Using real options theory to explain patterns in the valuation of research and development expenditures

An alternative way to measure the value of a firm’s real options is the difference between the market value of the firm (MV) and the value of the assets in place (AIP) (see, Hu et al. 2013; Makrominas 2017). Unfortunately, this method of measuring the value of a firm’s real options cannot be used in a test of value relevance (or in a regression model involving stock returns or stock prices) as it is mechanically related to the dependent variable. However, this concept can be used to test the validity of other real options proxies. Following Hu et al. (2013), I estimate a market based real option proxy as MV _t  − AIP _t . Hu et al. (2013) measure the value of the assets in place as the present value of future cash flows generated by existing resources, assuming that the future cash flows are a perpetuity. I define AIP in two different ways: cash flow from operations in year t−1 less net capital expenditures in year t−1, and as the mean analyst forecast of year t earnings. I estimate the correlation between this market based real option proxy and the variable of interest in this paper—R&D expenditures. As expected, R&D expenditures are correlated with the market based real option proxy for the entire sample (0.13, p < 0.0001), and the correlation is stronger for loss firms (0.38, p < 0.0001) and highly R&D intensive firms (0.21, p < 0.0001).

More analysts cover R&D intensive firms and private information gathering is higher for R&D intensive firms (Barth et al. 2001; Barron et al. 2002; Ho et al. 2007; Palmon and Yezegel 2012). Analysts following these firms have incentives to forecast the future benefit from R&D activities and determine how these benefits will impact future earnings.

The use of book value as a proxy for the value of real options is also complicated by the fact that for loss firms, book value acts as both an abandonment option and a proxy for expected future earnings (see, e.g., Collins et al. 1999; Pinnuck and Lillis 2007).

A recent study has also documented that recognized intangible assets are positively associated with the market’s perception of real growth options (Makrominas 2017).

In the Feltham and Ohlson 1995 model, the current operating assets variable (oa) captures the adjustment necessary to predict future profitability in the presence of conservative accounting (p. 706). Following Franzen and Radhakrishnan (2009), I substitute R&D expenditures for oa because they are the focus of this study and they also reflect conservative accounting due to their immediate expensing. This is consistent with previous papers that have used current R&D expenditures to capture the stock of R&D capital (see, e.g., Sougiannis 1994; Green et al. 1996). Other studies use capital expenditures to capture the conservative accounting for property, plant and equipment (see, e.g., Myers 1999). As a sensitivity analysis, I include capital expenditures in all regressions and all results are consistent.

As discussed in the sample selection section, the sample is limited to December 31 year-end firms to ensure that analysts had access to the same microeconomic and industry data for all firms at the time that the forecast was made.

Some studies examining whether the market values R&D expenditures deflate by the book value of equity (BV _i,t ) (e.g., Sougiannis 1994; Green et al. 1996). As a sensitivity analysis, all variables were deflated by BV _i,t and the results were consistent.

I/B/E/S reports earnings forecasts on a per share basis and a stock split adjusted basis. To convert the forecast per share to an undeflated forecast, the I/B/E/S forecast is converted to the actual, pre-split forecast and then multiplied by the shares used in the EPS calculation as reported by Compustat.

Dechow et al. (1999) and McCrae and Nilsson (2001) estimate ω for multiple lags of abnormal earnings and find that the additional lags have only a trivial impact on explanatory power. Accordingly, I only include one lag.

All analyses were also done using four and five groups ranked on R&D intensity and all results were consistent.

Industry-level sales were also measured using 3-digit SIC codes, 4-digit SIC codes, and the Hoberg-Phillips text-based fixed industry classifications (see Hoberg and Phillips 2010, 2016). All results were consistent.

All analyses were also done using data from two later time periods: 1994–2011 and 1994–2007 (in order to eliminate any distortions in market prices due to the financial crisis beginning in 2008). All results were consistent.