the doubly-censored tobit model

where ${\tilde{y}}_{it}$ is the dividend payout ratio for firm i in year t when it lies between 0 and 100 percent, x^′_it is a 1 × K vector of the above explanatory variables including industry-specific dummy variables, θ_t is a time-specific component capturing the effects of all measurable macroeconomic fluctuations, β is a K × 1 vector of coefficients and u_it is an idiosyncratic shock.¹⁴ I restrict attention to dividend payout ratios between 0 and 100 percent because these are typical instances where a firm makes a profit and then decides to pay out a proportion of that profit as a dividend. Atypical instances where a loss-making firm declares a dividend or a profit-making firm declares a dividend in excess of its annual profit are more difficult to understand and are beyond the scope of this analysis.

To test for the effect of cash holdings on the likelihood of a firm declaring dividends, I estimate the linear probability model for whether or not the firm declares dividends in a given year following Denis and Osobov (2008):

where II is the indicator function, y_it is the (uncensored) dividend payout ratio for firm i in year t, θ_t is a time-specific component capturing the effects of all macroeconomic fluctuations, c_i is a firm-specific fixed effect and x^′_it is a vector of explanatory variables without industry-specific dummy variables.^{15, 16}

Measuring corporate risk aversion

In choosing between a random normally distributed risky return r_it and a risk free return ${\overline{r}}_t$,, an investor with preferences described by a constant absolute risk aversion utility function with risk aversion parameter a allocates a proportion

of her wealth in the risky asset and the remainder 1 − w_it in the risk free asset, where E denotes the expectation operator and V denotes the variance operator.

By observing the allocation 1 − w_it of the assets of firm i in year t to cash, the return on which is approximately risk free and observable, and by estimating the mean and variance of the return to firm i in year t of allocating its assets to other, productive assets, one can infer the risk aversion parameter a from equation (4) above. By re-arranging equation (4) to

one obtains the useful interpretation of parameter a as the ratio of the amount that a risk-neutral agent would allocate to the risky asset to the amount that the firm actually allocates to the risky asset. If the firm allocates less to the risky asset than the risk neutral agent would, then the firm is risk averse. Conversely, the parameter a can be interpreted simply as the amount that the firm allocates to cash in excess of what the above risk neutral agent would allocate to cash.

I estimate E(r_it) and V(r_it) using the mean and variance of returns on assets for all firm—years in the same first-digit ICB industry as firm i between years t − 2 and t + 2 inclusive. The use of future-dated returns allows for forward-looking expectations and pooling returns across firms within an industry limits the means and variances of returns to reasonable values. I estimate the return on the risk free asset ${\overline{r}}_t$ by principal interest rates on ordinary deposits and savings at banks in year t published by the Bank of Japan, which are common to all firms in any given year.

Nonfinancial firms are not utility-maximizing investors and to the extent that they choose their asset allocation, this decision reflects more considerations than simply the risk-adjusted expected return on productive assets in excess of the return on cash. Nevertheless observers often comment that Japanese managers are more risk averse than their US counterparts, so even a highly stylized measure of risk aversion could help us test whether preferences can explain some of the variation in cash holdings in Japan.

A. Determinants of Cash Holdings

I present fixed effect and system generalized method of moments regression estimation results in Table 1. The baseline specifications following Opler et al. (1999) and Bates et al. (2009) are in columns (1) and (2). My contribution to the baseline specification is (i) to add a lagged dependent variable to capture the persistence of cash-to-asset ratios over time, and (ii) to add a control variable for survey expectations of the demand outlook two quarters ahead, which I match to firm—years using industry and size characteristics.

Table 1.

Fixed effect and system GMM regressions on a firm—year panel of Japanese nonfinancial firms where the dependent variable is the cash-to-asset ratio. Estimations without lagged dependent variables, instruments and Hansen test p-values are fixed effect estimations. Fixed effects estimations are reported with heteroskedasticity-robust standard errors and system GMM estimations are reported with Windmeijer-corrected standard errors, in parentheses. Coefficients or standard errors of zero are only zero to two decimal places; such coefficients are not economically significant. All estimations include time-specific fixed effects to control for all macroeconomic fluctuations. In each estimation, the reported R² is calculated as the squared correlation between actual and fitted values of the dependent variable.

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 1.

