RDP 2016-03: Why Do Companies Hold Cash? Appendix C: Estimates of Sample Selection Bias

The D&B company database is not a census of companies. Rather, the sample of companies may be biased given that companies are typically selected into the database because they have applied to D&B for a credit report. It follows that sample selection bias could be a problem if the cash management behaviour of companies that apply for credit is different to those companies that do not.

To gauge the importance of this bias, we match the sample of publicly listed companies in the D&B database to the sample of publicly listed companies in the Morningstar database. We then look for differences in the sample composition and, more importantly, differences in cash management behaviour that could affect our main results. This test relies on two key assumptions: 1) the Morningstar companies are assumed to be representative of the population of listed companies; and 2) the selection rules for companies appearing in the D&B database are the same for listed and unlisted companies.

Summary statistics of the two separate samples are shown in Table C1.

Table C1: Summary Statistics for Publicly Listed Companies
2005 to 2014
  D&B Morningstar
Mean Median Mean Median
Size 3.6 3.3 3.4 3.1
Cash ratio (%) 22.9 11.9 27.5 16.2
Age (years) 19 13 17 12
Growth (%) 10.5 6.1 12.3 7.1
Number of observations 10,619 13,957
Number of companies 1,830 1,900

Note: Size measured as the natural logarithm of total assets and growth measured as the annual change in the logarithm of total sales

Sources: Authors' calculations; D&B; Morningstar

Morningstar listed companies are very similar to their D&B counterparts in terms of size, age and sales growth, on average. The only notable difference is that Morningstar companies appear to have slightly higher cash-to-asset ratios on average. To more formally test for selection bias, and differences between the two samples, we estimate the following regression model:

where the dependent variable is the cash-to-assets ratio of company i in industry j in year t. Our key explanatory variable is a dummy variable (MORNING) that is equal to one if the listed company is in the Morningstar database and is zero otherwise. The estimated coefficient on this dummy variable captures the mean difference in cash holdings between the Morningstar and D&B companies. It therefore directly captures any selection bias inherent in the D&B database, to the extent that the Morningstar database is representative of the population of Australian companies. We also include a set of control variables (X), including size, age and sales growth, as well as industry dummies (μ) and year dummies (λ). We also interact the dummy MORNING with each of the control variables to gauge whether the relationship between cash holdings and its determinants varies across the two samples.

As Table C2 highlights, the cash-to-assets ratio of Morningstar companies is larger than that of comparable D&B companies by about 5 percentage points on average (column (1)). However, this result disappears when we control for observed company characteristics (column (2)). The sensitivity of cash holdings to sales growth varies across the two samples, but otherwise there are no significant differences between the two samples. This gives us some comfort that our results are not particularly sensitive to sample selection bias.

Table C2: Sample Selection Bias Estimates
  (1) (2)
MORNING 0.05*** −0.017
SIZE   −0.04***
AGE   −0.00
GROWTH   0.01***
MORNING *GROWTH   −0.02***
Constant 0.23*** 0.38***
Number of observations 24,576 14,264
Within R2 0.01 0.21
Industry fixed effects Yes Yes

Notes: Robust standard errors clustered at the company level used to accommodate within-company serial correlation; ***, **, and * denote significance at the 1, 5 and 10 per cent level, respectively

Sources: Authors' calculations; D&B; Morningstar