Returns on research and development investments : which sector benefits the most and when?

Returns on Research and Development Investments:

which sector benefits the most and when?

Natasja van der Brug (10650830) University of Amsterdam

Economics and Business Bachelor Thesis Specialization: Finance and Organization

June 2016

Abstract

This paper focuses on investments in Research and Development (R&D) and their influence on firm performance. Specifically, it focuses on the time frame in which these investments pay off, differences in returns on investments between different industry sectors and differences between the period before the financial crisis and the period since the beginning of the crisis. The study is based on annual data on 6,924 companies from 2001-2015 from the Compustat database. The effects are estimated by means of a panel regression model. First of all, the results suggest that investments in R&D are overall significant positively correlated with firm performance one year later. The relationship with two-, three- or four year lags are found to be unstable. Therefore, this paper focuses on the first year lag. Second, the positive effect of R&D-investments turned negative after the financial crisis started in 2007. Third, most sectors experienced a negative switch. However, sectors IT, Materials, Industrials and Telecommunication Services experienced a positive switch. This result was most striking for the IT sector, and it is contrary to theoretical expectations. For the Materials, Industrials and Telecommunication Services, on the other hand, this positive change might be the result of investments of local economies in the period after the crisis.

Keywords: R&D investment, employment growth, financial crisis of 2007, lags, industry sectors

Statement of Originality

This document is written by Student Natasja van der Brug 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.

1. Introduction

The Research and Development (R&D) department of companies specialize in the innovation and improvement of their products and/or services. As Hall et al. (2009) already stated, returns on investments in R&D are important for the obvious reason that these investments are expensive. Therefore, investors in R&D and innovation would like to see a positive return. Previous research has already shown that higher R&D investments have a positive correlation with firm growth because they lead to better products and therefore higher sales. However, there are several gaps in our knowledge on the effects of R&D investments on the performance of firms, which this thesis addresses.

First, most existing studies on the topic estimate the effect of R&D investments on the basis of a one-year or two-year lag (Kancs& Siliverstovs, 2016). However, Ravenscraft and Scherer (1982) suggest that R&D investments take time to implement. According to them, it takes three years on average to finish an R&D investment, so that it will take at least three years before R&D-investments can be expected to affect the company’s performance. This paper will therefore also look at three- and four-year lags.

Second, previous studies have studied the impact of R&D investments on total firm performance, without distinguishing between different types of industries (Falk, 2012; Brouwer et al., 1993). However, in some sectors R&D investments may be relatively more important than in other sectors. Looking at the IT sector, a company like Apple has to keep innovating if they want to stay at the top of their industry and therefore its performance may depend more on new

innovations than the performance of companies in other industries. In other sectors (e.g. the Health Care sector), investments in innovation might be crucial as well. However, in more service oriented sectors, such as the financial sector, investments in R&D may be less important. Earlier research focused only on high-, medium- or low-tech industries (Kancs& Siliverstovs, 2016), or on research-intensive industries versus non-research-research-intensive industries (Morbey, 1988). Hence, in this study a distinction will be made between ten types of sectors, to find whether R&D investments lead to a higher overall firm performance for some sectors more than for other sectors.

Finally, previous research about the effect of R&D investments on firm performance focused solely on 2006 or earlier, while there was a worldwide financial crisis in 2007/2008 (Kancs& Siliverstovs, 2016; Falk, 2012). In the context of overall economic growth firms may find it easier to benefit from new inventions than in the context of an economic downturn, i.e. during a financial crisis. The reason for this seems obvious: economic growth implies higher productivity and higher sales. Let us return to our previous example of Apple. Suppose R&D-investments lead to the

gadget and return on investments may be very good. However, in the middle of an economic crisis, these people may decide to use their money for different purposes , so that the R&D-investment may not pay off. Hence, in the context of overall economic growth it is easier to sell new inventions and benefit from these. Therefore, this study compares the effects of R&D investment on the performance of firms before the financial crisis that started in 2007 with those effects after 2007.

To summarize, this paper will focus on how R&D investments influence firm performance. First, it focuses on how long it takes for R&D investments to pay off. Therefore, instead of solely focusing on a one or two year time lag, this study will focus on three and four year time lags as well. Second, the study will compare between the period before the financial crisis (2001-2007) and the period since the beginning of the crisis (2008-2015). Finally, it will focus on the effects of R&D investments on firm performance for 10 different industry sectors.

To study the effects of R&D investments on firm performance, data are obtained from the Compustat database. It contains the relevant annual data of 6,924 global companies for a 15-year period, from 2001-2015. The relationships between R&D investments and firm performance are estimated by means of panel regression models, in three steps. First, the overall effect of R&D investments based on different lags is tested (a one-, two-, three- and four-year lag). Then a

distinction is made between the period before the financial crisis (between 2001 and 2007) and the period after the crisis kicked in (2008 to 2015). Finally in the last step, a distinction is made between ten different industry sectors.

The main results show that investments in R&D are overall significant positively correlated with firm performance at the 1% significance level. However, taking a two-, three- or four-year lag showed generally minimal significant results relative to the results found when a one-year lag was used. Secondly, the results show a positive correlation between R&D investments and firm performance from 2001-2007 which switched to negative in the period after the financial crisis of 2007. In contrast to this, four sectors experienced a positive change in correlation from 2001-2007 to 2008-2015. Specifically, these sectors were the IT-, Materials-, Industrials- and

Telecommunication Services sector.

The paper is structured as follows. First, the relevant literature and a number of hypotheses to be tested in this study are discussed. Then follows an explanation about the data used in this research and the research design. This is followed by a section in which the main results are presented. Finally in the concluding section, the implications of those results are discussed.

2. Literature review and hypotheses

As earlier mentioned, R&D investments are costly for firms. Therefore, it is important for firms that they receive a positive return on these investments. Several studies have already been done to find a correlation between the R&D intensity and the effect of these investments on firm performance. Some previous studies have found a positive significant relationship (Falk, 2012; Kancs& Siliverstovs, 2016) while others found a significant negative correlation (Brouwer et al., 1993). The difference in these previous findings may, as argued in the introduction, depend on (macro-) economic factors, such as economic crises.

