Data used in this study are from the Taiwan Economic Journal. Our data sample contains all firms listed on the Taiwan Stock Exchange (TSE) from January 1996 to December 2005. The firms traded over-the-counter (OTC) are excluded in our analysis. Financial firms and firms with negative book values on each formation date are also excluded from the sample.
During the sample period we cover from 1996 to 2005, Taiwan’s economy and stock market went through tremendous changes. In the middle of the 1990s, the Taiwan Weighted Stock Index skyrocketed from a low 4,700 level in January 1996 to
nearly the 10,000 level in July 1997. The peak lasted for approximately three years.
The summit was then followed by a sharp decline form a high of 9,854 in April 2000 to a low 3,637 level in September 2001 during less than two years. Afterward, the index traded up and down around the 6,000 level until December 2005. Obviously, both the market conditions and the business cycles were greatly different over the entire sample period, and so how the relationship between stock returns and R&D intensity evolved over time is an interesting issue by itself.
For the sake of analysis, we divide the entire sample into three subperiods: the bubble-forming period from January 1996 to March 2000, the burst-of-bubble period from April 2000 to September 2001, and the post-bubble period from October 2001 to December 2005. The final sample consists of 635 firms including 56,418 observations that exclude invalid and insufficient data.
In this paper we use two measures of R&D intensity: R&D expenditure relative to total assets and that relative to the book value of equity. It is conceivable to use the relative amount rather than the absolute amount. These definitions are standardized in the literature. There are still other possibilities of normalization in defining R&D intensity. For instance, aside from using total assets and book value of equity, Chan, Lakonishok, and Sougiannis (2001) also take the market value of equity, sales or earnings as the denominator. The reason we choose total assets as a measure from these candidates is because sales and earnings are more changeable over time. We try to find a relatively stable measure to reflect a comparatively long-term strategy of the firm. Using sales and earnings as the denominator to normalize R&D expenditure will make the consequent variables of R&D intensity too volatile than they actually are. Furthermore, according to many other studies, for example, Chiao and Hung (2006), the book value of equity is another appropriate measure. Therefore, for the sake of a robustness check, we also perform the analysis
using R&D expenditure relative to book value of equity as the R&D intensity. The two measures of R&D intensity are denoted as RD/A and RD/BE, respectively.
From Table 1 we see that only 65% of the sample firms carry out R&D activities every year. The ratio is especially low for the bubble-forming period from 1996 to 1999. After 2000, the number of samples with positive R&D expenditure climbs to 3,315, and since then the ratio has continued to increase gradually. Table 1 also provides statistics of R&D intensity at the aggregate level. It is obvious that R&D intensities are on the rise over the entire sample period whether we observe from the view of RD/A or RD/ME. In the late 1990s, the value-weighted average R&D expenditure was below 0.3% of total assets or less than 0.5% of the book value of equity, while in the 2000s the value-weighted average R&D expenditure increased to more than 0.4% of total assets or near 0.65% of the book value of equity. The equally-weighted R&D intensities also show an upward trend, indicating that R&D expenditures in both large and small firms have increased. The figures here are less than the ones in Japanese firms presented by Xu and Zhang (2003).
Although our focus in this paper is on the relation between stock returns and R&D intensity, we cannot say that R&D intensity is the only variable. According to the Capital Asset Pricing Model (CAPM), investors will be rewarded with a higher expected return on stocks with a higher systematic risk as measured by the market beta of the stocks. We therefore include market beta as an additional explanatory factor for expected returns. In addition, Fama and French (1992) document that, for US stocks, firm equity size and book-to-market ratio are two important variables that have predictive power on stock returns while the market beta based on the CAPM does not have much power. A similar situation could be found in the study of other countries. Chan, Hamao, Laconishok (1991) relate cross-sectional differences in returns on Japanese stocks to the underlying behavior of four variables including the
size and the book-to-market ratio. They find that the book-to-market ratio plays a significant role in explaining the cross-sectional changes of stock returns in the Japan stock market. Size effect also exists but the statistical significance of size is sensitive to the specification of the model. These results are confirmed by Kubota and Takehara (1996, 1997), Chui and Wei (1998), Jagannathan, Kubota, and Takehara (1998) and Daniel, Titman, and Wei (2001). In this paper we will examine the explanatory power of R&D intensity with and without size, the book-to-market ratio and the market beta respectively.
