• No results found

An emperical study of the relation between R&D intensity and cross-sectional stock returns

N/A
N/A
Protected

Academic year: 2021

Share "An emperical study of the relation between R&D intensity and cross-sectional stock returns"

Copied!
37
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

1

MSc Business Economics, Finance

Master Thesis

A

N

E

MPIRICAL

S

TUDY OF THE

R

ELATION

B

ETWEEN

R&D

I

NTENSITY AND

C

ROSS

-SECTIONAL

S

TOCK

R

ETURNS

Supervisor L. Zou

Lingjing Li

10067027

July 2016

(2)

2

Statement of Originality

This document is written by Student LINGJING LI who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

Table of Contents

Table of Contents ... 3 Introduction ... 4 Section I Related Prior Literature ... 7 Section II Methodology ... 12 2.1 R&D Capitalization Method ... 12 2.2 Fama-MacBeth Regression ... 13 Section III Data and Description Statistics ... 15 3.1 Data Selection ... 15 3.2 Descriptive statistics ... 17 Table 3.2.1 Description of Denotes ... 17 Table 3.2.2 Descriptive Statistics - three sets of data with different time lags ... 18 3.3 Correlation ... 20 Table 3.3 Correlation analysis ... 20 Section IV Empirical Results ... 21 4.1 Comparison of Three Time Lags ... 21 Table 4.1 Lagged cross-sectional regression of returns on R&D relative to market value of equity ... 22 4.2 Comparison of Two Time Periods within One Lag ... 24 Table 4.2.1 Half-year lagged cross-sectional regression of returns on R&D relative to market value of equity ... 25 Section V Robustness Check ... 27 5.1 Sample without Zero R&D Expenses ... 27 5.2 Taking R&D Capital Relative to Total Assets as Independent Variable ... 28 Section VI Conclusion ... 29 Appendix ... 32 Table 4.2.2 One-year lagged cross-sectional regression of returns on R&D relative to market value of equity ... 32 Table 4.2.3 One-and-half year lagged cross-sectional regression of returns on R&D relative to market value of equity ... 33 Table 5.1 Cross-sectional regression of returns on R&D relative to market value of equity, excluding zero R&D expenditures. ... 34 Table 5.2 Cross-sectional regression of returns on R&D relative to total assets ... 35 Reference ... 36

(4)

4

Introduction

In this paper, I will provide an empirical insight of the relation between firm’s R&D investment intensity and its stock returns. Stock return is the very first thing concerned by the shareholders and investors. Until today, the study of influential factors of stock return still is an attractive research area. A numerous amount of research has been done to investigate the determinants of stock return, yet it is still a hot topic to discuss about and there are questions left unanswered. Throughout times the mechanism of its fluctuation has become more complicated since any development in the financial sector as well as other industries could have effect on firms’ stock returns. With the rise of high-tech industry, firms nowadays pay more attention to the research and development (R&D) department to strengthen their core competitiveness, this thesis will be focused on this specific factor – R&D investment that is considered influencing firm’s stock return in a particular way. Does R&D investment have a positive or negative effect on firm’s stock returns? This thesis will try to answer the question by conducting empirical tests. There is a large amount of research has been done to investigate whether R&D expenditures have influence on firm’s stock performances. Chan, Lakonishok & Sougiannis (2001) and Eberhart, Maxwell & Siddique (2004) did two remarkable studies of the R&D-stock returns relationship. Chan et al (2001) provide evidence of the positive relation between R&D capital and future stock returns and develop a method to calculate cumulative R&D capital from current and previous R&D expenditures. The paper also discusses that there is a lagged market reaction towards the R&D information provided by firms. Eberhart et al (2001) study the effect of an unexpected R&D increase on long-term abnormal returns by adopting event-study methodology. They find that R&D increase could lead to a positive abnormal returns in the future.

(5)

5 Based on the previous research findings, to start up my study, I propose the R&D investment intensity has a positive relation with stock returns. R&D expenditures have grown over the last decades. Firms invest in R&D activities to differentiate themselves and mitigate competition. Empirical results show that R&D investments, actually, create value for the firm. The elaborate model will be discussed in the following sections. I use a specific method that is commonly adopted in previous research to measure the R&D investment and scale it down to a relative ratio to fit into the model. Firstly, I retrieve R&D expenses from Compustat database, then I calculate firm’s cumulative R&D capital (RDC) based on the expenses by using the five-year straight-line depreciation method of Chan et al (2001), this step of capitalization is essential because R&D expenses in previous fiscal periods are believed to have a continuous effect on firm’s performance in the future. To create a ratio proxy to measure R&D investment intensity (RDS), I use firm’s market value of equity (MV) as denominator under RDC. I use monthly holding period return to indicate stock return. As all the other sample data is quarterly, monthly holding period returns are calculated to quarterly holding period returns using the quarterly geometric means. The selection of control variables for the model will be discussed in the methodology section. I choose two time periods to analyze in the empirical study, they are the entire sample of year 1989 to 2015 and the subsample starting from year 2007 to 2015. The reason for the time selection is that, although I intended to look into longer time span (1975-2015), after the elimination of data, the available continuous sample starts only in 1989. Nevertheless, the number of observation is large enough to conduct the tests. I deliberately choose year 2007 as the start year of the subsample to have a further look into the performance of the model during the recent crisis. I take a few extra steps to add some interesting elements to the research area. As pointed out by previous research, stock market will have a lagged reaction to revise and

(6)

6 incorporate the R&D expense information, it occurred to me that to answer a question of how long the lag is exactly would be something constructive to the R&D related financial research. To have a glance at this topic, I conducted three homogeneous tests with different time lags – half year, one year and one and half years. The empirical results do appear differently due to the time lag changes, the detailed discussion will be provided in the section of empirical study. Concerning firms with zero R&D expenditures would contaminate the results of the test, a robustness check is conducted after the discussion of empirical results. I exclude all the zero quarterly R&D expenses in the sample and keep firms with continuous positive R&D expenses for the tests. The same regression method is applied, the robustness tests show very similar results to the original model, which implies the original tests have a reliable strength. Also, I used R&D capital to total assets for proxy as the independent variable to double test if the test results will be consistent. The paper consists of other six sections besides this introduction. In Section I is I will review relevant literatures done by other researchers throughout the recent decades; Section II is focused on the detailed description and demonstration of the methodology I use to conduct the empirical tests; In Section III readers will be provided the data selection procedures and descriptive statistics to have a clear picture of what types of data and variables are tested; Section IV is the discussion of empirical results of the tests, including the interpretation of coefficients, general findings and reasoning; A robustness check will be provided in Section V to strengthen the convincing power of the tests; Lastly, a general conclusion will be presented in Section VI.