In addition, the difference may also be caused by the fact that different countries and time frames are being studied. Ravenscraft and Scherer (1982) suggested that it takes on average three years to finish an R&D investment. Their results show strong evidence that a three-year lag is too short to find firm performance returns from R&D investments. Moreover, they found that the lag structure for returns to R&D is roughly bell-shaped, with a mean lag from four to six years. However, Falk (2012) used up to four-year lags in his research but noticed that his results hardly changed when different lags were included.

Falk (2012) also shows, on the basis of data on Austrian firms in the period between 1996 and 2006, that initial R&D intensity has a positive and significant impact on firm performance in the two subsequent years. These findings are confirmed by Kancs and Siliverstovs (2016), on the basis of a combination of two data sets, the EU industrial R&D investment Scoreboard and the Orbis

worldwide company information (BvDEP) in a relatively very short time frame of 2006-2007. Due to these results, first two hypotheses are expressed as follows:

Hypothesis 1a: R&D investments and firm performance are in general positively correlated. Hypothesis 1b: When including more lags this positive correlation will increase due to the increase in the impact of R&D investments on firm performance over time.

Falk’s (2012) and Kancs and Siliverstovs’ (2016) studies, which both showed a positive effect of R&D investments, are both conducted in the context of an economic boom. Brouwer et al. (1993), on the other hand, found a negative correlation between innovation investments and firm

performance. Even though the effect found was small, it was still negatively significant. Brouwer et al. used data of 859 Dutch firms from 1983-1988. However, in this period the Netherlands were still recovering from the worldwide economic debt crisis, which lasted from 1980 to 1983 (Krueger, 1987). Consequently, this was a period of mass unemployment and slow economic growth in the Netherlands (Scharpf& Schmidt, 2000). Therefore, the effect of R&D investments may depend on

macro-economic conditions. Investments in R&D may pay off well in times of economic booms, but may not benefit the company in times of crises. Hence, the second hypothesis is:

Hypothesis 2: R&D investments have a positive effect on firm performance during times of economic growth. After an economic crisis, this positive effect switches to negative due to the recovery period after an economic downfall.

Morbey (1988) studied the relationship between R&D investments and firm performance for the period 1976-1985, where he analyzed 31 different industries. His results suggest that, in general, there is a strong correlation between R&D intensity and growth in sales in the subsequent years. Moreover, he found that companies that invested >3% of their revenue in R&D had a relatively large chance (80%) of long-term growth. However, his results found for the different industry-levels where not significant.

Furthermore, Morbey focused on the correlation between R&D and sales growth for

research-intensive industries (similar to Health Care sector in this study) in particular. Morbey found a strongly positive correlation for research- intensive industries, and no significant correlation for non-research intensive industries. In addition, this relationship for research-intensive companies strengthens when the time period lengthens.

Moreover, Kancs and Siliverstovs (2016) also found inter-sectoral differences. They focused on high-tech sectors instead of research-intensive industries, which are similar to the IT sector in this study. Kancs and Siliverstovs found that high-tech sectors generally invest more in R&D. Additionally, these investments resulted in higher productivity gains, and thus had a positive correlation with firm performance. Therefore, expected is that the IT sector and Health Care sector have the highest positive correlation on their R&D investments in relation to the other sectors. The third hypothesis is:

Hypothesis 3: R&D investments in the IT-sector and the Health Care sector experience the highest return in payoff in terms of firm performance in relation to the other eight industry sectors.

Moreover, as argued in the introduction, in times of economic crisis people may decide to use their money for different purposes than buying new products that are not essential to them (e.g. gadgets). As a result, the crisis will affect different industries in different ways. The Health Care sector may be relatively unaffected by the crisis, as new medicines will often be paid by insurance

companies. However, the IT-sector seems more vulnerable, because consumers may not buy new gadgets in times of a crisis. Therefore, investing in R&D might not lead to higher sales and higher productivity in this sector. Hence, the fourth hypothesis is:

Hypothesis 4: The IT-sector will experience the biggest drop in payoff when the financial crisis of 2007 kicked in compared to the other sectors.

In this study one control variable will be added to the models, which is a dummy for size of the firm. Cohen et al. (1987) found a relatively small insignificant effect of overall firm size on R&D intensity, when they corrected for fixed industry effects. However, when did not correct for these effects and solely regressed R&D intensity on firm size, they found a positive correlation between firm size and R&D intensity. This result was confirmed by Archibugi et al. (2013), who found that a small group of firms significantly increased their R&D investments during the crisis of 2007. These where either dynamic firms or relatively new, small firms.

However, Brouwer et al. (1993) found a negative relationship in their study. Firm size, which they measured in terms of employees, had a negative impact on annual firm growth, and thus on firm performance. From this result they concluded that smaller firms in their dataset had a relatively better development of employment than larger ones. And because firm size and firm age are

correlated, they continue, their results can also imply that younger firms have a relatively better development than older firms. Hence, just as Archibugi et al. found, Brouwer et al.’s result show that new firms experience relatively better returns than older firms to their R&D investments. This confirmed that it is important to control for firm size in this study.

Different studies have used different measures to indicate firm performance. Some studies measure a firm’s performance by growth in the number of employees, while other studies focus on sales growth. For instance, Brouwer et al. (1993) used employment growth as a measure for firm performance, while Morbey (1988) used growth in sales. However, there are no valid theoretical reasons why one measure would be better than the other. To my knowledge, Falk’s research (2012) is the only study in which both measures of firm performance are used (sales growth and

employment growth). He shows that the results are relatively very similar for the two measures, both in terms of effect size as well as in terms of significance. Due to limitations in the data, this study will focus on just one measure of firm performance: employment growth.

3. Data

The dataset used in this study is obtained from the Compustat Global database from the Wharton Research Data Service. The data is normalized to provide comparability across a wide variety of global accounting standards and practices. The dataset contains annual data of 6,924 global

companies from 2001-2015. The original dataset contains roughly 62,000 observations, where some observations are missing for some years. In the analyses all observations without information on the value of Total Revenue and R&D Expenses were dropped. Secondly, all observations without

information of the number of employees and of industry sectors were excluded as well. Thirdly, outliers (observations) of the variable ‘R&D ratio’ where dropped, which will be discussed below. Eventually leaving us with a total of 58,195 observations.

