R&D information quality and stock returns

Abstract

Investors demand higher premiums from firms whose future R&D performance is difficult to evaluate. We construct a measure of R&D information quality (RDIQ) by linking a firm's historical innovation input (R&D expenditures) and innovation outcome (sales) and find significant evidence that expected excess returns decrease with RDIQ. We find that the high-minus-low RDIQ hedge portfolio earns excess returns of −23 (−25) bps per month in value-weighted (equal-weighted) returns. We also find that the RDIQ effect is weakly correlated with commonly used risk factors, is stronger for firms with greater uncertain business environment, and exhibits incremental pricing power.

Introduction

Information contained in a firm's research and development (R&D) plays a critical role in guiding investors' evaluation of its future prospects. However, it is usually difficult for investors to process and evaluate such information for two reasons: first, different from a firm's tangible assets, R&D is intangible and is associated with future-oriented long-term activities in science and technology, whose outcomes are hard to predict; second, the lack of accounting disclosure suggests that investors may not be fully informed of all information related to a firm's R&D activities, creating problems of asymmetric information (Aboody and Lev, 2000). Therefore, it is natural to believe that investors may regard R&D information quality as an important factor when making investment decisions. Even though numerous studies have investigated whether a firm's R&D information is fully impounded in its stock prices (e.g., Chan et al., 2001; Eberhart et al., 2008; Li, 2011; Cohen et al., 2013), little is known how stock prices react to R&D information quality.

The theoretical predictions on how general information quality affects asset returns are inconsistent. Veronesi (2000) considers a pure exchange economy with power utility preferences and shows that the equity risk premium increases with information quality. However, Ai (2010) and Brevik and d’Addona (2010) find that when investors have high information quality, equity premiums decrease. Epstein and Schneider (2008) present an ambiguity-aversion model and suggest that investors require compensation for holding assets with low information quality. In this paper, we empirically explore the relationship between R&D information quality and expected stock returns and examine whether investors demand higher premiums from firms whose future R&D performance is more difficult to evaluate.

To implement the empirical analysis, we develop a measure of R&D information quality (RDIQ), which captures the extent to which the variation of a firm's fundamentals can be explained by its R&D expenditures. Specifically, a firm's R&D information quality is measured based on the R-square generated from a regression of its sales growth on its realized R&D capital. The smaller the R-square, the lower the R&D information quality should be, suggesting that investors face great uncertainty in evaluating R&D information. When constructing the RDIQ, we take into account the fact that different firms may have different R&D lifespans and that each year's R&D expenditures may have different effects on future sales growth. Using all firms listed on the NYSE, AMEX, and NASDAQ with valid accounting and returns data over the period ranging from July of 1981 to June of 2015, we find that the one-year-apart persistence of RDIQ is 0.54 (t = 44.4) and is not correlated with other firm-specific variables (e.g., size, book-to-market, cash holdings, return on assets, return on equity, and certain innovation-related variables). This suggests that RDIQ is distinct from well-known firm characteristics and may contain different information.

We hypothesize that when the firm's past track record indicates low R&D information quality (a small R-square), investors face a high degree of uncertainty in evaluating its R&D information and are unwilling to make investments in its future R&D activities: they would require high premiums to make such an investment. To test our hypothesis, we conduct a portfolio analysis similar to Fama and French (1996). At the end of June of each year, we sort all firms into three RDIQ portfolios (low, middle, and high) based on the 30th and 70th percentiles of RDIQ in the previous year and construct a hedge portfolio that longs the high RDIQ portfolio and shorts the low RDIQ portfolio. We hold these portfolios over the next 12 months and compute their value- and equal-weighted monthly returns. We find that the average excess portfolio returns decrease with RDIQ. For example, the low RDIQ portfolio earns 125 bps (t = 4.35) per month in value-weighted returns and 126 bps (t = 4.33) per month in equal-weighted returns, whereas the high RDIQ portfolio earns only 102 bps (t = 2.99) per month in value-weighted excess returns and 101 bps (t = 3.51) per month in equal-weighted excess returns. Furthermore, the average monthly returns on the hedge portfolio are economically substantial and statistically significant, yielding −23 bps (t = −3.76) in value-weighted excess returns and −25 bps (t = −3.88) in equal-weighted excess returns. The same pattern also holds for characteristic- and industry-adjusted returns.

The alphas from the different factor models also decrease with RDIQ. More specifically, in the Fama and French (1993) three-factor model, the alpha for the low RDIQ portfolio is positive and statistically significant, whereas the alpha for the high RDIQ portfolio becomes small and insignificant. The alpha for the high-minus-low RDIQ hedge portfolio is −23 bps (t = −3.61) per month in value-weighted excess returns and is −25 bps (t = −3.79) per month in equal-weighted excess returns. The pattern for the estimated alphas for the low, middle, and high RDIQ portfolios and the hedge portfolio is the same in the Carhart (1997) four-factor model, in the q-factor model (Hou et al., 2015), and in the mispricing-factor model (Stambaugh and Yuan, 2017). These results suggest that investors are less certain about the prospects of low RDIQ firms’ future R&D activities and therefore require higher premiums when making such investments.

We also perform Fama-MacBeth cross-sectional regressions that allow us to control for a large number of variables, including size, book-to-market, momentum, leverage, idiosyncratic volatility, illiquidity, and innovation-related variables. Despite such extensive controls, we find that the coefficient on RDIQ is always negative and statistically significant. This finding provides further evidence in support of our hypothesis that expected excess returns decrease with RDIQ.