Fixed effect and system GMM regressions on a firm—year panel of Japanese nonfinancial firms where the dependent variable is the cash-to-asset ratio. Estimations without lagged dependent variables, instruments and Hansen test p-values are fixed effect estimations. Fixed effects estimations are reported with heteroskedasticity-robust standard errors and system GMM estimations are reported with Windmeijer-corrected standard errors, in parentheses. Coefficients or standard errors of zero are only zero to two decimal places; such coefficients are not economically significant. All estimations include time-specific fixed effects to control for all macroeconomic fluctuations. In each estimation, the reported R² is calculated as the squared correlation between actual and fitted values of the dependent variable.

	baseline		controlling for risk aversion
	(1) cash-to-assets (percent)	(2) cash-to-assets (percent)	(3) cash-to-assets (percent)	(4) cash-to-assets (percent)
cash-to-assets (percent) = L,		0.74***		0.71***
		(0.03)		(0.03)
cash flow to assets	5.16**	10.46***	412**	11.04***
	(2.01)	(3.72)	(2.06)	(3.93)
firm cash flow 5yr vol. to assets	0.02***	0.03***	0.01**	0.03***
	(0.01)	(0.01)	(0.01)	(0.01)
capital exp. (percent assets)	−0.23***	−0.20***	0 22***	−0.27***
	(0.04)	(0.06)	(0.04)	(0.05)
operating exp. (percent assets)	−0.03***	−0.01	−0.03***	−0.00
	(0.01)	(0.00)	(0.01)	(0.00)
R&D exp. (percent assets)	−0.09	0.11	−0.01	0.25**
	(0.19)	(0.11)	(0.12)	(0.10)
borrowing rate (percent)	0.00	−0.00	0.00	0.00
	(0.00)	(0.00)	(0.00)	(0.00)
total assets (log)	−1 95***	−0.09	−1.57**	−0.10
	(0.66)	(0.11)	(0.65)	(0.12)
Leverage	0.00	0.00	0.00	0.00
	(0.00)	(0.00)	(0.00)	(0.00)
dividends (percent total assets)	0.47	0.22	0.26	0.44
	(0.29)	(0.29)	(0.27)	(0.33)
market-to-book ratio	−0.05**	−0.03	−0.03	−0.02
	(0.02)	(0.03)	(0.02)	(0.03)
international assets (percent)		1.08	−3.63***	1.84**
	(0.80)	(0.73)	(0.86)	(0.75)
size-sector demand outlook	−0.01	−0.07***	−0.05***	−0.07***
	(0.01)	(0.01)	(0.01)	(0.01)
risk aversion			004*** (0.01)	0.00 (0.02)
Observations	14,346	14,346	11,574	11,574
Number of firms	1,755	1,755	1,403	1,403
R2	0.781	0.786	0.794	0.769
Number of instruments		1328		1145
Hansen test p-value		0.385		0.111

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

View Table

Table 1.

Fixed effect and system GMM regressions on a firm—year panel of Japanese nonfinancial firms where the dependent variable is the cash-to-asset ratio. Estimations without lagged dependent variables, instruments and Hansen test p-values are fixed effect estimations. Fixed effects estimations are reported with heteroskedasticity-robust standard errors and system GMM estimations are reported with Windmeijer-corrected standard errors, in parentheses. Coefficients or standard errors of zero are only zero to two decimal places; such coefficients are not economically significant. All estimations include time-specific fixed effects to control for all macroeconomic fluctuations. In each estimation, the reported R² is calculated as the squared correlation between actual and fitted values of the dependent variable.

	baseline		controlling for risk aversion
	(1) cash-to-assets (percent)	(2) cash-to-assets (percent)	(3) cash-to-assets (percent)	(4) cash-to-assets (percent)
cash-to-assets (percent) = L,		0.74***		0.71***
		(0.03)		(0.03)
cash flow to assets	5.16**	10.46***	412**	11.04***
	(2.01)	(3.72)	(2.06)	(3.93)
firm cash flow 5yr vol. to assets	0.02***	0.03***	0.01**	0.03***
	(0.01)	(0.01)	(0.01)	(0.01)
capital exp. (percent assets)	−0.23***	−0.20***	0 22***	−0.27***
	(0.04)	(0.06)	(0.04)	(0.05)
operating exp. (percent assets)	−0.03***	−0.01	−0.03***	−0.00
	(0.01)	(0.00)	(0.01)	(0.00)
R&D exp. (percent assets)	−0.09	0.11	−0.01	0.25**
	(0.19)	(0.11)	(0.12)	(0.10)
borrowing rate (percent)	0.00	−0.00	0.00	0.00
	(0.00)	(0.00)	(0.00)	(0.00)
total assets (log)	−1 95***	−0.09	−1.57**	−0.10
	(0.66)	(0.11)	(0.65)	(0.12)
Leverage	0.00	0.00	0.00	0.00
	(0.00)	(0.00)	(0.00)	(0.00)
dividends (percent total assets)	0.47	0.22	0.26	0.44
	(0.29)	(0.29)	(0.27)	(0.33)
market-to-book ratio	−0.05**	−0.03	−0.03	−0.02
	(0.02)	(0.03)	(0.02)	(0.03)
international assets (percent)		1.08	−3.63***	1.84**
	(0.80)	(0.73)	(0.86)	(0.75)
size-sector demand outlook	−0.01	−0.07***	−0.05***	−0.07***
	(0.01)	(0.01)	(0.01)	(0.01)
risk aversion			004*** (0.01)	0.00 (0.02)
Observations	14,346	14,346	11,574	11,574
Number of firms	1,755	1,755	1,403	1,403
R2	0.781	0.786	0.794	0.769
Number of instruments		1328		1145
Hansen test p-value		0.385		0.111

