The 2009 annual North American Compustat database provides the data for this study, which includes observations from 1989 through 2008.9 Panel A from Table 1 provides the details for the sample selection process, which begins with a total of 177,280 firm-years with nonmissing values for total assets (data6), sales revenue (data12), earnings before extraordinary items (data18) and operating cash flow (data308). A close inspection of the Compustat database indicates that a total of 4931 observations represent duplicates (i.e., same sales and total assets as the previous year), which are omitted from the sample. Surprisingly, the Compustat database contains observations with unaudited data, which necessitates the exclusion of 929 firm-years. Similar to Barth et al. (2001) and Cheng et al. (2008), the sample selection process requires each observation to maintain at least $1 million of total assets and $1 million of sales revenue to minimize the possibility that very small firms are driving the empirical results. An additional benefit for requiring the minimum amount of sales revenue is that the Compustat database contains observations with zero and negative values for data item 12 (i.e., the sales revenue variable). Excluding observations that fail to meet the $1 million minimum requirement for total assets and sales revenue results in the exclusion of 14,918 firm-years, and including these observations does not affect the inferences in the study. The sample also excludes all firm-years from the financial services industry (SIC 6000–6999) due to the nature of their operations and the related classification problems for the cash flow statement. Excluding the financial services industry firm-years reduces the sample by 20,980 observations. Changes in a firm's fiscal year-end result in truncated or extended reported periods, and possibly distort the comparability of observations within the sample. Excluding firm-years with changes in the fiscal year-end results in the omission of 249 firm-years. Extreme observations may distort the results from regression analyses, and the current study excludes the top and bottom 1% from the distributions for the dependent variable (i.e., earnings for time t + 1) and the two primary independent variables (i.e., operating cash flow and operating accruals variables for time t). This procedure excludes 5126 firm-years, or approximately 5% of the sample. One alternative method for managing extreme observations is to omit firm-years where the Studentized residuals exceed two in absolute value (see Belsley, Kuh, & Welsch, 1980). The application of this alternative method does not alter the inferences of the study. A total of 33,533 firm-years contain inadequate (i.e., missing) lagged values, and excluding these observations results in a final sample of 96,614 firm-years.
Panel B from Table 1 indicates that the sample contains a significant number of FPIs. Specifically, Panel B indicates that approximately 15% of the firm-years and 19% of the firms in the sample belong to FPIs. This suggests that the North American Compustat database contains a noteworthy proportion of foreign private issuer observations.
Table 2 reports the industry membership separately for the FPIs and U.S. firms using 2-digit Standard Industry Classifications (SIC). The last column in Table 2 contains the percentage of FPI firm-years relative to the U.S. firm-years in each industry. For example, the Metal Mining industry (SIC 10) contains nearly four times as many FPI firm-years (846) as U.S. firm-years (242), where the percentage of FPI firm-years relative to the U.S. firm-years is approximately 350%. This suggests that although mining opportunities outside the U.S. are significant, the supply of foreign investment capital for these opportunities is inadequate and/or prohibitively expensive. Hence, FPIs appear to look to U.S. investors for additional capital. Similar results appear in the Water Transportation industry (SIC 44), where the percentage of FPI firm-years relative to the U.S. firm-years is approximately 143%. Another important feature of FPIs is that many of these firms belong to industries which require significant capital expenditures, including investment in research and development (R&D) activities. For example, approximately 50% of the FPI firm-years belong to just six industries, which represent the Metal Mining industry (SIC 10, 5.8%), the Oil and Gas Extraction industry (SIC 13, 9.7%), the Chemical and Allied Product industry (SIC 28, 7.2%), the Other Electrical Equipment industry (SIC 36, 8.6%), the Communications industry (SIC 48, 8.1%) and the Business Services industry (SIC 73, 11.0%). Overall, the sample spans more than 60 industries, and represents a wide cross-section of firms.
Table 3 reports descriptive statistics for the sample, with separate results for the FPI and U.S. firms. A comparison of the results for the FPIs in Panel A with the results for the U.S. firms in Panel B indicates that the FPI firms are much larger than the U.S. firms. For example, the median values for total assets for the FPI and U.S. samples are approximately $521 million and $178 million, respectively. Similarly, the median values for sales revenue for the FPI and U.S. samples are $344 million and $179 million, respectively. However, the scaled values for earnings, operating cash flow and operating accruals are more similar in magnitude. For example, earnings scaled by average assets are 0.032 for the FPI sample, and 0.028 for the U.S. sample, where the larger value for scaled earnings for FPIs is likely the result of larger firms relative to the U.S. sample.10 The implicit median asset turnover ratio in Table 3 (i.e., median sales revenue ÷ median total assets) is slightly lower for the FPI sample than for the U.S. sample (i.e., 0.66 vs. 1.00). This result is likely attributable to the concentration of FPI firms in industries that require large capital investment (e.g., SIC 48), which reduces the asset turnover ratio. Overall, the descriptive statistics suggest that foreign firms which raise capital in U.S. markets are large firms.