The size of a firm for month t is measured as the market value of equity at the end of the last month. The book-to-market equity for month t uses a firm’s latest available book value of equity divided by its market value in the same month. A natural logarithm is taken on both size and the book-to-market ratio as standard in the literature. Using the Taiwan Weighted Stock Index as a proxy for the market, we estimate the market beta for month t from the most recent 60 months of return data for each stock. For a given month t, a firm’s R&D intensity is calculated using its most recent accounting numbers before month t.
R&D activities are important and beneficial for the development of the economy, but for every individual firm, especially firms in traditional industries, the R&D intensity may differ across different industries. High-tech industries naturally require more R&D activities than low-tech industries. Whether or not a firm is regarded to have invested more in R&D expenditure may depend on the industry it belongs to. Table 2 lists the R&D intensities of the electronics industry and the non-electronics industry respectively by year over the entire sample period. We see that the R&D intensities of the electronics industry are much higher than those of the non-electronics industries every year. Table 3 shows the industrial R&D intensities for included industries in 2005. Except for automobile and electronics, there is no
one industry with R&D intensity over 0.5% according to the RD/A measure.
3.2 Preliminary Analysis
Table 4 reports some descriptive statistics of variables we will use later. First, we calculate the cross-sectional average of monthly returns for each month. We then report the mean and the standard deviation (S.D.) of the return averages over time.
Similarly, we report the statistics for size, book-to-market ratio, and R&D intensity.
Considering that firms with or without R&D activities may present differences in sample characteristics, we report the statistics for the two categories separately: one includes firms with no R&D expenditure and the other includes firms with positive R&D expenditure. As shown in Panel A, for the entire sample, the average monthly return of stocks with positive R&D expenditure is larger than that of stocks with no R&D expenditure, with a difference of 0.27%, but the standard deviation of return average is smaller than that of stocks with no R&D expenditure. The result is contrary to the argument that firms with higher returns should endure higher risk.
The size ln(ME) of stocks with positive R&D expenditure is nearly close to that of stocks with no R&D expenditure, which indicates that larger firms have no tendency to invest more in R&D projects. However, there is a strong difference in the book-to-market ratio between the two categories.
Panels B to D report the same statistics for the three subperiods. In the bubble-forming period from January 1996 to March 2000, the average returns are high and standard deviations are low, relatively speaking. Returns are higher for firms with R&D expenditure than for firms without R&D expenditure. The standard deviation of return average for firms with R&D expenditure is also larger than for firms without R&D expenditure. During the burst-of-bubble period, the return averages of both categories decline sharply, but the one with R&D expenditure declines more. On the contrary, the standard deviation of return average in this
period is larger than the first one.
Taiwan’s economy entered into a prolonged adjustment after 2002. During the post-bubble period, Taiwan encountered a global economic recession and unprecedented domestic political chaos. It happened that the price of raw materials and metals rose up rapidly. We find that firms without R&D expenditure have higher returns as well as standard deviation of return on average than ones with R&D expenditure. We explain the phenomenon that most firms with R&D expenditure are electronics industries whose stock performance is easily affected by the preceding factors. In conclusion: first, the average returns of stocks with positive R&D expenditure is smaller than that of stocks with no R&D expenditure, and the standard deviation of return average is also smaller for the for the category of positive R&D firms; second, the size ln(ME) of stocks with or without R&D expenditure makes no difference; finally, the book-to-market ratios of firms with R&D expenditure are higher than those with no R&D expenditure for each subperiod.
Table 5 provides pairwise correlations of these variables for the entire sample and for the three subperiods separately. For the entire sample, the average return is higher for R&D intensity based on total assets than on the book value of equity. The subperiod analysis reveals that most of the R&D effect comes from the bubble-forming period and the burst-of-bubble period. In the post-bubble period, R&D intensity is in fact negatively correlated with the average return on the basis of RD/A or RD/ME. The correlations between the stock return and the R&D intensities are typically small, partially because the size of the cross-sections is large and some of the firms actually have zero R&D expenditure. The main findings of Table 5 can be summarized as follows: first, in the post-bubble period, the R&D expenditure is not advantageous to stock performance; second, the positive correlation between the stock return and two R&D measures, though not large, suggests that R&D might be
another potential explanatory variable for stock returns; finally, the correlation between size, book-to-market ratio and two R&D measures is not large, and so potential collinearity in the later regression analysis will not be problematic.