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

The signs and magnitudes of the statistically significant coefficients in columns (1) and (2) agree with those found in the empirical literature and can be rationalized in terms of the above explanations. In this sense, they generally support the transaction cost and precautionary demand theories of cash holdings. The statistically significant coefficients also agree between fixed effect and system GMM estimations. The preferred baseline specification is the system GMM specification with a lagged dependent variable in column (2) because the exogeneity assumptions required for consistent estimation of the coefficients in the latter model are more plausible.

The coefficient on the lagged dependent variable is positive, less than unity and statistically significant, indicating stationarity and persistence in cash-to-asset ratios over time. The persistence could reflect the costs to the firm of adjusting its cash holdings from one year to the next. Alternatively the firm’s cash holdings could be viewed as an unplanned outcome of cash inflows and outflows, in which cash the persistence in cash-to-asset ratios reflects the degree to which Japanese nonfinancial firms do not actively manage their cash holdings.

The coefficient on operating cash flow is positive, reflecting the propensity of firms to save positive cash flow due to financing constraints. The coefficient on rolling standard deviation of cash flow is positive, so that firms with riskier cash flows tend to hold more cash to self-insure against cash shortfalls. I obtain a negative coefficient on capital expenditures, which could reflect the extent to which firms choose to finance their investment spending out of cash holdings, due to financing constraints or asymmetric information about the nature of investment projects between management and financiers. Bates et al. (2009) explain this negative coefficient as the reduced financing constraints associated with the increased collateral that follows increased investment spending. Another interpretation is that cash is not the outcome of a conscious decision by the firm. In this case, managers invest when they have cash on hand, and the effect of investing is to drawn down on cash holdings. The coefficient on the firm’s expectations for future demand is significantly negative, consistent with the interpretation that firms deploy cash in response to improving demand conditions.

I add a control variable for risk aversion in specifications (3) and (4) of Table 1, estimated using fixed effects and system GMM respectively. I describe the construction of this variable above. The risk aversion variable is self-referential in that its calculation depends on the very cash-to-asset ratio that it is purporting to explain. The rolling mean and standard deviations of returns that are used in the calculation exhibit limited time variation and are hence difficult to distinguish from the firm-specific fixed effects. Nevertheless there is some evidence for a positive coefficient on this variable, but only in the fixed effect specification. A positive coefficient could reflect that even when holding the riskiness of cash flows constant, firms with more risk averse managers tend to hold higher levels of cash.

Table 2 presents the estimation results for the baseline system GMM specification with explanatory variables for corporate governance obtained from Bloomberg. I consider seven measures of corporate governance. Data limitations seriously restrict the available sample for estimation, so the standard errors on coefficients are relatively large. In specification (1) with a full set of governance variables, only the number of audit committee meetings is even marginally statistically significant. This coefficient has a positive sign, indicating that for the average firm in the sample, one additional audit committee meeting per year tends to results in about a 0.3 percent increase in the cash-to-asset ratio, which may be a sign that Japanese nonfinancial firms engage in too many audit committee meetings on average. However, the Hansen test p-value in specification (1) is less than 10 percent, indicating some mis-specification.

Table 2.

System GMM regressions for cash-to-asset ratios on Japanese nonfinancial firm—year panel data. All estimations include the control variables from the baseline specification in Table 1 model (2) above, including a lagged dependent variable and firm- and time-specific fixed effects, coefficients on which are not reported to save space. Standard errors are reported in parentheses and allow for the Windmeijer correction. In each estimation, the reported R² is calculated as the squared correlation between actual and fitted values of the dependent variable.