The database did not collect data for all the companies for all the years from 2001-2015. But as we can see in Table 1, the frequency per year is stable especially from 2006 and on. So expected is that differences between frequency of observations will not form a problem in this study.

Table 1

Observation frequency per year, from 2001-2015

Date (year) Frequency Percent Cum.

2001 1,779 3.06 3.06 2002 3,225 5.54 8.6 2003 3,408 5.86 14.45 2004 3,597 6.18 20.64 2005 3,157 5.42 26.06 2006 4,061 6.98 33.04 2007 4,208 7.23 40.27 2008 4,180 7.18 47.45 2009 4,185 7.19 54.64 2010 4,263 7.33 61.97 2011 4,358 7.49 69.46 2012 4,312 7.41 76.87 2013 4,376 7.52 84.39 2014 4,461 7.67 92.05 2015 4,625 7.95 100.00 Total 58,195 100.00

The dataset distinguishes ten different sectors, which are the following: Energy- , Materials-, Industrials-, Consumer Discretionary-, Consumer Staples- , Health Care-, Financials-, Information Technology-, Telecommunication Service- and the Utilities sector. The sectors are relatively broad and therefore contain different industry groups. They are summarized in Table 2. This study will focus on industry sectors only due to the number of industry groups.

Table 2

The ten types of industry sectors and their frequency qua observations (companies*year)

Industry sector Industry groups Frequency Percentage

Energy Energy 1,169 2.01

Materials Materials 8,163 14.03

Industrials Capital Goods, Commercial Services, Transportation, Airlines, Marine

14,932 25.66

Consumer Discretionary

Automobiles & Components, Consumer Durables, Consumer Services, Media, Retailing

8,874 15.25

Consumer Staples Food & Staples Retailing, Food Beverage & Tobacco, Household & Personal products

4,516 7.76

Health Care Health Care Equipment & Services, Pharmaceuticals, Biotechnology & Life Sciences

4,919 8.45

Financials Banks, Diversified Financials, Insurance, Real Estate 287 0.49 Information

Technology

Software & Services, Technology Hardware & Equipment, Semiconductors

13,712 23.56

Telecommunication Services

Telecommunication Services 572 0.98

Utilities Utilities (Electric, Gas, Water) 1,051 1.81

Total 58,195 100

Notes: information source for industry groups is obtained from WRDS

From previous studies it is clear that firm performance is measured in two different ways: employment growth and sales growth. Because of data limitations, in this study employment growth is used as measure for firm performance. Falk’s (2012) study strongly suggests that this does not threaten the validity of the results. Employment growth is expressed in terms of a change variable. The variable ‘Employees’ is used to calculate this growth. It is obtained from the database, and defined as:

Employees = this item represents the number of people employed by the company and its consolidated subsidiaries in thousands.

Because the dataset also contained ‘Employee’ values of 0.001, a new variable ‘Number of

employees’ is generated which is the variable ‘Employees’ multiplied by 1000. The latter is used to compute the employment growth.

Additionally,

𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝑔𝑔𝑔𝑔𝐸𝐸𝑔𝑔𝐸𝐸ℎ =

.

To show whether the variable is normally distributed a histogram is made, with employment

growth on the x-axis and the density on the y-axis.

From figure 1 it is clear that the employment growth variable is normally distributed. In addition, all the variables are summarized in Figure 4.

Next, as main independent variable the R&D investments are used. However, the dataset contains both small and very large companies, which result in relatively big differences in Total Revenue and R&D Expenses between firms. To control for these big differences an R&D ratio per observation is used. The variables used to calculate this ratio were provided by the database, and are defined as:

- Revenue – Total = represents the Gross Income received from all divisions of the company. It includes revenue from Banking Operations, Broker/Dealer Operations, Insurance

Operations, Nonfinancial Services Operations, Other Financial Services Operations, Real Estate Operations and Sundry Revenue. It excludes Exceptional Income, Foreign Exchange Income and Provisions Transferred from reserves.

- Research and Development expense = represents all costs incurred during the year that relate to the development of new products or services. It includes Software Expenses and

Figure 1

Frequency of employment growth, obs. emplgrowth≤10

Notes: [a] employment growth has a few outliers with a value >10

0 1 2 3 4 D en s it y 0 5 10 emplgrowth

Amortization of Software Costs. But excludes Customer or Government-Sponsored Research and Development, Extractive Industry Activities, Engineering Expense, Inventory Royalties and Market Research and Testing.

Additionally, the variable is computed by 𝑅𝑅&𝐷𝐷𝑔𝑔𝐷𝐷𝐸𝐸𝐷𝐷𝐸𝐸 =𝑅𝑅𝑛𝑛𝑒𝑒𝑛𝑛𝑅𝑅𝑛𝑛𝑅𝑅ℎ 𝑅𝑅𝑛𝑛𝑎𝑎 𝐷𝐷𝑛𝑛𝐷𝐷𝑛𝑛𝑒𝑒𝑜𝑜𝑒𝑒𝑛𝑛𝑛𝑛𝑛𝑛𝐷𝐷 𝑛𝑛𝑒𝑒𝑒𝑒𝑛𝑛𝑛𝑛𝑒𝑒𝑛𝑛𝑒𝑒_{𝑇𝑇𝑜𝑜𝐷𝐷𝑅𝑅𝑒𝑒 𝑅𝑅𝑛𝑛𝐷𝐷𝑛𝑛𝑛𝑛𝑛𝑛𝑛𝑛} . This ratio should have a value between 0 and 1 (0≤R&D ratio≤1). Outliers contain small firms (with <1000 employees) in the dataset, which have relatively high R&D expenses but almost no revenue or small negative revenue. The motive of these high expenses for small firms might be to rapidly stimulate firm growth. However, this is not representative for this study. Because these outliers influence the regression all observations with a R&D ratio>1 and R&D ratio<0 are dropped from the dataset. As shown in Figure 2, the density decreases as the R&D ratio increases.

Furthermore, it is important to take the size of the firm into account. Not solely because of the relatively large differences between R&D Expenses and Total Revenue, but also because of earlier research about the relationship between firm size and R&D investments. As argued in the literature section, Cohen et al. (1987) found a the positive correlation between firm size and R&D intensity. However, Brouwer et al. (1993) found the exact opposite in their study. Their results showed a negative correlation between firm size and annual firm growth. From this result they concluded that smaller firms in their dataset had a relatively better development of employment than larger ones. Therefore, a company size dummy is made to correct for the size differences between companies. This dummy has a value of 0 if the company has less than/or 10,000 employees, and a value of 1 if it has more than 10,000 employees.

