ABSTRACT Three essays consider alternatives to agency theory explanations for the diversification discount, as discussed in the introduction (chapter one). The two empirical studies use extensive data on firm knowledge assets in the form of patents. In the first essay (chapter two), a sample of 237 firms that diversify between 1980 and 1992 is used to test hypotheses derived from theories related to the resource-based view of the firm. Contrary to expectation, I find the diversifying firms actually outperform their close competitors immediately prior to diversification, as measured by Tobin's q. Two years after the event, a mean discount of 14% has appeared. However, the diversifying firms maintain a mean q near one, suggesting the firms may still be pursuing positive NPV projects. The core segments of the firms are not harmed. Rather, the performance drop apparently comes from extremely low returns on assets in new segments over the first several years of their existence. As hypothesized, firms that stay focused have higher R&D intensity and larger citation-weighted patent counts. Other patent-based measures show that focused firms have more specialized patent portfolios. An exploratory analysis suggests it is the better performing, but less innovative firms in poor industries that diversify. ii
Most of the empirical tests of the diversification discount and of the effect of relatedness in diversification have been cross-sectional studies. Therefore, the second essay (chapter three) tests hypotheses derived from theories related to the resource-based view of the firm in a cross-sectional context. Models explaining the diversification discount apply controls to account for information about intangible assets that is missing from the usual ways of measuring Tobin's q and ROA. Relatedness is measured through a firm-specific analysis of technological resources in the form of patents. Data on 179 diversified firms and 525 single-segment firms in 1990 shows patent citation data adds information about firms' Tobin's q even after controlling for R&D investment. Consistent with the longitudinal study, diversified firms have lower intangible asset intensities and broader-based technology than portfolios of matched, single-segment firms. As in several finance studies, the diversification discount is substantial when the dependent variable is Tobin's q or market-to-book value. No discount is evident using ROA as the measure of performance. Contrary to the hypotheses, in this sample the discount is not mitigated by the patent variables, nor correlated with the patent-based measure of relatedness. However, there is a relationship between industry-level patenting characteristics and the discount. Diversified firms with the main segment in an industry with high applicability of patents across other industries have a higher mean discount than diversified firms in more specialized industries. This unexpected result is shown to occur because focused firms in those industries maintain a mean q well above one. The mean q for diversified firms in industries with broadly applicable technology is also above one, indicating these firms may be pursuing optimal investment to use excess iii
capacity. In other sectors of the economy, the mean q for both focused and diversified firms is below one. The last essay (chapter 4) addresses an important issue in empirical work within the resource-based paradigm. Since measures of performance do not account well for the replacement cost of intangible assets, correlation between levels of a resource and performance indicate the presence of competitive advantage, but not necessarily economic rents. I argue Tobin's q is superior to accounting ratios as a measure of firm performance for tests of the resource-based view of the firm. Previous research indicates several controls to include in any regression with q as the dependent variable, including estimates of the firm's stock of human capital and patented innovations. Although the regression approach has its problems, it is superior to other approaches such as adjusting q by industry or using change in q over time. Models testing hypotheses about economic rents should include measures of the firm's actual investment in a strategic, firm-specific asset, as well as the firm's current stock of that asset. An example from my research using patent data illustrates the argument. iv