If our RDIQ measure really captures investors' evaluation of a firm's R&D information quality, it should have little effects on its fundamentals and on uncertainties of its fundamentals. For this purpose, we conduct panel regressions of firms' future operating performance as measured by return on assets, sales, cash flows, and performance, and their uncertainties on RDIQ. After controlling for standard variables in the regressions such as size, book-to-market, leverage, capital expenditures, net property, plant, and equipment, and certain innovation-related variables, we find that for each of the four proxies of fundamentals and their uncertainties, the coefficient on RDIQ is statistically insignificant.

Our hypothesis is that investors require high premiums to invest in firms with low RDIQ. We therefore posit that the RDIQ-return relationship should be stronger for firms with smaller market capitalization, younger firms, firms with greater analyst forecast dispersion, and/or firms with higher fundamental volatility. These firms may have more uncertain business environments and investors face more information asymmetries and/or are more ambiguous about their future prospects. To test this hypothesis, we perform double sorts on RDIQ and size, age, analyst forecast dispersion, or fundamental volatility. We find that the high-minus-low RDIQ hedge portfolio earns −29 (t = −2.09) bps per month for small firms, whereas it earns only −12 (t = −1.43) bps per month for large firms. The alphas from the Fama-French three-factor model, the Carhart four-factor model, the q-factor model, and the M-factor model for the hedge portfolio are −37 (t = −2.60), −44 (t = −2.80), −45 (t = −2.76), and −49 (t = −2.69) bps per month, respectively, for small firms, whereas they become small and insignificant, −10 (t = −1.14), −10 (t = −1.16), −6 (t = −0.56), and −12 (t = −1.25) bps per month, respectively, for large firms. Our tests on age, analyst forecast dispersion, and fundamental volatility present the same implications.

To further explore the relationship between R&D information quality and future stock returns and to examine whether the RDIQ effect reflects commonalities in returns that are not captured by existing factors, we construct a factor-mimicking portfolio for R&D information quality following a methodology that is similar to the one in Fama and French (1993). At the end of June of each year, we sort firms independently into two groups based on size (small “S” and big “B”) and into three RDIQ groups (low “L”, middle “M”, and high “H”). The intersection of these portfolios forms six size-RDIQ portfolios (S/L, S/M, S/H, B/L, B/M, and B/H). The RDIQ factor (IQF) is constructed as (S/L + B/L)/2 − (S/H + B/H)/2. We find that the IQF is weakly correlated with commonly used factors such as the Fama-French five factors. We also find that IQF captures a different pricing factor that is distinct from the existing factors through constructions of tangency portfolios. For example, adding IQF to the Fama-French three factors improves the ex post Sharpe ratio by 33.3% and adding IQF to the Fama-French five factors improves the ex post Sharpe ratio by 14.3%.

Our study relates and contributes to two strands of literature. On the one hand, many studies examine whether asset prices fully impound the information contained in the innovation process. Chan et al. (2001) find that R&D intensity measured as R&D expenditures relative to the market value of equity has the ability to predict future returns. However, its predictability power disappears when the ratio of R&D expenditures to sales is used. Eberhart et al. (2008) report significantly positive long-term abnormal stock returns following unexpected and economically significant increases in R&D and suggest that increases in R&D investments are beneficial expenditures, but the market underreacts to this benefit. Li (2011) suggests that the positive relationship between R&D intensity and stock returns exists only in financially-constrained firms, and this relationship is robust to measures of R&D intensity. Cohen et al. (2013) demonstrate that firm-level innovation is persistent and predictable, but the market appears to ignore the publicly available information in R&D when valuing future innovation. Gu (2005) finds that changes in patent citations relative to total assets are positively related with firm's future earnings and stock returns. Pandit et al. (2011) show that a firm's patent citations are positively associated with its future operating performance. Hirshleifer et al. (2013, 2017) construct an innovative efficiency (IE) measure and an innovative originality (IO) measure, respectively, using the number of patents and patent citations of a firm and find that both IE and IO positively predict future stock returns. They mainly attribute this positive IE/IO-return relationship to limited investor attention. However, unlike the studies above, we focus on information quality/uncertainty in the innovation process by linking innovation input (R&D) to innovation outcome (sales).

On the other hand, how general information quality/uncertainty affects asset returns has attracted considerable attention. Veronesi (2000) finds that the equity risk premium increases with information quality, whereas Ai (2010) and Brevik and d’Addona (2010) find an opposite result. Chen and Epstein (2002) show in a theoretical model that excess return should be composed of a risk premium and a premium for Knightian uncertainty (ambiguity), and Epstein and Schneider (2008) make a further refinement and present a model in which investors demand a premium for holding assets with low quality information. Zhang (2006) investigates the relation between information uncertainty and stock returns. He finds that greater information uncertainty leads to higher expected excess returns following good news but lower returns following bad news. In this paper, we focus on this seemingly contentious issue using information contained in a firm's R&D activities and investigate the relation between R&D information quality and future stock returns. Our results show that expected excess returns contain a premium for R&D information quality and that the higher information quality is, the smaller the future excess returns will be.

The rest of the paper is organized as follows. In Section 2, we introduce a measure of R&D information quality and provides data and summary statistics. In Section 3, we investigate the relation between R&D information quality and future stock return. In Section 4, we provide additional evidence on the return predictive power of R&D information quality. In Section 5, we construct an R&D information quality factor. We conclude in Section 6.