¹ This rationale complements the argument that international capital flows occur as investors try to lower risk. See Grubel, H. G., “Internationally Diversified Portfolios: Welfare Gains and Capital Flows,” American Economic Review, vol. 58 (December 1968), pp. 1299–1314.Google Scholar

² Caves, R. E., “International Corporations: The Industrial Economics of Foreign Investment,” Economica NS, vol. 38 (February 1971), p. 5.CrossRefGoogle Scholar

³ Aliber, Robert Z., “A Theory of Direct Foreign Investment,” in The International Corporation, edited by Kindleberger, C. P. (Cambridge: MIT Press, 1970), pp. 17–34.Google Scholar

⁴ Hymer, Stephen and Rowthorn, Robert, “Multinational Corporations and International Oligopoly,” in The International Corporation, edited by Kindleberger, C. P. (Cambridge: MIT Press, 1970), pp. 57–94.Google Scholar

⁵ Gruber, William, Mehta, Dileep, and Vernon, Raymond, “The R&D Factor in International Trade and International Investment of United States Industries,” Journal of Political Economy, vol. 75 (February 1967), pp. 20–37.CrossRefGoogle Scholar

⁶ Wolf, Bernard M., “Internationalization of U. S. Manufacturing Firms: A Type of Diversification” (unpublished doctoral dissertation, Yale University, 1971)Google Scholar. Wolf's study is based on industry data, though at a finer degree of disaggregation than that used by Gruber, Mehta, and Vernon.

⁷ Horst, Thomas, “Firm and Industry Determinants of the Decision to Invest Abroad,” Review of Economics and Statistics, vol. 54 (August 1972), p. 261 (emphasis in original). See also pp. 258–266.Google Scholar

⁸ For evidence on this effect in the short run, see Severn, Alan K., “Investment and Financial Behavior of American Direct Investors in Manufacturing,”in Conference on International Mobility and Movement of Capital, edited by Machlup, F. et al. (Columbia University Press, 1972), pp. 367–396.Google Scholar

⁹ For weak evidence on this line of causation, see Mueller, Dennis, “The Firm Decision Process: An Econometric Investigation,” Quarterly Journal of Economics, vol. 81 (February 1967), pp. 58–87.CrossRefGoogle Scholar

¹⁰ To the extent that implicit interest cost on accounts payable is included in the nominal price of goods purchased, EBIT and EBIAT are understated. But as Weston and Mansingka point out, the percentage of non-interest-bearing liabilities is generally small for manufacturing firms. See Weston, J. F. and Mansingka, S., “Tests of the Efficiency Performance of Conglomerate Firms,” Journal of Finance, vol. 26 (September 1971), p. 925CrossRefGoogle Scholar. EBIAT is biased upward for leveraged firms, because all of interest is added back to net after-tax profits, rather than only interest less the tax saving.

¹¹ In addition, the U. S. Revenue Act of 1962 did not have the same effect on all firms, because of the restriction on “unreasonable accumulations” of earnings. See Krause, L. B. and Dam, K. W., Federal Tax Treatment of Foreign Income (Washington, D. C.: Brookings Institution, 1964).Google Scholar

¹² “Direct investment” is defined as the median of the percentage of total firm sales, earnings, assets, employment and production outside the United States. See Bruck, Nicholas K. and Lees, Francis A., “Foreign Investment, Capital Controls, and the Balance of Payments,” The Bulletin, Institute of Finance, New York University, Graduate School of Business Administration, no. 48–49 (April 1968), p. 69 and p. 17 for details of this definition.Google Scholar

¹³ This sample was drawn from the sample of firms described in Severn, “Investment and Financial Behavior,” by omitting those which did not meet the 10 percent criterion and those which were not in the Fortune 500.

¹⁴ This control sample was randomly selected from among those firms in the Fortune 500 which did less than 10 percent of their business abroad in 1965.

¹⁵ These ratios were obtained from National Science Foundation, Basic Research, Applied Research, and Development in Industry, 1962, (Washington, D. C.: U. S. Government Printing Office, 1963), p. 120Google Scholar. While the use of industry averages may be suspect, limited tests based on company-level data shown below, cause no important changes in the interpretation of the results.

¹⁶ A lag of three years was used because a previous study has suggested such a lag between R&D and peak results in profitability. See Scherer, F. M., “Corporate Inventive Output, Profits, and Growth,” Journal of Political Economy, vol. 73 (June 1965), pp. 290–297.CrossRefGoogle Scholar

¹⁷ This procedure builds in a potential bias, since the book value of assets is an imperfect proxy for the current reproduction cost of such assets. When two variables are normalized, i.e., divided by a variable containing a random error, the resulting ratios may be correlated even if the two original variables are uncorrelated. See Meyer, John R. and Kuh, Edwin, The Investment Decision (Cambridge, Mass.: 1959), pp. 258–267Google Scholar, for a discussion of normalization bias. To test for normalization bias, we calculated (a) the homogeneity of the dollar amount of the profits and of R&D expenditures with respect to assets, and (b) the coefficient of variation of assets. The results were acceptable when we removed one firm from our original sample, leaving a sample of 48 firms. Our results were virtually unchanged when the offending firm was returned to the sample.