(7)

7

Section I Related Prior Literature

In the past few decades, researchers have noticed the growing impact of R&D have in various aspects of firms’ performance, such as operating performance, accounting assessment, capital market performance, stock returns, and productivity. Hall (1993) reported that in the US during the 1980’s, the stock market’s valuation of the R&D investments in the manufacturing sector had fallen abruptly. Hall concluded that this rapid deterioration can be possible due to first, the indeed fall of private rate of return to R&D; second, R&D capital depreciated much more than previously; finally, the large amounts of merges and leveraged buyouts happened in the 1980’s. Hall’s finding seems to conflict the impression we have nowadays of firms who invest heavily in the R&D department. To explain this conflict, there are many arguments of investors and market under-overreact to the available information, which might be the case that firm’s stock is mispriced after the release of R&D information. Daniel, Hirshleifer and Subrahmanyam (1998) developed a theory based on investor overconfidence and on changes in confidence resulting from biased self-attribution of investment outcomes. Finding from the theory is that investors overreact to private information and underreact to public information. Lev and Sougiannis (1996) found in their paper that R&D capital did not appear to be reflected by current stock prices, since the effect it had will be on the subsequent stock returns. This indicates that market has underreaction to R&D information and needs a period of time to incorporate the information fully into the stock prices. Besides the lagged reaction of the stock market to R&D expenditure information, the firm’s market concentration and the specific feature of the time period that we look at

(8)

8 might also be influential factors for opposite movements of R&D spending and stock returns. The study of Su, Martin and Kensinger (1990) increased the evidence against market myopia with innovation. The study focused on event study of 95 R&D spending announcements during 1979 to 1985. The main finding is that there was statistically significant positive cumulative abnormal returns of 1.38% for the entire sample. But when the sample was split into high-tech and low-tech firms, the stock price response were significantly positive to high-tech firms and negative to low-tech firms. Doukas and Switzer(1992) provided an empirical investigation of the dynamic relationship between firm’s R&D expenditures and its stock market value. The authors found that R&D expenditure information did have a causal relationship with the capital market. The major finding of the paper is that how the capital market reacts to the R&D expenditures announcements depended highly on the firm’s market concentration. In other words, R&D increase would have a positive and significant impact on firm’s common stock price when the firm is highly-concentrated on technology, and when the firm has low concentration of technology, the impact would be significantly negative. The paper used empirical approach to justify the causal relationship. Xu and Zhang (2004) studied the explanatory power of R&D for the stock returns in Japan during 1985-2000. In the paper, an elaborate empirical research is conducted and a lagged Fama-MacBeth regression model is used. Due to the specific historical economy pattern of Japan, the authors also looked into sub-periods, especially during late 80’s and early 90’s which is the bubble-forming period. The main finding in the paper is that R&D intensity was helpful in explaining the expected stock returns on average, but the association was weak. During the bubble period, R&D intensity had a negative effect on stock returns, reflecting the speculative nature of the price appreciation during that time. For the Japanese stock market, the R&D effect does not differ much from high-tech and low-tech firms.

(9)

9 Apart from these investigations of the negative effect R&D has on the firm, they are still partially agree on the “good side” of R&D investment. After we entered 21st century, the technology booming and salient performances of high-tech firms have caught our eyes. More investors and researchers start to believe that firms with high orientation of technology innovation tend to have a brighter prospect, and there are numerous academic papers that support this argument. Chauvin and Hirschey (1993) provided evidence that advertising and R&D expenditures had large, positive and consistent influence on the market value of firms. The authors also reached the conclusion by empirical approach, current cash flow, growth, risk and market share are also taken into consideration to be related to market value of firms. Lev and Sougiannis (1996) addressed in their paper the issues of reliability, objectivity and value-relevance of R&D capitalization. Some of their major conclusions stated that R&D capital were found to be strongly related to stock prices and returns, which indicates the capitalization of R&D expenditures yields valuable information to investors; One of the most referred paper, Chan, Lakonishok and Sougiannis (2001) argued that generally accepted accounting principles in the US might cause stock mispricing for firms that have large amounts of intangible assets such as R&D investments. R&D expenses are usually treated expensed rather than capitalized in the US, however, R&D investments can possibly generate benefits in the future, Chan et al (2001) tested stock returns of technology-concentrated firms with intensive R&D expenses and firms with no R&D expenses to see if the stock prices incorporated R&D investments. In the paper, R&D capital was calculated with linear depreciation. The authors based their method on the literature of Lev and Sougiannis (1996) and assumed that R&D expenses have a useful life of five years, and each expense depreciates equally 20% each year. The cumulative R&D capital is calculated as the summation of undepreciated previous R&D expenses and the current year R&D expense.

(10)

10 Two measures of R&D intensity were provided in the paper, R&D expenses to sales and to market value of equity. The paper does not find a direct link between R&D expenditures and future stock returns, nevertheless, it provided some insights of the connection between the two. The test results of first measure showed that, historically, it didn’t appear that high R&D firms outperformed others. When stock returns are tested against R&D intensity to market value, it became clearer that high R&D plays a distinctive role in higher excess returns. Interestingly, high-R&D stocks tended to perform poorly in the past, this indicates that market has a lagged reaction to R&D investment information. Using another approach, which is event study to investigate the relation, Eberhart, Maxwell and Siddique (2004) found that unexpected R&D increases could lead positive results of long-term abnormal stock returns and operating performance. 8313 firms are examed between 1951 and 2001. By adopting Fama and French three-factor model to test against abnormal stock returns, for the 5-year period following their R&D increases, consistent evidence was found that shareholders experience significantly positive abnormal stock returns. There was also strong evidence of that firms experience significantly positive abnormal operating performance. In the paper, there were also findings with evidence that firm attributes reveal mispricing that the market takes years to correct, market is slow to incorporate the information contained in corporate events. Their results provide strong evidence that investors systematically underreact to the benefit of an R&D increase. This finding strengthens the assumption of market’s lagged reaction to R&D information. Li (2011) studied the puzzles between financial constraints and positive R&D-return relation in her paper. The method of Chan et al (2001) was used to calculate cumulative R&D capital. In the Fama-MacBeth regressions, holding period return is the proxy for stock return, market capitalization and book-to-market ratio are used as control variables. Li’s finding states that a strong constraints-return relation exists among high-tech firms, and the positive R&D-return relation exists only among financially constrained firms.

(11)

11 Lin (2012) used a neoclassical model to explain some regularities in the cross-section of stock returns with endogenous technological progress driven by R&D investment. The results showed that R&D investment leveled up the expected marginal capital and lowered the marginal cost, hence increased the expected returns. Lin also pointed out that value firms are more risky than growth firms since they are more conservative in R&D investment which results in less efficiency improvement, and this disadvantage will display especially in bad times. Hirshleifer, Hsu and Li (2013) examine the relation between innovative efficiency and subsequent stock returns as well as operating performance. Using Fama-MacBeth cross-sectional regressions, the authors find that firms with averagely higher innovation efficiency have higher current and future market valuations and will have better operating performances and stock returns. Hou, Hsu, Watanabe and Xu (2016) used Fama-MacBeth regressions and conducted both cross-country analysis and firm-level analysis, reaching the conclusion that in the international equity markets, firms with higher R&D intensity, are experiencing higher stock returns subsequently, this conclusion reminds us the fundamental important role of intangible R&D investments for the asset pricing. Considering the recent crisis, which will also be an aspect that I study in this paper, firms with intensive R&D spending were very likely to have less capital during the bad times, how did they survive or did they? This has become another hot topic among researchers. Lome, Heggeseth and Moen (2016) provide empirical evidence of R&D and growth in turbulent times. The authors have found the connection of R&D activities and firm’s future revenues, and stated that there is a time gap in between when R&D investments starts to benefit firm’s revenue streams. The importance of R&D activities is even more accentuated in financial crisis. The paper suggests that managers should set a strategic vision for R&D investment because it helps to bolster the firms in crisis times.