Another dummy variable made is the ‘aftercrisis’ dummy, to measure the difference in the

Figure 2

relation between R&D investments and firm performance from 2001-2007 to the period from 2008-2015. Therefore, the dummy has a value of 0 for the period before the crisis (≤2007) and 1 for the period after the crisis (≥2008).

The last variable is the return on assets (ROA). The reason for this is that the number of employees may grow due to an increase in the profit. This increase in profit is captured by the return on assets.

On the other hand, ROA is frequently used as firm performance measure (Dess& Robinson, 1984). However, as we see in Figure 3, the return on assets and employment growth have almost no correlation. As a result, we don’t have to omit this variable due to multicollinearity.

ROA Employment growth

ROA 1.000

Employment growth -0.001 1.000

Figure 3

Correlation of variable Return on Assets and variable Employment Ratio

Observations Mean Standard Deviation Min Max

Employment growth 45,216 0.057 0.82 -1 138.38 R&D ratio 58,195 0.044 0.094 0 1 Sizecomp 58,195 0.1437 0.351 0 1 Aftercrisis 58,195 0.597 0.490 0 1 ROA 56,290 0.013 1.024 -18.607 234.147 Figure 4

4. Methodology and predictions

The analyses in this study are conducted by means of panel regression models, where employment growth is the dependent variable and R&D ratio the main independent variable. This study does not only focus on the impact of the R&D ratio, but particularly on differences before and after the financial crisis and differences between ten industries. Therefore a three step approach is used.

The first step is meant to confirm that there is indeed a correlation between R&D ratio and employment growth. More importantly, however, these analyses focus on how many years it takes for firms to get returns on R&D investments. Therefore, the analysis starts with a simple regression model in which the following regression model is used:

𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝑔𝑔𝑔𝑔𝐸𝐸𝑔𝑔𝐸𝐸ℎ = 𝛼𝛼 + 𝛽𝛽 ∗ 𝐸𝐸(𝐸𝐸𝐸𝐸𝐷𝐷𝑔𝑔). 𝑅𝑅&𝐷𝐷𝑔𝑔𝐷𝐷𝐸𝐸𝐷𝐷𝐸𝐸 + 𝜉𝜉 , 𝑓𝑓𝐸𝐸

Where the variable l.(year) is defined as the lag per year {l1.-l4.}, and fe stands for fixed effects. Panel regression with correction for fixed effects indicates that you only analyze the impact of variables that vary over time and do not take differences between companies into account.

The reason for estimating a ‘fixed effects’ model is that it provides the strongest evidence of effects, as differences between companies (that may be due to other factors) are controlled for. Moreover, there may be R&D spillovers between companies that influence inter-sectoral differences. Meaning that high R&D investments in one firm, positively influence the firm

performance of other firms. Hall et al. (2009) state that there are two kinds of spillovers between firms: rent spillovers and knowledge spillovers. These R&D spillovers can be the result of R&D investment by four different institutions, they continue. These four institutions are: companies in the same sector, companies in other sectors, public research laboratories and universities, and companies, laboratories and governments in other countries.

However, by estimating the models by means of fixed effects, possible important differences between companies are overlooked. Companies that invest much in R&D across these 15 years may do better on average than companies that invest less in R&D. These differences between companies therefore give important information on consequences of R&D investments. As a result, in this study each regression is performed with both correcting for fixed effects and without.

The second step of the approach is a test whether there’s a difference between R&D investments and firm performance in the period before and the period after the financial crisis. To measure this, the dummy variable ‘aftercrisis’ is added to the regression. It has a value of 0 for the period before the crisis (≤2007) and 1 for the period after the crisis (≥2008). In the second step, the following regression model is used:

𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝑔𝑔𝑔𝑔𝐸𝐸𝑔𝑔𝐸𝐸ℎ = 𝛼𝛼 + 𝛽𝛽 ∗ 𝐸𝐸(𝐸𝐸𝐸𝐸𝐷𝐷𝑔𝑔). 𝑐𝑐. 𝑅𝑅&𝐷𝐷𝑔𝑔𝐷𝐷𝐸𝐸𝐷𝐷𝐸𝐸##𝐷𝐷. 𝐷𝐷𝑓𝑓𝐸𝐸𝐸𝐸𝑔𝑔𝑐𝑐𝑔𝑔𝐷𝐷𝑎𝑎𝐷𝐷𝑎𝑎 + 𝜉𝜉 , 𝑓𝑓𝐸𝐸

The variable 𝐸𝐸(𝐸𝐸𝐸𝐸𝐷𝐷𝑔𝑔). 𝑐𝑐. 𝑅𝑅&𝐷𝐷𝑔𝑔𝐷𝐷𝐸𝐸𝐷𝐷𝐸𝐸##𝐷𝐷. 𝐷𝐷𝑓𝑓𝐸𝐸𝐸𝐸𝑔𝑔𝑐𝑐𝑔𝑔𝐷𝐷𝑎𝑎𝐷𝐷𝑎𝑎 is defined as the lag per year {l1.-l4.} of the interaction between R&D ratio and the ‘aftercrisis’ dummy. ‘c’. denotes that the variable is a continuous distributed variable, and ‘i.’ denotes that the variable is numerically distributed.