¹⁸ The mean value of industry-average R&D was much higher for the direct-investor sample than for the control group; the difference between these two means was significant at the 1 percent level. Further evidence that firms in our control group were not concentrated in research-intensive industries is that, of the 70 control firms and 48 direct investors in our samples, we were able to find only 13 pairs (of one direct investor and one control-group firm) which were in the same three-digit industry. Evidence on the difference, among samples, in profitability is shown in Table 1.

¹⁹ Although firm size and direct investment are related (see Horst, “Firm and Industry Determinants”), the direct investment dummy remains significant when firm size (measured by total assets) is added to the equation.

²⁰ A negative coefficient would be plausible, since international diversification may reduce the variance of a firm's earnings and thereby improve the well-being of its stockholders. Note that after 1963, the Interest Equalization Tax prevented them from obtaining international diversification directly.

²¹ In this test, R&D expenditures were lagged only two years, rather than three as before, because of data availability. However, fuller data for a handful of firms suggest that the relationship of a firm's R&D/sales ratio to the ratio for its industry tends to be stable over time.

²² For evidence that advertising has the same effect, see Comanor, W. S. and Wilson, T. A., “Advertising Market Structure and Performance,” Review of Economics and Statistics, vol. 49 (November 1967), pp. 423–440.CrossRefGoogle Scholar

²³ It is interesting to note that with improved data, rises to .44 when only R&D variables are used. This compares to an of .34 obtained by Comanor and Wilson, Ibid., with the aid of seven structural variables. Since R&D is not among their seven variables, it is likely that a comprehensive set of data would explain most of interfirm differences in profitability.

²⁴ Our test of firm-level R&D data suggests that the R&D coefficients in Table 2 approximate the true effect of a firm's R&D effort on its profitability. The data for this test suggested that firm-level data averaged 0.8 percent higher than industry-average data. Let us assume that industry-average R&D data are an unbiased proxy for firm-level data in the control group. Then the coefficients of the direct-investment dummy in Table 1 should be reduced by 0.8 times the coefficients of the R&D variable in Table 2. With this adjustment, the coefficients of the direct-investment dummy in Table 1 are still positive and significant.

²⁵ See Vernon, Raymond, “International Investment and International Trade in the Product Cycle,” Quarterly Journal of Economics, vol. 80 (May 1966), pp. 190–207.CrossRefGoogle Scholar

²⁶ These conclusions are not invalidated by the use of a direct-investment variable which is defined as the ratio of foreign sales to total sales. While nonsignificant, this variable has an algebraically higher bivariate correlation with R&D, consistent with the exporting of new products. This finding accords with that of Horst, who concluded that the technological intensity of a U. S. industry was related to the sum of exports and subsidiary sales to the Canadian market; U. S. participation in that market depends on comparative advantage defined in terms of technology. See Horst, Thomas, “The Industrial Composition of U. S. Exports and Subsidiary Sales to the Canadian Market,” American Economic Review, vol. 62 (March 1972), pp. 37–45.Google Scholar

²⁷ This finding confirms and strengthens Horst's finding that R&D is uncorrelated with direct investment, once firm size and industry are taken into account. In other words, our direct-investor sample had higher R&D because firms in that sample were in different industries than were firms in our control group.

²⁸ This statement relies on the usual convention that zero raised to the power zero equals unity.

²⁹ Letting n^⇒∞ requires the assumption of a constant proportion between knowledge and reported assets; otherwise we would have no firms possessing both knowledge and assets. This assumption in turn implies that the ratio of knowledge to assets is equal to r(l-d)/(g+d) at all times, as can be observed from the denominator of this equation.

³⁰ See various issues of National Science Foundation, Basic Research, and Securities and Exchange Commission, Quarterly Financial Report for Manufacturing Corporations.

³¹ See Laurence, Martin M. and Severn, Alan K., “The Relative Growth Performance of American Multinational Manufacturing Firms, 1959–1965”(paper prepared for delivery at the meetings of the American Academy of Management, Boston, Mass., August 20–22, 1973).Google Scholar

³² These are, respectively, the 1960 sample mean of industry-average data, and our best guess as to the true, firm-level data for 1965.

³³ This would represent a service life of 25 years with double declining balance, or 18.75 years with 150 percent declining balance—far longer even than the nominal patent protection of 17 years.