(12)

12 Hud and Hussinger (2015) in their paper, find that government’s R&D subsidies for German SMEs have provided these firms recovery power during and after the crisis. From a broad reading of literatures, most researchers tend to capitalize R&D expenditures and use R&D capital to implement the empirical study. Oswald and Zarowin (2007) examine whether R&D capitalization is more associated with stock price than R&D expenditures and their finding provides affirmative results to the hypothesis. For the control variable selection for the model, as many of the papers mentioned here, Bens, Hanna and Zhang (2003) investigate the effect of capitalized R&D expenditures on the cross-sectional variation of stock returns. The authors examine the difference between alternative measures of capitalized R&D expenditures and how this measurement choice can affect conclusions of R&D explaining the cross-section of stock returns. The finding of the cross-sectional regressions is that firm size, book-to-market ratio and R&D intensity are all strongly associated with firms’ stock returns.

Section II Methodology

2.1 R&D Capitalization Method In common with the majority of the R&D literature, I base the proxy for R&D intensity on a capitalized R&D variable, namely cumulative R&D capital. The procedures to obtain R&D capital involves getting R&D expenses data and calculating cumulative R&D capital using a depreciable method, the final step to obtain R&D intensity is deflating R&D capital to relative amounts. The R&D expenses can be retrieved from Compustat - Fundamentals Quarterly directly. In order to capitalize R&D expenses, I use the method from Chan et al (2001) which assumes that R&D assets have an estimated useful life of five years and

(13)

13 depreciate in a straight-line manner. The cumulative R&D capital (RDC) of a firm in current fiscal quarter is calculated as: 𝑅𝐷𝐶$% = 𝑅𝐷𝐸$%+ 0.8 ∗ 𝑅𝐷𝐸$%-.+ 0.6 ∗ 𝑅𝐷𝐸$%-0+ 0.4 ∗ 𝑅𝐷𝐸$%-2+ 0.2 ∗ 𝑅𝐷𝐸$%-4 The time lag in this capitalization model is fixed - only “quarter”, since a lagged model will be adopted in the time series regression as well, I would hereby notify that do not mistake one for the other. After obtaining the cumulative R&D capital (RDC), I convert it to a relative amount by dividing RDC by market value of equity (ME), the quotient is used to present as R&D intensity (RDS), which will be the independent variable in the following time series regression model. RDC/ME is widely used as proxy for R&D intensity in academic research, I take it as the main indicator for R&D intensity in this paper. In order to assure the robustness of the test, I will discuss using other forms of R&D intensity in the robustness check section 2.2 Fama-MacBeth Regression As proposed previously, the hypothesis of the test is that firm’s R&D investment intensity has a positive relationship with its stock returns. Following Xu and Zhang (2004) and other succeeding researchers, I use the well-known Fama-MacBeth Regression to test the hypothesis statistically, I construct the following model: 𝑟$%67 = 𝛿9+ 𝛿.𝑅𝐷𝑆$%+ 𝜃. ∗ ln 𝑀𝐸 $% + 𝜃0∗ ln 𝐵𝐸 𝑀𝐸 $% + 𝜃2 ∗ β$%+ 𝜀$%67 The Fama-MacBeth Regression is a method which can adopt panel data analysis, and is commonly used to estimate the coefficients of risk factors for asset prices. The biggest

(14)

14 advantage of this approach is that any risk factors can be included in the regression as variables and after the regression, the effect of them on the asset price will be presented by the statistics of the factors. In the equation stated above, where 𝑟$%67 on the left side of the equation stands for stock returns with n time lag, 𝑅𝐷𝑆$% for R&D intensity. The rest of the variables all act as control variables: ln (𝑀𝐸)$% is the natural logarithms of market value of equity, it functions as control variables for the firm size. ln (GFEF)$% is the natural logarithm of firm’s book-to-market ratio and β$% is the firm’s stock beta. As indicated previously, R&D intensity is tested in the model. Let’s denote R&D capital to to market value of equity as RDCME, since market value is incorporated into the independent variable, in order to avoid omitted variable biases, I include ln (𝑀𝐸)$% as one of the control variables. Equations below give the elaboration of the regression diversion. 𝑟$%67 = 𝛿9+ 𝛿.𝑅𝐷𝐶𝑀𝐸$%+ 𝜃.∗ ln 𝑀𝐸 $%+ 𝜃0∗ ln 𝐵𝐸 𝑀𝐸 $%+ 𝜃2∗ β$%+ 𝜀$%67 Book-to-market ratio is frequently used by finance literatures as an indicator of firm’s growth condition and expected return, hence, I include it for one of the control variables as well. As the findings of Lev and Sougiannis (1996), Chan et al (2001) and Eberhart et al (2004), the market needs time to revise the reaction to the R&D expenditure information of firms, which leads to lagged stock prices that incorporate the R&D investment information and lagged stock returns accordingly. During the sorting process of the data, I found out that using different length of time lags will result in different significance level of coefficients of R&D intensity, and, more interestingly, the discrepancies seem to follow a certain trend. To form a more in-depth investigation of this issue, I conduct homogeneous tests using three different time lags, namely half-year, one-year and one and half-year to show how

(15)

15 the explanatory power of R&D intensity varies over time. Therefore, the n of 𝑟$%67 equals half year (2 quarters), one year (4 quarters) and one and half years (6 quarters) respectively. In summary, there will be three sets of data with different time lags to test, for each set, there will be two selections of time periods available to compare.

Section III Data and Description Statistics

3.1 Data Selection In the recent years, firms have realized the importance of investing in R&D and being innovative in their own industry, R&D concentration is not a unique thing for high-tech industries nowadays. In this thesis, I will unify all the industries in my sample and have a overview of effect of R&D intensity on all types of firms. By saying so, I will still exclude financial institutions and utility companies. The reason for the exclusion is that according to Fama and French (1992), both the industries have extremely high leverage and were strictly regulated during the sample period. The sample selection includes all NYSE, AMEX and NASDAQ firms in the CRSP and Compustat data base from 1975 to 2015, excluding financial institutions (SIC 6000-6999) and utilities (SIC 4900-4999). I retrieve quarterly financial reporting data from Compustat-North America-Fundamentals Quarterly. The main variables are namely: total assets (TA), common shares outstanding (CSHO), close prices (PRCC), shareholders’ equity (SEQ) and R&D expenditures (RDE). From these variables, I calculate, in quarterly basis, the firm’s market value of equity (MV=PRCC*CSHO), firm’s book-to-market ratio (SEQ/ME) and firm’s R&D capital (RDC) by using the capitalization approach stated in the methodology. The stock return data is retrieved as the holding period returns from CRSP-Stock/Security Files-Monthly Stock Files with the same time dimension. In order to match the quarterly

(16)

16 reporting cycle as the Compustat data, I calculate geometric means of every three months in a calendar quarter (RET). The data set of stock beta can be retrieved from CRSP - Stock / Portfolio Assignments – Beta Deciles. After obtaining all the three data sets, I do the filtering process regarding the following criteria and merge the three data sets after filtering. The criteria of filtering data include: • year range is 1975-2015 for all observations in Compustat and CRSP; • In Compustat sample, keep the observations with positive quarterly R&D expenses; • In Compustat sample, keep the observations with positive common shares outstanding and valid close prices, this will ensure that a valid market value of equity can be calculated; • In Compustat sample, keep the observations with positive shareholders’ equity, this will ensure that a valid book-to-market ratio can be calculated; • In CRSP sample, keep the observations with valid monthly returns for calculating quarterly returns by geometric means; • In Beta sample, keep the observations with valid beta; • In addition, the keys of the merge of the data samples are gvkey (global company key), year-and-quarter and permno (a unique permanent security identification number). The meaning of “valid” in the stated criteria is that the availability of a certain number of the data, explicitly not a “.” In the cell. After the elimination, the data availability changes. The time period shortens from 1975 to 2015 to 1989 to 2015. the main cause of this could be due to the elimination of companies with unavailable R&D expenses. In the 1970’s and early 1980’s, research and development was not recognized as an essential statement on the balance sheet, it is not unexpected