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 2.

System GMM regressions for cash-to-asset ratios on Japanese nonfinancial firm—year panel data. All estimations include the control variables from the baseline specification in Table 1 model (2) above, including a lagged dependent variable and firm- and time-specific fixed effects, coefficients on which are not reported to save space. Standard errors are reported in parentheses and allow for the Windmeijer correction. In each estimation, the reported R² is calculated as the squared correlation between actual and fitted values of the dependent variable.

	(1) cash-to-assets (percent)	(2) cash-to-assets (percent)	(3) cash-to-assets (percent)	(4) cash-to-assets (percent)	(5) cash-to-assets (percent)	(6) cash-to-assets (percent)	(7) cash-to-assets (percent)	(8) cash-to-assets (percent)
audit
committee
meetings	0.30*	0.05
	(0.17)	(0.06)
board average
age	0.04		−0.06
	(0.14)		(0.08)
CEO duality	0.68			1.04**
	(0.63)			(0.52)
ethics policy	2.25				−2.24
	(2.80)				(2.43)
disclosure
score	0.07					0.01
	(0.15)					(0.02)
independent
directors
(percent)	−0.01						0.05**
	(0.06)						(0.02)
nonexecutive
directors
(percent)	−0.02							0.02
	(0.05)							(0.02)
Observations	2,528	2,528	2,528	2,528	2,528	2,528	2,528	2,528
Number of
firms	825	825	825	825	825	825	825	825
R2	0.711	0.818	0.821	0.816	0.819	0.820	0.813	0.819
Number of
instruments	234	801	499	491	497	499	488	494
Hansen test p-value	0.0812	0.996	0.284	0.236	0.349	0.325	0.392	0.236

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

View Table

Table 2.

System GMM regressions for cash-to-asset ratios on Japanese nonfinancial firm—year panel data. All estimations include the control variables from the baseline specification in Table 1 model (2) above, including a lagged dependent variable and firm- and time-specific fixed effects, coefficients on which are not reported to save space. Standard errors are reported in parentheses and allow for the Windmeijer correction. In each estimation, the reported R² is calculated as the squared correlation between actual and fitted values of the dependent variable.

	(1) cash-to-assets (percent)	(2) cash-to-assets (percent)	(3) cash-to-assets (percent)	(4) cash-to-assets (percent)	(5) cash-to-assets (percent)	(6) cash-to-assets (percent)	(7) cash-to-assets (percent)	(8) cash-to-assets (percent)
audit
committee
meetings	0.30*	0.05
	(0.17)	(0.06)
board average
age	0.04		−0.06
	(0.14)		(0.08)
CEO duality	0.68			1.04**
	(0.63)			(0.52)
ethics policy	2.25				−2.24
	(2.80)				(2.43)
disclosure
score	0.07					0.01
	(0.15)					(0.02)
independent
directors
(percent)	−0.01						0.05**
	(0.06)						(0.02)
nonexecutive
directors
(percent)	−0.02							0.02
	(0.05)							(0.02)
Observations	2,528	2,528	2,528	2,528	2,528	2,528	2,528	2,528
Number of
firms	825	825	825	825	825	825	825	825
R2	0.711	0.818	0.821	0.816	0.819	0.820	0.813	0.819
Number of
instruments	234	801	499	491	497	499	488	494
Hansen test p-value	0.0812	0.996	0.284	0.236	0.349	0.325	0.392	0.236

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Specifications (2) to (8) estimate the baseline system GMM specification with only one corporate governance variable included in each case. None of the Hansen tests reject the over-identifying exogeneity conditions.¹⁷ Evidence in specification (4) suggests that having the CEO act in a dual capacity also as the chairman of the board tends to increase the firm’s cash by 1.04 percent of its assets, which is consistent with the interpretation of dual CEOs exerting control over the board of directors at the expense of shareholders.

In specification (7), increasing the proportion of independent directors on the executive board by 1 percent tends to result in a 0.05 percent increase in the share of the firm’s assets allocated to cash. This finding suggests that the inexperience or complacency of independent directors in Japanese nonfinancial firms outweighs their impartiality and allows mangers to accumulate cash at the expense of shareholders. This positive relationship between cash holdings and independent director representation suggests we should be skeptical of the view that current reforms in Japan to improve the share of independent director representation on executive boards could be expected to reduce the shares of nonfinancial firms that are allocated to cash.