Thirdly, the regression is extended with the industry sectors to find if there is a correlation between R&D investments and firm performance for all the ten sectors, in the period before and after the crisis. The following regression model is used in the final step of the approach:

𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸𝐸 𝑔𝑔𝑔𝑔𝐸𝐸𝑔𝑔𝐸𝐸ℎ

= 𝛼𝛼 + 𝛽𝛽 ∗ 𝐸𝐸(𝐸𝐸𝐸𝐸𝐷𝐷𝑔𝑔). 𝑐𝑐. 𝑅𝑅&𝐷𝐷𝑔𝑔𝐷𝐷𝐸𝐸𝐷𝐷𝐸𝐸##𝐷𝐷. 𝐷𝐷𝑓𝑓𝐸𝐸𝐸𝐸𝑔𝑔𝑐𝑐𝑔𝑔𝐷𝐷𝑎𝑎𝐷𝐷𝑎𝑎##𝐷𝐷. 𝐷𝐷𝐸𝐸𝑖𝑖𝑖𝑖𝑎𝑎𝐸𝐸𝑔𝑔𝐸𝐸 + 𝑎𝑎𝐷𝐷𝑠𝑠𝐸𝐸𝑅𝑅𝑜𝑜𝑛𝑛𝑒𝑒 + 𝐸𝐸(𝐸𝐸𝐸𝐸𝐷𝐷𝑔𝑔). 𝑅𝑅𝑅𝑅𝑅𝑅 + 𝜉𝜉 , 𝑓𝑓𝐸𝐸

The variable 𝐸𝐸(𝐸𝐸𝐸𝐸𝐷𝐷𝑔𝑔). 𝑐𝑐. 𝑅𝑅&𝐷𝐷𝑔𝑔𝐷𝐷𝐸𝐸𝐷𝐷𝐸𝐸##𝐷𝐷. 𝐷𝐷𝑓𝑓𝐸𝐸𝐸𝐸𝑔𝑔𝑐𝑐𝑔𝑔𝐷𝐷𝑎𝑎𝐷𝐷𝑎𝑎##𝐷𝐷. 𝐷𝐷𝐸𝐸𝑖𝑖𝑖𝑖𝑎𝑎𝐸𝐸𝑔𝑔𝐸𝐸 is defined as the lag per year {l1.-l4.} of the interaction between R&D ratio, the industry dummies and the ‘aftercrisis’ dummy.

As we can see, a few control variables are added due to omitted variable bias. The first control variables added are nine sector dummies (one is excluded due to dummy bias). However, due to multicollinearity these control variable dummies will be omitted from the regression when we correct for fixed effects.

The second added control variable is a company size dummy, to correct for the size

differences between small and big companies. This dummy has a value of 0 if the company has less than/or 10,000 employees, and a value of 1 if it has more than 10,000 employees.

The last variable added to the regression is the return on assets. As argued in the data section, this variable doesn’t have to be omitted due to multicollinearity. Moreover, adding the return on assets to the regression makes sense for the simple reason that an increase in profit per sector (thus an increase in ROA for a company) might influence the number of employees positively. Forcing the growth in employment to increase as well.

5. Results

This section will present the results of the analyses, following the order of the four different hypotheses.

The first step of the analysis focuses on hypotheses 1a and 1b, which test when R&D investments have a significant effect on firm performance. Hence, the only independent variable used is the R&D ratio, but different lags are used. This simple regression is done eight times: with correcting for fixed effects and without, and for four different time lags. The results are shown in Table 3.

Table 3

Panel regression estimates of the correlation of R&D investment and employment growth Fixed effects Without fixed effects

R&D ratio Constant R&D ratio Constant Number of observation Number of companies 1 year lag 0.191* (0.116) 0.048*** (0.002) 0.218*** (0.046) 0.047*** (0.004) 45,010 6,864 2 year lag 0.286** (0.145) 0.038*** (0.007) 0.268*** (0.068) 0.044*** (0.007) 37,008 5,843 3 year lag 0.206*** (0.077) 0.033*** (0.004) 0.19*** (0.042) 0.039*** (0.005) 31,535 5,132 4 year lag -0.15** (0.072) 0.043*** (0.003) 0.14*** (0.033) 0.034*** (0.003) 27,283 4,675

* significant at the 10% level ** significant at the 5% level *** significant at the 1% level

From these results we can conclude that investments in R&D are overall significantly positively correlated with firm performance in the following three years. These results confirm the hypothesis 1A, because the results show that R&D investments and firm performance are in general positively correlated. However, with the correction for fixed effects, we find that R&D ratio and firm performance are negatively correlated at the 5% level when using a four-year lag. This strikes with the results obtained in all the other models. In the model with correcting for fixed effects, we see that R&D investments are significantly correlated with firm performance using a two- or three-year lag while this was only significant at the 10% level using a one-year lag.

Moreover, when including more lags we do not see an increasing positive correlation. The correlation is even dropping when increasing the lag after the second-year lag. Therefore, hypothesis 1B is not supported.

In the results of the analyses conducted below, very similar results are obtained using the one-year, two-year or three-year lag. Hence, in the next two steps of the analysis only the results of one-year lag analyses are discussed in this study. The results of including more lags are presented in Appendix 1 and 2.

Secondly, to test hypothesis 2, the models displayed in Table 4 test whether there is a difference in the correlation between R&D and firm performance in the time before the financial crisis of 2007 and after this crisis. To control for this a dummy is used (‘aftercrisis’) which has a value of 0 if the observation is from the period 2001-2007, and a value of 1 from the period 2008-2015. The important variable in this regression is not solely the R&D ratio, but the interaction between the R&D ratio and the aftercrisis dummy. The interaction effect tells us whether there is a difference in returns on R&D investments before the crisis and after the crisis. The results are summarized in Table 4.

Table 4

Panel regression estimates of the correlation of R&D investment and employment growth after the crisis (one-year lag)

R&D ratio Aftercrisis R&D ratio* aftercrisis Constant Number of observations Number of companies Fixed effects 1 year lag 0.243* (0.139) -0.032*** (0.010) -0.094 (0.126) 0.068*** (0.009) 45,010 6,864

Without fixed effects 1 year lag 0.35*** (0.082) -0.001 (0.009) -0.19* (0.099) 0.047*** (0.007) 45,010 6,864

* significant at the 10% level ** significant at the 5% level *** significant at the 1% level

The results in Table 4 show that the findings are quite similar in both models (with and without fixed effects). The most important difference is that when we only focus on differences over time (the fixed effects model), the coefficient of the interaction of the dummy with the R&D

investments is not significant anymore. Moreover, the coefficient of the R&D investments is only significant at the 10% level. This suggests that the fixed effects correction might be too strict. Differences between companies in their R&D investments may be as relevant as changes over time in these investments.