(17)

17 that there was limited amount of data resource available. Meanwhile, since the lack of R&D recognition during that period, it will not influence the test results if not included. Hence, the entire sample is from year 1989 to 2015. I use R&D capital relative to market value of equity (RDC/ME) as the measure to indicate R&D intensity in this paper. This measure is widely used in various literatures. Chan et al (2001) used this proxy as one of the R&D intensity indicators, Lev and Sougiannis (1996) focused on deflating R&D capital by market value of equity as well. 3.2 Descriptive statistics

Table 3.2.1 Description of Denotes

VARIABLE DESCRIPTION

gvkey Global Company Key

RET Quarterly Returns

RDC Quarterly (Cumulative) R&D Capital

TA Quarterly Total Assets

SEQ Quarterly Stockholders Equity

ME Quarterly Market Value of Equity

BTM Quarterly Book-to-Market Ratio

RDCME Quarterly R&D Intensity (ta)

RDCTA Quarterly R&D Intensity (mv)

LNME natural logarithm mv

LNBTM natural logarithm btm Beta Beta Table 3.2.1 provides the description of all the denotes and abbreviations readers might encounter in the paper to clear up ambiguity. TABLE 3.2.2 presents the descriptive statistics of the variables. I obtain the quarterly stock returns (RET) by calculating geometric averages of every three months in a quarter. After the filtering process, I have 59145-66049 quarterly observations for 1864-2020 firms in North America. The means, standard deviations, minimums and maximums are then reported for the quarterly returns. Similarly, the descriptive statistics for firm sizes in terms of market value (ME), balance sheet items including total assets (TA),shareholders’

(18)

18 equity (SEQ), ratio item of book-to-market ratio (BTM), the R&D items of quarterly cumulative R&D capital (RDC) and R&D intensity (RDCME) and firm’s stock beta (Beta) are reported in Table 3.2.2.

Table 3.2.2 Descriptive Statistics - three sets of data with different time lags

This table presents the number of counts, mean, standard deviation (std), minimum and maximum of the main data I collected for the regression analysis. There are three sets of them, belonging to three types of time lags: half-year, one-year and one-and-half years.

VARIABLES Number mean std min max

Half-year lagged RET 66,049 0.0101 0.0368 -0.0981 0.107 ME 66,049 2,011 16,251 0.0932 717,000 TA 66,049 822.3 5,741 0.531 261,894 SEQ 66,049 491.0 3,428 0.00900 135,490 BTM 66,049 0.596 0.653 0.000427 22.19 RDC 66,049 46.20 307.9 0 8,986 RDCME 66,049 0.0693 0.136 0 8.141 Beta 66,049 0.915 0.644 -3.072 7.361 Number of gvkey 2,020 One-year lagged RET 62,535 0.0105 0.0368 -0.0981 0.107 ME 62,535 2,014 16,008 0.195 626,550 TA 62,535 812.6 5,527 0.531 231,839 SEQ 62,535 489.0 3,360 0.00900 135,490 BTM 62,535 0.589 0.636 0.000427 22.19 RDC 62,535 46.06 303.6 0 8,683 RDCME 62,535 0.0666 0.121 0 6.266 Beta 62,535 0.923 0.644 -3.072 4.886 Number of gvkey 1,958 One and half-year lagged

RET 59,145 0.0105 0.0372 -0.0981 0.107 ME 59,145 2,016 15,851 0.195 626,550 TA 59,145 799.5 5,299 0.531 225,184 SEQ 59,145 485.4 3,281 0.00900 135,490 BTM 59,145 0.585 0.626 0.000427 22.19 RDC 59,145 45.79 298.6 0 8,431 RDCME 59,145 0.0651 0.116 0 6.266 Beta 59,145 0.926 0.642 -1.648 4.214 Number of gvkey 1,864

(19)

19 Readers might notice that the numbers of observations decrease with the increase of time lag lengths. This issue is due to the fact that, since lagged stock returns are paired with other “current” data, with the fixed total numbers of observations, the longer the time lag is, there causes more mismatched stock returns/other data, hence, the number of valid observations decrease, approximately 3000 observations are eliminated for each half-year increase of time lag. Nevertheless, since the decreases are relatively minor and the statistics of the entries vary in a small range, the discrepancy would not be expected to cause any significant biases. Since the variation of the descriptive statistics with different time lags is quite small, let’s take a look at the half-year lag data set only to have a brief idea of the magnitude of the data. The quarterly holding period return has a mean of 0.0101, a positive number which is inspiring since it indicates over time for the investors the returns of stocks are positive. The market value of equity varies from 0.0932 million to 717000 million which is very volatile, but it also shows that the sample covers companies with different firm sizes, the mean of market value is 2011 million, which is a common amount for a young high-tech firm. Similar findings can be seen with total assets and shareholder’s equity entries. I notice the shareholders’ equity/total assets ratio is very close to 0.5, meaning the firms included in the data sample have a mean of financial leverage ratio of approximately 1.0, this is a reasonable number for most industries. The mean of book-to-market ratio is 0.596, indicating that averagely the firms in the stock market are overvalued. This is a common argument among researchers and investors since the last crisis. There are academic papers pointing out this issue even before 2007 when the crisis burst out. Miller, Weller and Zhang (2002) and Jensen (2005) noticed the moral hazard problem in the equity market and advocated to put it to cease. Until today, US stock market is still under overvaluation, fortunately, extensive regulations and controlling have been set and taken to maintain the health of the financial market nowadays.

(20)

20 Moving onto the R&D items, apart from the amount of cumulative R&D capital and the R&D intensity ratio, I notice there are zero R&D spending in the sample, which indicates there are firms that do not implement any R&D movements and they can be the outliers in the sample. Due to this reason, a robustness check with non-zero R&D sample will be carried out in the according section. The mean of stock beta is 0.915, a number close to 1.0. It is good to see since the sample should be able to stand for the market, which means with a weighted average beta of 1.0. 3.3 Correlation

Table 3.3 Correlation analysis

Table 3.3 presents the brief correlation check for the data mentioned above. Since I care about the relationship of R&D and stock return, it is consistent to my hypothesis that both the correlations between stock return and the two R&D items are positive, specifically 0.0026 and 0.0694. The positive correlation is stronger between stock return and R&D intensity, which will be the variable inputs for the time series regressions, this gives a optimistic hint of the hypothesis to be true, but the final outcome needs to be derived from next section. After analyzing the descriptive statistics of the sample, I conclude the sample could be considered a representative of the US stock market in general, the regression tests are following after this analysis.