B. Normal and Excess Cash Holdings in Japan

From Table 1 above, the baseline system GMM regression is able to explain close to 80 percent of the variation in cash-to-asset ratios across 1,755 firms and 19 years. This proportion can be interpreted as the variation in Japanese nonfinancial firm cash holdings that is explainable by the transaction cost and precautionary demand theories.¹⁸ In this sense, the fitted values from this baseline estimation can be interpreted as ‘normal’ levels of cash holdings, while the residuals can similarly be interpreted as ‘excess’ cash holdings.¹⁹

Figure 5 presents the evolution of the median actual and baseline fitted cash-to-asset ratio over time for the full sample of 4,037 Japanese nonfinancial firms. The plot suggests that the transaction cost and precautionary demand theories can explain most of the increase in Japanese nonfinancial firm cash-to-asset ratios between 1999 and 2011. Between 2011 and 2013 however, Japanese nonfinancial firms continued to accumulate cash rapidly while their characteristics and past behavior suggest that cash-to-asset ratios should have declined over this period. The main factor contributing to the expected decline in cash-to-asset ratios between 2011 and 2013 is an improving demand outlook.

View Full Size

Median across firms in each year of the actual cash-to-asset ratio and fitted cash-to-asset ratio according to the baseline system GMM specification.

Citation: IMF Working Papers 2014, 221; 10.5089/9781498322171.001.A001

• Download Figure
• Download figure as PowerPoint slide

One can infer the implied target cash-to-asset ratios for firm i in year t using

in which case the change in implied target cash-to-asset ratios for firm between years s and t is

and the average change in target cash-to-asset ratios across all firms i = 1, …, N can be written in terms of the average change in explanatory variables as

where x^′_it is the × K vector of explanatory variables for firm i at time t, $\hat{\beta }$ is the K × 1 vector of coefficients estimated according to system GMM in Table 1 model (2) above, and ${\hat{\theta }}_t$ is the estimated time-specific fixed effect at time t.

Figure 6 shows a bar plot decomposing this average increase in target cash-to-asset ratios ${\hat{y}}_{it}^{*}$ into the average changes in explanatory variables. Of the 2.5 percent increase in average target cash-to-asset ratios between 1999 and 2011, roughly equal proportions of the increase can be explained by each of:

View Full Size

Average across firms of the change in implied target cash-to-asset ratios ${\hat{y}}_{it}^{*}$ from the baseline regression in specification (2) of Table 1 between 1999 and 2011. The average change in baseline target cash-to-asset ratio is decomposed into the average change in the explanatory variables multiplied by their respective coefficients. The contributions from time fixed effects are small and have been surpressed in this plot to save space.

Citation: IMF Working Papers 2014, 221; 10.5089/9781498322171.001.A001

• Download Figure
• Download figure as PowerPoint slide

• financially constrained optimizing firms choosing to save their cash inflows, or nonoptimizing firms unintentionally letting their cash reserves accumulate

• financially constrained optimizing firms choosing to hold higher levels of cash due to rising uncertainty in operating cash flow

• nonoptimizing firms cutting back on capital expenditure and hence allowing their cash reserves to accumulate, or optimizing firms cutting back on capital expenditure and hence choosing to hold higher levels of cash due to their lower levels of collateral and higher borrowing constraints

This overall 2.5 percent increase in average target cash-to-asset ratios occurs despite an improving demand outlook which contributes to lower target cash-to-asset ratios. The positive contributions of cash flow, cash flow uncertainty and capital expenditure more than offset the negative contribution from the improving demand outlook.²⁰

Figure 5 shows that the transaction cost and precautionary demand theories can explain much of the variation over time in the allocations of firm assets to cash. These theories can also explain the allocation of firm assets to cash across firms at a point in time. Figures 7−9 show that the excess cash assets held in Japanese nonfinancial firms in 2013 are concentrated in smaller firms, in the tradable sector and among exporters. Smaller firms tend to hold a higher share of their assets in cash because they are more financing constrained than larger firms, but given the government’s expansionary monetary policy it is hard to believe that financing constraints worsened between 2011 and 2013. The more compelling explanation is that smaller firms have unintentionally let their cash holdings accumulate over this period.

View Full Size

Median across firms of the actual and baseline fitted cash-to-asset ratios by firm size in 2013.

Citation: IMF Working Papers 2014, 221; 10.5089/9781498322171.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Median across firms of the actual and baseline fitted cash-to-asset ratios by firm sector in 2013.

Citation: IMF Working Papers 2014, 221; 10.5089/9781498322171.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Median across firms of the actual and baseline fitted cash-to-asset ratios by degree of a firm’s foreign activity in 2013.