Looking further at the results from the regression without correcting for fixed effects gives us a significant negative correlation between R&D investments and firm performance after the crisis,

with a p-value of 0.054. This is in line with the expectation that the return on R&D will affect the firm performance negatively after the crisis. Moreover, the corresponding coefficient of the R&D ratio is positively significant. This shows that in the period before the financial crisis firms with high R&D investments were performing relatively well in the years thereafter. The coefficient shows that if the R&D ratio goes up with one percent, the firm performance will increase with 0.35 percent in the next year. After the crisis this R&D effect was just 0.159 percent, which implies a downfall of -0.191 percent. This result is perfectly in line with the expectations presented in the second hypothesis. However, if we look at Table 7 included in Appendix 1, this is the only observation where both the main effect of R&D ratio and the interaction between R&D ratio and the ‘aftercrisis’ dummy are significant. Hence, hypothesis 2 is supported, however we cannot conclude that this finding is very robust.

In the third step of the analysis, a panel regression is performed to estimate the effect of R&D investments before and after the crisis on firm performance for all ten different industry sectors. In order to reduce the risk of omitted variable bias, a few control variables are added to this model. The first control variable is a company size dummy, which is added to the regression to correct for the size differences between small and big companies. Secondly, the return on assets is added because an increase in profit per sector (thus an increase in ROA for a company) might influence the number of employees positively. Thereby, forcing the growth in employment to increase as well.

The most important variables in this regression are the interactions between the R&D ratio and the nine industry dummies. These main effects of R&D ratio estimate the returns on R&D investment before the crisis. On the other hand, the interactions between the R&D ratio, nine industry dummies and the ‘aftercrisis’ dummy are equally important. Specifically, the effect of the interaction measures the change in returns on R&D investment after the crisis (compared to the effect before). Combining these results will make it possible to explain if there is a correlation between the investments in R&D before and after the crisis with overall firm performance, for the ten different industries. They will also make it possible to show if there is a change in return in terms of firm performance of R&D investments.

The most important results of this regression are summarized in Table 5. Again, only the results of the one-year lag analyses are discussed. The results of the other analyses are included in Appendix 2. In Table 5 two kinds of interaction effects are included with the ‘company sectors’, with the dummy for the economic crisis and without. The R&D ratio coefficient gives us the effect for investing in R&D for companies in the energy sector on the firm performance, for the period before the crisis. The first interaction between the ‘aftercrisis’ dummy and the R&D ratio gives us the same

effect, but for an energy sector company in the period after the crisis. The same holds for all the other coefficients showed in Table 5, but then in line with their sector dummy.

Table 5

Panel regression estimates of the effect of R&D investment (1 year lag) on employment growth before and after the crisis for the industry sectors, for both with correction for fixed effects and without

Fixed effects Without fixed effects

2001-2007 Interactions with 2008-2015 2001-2007 Interactions with 2008-2015 R&D ratio 19.2*** (2.01) 5.87*** (1.0) Aftercrisis -0.02 (0.08) -0.09 (0.07) Aftercrisis* R&D ratio -16.69*** (2.06) -6.08*** (1.18) Industry dummy* R&D ratio Materials -17.16*** (2.26) 18.95*** (2.26) -4.97*** (1.19) 7.35*** (1.37) Industrials -20.97*** (2.09) 17.32*** (2.14) -5.71*** (1.08) 5.97*** (1.27) Consumer Discretionary -13.07*** (2.13) 11.59*** (2.19) -1.3 (1.11) 1.7 (1.29) Consumer Staples -17.85*** (2.62) 16.20*** (2.75) -5.03 (1.61) 5.19*** (1.88) Health Care -19.31*** (2.03) 16.64*** (2.08) -5.93*** (1.02) 6.13*** (1.19) Financials -18.37*** (2.96) 15.9*** (3.13) -5.71*** (1.25) 5.61*** (1.55) IT -19.23*** (2.02) 16.82*** (2.1) -5.76*** (1.01) 6.06*** (1.19) Telecomm. -19.16*** (3.43) 16.88*** (4.91) -6.86*** (1.57) 6.95*** (1.95) Utilities -18.64*** (2.26) 15.73*** (2.45) -5.14*** (1.30) 5.16*** (1.55) Size company 0.02*** (0.04) 0.02*** (0.04) -0.0003 (0.01) -0.0003 (0.01) ROA 0.06*** (0.02) 0.06*** (0.02) 0.04** (0.02) 0.04** (0.02) Constant 0.04*** (0.01) 0.04*** (0.01) 0.15*** (0.06) 0.15*** (0.06) Number of observations 43,667 43,667 43,667 43,667 Number of companies 6,644 6,644 6,644 6,644

* significant at the 10% level ** significant at the 5% level *** significant at the 1% level

To facilitate the interpretation of the results, Table 6 presents the impact of R&D

investments on firm performance per industry sector before and since the crisis started in 2007. As displayed in Table 5, when the correction for fixed effects is not included, the effect of the control variable ‘size company’ is not significant anymore. Moreover, even though both correcting for fixed effects and without correcting find significant results, the results found under the analysis with the correction gives us relatively more significant results. Hence, this analysis is summarized in Table 6.

Table 6

Summary of the effect of R&D investments before and after the crisis on employment growth for all the industry sectors (one year lag), with correcting for fixed effects

2001-2007 2008-2015 Change between periods Energy 19.2 2.49 -16.71 Materials 2.04 4.28 2.24 Industrials -1.77 -1.16 0.61 Consumer Discretionary 6.13 0.23 -5.9 Consumer Staples 1.35 0.84 -0.51 Health Care -0.11 -0.18 -0.07 Financials 0.83 0.02 -0.81 IT -0.03 0.08 0.11 Telecomm. 0.04 0.21 0.17 Utilities 0.56 -0.42 -0.98

Notes: [a] all the results displayed can be interpreted as percentages. [b] all the results displayed in the table are significant.

The results displayed in Table 6 show the return in terms of firm performance when the R&D ratio changes with one unit. In the period before the financial crisis, an change of 1 percent in the R&D ratio of an Energy company resulted in an employment growth of 19.2 percent. In this period, investments in R&D thus had a relatively big impact on firm performance in this sector. The same way of interpretation holds for all the results in Table 6.

What strikes is that both the Health Care sector and IT sector had a relatively small negative return on R&D investment before the crisis, while expected was that these sectors would have large positive returns on their investments relative to the other sectors. Hence, the results found do not support hypothesis 3.