RET ME TA SEQ BTM RDC RDCME Beta

RET 1.0000 ME 0.0001 1.0000 TA 0.0012 0.8607 1.0000 SEQ 0.0007 0.8854 0.9762 1.0000 BTM 0.0719 -0.0668 -0.0463 -0.0468 1.0000 RDC 0.0026 0.7762 0.8660 0.8813 -0.0560 1.0000 RDCME 0.0694 -0.0420 -0.0383 -0.0379 0.3199 -0.0159 1.0000 Beta -0.0121 0.0230 0.0157 0.0203 -0.1707 0.0269 -0.0234 1.0000

(21)

21

Section IV Empirical Results

4.1 Comparison of Three Time Lags To give a clear comparison of the time lag models, Table 4.1 presents the comparisons of complete regressions parameters with t-statistics of the three time lags in both selections of time spans. Panel A presents entire time period and Panel B presents the crisis period; Column (1) to Column (3) present time lags of half-year to one-and-half year. From a quick glance at the table, we can see that except for the third column in Panel B, which is the regression of crisis time with one-and-half-year lag, all the other regression results show that stock returns have a positive relationship with R&D intensity, this can be demonstrated by the positive coefficients of R&D intensity of the regressions. All of the t-statistics of the coefficients present prominent significance, which gives evidence that the relationship between stock returns and R&D intensity is statistically significant and the control variables have sufficient explanatory powers to mitigate omitted variable bias. The R-squared value decreases quite evenly from 1.4% to 0.8% to 0.2%, the value tends to be low due to the cross-sectional type of study I do, since the coefficients are significant, the model is valid for analyzing. In spite of the significant coefficients, as expected, the statistics vary along the increase of the time lag. All of the absolute values of the coefficients’ follow a decreasing pattern while time lag increases, for R&D intensity, specifically, the coefficient value decreases from 0.0109 to 0.00552 to 0.00360. What’s more, the t-statistics drop drastically from the first jump of time lag, it starts from 6.332 to 2.194 to 2.330, although the last value does not decrease, but compared to the first value in the half-year lag model, it has much less power to determine the relation. For the control variables, the t-statistics decrease in a similar manner to the R&D intensity coefficients’.

(22)

22

Table 4.1 Lagged cross-sectional regression of returns on R&D relative to market value of equity This table presents the comparison of cross-sectional regressions of the three time

lags. Column 1 to Column 3 present half-year lag to one-and-half-year lag; Panel A presents the entire time period and Panel B presents the crisis period

(1) (2) (3)

VARIABLES RET RET RET

Panel A: Entire sample (1989Q1 – 2015Q4)

RDCME 0.0109*** 0.00552** 0.00360** (6.332) (2.194) (2.330) LNME -0.00297*** -0.00248*** -0.00154*** (-18.53) (-16.51) (-13.24) LNBTM 0.00278*** 0.00225*** 0.000852*** (12.17) (9.544) (3.732) Beta 0.00241*** 0.00201*** 0.000864*** (11.49) (6.662) (4.312) Constant 0.0248*** 0.0230*** 0.0182*** (29.77) (25.69) (27.99) Observations 66,049 62,535 59,145 R-squared 0.014 0.008 0.002 Number of gvkey 2,020 1,958 1,864

Panel B: Crisis period (2007Q1-2015Q4)

RDCME 0.0227*** 0.00339 -0.00125 (4.919) (0.826) (-0.297) LNME -0.00826*** -0.00568*** -0.00392*** (-9.996) (-8.447) (-6.291) LNBTM 0.00901*** 0.00735*** 0.00689*** (9.405) (8.546) (8.824) Beta 0.00515*** 0.00309*** 0.00173*** (8.410) (3.227) (3.180) Constant 0.0530*** 0.0420*** 0.0351*** (11.59) (11.16) (9.884) Observations 19,664 17,957 16,299 R-squared 0.062 0.024 0.014 Number of gvkey 941 892 825

Robust t-statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1

(23)

23 Moving onto Panel B, the values of the coefficients of the R&D intensity drop drastically as well, especially between the half-year-lag model and one-year-lag model, the coefficient decreases from 0.0227 to 0.00339 and ends up a negative value of -0.00125 for the one-and-half-year lag model. The t-statistics have the same movement trend along the increase of time lag compared to Panel A, the t-statistics vary from 4.919 which is statistically significant to the insignificant 0.826 and end up with -0.297. Hence, the R&D intensity coefficient of Column (2) and Column (3) in Panel B are considered not statistically significant and not valid to be interpreted. The control variables in Panel B seem to have a better performance than R&D intensity, all of them still remain statistically significant in t-statistics despite of the same decrease with time lag increase. This gives more confidence for the control variables being good controls for the model and conversely implies that the insignificance of the R&D coefficient is convincing. From the analysis above, I assume out of the three homogeneous models, model with half-year lag gives the most significant results to interpret. Additionally, it implies that the fitness of the data to the model, as well as the explanatory power of R&D intensity and other variables seem to diminish with the extension of the time lag. Nevertheless, I will keep looking at all of the three models in the following discussion. This phenomenon at this stage is merely drawn from observation, being limited to research resources, I cannot investigate further into this issue, yet it will be something interesting for future researchers to take a deep dive to see how long it will take R&D investments to be fully incorporated into the stock prices and sequently influence the stock returns. This can be an added value for firms and policy makers, as well as economists to have a clearer picture of business cycles in R&D aspect.

(24)

24 In summary, the strongest test results are derived in the half-year-lag model, the values of coefficients and significance are both large, with time lag increases to one year and one-and-half year, the significance decreases until lastly, not significant. Due to the finding stated above, I use the half-year lagged model to mainly justify the hypothesis while still keeping other two model regression results presented for comparison. 4.2 Comparison of Two Time Periods within One Lag Table 4.2.1 provides the half-year-lag model regression results with t-statistics of stock returns on R&D intensity, firm size, book-to-market ratio and stock beta taking R&D intensity in terms of market value of equity (RDCME) as independent variable. Again, Table 4.2.1 consists of two panels, Panel A presents the results of the entire sample from year 1989 to 2015 and Panel B contains sample from the recent crisis which started in 2007 to 2015. In Panel A, from the individual regressions, we can see each variable has significant t-statistics to the dependent variable which is stock return, therefore, combined with the discussion in the previous session, I can draw the conclusion that the control variables are helpful for explaining the stock returns. In the last column of Regression (7), all the variables are included to present the complete model regression results, for time span 1989-2015, 2020 firms with 66049 observations are valid to test. The firm size effect is negative with a coefficient of -0.00297, the book-to-market ratio effect and the beta effect are positive with coefficients of 0.00279 and 0.00241. With the effects of control variables, R&D intensity presents a positive relationship with stock return with a significant positive coefficient of 0.0109. The value of the coefficient appears small due to the small magnitude of both dependent and independent variables, since the purpose of this paper is to find whether positive or negative relationship is between stock returns and

(25)

25 R&D intensity, the significance of the positive coefficient (t=6.304) tends to be more meaningful to the study.

Table 4.2.1 Half-year lagged cross-sectional regression of returns on R&D relative to market value of equity

(1) (2) (3) (4) (5) (6) (7)

VARIABLES RET RET RET RET RET RET RET

Panel A: Entire sample (1989Q1 – 2015Q4)

RDCME 0.0254*** 0.0140*** 0.0108*** 0.0109*** (11.37) (7.994) (6.256) (6.332) LNME -0.00435*** -0.00364*** -0.00285*** -0.00297*** (-27.19) (-22.88) (-18.50) (-18.53) LNBTM 0.00526*** 0.00255*** 0.00278*** (25.58) (11.39) (12.17) Beta 0.000725*** 0.00241*** (3.752) (11.49) Constant 0.00830*** 0.0323*** 0.0147*** 0.00939*** 0.0277*** 0.0262*** 0.0248*** (53.61) (39.46) (80.82) (53.08) (31.94) (32.48) (29.77) Observations 66,049 66,049 66,049 66,049 66,049 66,049 66,049 R-squared 0.006 0.010 0.008 0.000 0.012 0.013 0.014 Number of gvkey 2,020 2,020 2,020 2,020 2,020 2,020 2,020

Panel B: Crisis period (2007Q1-2015Q4)