Citation: IMF Working Papers 2014, 221; 10.5089/9781498322171.001.A001

• Download Figure
• Download figure as PowerPoint slide

Similarly, exporters of tradable goods experienced high profits between 2011 and 2013 due to a depreciating yen. Since I control for these profits through operating cash flows in the baseline specification, it must be the case that these exporters of tradable goods saved a higher share of their cash inflows than they did before 2011. This increase in the rate of saving could have been intentional or unintentional. Almeida et al. (2004) argue that a firm’s financing constraints determine the extent to which it saves cash out of cash flow. Again given the expansionary monetary policy over the period, it is hard to argue that financing constraints have risen between 2011 and 2013. The more compelling explanation therefore is that exporters of tradable goods have unintentionally let the increased profits from a depreciating yen add more to their share of assets allocated to cash than they would have allowed in previous years.

C. Consequences of High and Rising Cash Holdings in Japan

Figure 1 above suggests that Japanese nonfinancial firms’ accumulation of cash in the last decade has been at the expense of investment. In this section I show that this interpretation also holds at the firm level after controlling for other factors thought to determine firm-level investment.²¹ I estimate a standard model for firm-level investment following Syed and Lee (2010) and Kang (2014) that explains investment in terms of:

• Tobin’s Q, measured by the market value of equity divided by the book value of equity, as a forward-looking indicator of investment opportunities

• Liquidity, measured by the firm’s operating cash flow divided by total assets, which should be more important for investment if the firm is more financing constrained

• Leverage, measured by the ratio of the book value of assets to the book value of equity, to capture the effect of the firm’s debt burden on its investment behavior

• Uncertainty, measured by the average across firms within each sector (defined by first digit ICB code) of the 5-year rolling standard deviation of returns on assets, to capture the option value of delaying investment at times of high uncertainty.

I add the firm’s cash holdings as a share of total assets as an additional explanatory variable to test whether there is a trade-off between investment and cash holdings after controlling for other factors thought to affect the firm’s investment decision. I also add a lagged dependent variable to allow for persistency in capital expenditure that could arise, for example, through adjustment costs.

Table 3 reports the estimation results using fixed effects (1) and using system GMM with a lagged dependent variable (2). The coefficients of the estimation results under fixed effects and system GMM agree with each other in signs and in approximate magnitude, and they agree with predictions from theory. The Hansen test is rejected for the system GMM specification, suggesting that focusing on the fixed effects estimation results is preferable. The coefficient on cash holdings is statistically significant and negative, providing evidence for the trade-off between Japanese nonfinancial firms’ investment spending and their cash stocks. In particular, the fixed effects estimation suggests that a 1 percentage point increase in the ratio of a firm’s cash to total assets could reduce its capital expenditure by 0.06 percent of total assets. Policies to release some of these cash stocks could therefore result in higher investment.

Table 3.

Fixed effect (1) and system GMM (2) estimation results for the ratio of capital expenditures to total assets in a firm—year panel of Japanese nonfinancial firms. Standard errors in parentheses are heteroskedasticity-robust in (1) and Windmeijer-corrected in (2). Two-way firm and year fixed effects are included in all models. In each estimation, the reported R² is calculated as the squared correlation between actual and fitted values of the dependent variable.

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 3.

Fixed effect (1) and system GMM (2) estimation results for the ratio of capital expenditures to total assets in a firm—year panel of Japanese nonfinancial firms. Standard errors in parentheses are heteroskedasticity-robust in (1) and Windmeijer-corrected in (2). Two-way firm and year fixed effects are included in all models. In each estimation, the reported R² is calculated as the squared correlation between actual and fitted values of the dependent variable.

	(1) capital exp. / assets (percent)	(2) capital exp. / assets (percent)
capital exp. / assets = L,		0.0860**
		(0.0414)
market-to-book ratio	0.0167	0.143***
	(0.0388)	(0.0478)
cash flow from operations / assets	28.09***	30.27***
	(5.925)	(5.987)
leverage	−0.00115	−0.0146***
	(0.00427)	(0.00507)
sector ROA 5-yr vol.	−0.0207	−0.0468**
	(0.0196)	(0.0233)
cash-to-assets (percent)	−0.0623***	−0.0552***
	(0.00802)	(0.0185)
Observations	22,331	23,844
Number of firms	2,121	3,400
R2	0.655	0.278
Number of instruments		803
Hansen test p-value		< 0.01

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

View Table

Table 3.