In addition, the IT sector experienced a relatively small positive change between the two periods in relation to the other sectors. As expressed in hypothesis 4, expected was that this sector would experience the biggest negative change in terms of return on R&D investments as a results of the crisis. Hence, these results found do not support hypothesis 4 as well.

The same positive change in return on R&D investments holds for sectors Materials, Industrials and Telecommunication Services. A possible explanation for this positive change is that local economies started providing construction industry jobs in the period after the crisis (Parkinson, 2009). Even though Parkinson focuses in his paper on local economies in Great Britain, some of his statements are valid worldwide. According to him, the Materials- and Industrials sector were hit as one of the hardest during the financial crisis. Consequently, these were also the sectors in which many residents of cities were looking for work after the crisis kicked in in 2007, Parkinson states (2009). Thus even though these sectors were hit hard by the crisis, they were also the first ones to experience investments from local governments. As a results, this might explain the positive change in return of R&D investments on firm performance from the period before the financial crisis to the period afterwards.

Another thing that stands out is that companies in the Energy sector experienced a relatively high downfall in return of R&D investment, namely -16.7 percent. However, their return on

investment is still positive after the financial crisis. A possible explanation for this downfall is that this was caused by the stagnation in the construction industry. This stagnation obviously resulted in a decline in the need for energy as well. However, apparently the Energy sector was booming before the financial crisis. As a result, even though it was hit hard during and after the crisis, the return on R&D was still positive in the period form 2008-2015.

6. Conclusions

This paper focuses on the return in overall firm performance of investments in R&D. Moreover, it focuses on how long it takes for R&D investments to pay off. This is done by looking at how R&D investments influence firm performance for different industry sectors from 2001-2015. The study not only compares between different sectors, but also between the period before the financial crisis and the period since the beginning of the crisis.

The dataset used in this study is obtained from the Compustat Global database from the Wharton Research Data Service. It contains annual data of 6,924 global companies from 2001-2015, with a total of 58,195 observations. Furthermore, the data distinguishes ten different industry sectors and contains both small and large firms. To control for these relative big differences in firms, an R&D ratio and employment growth variable are used.

Analysis of the data is done by means of a panel regression model and a three step approach. Based on these analysis, six main results stand out.

First of all, the results in the first step of the approach suggest that investments in R&D are overall significantly positively correlated with firm performance in the following years at the 1% level. Which is in line with the expectation that investments in R&D and firm performance are positively correlated, and therefore supports the first sub-hypothesis (1a).

Second, when testing for the effect of R&D-investments over longer time periods (thus including two, three and four year lags), no increasing positive correlation was found. Moreover, the correlation is even dropping when increasing the lag after the second-year lag. In addition, the other analyses preformed in this paper did not contradict this finding. Taking a two year-, three- or four year lag even showed minimal significant effects in the last steps of the analyses. As more control variables were added in the last regression, we can place most confidence in these results. Hence, concluded is that taking the first year lag is the best option. Hypothesis 1b is not supported by the results.

Third, the results in the second step of the approach found only one significant observation where both R&D ratio and the interaction between R&D ratio and the ‘aftercrisis’ dummy where significant. It showed us that in the period before the financial crisis R&D investment had a positive impact on firm performance. After the crisis this R&D effect was smaller, which implied a downfall in return on firm performance. This result is perfectly in line with the expectations as well. Hence, hypothesis 2 is supported by the results.

Fourth, the results in the last step of the approach showed unexpectedly that there was a relatively small negative return on R&D investment for the Health Care sector and IT sector, in the period before the crisis. However, expected was that these sectors would have large positive returns

on their investments relative to the other sectors. Hence, the results found do not support hypothesis 3.

Fifthly, the results found for the IT sector even implied an significant positive change in return from the period before the crisis to the period after the crisis. The expectations expressed in hypothesis 4 were the exact opposite, thus is not supported by the results. Moreover, the same positive change holds for sectors Materials, Industrials and Telecommunication Services. A possible explanation is that although these sectors were hit hard by the crisis, they were also the first ones to experience investments from local governments. This positive change also goes against the

expectations, because a negative change in return was expected.

Finally, a distinction is made between analysis with correcting for fixed effects and without correcting for these effects. As argued in the data section, the reason for estimating a ‘fixed effects’ model is that it provides the strongest evidence of effects, as differences between companies are controlled for. However, by estimating the models by means of fixed effects, possible important differences between companies are overlooked. In the first step and second step the regression analyses without correcting found more significant results, or even the only significant results. On the contrary, during the last step of the approach both correcting for fixed effects and without correcting for these effects found significant results. In addition, more control variables were added in the last step, thus we can place most confidence in these results. This implies that even though the regression with correction for fixed effect is a relatively more strict approach, it is the best option in this study.

The most outstanding result found in this study is the relatively small negative return on R&D investment for the Health Care sector and IT sector, in the period before the crisis. However, R&D projects in some sectors, might take more than 1 year to implement. Therefore, follow-up research might only include sectors in their dataset in which it takes companies the same time to implement R&D activities.

Additionally, follow-up research might check for differences in return on R&D investments for only small or large companies, and/or only for the period after the financial crisis.

Reference list

Archibugi, D., Filippetti, A., & Frenz, M. (2013). Economic crisis and innovation: Is destruction prevailing over accumulation?. Research Policy,42(2), 303-314.

Brouwer, E., Kleinknecht, A., & Reijnen, J. O. (1993). Employment growth and innovation at the firm level. Journal of Evolutionary Economics, 3(2), 153-159.

Cohen, W. M., Levin, R. C., & Mowery, D. C. (1987). Firm size and R&D intensity: A re-examination. Dess, G. G., & Robinson, R. B. (1984). Measuring organizational performance in the absence of

objective measures: the case of the privately-held firm and conglomerate business unit. Strategic management journal,5(3), 265-273.

Falk, M. (2012). Quantile estimates of the impact of R&D intensity on firm performance. Small Business Economics, 39(1), 19-37.

Hall, B. H., Mairesse, J., & Mohnen, P. (2009). Measuring the Returns to R&D (No. w15622). National Bureau of Economic Research.

Kancs, D. A., & Siliverstovs, B. (2016). R&D and non-linear productivity growth. Research Policy, 45(3), 634-646.