RDCME 0.0559*** 0.0264*** 0.0229*** 0.0227*** (5.306) (4.658) (4.788) (4.919) LNME -0.0148*** -0.0125*** -0.00798*** -0.00826*** (-22.63) (-15.90) (-9.504) (-9.996) LNBTM 0.0163*** 0.00878*** 0.00901*** (20.70) (9.064) (9.405) Beta 0.00267*** 0.00515*** (4.594) (8.410) Constant -0.000114 0.0908*** 0.0175*** 0.00138** 0.0751*** 0.0563*** 0.0530*** (-0.144) (23.69) (26.96) (2.356) (15.46) (12.01) (11.59) Observations 19,664 19,664 19,664 19,664 19,664 19,664 19,664 R-squared 0.025 0.047 0.043 0.001 0.051 0.059 0.062 Number of gvkey 941 941 941 941 941 941 941

Robust t-statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1

(26)

26 In Panel B, as for the crisis period (year 2007-2015), due to the shortening of time period, about half of the firms (941 firms) and one-third observations (19664 obs) from the entire sample are tested. The regression results show similar patterns, the t-statistic of each variable and combined model drops for each variable compared to the entire sample due to the decrease of data quantity, but the values of the t-statistics remain high hence the coefficients are still significant. The interesting finding is that the absolute values of the coefficients are all higher than those of the entire sample. The absolute values of coefficients of the control variables increase from 0.00297 to 0.00826 for firm size, from 0.00279 to 0.00901 for book-to-market ratio and from 0.00241 to 0.00515 for stock beta. Most importantly, the coefficient of R&D intensity increases from 0.0109 to 0.0227 with the different time spans. These comparisons indicate that all the factors in the model are having more influence on the stock returns with the time being more recent, on the other hand, this shows that stock returns are becoming more volatile to the change of these factors. Some empirical research has shown that stock volatility has increased over time. Campbell, Lettau, Malkiel and Xu (2001) suggested that there has been a noticeable increase in firm-level volatility relative to market volatility. One of the factors that might be responsible for this finding is the increasing idiosyncratic volatility. As R&D is also an idiosyncratic feature of a company, the stronger effect it has on the stock returns during more recent time period is plausible.

In the paper of Bartram, Brown and Stulz (2012), the authors found that stocks in US are more volatile than stocks in other foreign exchanges. Specifically, stock volatility is higher in the United States because it increases with firm-level investment in R&D and other factors. These factors are connected to better growth opportunities for firms. In the one-year-lag model (Appendix Table 4.2.2) the comparison between Panel A and Panel B is similar to Table 5.1, however, only the value of the coefficient of R&D intensity drops from 0.00552 to 0.00339 and becomes insignificant with a t-statistics of 0.826. With

(27)

27 the control variable still have enough explanatory power of stock returns (with significant coefficients), R&D intensity seems to lose eminent effect on the stock returns with one-year lag. In the one-and-half-year-lag model (Appendix Table 4.2.3), the positive effect of R&D capital on stock returns becomes even smaller (0.00360) for the entire time period, the insignificance of R&D intensity coefficient become larger (t=2.330), even the value turns negative (-0.00125) in the crisis time period. The reliability of the coefficient becomes questionable with a t-statistic of -0.297. Combining this diminishing explanatory power of variables with the finding from Section 4.1, instead of questioning the significance of relationship between R&D intensity and stock return, I would rather put a doubt in the choice of time lag. As expected, the models with longer time lag seem to give poorer performance with statistics of R&D intensity.

Section V Robustness Check

5.1 Sample without Zero R&D Expenses To support the empirical results of this paper and strengthen the convincing power of the findings, I conducted two types of robustness checks. First, I exclude the zero R&D expenses in the sample to prevent zero R&D from contaminating the regression results. The results can be found in Appendix Table 5.1. Table 5.1 presents the similar comparisons to Table 4.1 of complete regressions parameters with t-statistics of the three time lags in both selections of time spans. The numbers of observations and firms reduce due to the non-zero R&D expenditure elimination, nevertheless, from Table 5.1 (see in Appendix), we can see that the overall trend of the variance of the coefficients and significance of the t-statistics does not change, which supports the previous results. The statistics barely move from the results of Table 4.1, this might be due to the number of observations of zero R&D is relatively small compare to the sample, approximately 12000 compared to 66000, the zero-R&D

(28)

28 contamination is not significant enough to bias the test. Also, since the sample period starts in year 1989, firms gradually began to realize the technology innovation power and input more capital to research and development, ever since then R&D has been playing a bigger role. 5.2 Taking R&D Capital Relative to Total Assets as Independent Variable The second type of robustness check is replacing R&D capital relative to market value of equity for R&D capital relative to total assets. If R&D is invested when the market is expected to be successful, then the increase in market value works against the information contained in RDC (the numerator of the ratio). That is, the increase in the market value of equity due to expectations of future cash flow returns from the R&D will cause RDC/MV to fall, even if the firm is R&D intensive. Due to this reason, I take a balance sheet approach to deflate R&D capital by total assets (RDCTA), which is widely adopted by previous literatures as well. RDC is an estimate of asset and would appear on firm’s balance sheet if the accounting standards do not treat R&D as immediate expenses, hence R&D can be regarded as a proportion of total assets. This measure can avoid problems associated with deflating by market value of equity, it prevents the influence of the market’s response to the firm’ investments to the intangibles. There are other common approaches to obtain R&D intensity such as R&D expenditure to revenue/earnings (Chan et al. (2001)). I use R&D relative to total assets instead of revenues and earnings is due to the higher stability of total assets. Since R&D expenditure is a relative long-term stable expenses to most firms, while revenues and earnings can be fairly volatile over time, total assets is more suitable for presenting the stability of R&D capital. Similarly, Table 5.2 (see in Appendix) presents the comparisons of complete regressions parameters with t-statistics of the three time lags in both selections of time spans, taking R&D capital relative to total assets (RDCTA) as independent variable. Interestingly,

(29)

29 although the significance still decreases with the increase of time lag, surprisingly, for the crisis period in Panel B, the coefficients remain statistically significant. The value of coefficients remains positive from 0.0324 to 0.0214 to 0.0151, and the significance decrease from 5.150 to 3.392 to 2.603, large enough to make the coefficients have economical meanings. This could imply that for more recent research, R&D capital relative to total assets might be a better option for the independent variable. The reason, as also explained previously, might be due to the inconsistency of R&D capital to market value of equity ratio (RDCME). Since in recent years, firms start to recognize the added value of R&D investment for future and the boom of high-tech companies, the amount of intangible assets is growing among most firms, this business structure change can cause market value of equity have counter fluctuations to R&D intensity, treating R&D capital as a proportion of total assets can avoid this noise. Additionally, accounting rules have been refined that R&D expenditures should be capitalized more for the times instead of being treated as immediate expenses, this causes R&D capital stands more firmly on the firm’s balance sheet. Hence, deflate R&D capital with a balance sheet statement such as total assets, might be a more efficient approach for more up-to-date study. To sum up, both types of robustness check support the affirmative statement of the hypothesis, which is R&D intensity has a positive impact on stock returns.