Fixed effect (1) and system GMM (2) estimation results for the ratio of capital expenditures to total assets in a firm—year panel of Japanese nonfinancial firms. Standard errors in parentheses are heteroskedasticity-robust in (1) and Windmeijer-corrected in (2). Two-way firm and year fixed effects are included in all models. In each estimation, the reported R² is calculated as the squared correlation between actual and fitted values of the dependent variable.

	(1) capital exp. / assets (percent)	(2) capital exp. / assets (percent)
capital exp. / assets = L,		0.0860**
		(0.0414)
market-to-book ratio	0.0167	0.143***
	(0.0388)	(0.0478)
cash flow from operations / assets	28.09***	30.27***
	(5.925)	(5.987)
leverage	−0.00115	−0.0146***
	(0.00427)	(0.00507)
sector ROA 5-yr vol.	−0.0207	−0.0468**
	(0.0196)	(0.0233)
cash-to-assets (percent)	−0.0623***	−0.0552***
	(0.00802)	(0.0185)
Observations	22,331	23,844
Number of firms	2,121	3,400
R2	0.655	0.278
Number of instruments		803
Hansen test p-value		< 0.01

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

It is a stylized fact that Japanese firms pay lower dividends out of profits than firms in G7 peer countries. In addition, in Figures 3 and 4 below I show that aggregate dividends of Tokyo Stock Exchange 1^st Section firms are less sensitive to their aggregate profits than aggregate dividends of S&P 500 firms are to their aggregate profits. One might therefore expect dividend growth to respond to the recent two decades of profit growth more sluggishly in Japan than in the US, while nonfinancial firm cash holdings were accumulating.

View Full Size

Tokyo Stock Exchange 1^st Section, 2009Q3—2014Q2. Source: Haver

Citation: IMF Working Papers 2014, 221; 10.5089/9781498322171.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

S&P500, 2009Q3—2014Q2. Source: Haver

Citation: IMF Working Papers 2014, 221; 10.5089/9781498322171.001.A001

• Download Figure
• Download figure as PowerPoint slide

I investigate the firm-level evidence for a trade-off between paying dividends and accumulating cash. Following Denis and Osobov (2008) I construct the following explanatory variables that theory predicts should explain dividend payout ratios:

• Firm size, measured as the natural logarithm of total assets in yen. Larger firms are less financing constrained and hence have less reason to retain profits for financing projects later.

• Tobin’s Q, measured as the ratio of the market value of equity to the book value of equity is again a forward-looking indicator of investment opportunities. The greater the firm’s investment opportunities, the less the firm could be expected to pay in dividends.

• The proportion of retained earnings in book equity. To the extent that dividend payout ratios reflect the firm’s desire to smooth its free cash flow over time, the age of the firm is the most important determinant of its free cash flow (Allen and Michaely, 2003; DeAngelo, DeAngelo and Skinner, 2004; DeAngelo and DeAngelo, 2006). In the early part of the firm’s life cycle, it has ample investment opportunities and limited free cash, so it pays out less of its profits as dividends. In the later part of the firm’s life cycle, the firm has fewer investment opportunities and more free cash flow. The proportion of retained earnings in book equity is a measure of the maturity of the firm.

• Return on assets, measured as the ratio of net profit after tax to total assets. To the extent that dividend payout ratios are a device for conveying information about firm profitability to investors, higher returns on assets should be associated with higher dividend declarations.

In addition I introduce three further explanatory variables. The first directly measures cash flow, rather than approximating it through the ratio of retained earnings to common equity.

• The ratio of cash flow to total assets. If the firm seeks to use its dividend policy to smooth cash flows over time, its dividend payout ratios should be positively associated with operating cash flows.

• The ratio of the 5-year rolling standard deviation of firm cash flow to total assets. We would expect more uncertainty around the firm’s future liquidity position to result in lower dividends

• The ratio of capital expenditure to total assets. We would expect that more capital expenditure leaves less cash available to pay dividends and hence lower dividend declarations.

I present the results of the tobit estimation for dividend payout ratios in column (1) of Table 4, and I present the estimation results of the fixed effect and system GMM linear probability models in columns (2) to (3) of that table. From specification (1), the size of the firm’s cash stock is a statistically significant negative predictor of the magnitude of its dividends in normal times, even after controlling for other factors thought to determine the dividend declaration decision. However, from specifications (2) and (3) there is no evidence that the size of the firm’s cash holdings affects its decision of whether or not to declare dividends. The trade-off between cash holdings and dividend declarations therefore exists on the intensive but not the extensive margin.

Table 4.