Krueger, A. O. (1987). Origins of the developing countries' debt crisis: 1970 to 1982. Journal of Development Economics, 27(1), 165-187.

Morbey, G. K. (1988). R&D: Its relationship to company performance. Journal of product innovation management, 5(3), 191-200.

Parkinson, M. (2009). The credit crunch and regeneration: Impact and implications (Doctoral dissertation, City University).

Ravenscraft, D., & Scherer, F. M. (1982). The lag structure of returns to research and development. Applied economics, 14(6), 603-620.

Scharpf, F. W., & Schmidt, V. A. (Eds.). (2000). Welfare and work in the open economy: volume II: diverse responses to common challenges in twelve countries. OUP Oxford, 179-181. WRDS (Table 2 source):

https://wrds-web.wharton.upenn.edu/wrds/support/Data/_001Manuals%20and%20Overviews/_001Com pustat/_001North%20America%20-%20Global%20-%20Bank/_000dataguide/gicscodes2.cfm

APPENDIX 1:

Table 7

Panel regression estimates of the correlation of R&D investment and employment growth after the crisis R&Dratio Aftercrisis R&Dratio*

aftercrisis Constant Number of observations Number of companies Fixed effects 1 year lag 0.243* (0.139) -0.032*** (0.010) -0.094 (0.126) 0.068*** (0.009) 45,010 6,864 2 year lag 0.177 (0.187) -0.035*** (0.013) 0.149 (0.171) 0.062*** (0.011) 37,008 5,843 3 year lag 0.065 (0.098) -0.041*** (0.006) 0.189** (0.088) 0.063*** (0.006) 31,535 5,132 4 year lag -0.070 (0.102) -0.031*** (0.006) -0.115 (0.089) 0.067*** (0.006) 27,283 4,675

Without fixed effects 1 year lag 0.35*** (0.082) -0.001 (0.009) -0.19* (0.099) 0.047*** (0.007) 45,010 6,864 2 year lag 0.209* (0.126) -0.017 (0.011) 0.078 (0.138) 0.056*** (0.011) 37,008 5,843 3 year lag 0.150** (0.074) -0.03*** (0.006) 0.050 (0.077) 0.062*** (0.007) 31,535 5,132 4 year lag 0.20*** (0.075) -0.024*** (0.006) -0.068 (0.078) 0.053*** (0.006) 27,283 4,675

* significant at the 10% level ** significant at the 5% level *** significant at the 1% level

APPENDIX 2:

Table 8

Panel regression estimates of the correlation of R&D investment and employment growth after the crisis for, with correction for fixed effects

1 year lag 2 year lag 3 year lag 4 year lag R&Dratio 19.2*** (2.01) 3.79 (12.28) 5.27 (4.14) 9.6 (6.17) Aftercrisis* R&Dratio -16.69*** (2.06) -1.18 (12.21) -4.58 (4.12) -9.68 (6.14) Aftercrisis* Industry* R&Dratio Materials 18.95*** (2.26) 2.37 (12.27) 7.09* (4.16) 8.78 (6.17) Industrials 17.32*** (2.14) 4.44 (12.23) 6.98* (4.14) 9.39 (6.15) Consumer Discr. 11.59*** (2.19) 1.95 (12.24) 4.76 (4.15) 9.47 (6.16) Consumer Stapl. 16.20*** (2.75) 1.71 (12.51) 5.94 (4.40) 10.47* (6.33) Health Care 16.64*** (2.08) 0.81 (12.21) 4.49 (4.13) 9.91 (6.14) Financials 15.9*** (3.13) 4.16 (12.98) 3.3 (4.78) -10.68 (11.29) IT 16.82*** (2.1) 1.17 (12.21) 4.81 (4.13) 9.81 (6.14) Telecomm. 16.88*** (4.91) 1.94 (16.99) 6.86 (6.94) 13.02 (8.62) Utilities 15.73*** (2.45) -3.6 (12.38) 2.98 (4.27) 12.49 (11.72) ROA 0.06*** (0.02) 0.02 (0.02) -0.01 (0.01) 0.005 (0.002) Constant 0.04*** (0.01) 0.05*** (0.01) 0.05*** (0.01) 0.06*** (0.008) Number of observations 43,667 35,999 30,725 26,578 Number of companies 6,644 5,647 4,984 4,553

* significant at the 10% level ** significant at the 5% level *** significant at the 1% level

Table 9

Panel regression estimates of the correlation of R&D investment and employment growth after the crisis for, without correction for fixed effects

1 year lag 2 year lag 3 year lag 4 year lag R&Dratio 5.87*** (1.0) 0.84 (3.76) 4.95 (3.97) 9.5 (5.81) Aftercrisis* R&Dratio -6.08*** (1.18) -0.25 (3.84) -4.71 (3.97) -9.42 (5.82) Aftercrisis* Industry* R&Dratio Materials 7.35*** (1.37) 2.07 (3.98) 6.51 (4.0) 9.54 (5.84) Industrials 5.97*** (1.27) 1.64 (3.89) 6.1 (3.99) 9.18 (5.83) Consumer Discr. 1.7 (1.29) 0.71 (3.91) 4.88 (3.99) 8.83 (5.83) Consumer Stapl. 5.19*** (1.88) 0.48 (4.45) 5.86 (4.12) 10.59* (5.96) Health Care 6.13*** (1.19) 0.14 (3.85) 4.62 (3.97) 9.61* (5.82) Financials 5.61*** (1.55) 0.62 (4.41) 3.54 (4.11) 9.01 (5.87) IT 6.06*** (1.19) 0.23 (3.85) 4.81 (3.97) 9.49 (5.82) Telecomm. 6.95*** (1.95) 2.84 (5.2) 5.33 (6.35) 10.74 (7.85) Utilities 5.16*** (1.55) -4.17 (4.23) 3.03 (4.05) 9.61 (7.4) ROA 0.04** (0.02) 0.01 (0.02) -0.02** (0.01) 0.0004 (0.001) Constant 0.15*** (0.06) 0.11 (0.09) 0.02 (0.05) -0.02 (0.05) Number of observations 43,667 35,999 30,725 26,578 Number of companies 6,644 5,647 4,984 4,553

* significant at the 10% level ** significant at the 5% level *** significant at the 1% level