Section VI Conclusion

In this paper, in order to find the casual relationship of firm’s R&D intensity and stock returns, I conduct several empirical tests to justify the hypothesis that R&D intensity has positive effect on stock returns. The tests involve up to 2020 firms that are listed on NYSE, AMEX and NASDAQ stock exchanges from year 1989 to 2015, using lagged Fama-MacBeth

(30)

30 Regression model. Two time periods are considered in the tests, they are from year 1989 to 2015 and from year 2007 to 2015. Three time lags are selected, they are half-year, one-year and one-and-half year. The general findings are stated below. First, out of the three lags, the half-year lag model performs the best and gives the strongest test results. The positive coefficients of R&D intensity tell that the hypothesis is true. For both time periods, the significance level of the coefficients is large enough compared to the other two lagged models, and the value of the coefficients is also the largest which indicates that R&D intensity has the biggest impact on stock returns with half-year lag. The other two models show some significant results with decreasing t-statistics, but not for both time periods. Especially for the crisis period in which there are fewer firms and observations, the significance of the coefficients drops drastically until it is no longer statistically significant. This trend implies that the explanatory power of R&D intensity as well as the control variables diminishes with the increase of time lag. The question raised and might be worth investigating for further research is that how long it takes exactly for R&D investments to have full effect on the stock performance and what factors will influence this. Due to the finding above, I use half-year lag to compare the effect R&D intensity has on stock returns in the two time periods. The regression results show that R&D intensity has a positive relationship with stock returns, furthermore, in the crisis period (2007-2015), the positive effect is larger than the entire period (1989-2015), this is consistent with the findings of previous research that stock returns are more volatile, especially in the US, over time. This can be due to the idiosyncratic volatility increase of individual stock nowadays, of which one factor is the R&D investment extent. To add some convincing power of this empirical research, I do two types of robustness checks. The first one is to exclude all the zero R&D expenses in the sample to prevent

(31)

31 them from contaminating the test results, of which the results turn out very similar to the data set that contains zero R&D expenses and indicate the contamination should be very limited. Secondly, I use R&D capital relative to total assets as the indicator of R&D intensity to be put in the independent variable position and tested. The results show similar patterns as well, except for that the results are much more significant for the crisis period in the longer lagged models. For one-year lag and one-and-half year lag, the R&D intensity also presents a significant positive effect on firm’s stock returns. This phenomenon sends the message that for recent years, R&D capital relative to total assets might be a better proxy for R&D intensity, since business structure nowadays, for many companies, has changed to a more intangible-inclined manner. This paper provides some useful information on the relationship between R&D intensity and stock returns by empirical approaches. R&D investment, as one of the idiosyncratic features of firms, has been studied since 1980’s and still an interesting area to research. It has taken a bigger role in multiple industries since the technology boom started in the last decade. Stock return, as the most important and focused part for the investors when it comes to the security market, how it is influenced by R&D investment as a firm-specific factor, especially in this time that numerous high-tech companies are entering the stock market and having eye-catching performances, must be very attentive to them.

(32)

32

Appendix

Table 4.2.2 One-year lagged cross-sectional regression of returns on R&D relative to market value of equity

(1) (2) (3) (4) (5) (6) (7)

VARIABLES RET RET RET RET RET RET RET

Panel A: Entire sample (1989Q1 – 2015Q4)

RDCME 0.0196*** 0.00869*** 0.00534** 0.00552** (7.199) (3.491) (2.148) (2.194) LNME -0.00339*** -0.00298*** -0.00238*** -0.00248*** (-24.98) (-18.61) (-16.24) (-16.51) LNBTM 0.00405*** 0.00205*** 0.00225*** (20.30) (8.953) (9.544) Beta 0.000679** 0.00201*** (2.273) (6.662) Constant 0.00920*** 0.0279*** 0.0141*** 0.00987*** 0.0253*** 0.0242*** 0.0230*** (50.77) (40.02) (79.40) (35.81) (27.09) (28.16) (25.69) Observations 62,535 62,535 62,535 62,535 62,535 62,535 62,535 R-squared 0.003 0.006 0.005 0.000 0.007 0.008 0.008 Number of gvkey 1,958 1,958 1,958 1,958 1,958 1,958 1,958

Panel B: Crisis period (2007Q1-2015Q4)

RDCME 0.0384*** 0.00865** 0.00372 0.00339 (10.53) (2.249) (0.920) (0.826) LNME -0.00985*** -0.00912*** -0.00546*** -0.00568*** (-19.79) (-15.70) (-8.117) (-8.447) LNBTM 0.0115*** 0.00722*** 0.00735*** (17.20) (8.452) (8.546) Beta 0.00121 0.00309*** (1.161) (3.227) Constant 0.00328*** 0.0638*** 0.0155*** 0.00481*** 0.0589*** 0.0437*** 0.0420*** (12.49) (21.86) (28.23) (4.491) (16.54) (11.74) (11.16) Observations 17,957 17,957 17,957 17,957 17,957 17,957 17,957 R-squared 0.007 0.019 0.020 0.000 0.019 0.023 0.024 Number of gvkey 892 892 892 892 892 892 892

Robust t-statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1

(33)

33

Table 4.2.3 One-and-half year lagged cross-sectional regression of returns on R&D relative to market value of equity

(1) (2) (3) (4) (5) (6) (7)

VARIABLES RET RET RET RET RET RET RET

Panel A: Entire sample (1989Q1 – 2015Q4)

RDCME 0.0115*** 0.00487*** 0.00351** 0.00360** (6.913) (3.300) (2.284) (2.330) LNME -0.00194*** -0.00172*** -0.00150*** -0.00154*** (-18.21) (-14.45) (-12.91) (-13.24) LNBTM 0.00203*** 0.000767*** 0.000852*** (10.64) (3.396) (3.732) Beta 0.000166 0.000864*** (0.817) (4.312) Constant 0.00979*** 0.0205*** 0.0123*** 0.0104*** 0.0191*** 0.0187*** 0.0182*** (90.60) (37.39) (72.32) (55.12) (28.74) (29.72) (27.99) Observations 59,145 59,145 59,145 59,145 59,145 59,145 59,145 R-squared 0.001 0.002 0.001 0.000 0.002 0.002 0.002 Number of gvkey 1,864 1,864 1,864 1,864 1,864 1,864 1,864

Panel B: Crisis period (2007Q1-2015Q4)

RDCME 0.0305*** 0.00404 -0.00110 -0.00125 (7.548) (0.966) (-0.262) (-0.297) LNME -0.00767*** -0.00734*** -0.00381*** -0.00392*** (-16.69) (-13.16) (-6.114) (-6.291) LNBTM 0.00957*** 0.00683*** 0.00689*** (16.01) (8.767) (8.824) Beta 0.000394 0.00173*** (0.644) (3.180) Constant 0.00596*** 0.0532*** 0.0159*** 0.00771*** 0.0510*** 0.0361*** 0.0351*** (20.94) (19.69) (32.68) (12.30) (14.76) (10.24) (9.884) Observations 16,299 16,299 16,299 16,299 16,299 16,299 16,299 R-squared 0.004 0.010 0.013 0.000 0.010 0.014 0.014 Number of gvkey 825 825 825 825 825 825 825

Robust t-statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1

(34)

34

Table 5.1 Cross-sectional regression of returns on R&D relative to market value of equity, excluding zero R&D expenditures.

This table presents the comparison of cross-sectional regressions of the three time lags. The data that contain zero R&D expenditures are excluded. Column 1 to Column 3 present half-year lag to one-and-half-year lag; Panel A presents the entire time period and Panel B presents the crisis period.

