RDP 2016-03: Why Do Companies Hold Cash? 3. Identification

Gianni La Cava and Callan Windsor

May 2016 – ISSN 1448-5109 (Online)

We take two separate approaches to identify the determinants of company cash in Australia. First, we compare the cash management behaviour of public and private companies. This allows us to directly test the relative importance of the financing frictions and agency costs hypotheses. Second, we focus on the long-run cash management behaviour of publicly listed companies. This provides further insight into the drivers of the secular rise in aggregate cash and allows us to examine whether the long-run rise is in line with fundamentals or not.

3.1 Comparing Public and Private Company Cash

To the extent that private companies disclose less financial information than public companies, the lower transparency of private companies is likely to be associated with greater asymmetric information, and hence more costly external finance for private companies (Table 1). This, in turn, implies that private companies may face higher financing frictions than public companies, and hence have a higher precautionary (and speculative) demand for cash.

This give rise to our first hypothesis:

  • H1: Private companies should have higher levels of cash, on average, than public companies if there are financing frictions, all else equal.

Agency costs occur when corporate managers and shareholders have conflicting interests. Private companies should have fewer agency conflicts than public companies; they have more concentrated ownership structures and are typically more reliant on debt creation to fund investments. A greater reliance on debt effectively forces private companies to pay out future cash flows in interest and principal repayments and gives managers less discretion in the use of funds (Jensen 1986). Higher agency costs among public companies – resulting from a separation of ownership and control – could encourage the financial managers of public companies to hold more cash and to accumulate more cash in good times. Such behaviour would enable these companies to exercise more discretion in financing future investments.

This give rise to our second hypothesis:

  • H2: Private companies should have lower levels of cash, on average, than public companies if agency costs affect corporate cash policies, all else equal.

In the presence of agency costs and financing frictions it is unclear whether public or private companies should hold more cash. But, the finding that the cash holdings of public companies are higher and/or more sensitive to cash flows compared to private companies provides evidence in support of agency costs.

Our modelling approach and hypotheses are similar to that of Gao et al (2013) for the United States.[10] Although Gao et al focus on the level of cash holdings across public and private companies in the United States, in our extensions we also examine the sensitivity of cash holdings to cash flows across public and private companies.

To begin, we estimate the following company-level panel regression model:[11]

where the dependent variable is the cash-to-assets ratio (CASHit) of company i in year t. We define ‘cash’ as the stock of currency, deposits and other liquid securities, such as government bonds. This definition captures all financial instruments that a company can use to buffer against adverse shocks or to respond quickly to new investment opportunities.

The key explanatory variable in Equation (1) is a dummy variable (PUBLICi), which is equal to one if company i is public and is equal to zero if the company is private. This dummy variable captures the average difference in cash holdings for public and private companies, conditional on other observable company characteristics.

Our choice of control variables (CONTROLSit) is guided by theory and previous studies (Gao et al 2013). We include the following, which are all normalised by total assets except company size and age:[12]

  • Company size (SIZE): measured as the natural logarithm of real assets. This captures the transactions demand for cash; there are economies of scale in cash holdings such that larger companies tend to use cash more efficiently.
  • Company age (AGE): measured as the difference between the current reporting year and the year of registration. This should partly reflect financing frictions as younger companies are typically less well known and therefore more likely to find it difficult to raise external funds from creditors.
  • Cash flow (CASHFLOW): measured as earnings before interest, tax, depreciation and amortisation. This potentially captures financing frictions as financially constrained companies will be more sensitive to cash flows. It may also capture agency costs if managers aim to increase their decision-making control by saving cash out of earnings.
  • Industry cash flow risk (RISK): measured as a rolling standard deviation of cash flows divided by assets and averaged across industries.[13] This captures the precautionary demand for cash as companies operating in riskier industries are more likely to hold cash.
  • Leverage (LEVERAGE): measured as total debt divided by assets. According to the financing hierarchy, cash (internal finance) is a cheaper alternative to debt in financing investment, implying a negative correlation between cash and leverage.
  • Capital spending (CAPEX): measured as spending on property, plant and equipment. This captures the precautionary demand for cash, as companies with large capital spending commitments face larger costs if financing conditions deteriorate, so we would expect a positive correlation with cash holdings.[14]
  • Net working capital (WORKINGCAPITAL): measured as the stock of inventories and short-term receivables outstanding. Inventories and trade credit should be negatively correlated with cash to the extent that they are substitutable forms of liquidity.

As Gao et al (2013) recognise, sample selection is likely to be an issue in modelling cash holdings because companies are not randomly assigned to being either public or private. Rather, the decision to be a public company might be correlated with unobserved company characteristics that determine the level of cash holdings.

To the extent that the choice to be public is determined by company characteristics that are fixed over time, we control for this endogenous sample selection through the inclusion of company fixed effects (αi) in Equation (1). The company fixed effect controls for unobserved time-invariant company characteristics that explain why some companies hold more cash than others on average (e.g. the company's business model or its level of risk aversion).