Estimation results for (1) a tobit regression of the dividend payout ratio censored between 0 and 100 percent, (2) a fixed effect linear probability model for whether or not dividends are declared in a firm—year and (3) a system GMM linear probability model for whether or not dividends are declared in a firm—year. Two-way firm and year fixed effects are included in specifications (2) and (3), while year- and industry-specific fixed effects are included in (1) due to problems of consistent estimation in fixed effect tobit models. Standard errors in parentheses are clustered at the firm level in specifications (1) and (2), and they include a Windmeijer correction in (3). Coefficients or standard errors of zero are only zero to two decimal places; such coefficients are not economically significant. In each estimation, the reported R² is calculated as the squared correlation between actual and fitted values of the dependent variable.

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 4.

Estimation results for (1) a tobit regression of the dividend payout ratio censored between 0 and 100 percent, (2) a fixed effect linear probability model for whether or not dividends are declared in a firm—year and (3) a system GMM linear probability model for whether or not dividends are declared in a firm—year. Two-way firm and year fixed effects are included in specifications (2) and (3), while year- and industry-specific fixed effects are included in (1) due to problems of consistent estimation in fixed effect tobit models. Standard errors in parentheses are clustered at the firm level in specifications (1) and (2), and they include a Windmeijer correction in (3). Coefficients or standard errors of zero are only zero to two decimal places; such coefficients are not economically significant. In each estimation, the reported R² is calculated as the squared correlation between actual and fitted values of the dependent variable.

	(1) dividend payout ratio (percent)	(2) dividends indicator	(3) dividends indicator
total assets (log)	1.45***	0.05***	0.05***
	(0.29)	(0.01)	(0.01)
market-to-book ratio	−0.21	−0.00	−0.01***
	(0.18)	(0.00)	(0.00)
return on assets (percent)	1.31***	0.02***	0.01***
	(0.19)	(0.01)	(0.00)
retained earnings / book equity	0.66*	−0.00	−0.00***
	(0.38)	(0.00)	(0.00)
cash flow from operations /
assets	11.00	0.28	0.43
	(9.10)	(0.22)	(0.27)
firm cash flow 5-yr vol.
(percent assets)	−0.34*	0.00	0.00
	(0.18)	(0.00)	(0.00)
cash-to-assets (percent)	−0.15***	−0.00	0.00**
	(0.04)	(0.00)	(0.00)
Observations	25,671	25,671	25,671
Number left-censored	6037
Number right-censored	1573
Number uncensored	18061
Number of firms	3,720	3,720	3,720
R2		0.527	0.166
Number of instruments			1082
Hansen test p-value			0.00

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

View Table

Table 4.

Estimation results for (1) a tobit regression of the dividend payout ratio censored between 0 and 100 percent, (2) a fixed effect linear probability model for whether or not dividends are declared in a firm—year and (3) a system GMM linear probability model for whether or not dividends are declared in a firm—year. Two-way firm and year fixed effects are included in specifications (2) and (3), while year- and industry-specific fixed effects are included in (1) due to problems of consistent estimation in fixed effect tobit models. Standard errors in parentheses are clustered at the firm level in specifications (1) and (2), and they include a Windmeijer correction in (3). Coefficients or standard errors of zero are only zero to two decimal places; such coefficients are not economically significant. In each estimation, the reported R² is calculated as the squared correlation between actual and fitted values of the dependent variable.

	(1) dividend payout ratio (percent)	(2) dividends indicator	(3) dividends indicator
total assets (log)	1.45***	0.05***	0.05***
	(0.29)	(0.01)	(0.01)
market-to-book ratio	−0.21	−0.00	−0.01***
	(0.18)	(0.00)	(0.00)
return on assets (percent)	1.31***	0.02***	0.01***
	(0.19)	(0.01)	(0.00)
retained earnings / book equity	0.66*	−0.00	−0.00***
	(0.38)	(0.00)	(0.00)
cash flow from operations /
assets	11.00	0.28	0.43
	(9.10)	(0.22)	(0.27)
firm cash flow 5-yr vol.
(percent assets)	−0.34*	0.00	0.00
	(0.18)	(0.00)	(0.00)
cash-to-assets (percent)	−0.15***	−0.00	0.00**
	(0.04)	(0.00)	(0.00)
Observations	25,671	25,671	25,671
Number left-censored	6037
Number right-censored	1573
Number uncensored	18061
Number of firms	3,720	3,720	3,720
R2		0.527	0.166
Number of instruments			1082
Hansen test p-value			0.00

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1