(1) (2) (3)

VARIABLES RET RET RET

Panel A: Entire sample (1989Q1 – 2015Q4)

RDCME 0.0105*** 0.00403* 0.00292* (6.249) (1.676) (1.807) LNME -0.00307*** -0.00255*** -0.00146*** (-16.67) (-14.86) (-11.29) LNBTM 0.00283*** 0.00256*** 0.000900*** (10.87) (9.664) (3.507) Beta 0.00272*** 0.00193*** 0.000499** (11.61) (5.717) (2.207) Constant 0.0251*** 0.0240*** 0.0183*** (25.53) (22.66) (23.97) Observations 54,323 51,469 48,712 R-squared 0.016 0.009 0.002 Number of gvkey 1,653 1,591 1,518

Panel B: Crisis period (2007Q1-2015Q4)

RDCME 0.0214*** 0.00326 -0.00140 (5.325) (0.926) (-0.336) LNME -0.00807*** -0.00523*** -0.00321*** (-9.448) (-7.277) (-4.566) LNBTM 0.00937*** 0.00771*** 0.00692*** (9.038) (8.120) (7.846) Beta 0.00551*** 0.00201* 0.000772 (8.261) (1.925) (1.304) Constant 0.0516*** 0.0408*** 0.0321*** (11.00) (10.35) (8.025) Observations 16,821 15,344 13,906 R-squared 0.064 0.024 0.012

Robust t-statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1

(35)

35

Table 5.2 Cross-sectional regression of returns on R&D relative to total assets

This table presents the comparison of cross-sectional regressions of the three time lags. Column 1 to Column 3 present half-year lag to one-and-half-year lag; Panel A presents the entire time period and Panel B presents the crisis period.

(1) (2) (3)

VARIABLES RET RET RET

Panel A: Entire sample (1989Q1 – 2015Q4)

RDCTA 0.00727*** 0.00282** 0.00518 (4.737) (2.213) (1.603) LNME -0.00313*** -0.00256*** -0.00149*** (-18.99) (-18.20) (-10.45) LNBTM 0.00351*** 0.00258*** 0.00117*** (15.72) (11.84) (5.098) Beta 0.00244*** 0.00202*** 0.000895*** (11.64) (6.686) (4.439) Constant 0.0263*** 0.0239*** 0.0180*** (32.26) (32.85) (20.34) Observations 66,049 62,535 59,145 R-squared 0.014 0.008 0.002 Number of gvkey 2,029 1,958 1,864

Panel B: Crisis period (2007Q1-2015Q4)

RDCTA 0.0324*** 0.0214*** 0.0151*** (5.150) (3.392) (2.603) LNME -0.00835*** -0.00485*** -0.00310*** (-9.341) (-6.677) (-4.797) LNBTM 0.0111*** 0.00847*** 0.00758*** (10.51) (9.305) (9.420) Beta 0.00525*** 0.00314*** 0.00178*** (8.674) (3.282) (3.272) Constant 0.0537*** 0.0362*** 0.0292*** (11.02) (8.801) (8.080) Observations 19,664 17,957 16,299 R-squared 0.060 0.025 0.015 Number of gvkey 941 892 825

Robust t-statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1

(36)

36

Reference

Bartram, S. M., Brown, G., & Stulz, R. M. (2012). Why are US stocks more volatile?. The Journal of

Finance, 67(4), 1329-1370.

Bens, D. A., Hanna, J. D., & Zhang, X. F. (2003). Research and development, risk, and stock

returns. The University of Chicago: Working paper.

Campbell, J. Y., Lettau, M., Malkiel, B. G., & Xu, Y. (2001). Have individual stocks become more volatile? An empirical exploration of idiosyncratic risk.The Journal of Finance, 56(1), 1-43.

Chauvin, K. W., & Hirschey, M. (1993). Advertising, R&D expenditures and the market value of the firm. Financial management, 128-140.

Chan, L. K., Lakonishok, J., & Sougiannis, T. (2001). The stock market valuation of research and development expenditures. The Journal of Finance,56(6), 2431-2456.

Chan, S. H., Martin, J. D., & Kensinger, J. W. (1990). Corporate research and development expenditures and share value. Journal of Financial Economics,26(2), 255-276.

Daniel, K., Hirshleifer, D., & Subrahmanyam, A. (1998). Investor psychology and security market

under-and overreactions. the Journal of Finance, 53(6), 1839-1885.

Doukas, J., & Switzer, L. (1992). The stock market's valuation of R&D spending and market concentration. Journal of Economics and Business,44(2), 95-114.

Eberhart, A. C., Maxwell, W. F., & Siddique, A. R. (2004). An examination of long-term abnormal stock returns and operating performance following R&D increases. The Journal of Finance, 59(2), 623-650.

Fama, E. F., & French, K. R. (1992). The cross-section of expected stock returns. the Journal of

Finance, 47(2), 427-465.

Fama, E. F., & MacBeth, J. D. (1973). Risk, return, and equilibrium: Empirical tests. The journal of

political economy, 607-636.Hall, B. H. (1993). The stock market's valuation of R&D investment

(37)

37

Hirshleifer, D., Hsu, P. H., & Li, D. (2013). Innovative efficiency and stock returns. Journal of

Financial Economics, 107(3), 632-654.

Hou, K., Hsu, P. H., Watanabe, A., & Xu, Y. (2016). Corporate R&D and Stock Returns: International Evidence.

Hud, M., & Hussinger, K. (2015). The impact of R&D subsidies during the crisis. Research

Policy, 44(10), 1844-1855.

Jensen, M. C. (2005). Agency costs of overvalued equity. Financial management, 34(1), 5-19.

Lev, B., & Sougiannis, T. (1996). The capitalization, amortization, and value-relevance of R&D. Journal of accounting and economics, 21(1), 107-138.

Li, D. (2011). Financial constraints, R&D investment, and stock returns.Review of Financial

Studies, 24(9), 2974-3007.

Lin, X. (2012). Endogenous technological progress and the cross-section of stock returns. Journal

of Financial Economics, 103(2), 411-427.

Lome, O., Heggeseth, A. G., & Moen, Ø. (2016). The effect of R&D on performance: Do R&D-intensive firms handle a financial crisis better?. The Journal of High Technology Management

Research.

Miller, M., Weller, P., & Zhang, L. (2002). Moral Hazard and the US Stock Market: Analysing theGreenspan Put'. The Economic Journal, 112(478), C171-C186.

Oswald, D. R., & Zarowin, P. (2007). Capitalization of R&D and the informativeness of stock prices. European Accounting Review, 16(4), 703-726.

Xu, M., & Zhang, C. (2004). The explanatory power of R&D for the cross-section of stock returns: Japan 1985–2000. Pacific-Basin Finance Journal,12(3), 245-269.

Referenties

GERELATEERDE DOCUMENTEN

These examples show that the use of big data analytics may moderate existing R&amp;D resources and increase performance in the innovation process in a way that marginal returns to

The cross-sectional regressions confirm a positive competition-return relation for R&amp;D intensive firms, but as the general competition-return relation is negative firms

Under the cointegration frame work, in the Netherlands, stock prices in long run is an effective tool to hedge against the inflation even though in the short run it does not show

In Section 5 the results for the regressions run on the relationship between the conditioning variable and the business cycle, as well as those for the

This relation between R&amp;D and productivity in the European banking sector confirms the literature written on the positive relationship between innovation and sales and

•n Kandidaat moet vir minstens ses en hoogstens sewe vakke inskryf en die vakke moet uit dieselfde lys gekies word as vir die universiteitstoelating, soos

An alarming finding from our study is that a large proportion of COVID- 19 trials test the same treatments or drugs, creating a thicket of redundant, uncoordinated, and

the state, above all the culturalisation of citizenship, with a particular focus on educational policies and practices. The interest in this specific subject originated from