However, the inclusion of the company fixed effect creates an issue in estimation as it will be perfectly collinear with the dummy variable for whether a company is public or not. To circumvent this, we estimate a ‘correlated random effects’ (CRE) model (Mundlak 1978; Chamberlain 1982). Given the model is not commonly used, particularly in the corporate finance literature, it is useful to provide a detailed outline. Suppose the true model is:

where Xit is a set of time-varying explanatory variables, PUBLICi is the time-invariant explanatory variable (of most interest here) and αi is the unobserved fixed effect. Assume:

we can explicitly model the non-zero correlation between the unobserved fixed effect and the observed explanatory variables. Given the fixed effect only varies in the cross-section, if it is correlated with Xit in period t, then it will be correlated with Xis in period s (where s ≠ t). In this case, we need to model its correlation with the explanatory variables that do vary over time (e.g. SIZE). Assuming that the relationship is linear:

Furthermore, if we assume that the correlation between the fixed effect and the explanatory variables is the same in each period (i.e. that θ1 = θ2 = θ3 = θ) then the above reduces to:

where Inline Equation = ΣtXit / Ti is the temporal mean of the explanatory variable X and Ti is the total number of observations for company i (this can vary by company because of the unbalanced nature of the panel).

Now define a new (unobserved) fixed effect, Inline Equation ≡ αi − θInline Equation − χPUBLICi. Therefore:

By construction, this new fixed effect is not correlated with any explanatory variables. Equation (2) can be rewritten as:

where vit = Inline Equation + εit is the new error term. This modified model satisfies the assumption that the explanatory variables and the company fixed effect are uncorrelated (i.e. E(Inline Equation|Xit) = 0).[15] So we can run random effects on this model and obtain a consistent estimate of the effect of the binary indicator for whether the company is public or not.[16]

3.2 The Secular Rise in Publicly Listed Company Cash

We follow a more standard approach to examine the determinants of the long-run trend in corporate cash holdings (Opler et al 1999; Bates et al 2009). The model is estimated on a sample of publicly listed companies covering the period from 1990 to 2014. The regression model is specified as:

where the variables are defined similarly to Equation (1). In particular, the dependent variable is the cash-to-assets ratio (CASHit) for each listed company in each year. The set of control variables (CONTROLSit) includes a slightly broader range of variables than before as listed companies provide more detailed balance sheet information. In addition to the control variables listed earlier, we include two explanatory variables that are designed to capture growth opportunities (but are only available for listed companies):

  • Market-to-book ratio or Tobin's Q (TQ): companies with greater investment opportunities are thought to value cash more since it is costly for these companies to be financially constrained. Tobin's Q is measured as the company's market value (shares outstanding × share price) divided by the book value of net assets. A company's book value is assumed to capture the value of its existing assets, while the market value captures both the value of its existing assets and its growth opportunities.[17]
  • Research and development expenditure-to-assets ratio (R&D): we expect a positive correlation with cash holdings as research and development spending is typically correlated with growth opportunities and company-level risk.

Footnotes

Other studies have also examined the effect of agency costs on public companies' liquid asset holdings by exploiting variation in agency conflicts across different types of corporate governance, but with mixed results. Dittmar et al (2003) study cash holdings across different economies and find that in places where investor protection is lower companies hold more cash. On the other hand, Harford et al (2008) find that companies with more entrenched managers actually hold comparatively less cash than otherwise similar companies. [10]

All of the results are robust to the inclusion of time fixed effects that control for the overall business cycle. For simplicity, they are excluded from the notation here. [11]

As robustness, however, we employed several alternative definitions, including taking the natural logarithm of the variables and normalising them by total assets less cash. The results (available upon request) are not affected in a material way when using these alternatives. [12]

The variable RISK is constructed in two steps. First, for each company, a rolling standard deviation of cash flows (net cash flows from operating activities) to assets is calculated using the previous two years of data. Second, for each year, these company-specific measures of cash flow risk are averaged by industry. [13]

If capital spending creates assets that can be used as collateral, then capital spending could increase the borrowing capacity of the company and reduce the demand for cash. [14]

As noted, this model relies on the assumption that the correlation between the fixed effect and each of the explanatory variables does not change over time. If the correlation did vary over time, a key assumption of the CRE model would be violated. To address this possibility, we also estimate a Hausman-Taylor (HT) model in Appendix D. [15]

In this example, the binary variable PUBLICi has a direct effect (captured by β) and an indirect effect (captured by χ) on cash holdings. [16]

We recognise that the literature is divided as to whether Tobin's Q is a good measure of corporate performance. Some studies suggest that it is an appropriate indicator after correcting for measurement error (Erickson and Whited 2012), while others suggest it is not a good measure (Dybvig and Warachka 2015). We have experimented with an alternative measure of growth performance based on current sales growth and our key results are unaffected. [17]