3 Temporal Structure of Firm Growth
3.3 Methodology and data
3.3.1 Data source and firm sample
We use the European Industrial R&D Investment Scoreboard as data source. This data source includes 1000 European companies (see Table 3.8 in the appendix) with information on employees, turnover, sector affiliation and details on capital expenditure and R&D expenditure. These data are available for the time frame from 2003 to 2006. Hence, we are
49
able to generate the growth rates for three time periods. It is important to mention that we exclude outliers in terms of turnover growth. The descriptive analysis shows that the data set faces a high frequency of extreme growth events which would strongly influence OLS estimates. We therefore exclude these extreme values. Additionally, we cannot distinguish organic from acquired growth. Hence, we study total growth. Finally, the sample consists of 978 firms. Table 3.8 (in the appendix) shows the firm sample differentiated by country. The highest number of observations can be found in the UK with 310 observations, which is 31.8 percent of the firm sample. Table 3.1 displays our sample within firm size classes measured by the number of employees. These size classes (by number of employees) are derived from European Commission’s definition of SMEs (see European Commission (2003)).
In order to examine differences between industries, we analyse a number of industries separately. Since innovation processes are of more importance in the manufacturing sector than in the service sector, we focus on manufacturing industries. However, our firm sample contains a large number of firms from the real estate industry. Therefore, we also examine the real estate firms in comparison to the sample from the manufacturing industries. All other service industries are ignored because the sample contains only few firms of each of these industries and they have less relevance for an analysis of the effects of R&D activities. Table 3.2 presents the firm numbers for those industries that are studied separately (manufacturing industries with low numbers of firms are also not considered, such as Textiles (17), leather articles (19) or paper (21)). Furthermore, we build two further subpopulations containing all firms in high-tech manufacturing industries and all firms in low-tech manufacturing industries1 (see Table 3.10 in the appendix). Each of these industry and technology classes is used as estimation basis of the regressions conducted below. The number of firms (final column in Table 3.2) considered in this way does not add up to the 978 firms in the sample because some are considered twice (according to their high- and low-tech classification and for a separate industry study). The descriptive stats for the individual variables are presented in the appendix (see Tables 3.12 – 3.15).
Table 3.1 Firm size in terms of number of employees (SIZE)
SIZE* Cut-off points Frequency Percentage
SME 5 ≤ x ≤ 250 144 14.8
large 250 < x ≤ 1000 182 18.6
very large >1000 652 66.6
Total 978 100
Table 3.2 Type of industry within manufacturing
1 The classification is based on the study by Legler and Frietsch (2006). They allocate firms into high-tech (knowledge intensive) industries and low-tech industries (2-digit industrial classes).
Industry NACE code (2-digit NACE classification) obs.
high-tech industries 24+29+30+31+32+33+34+35 408
low-tech industries 15+16+17+18+19+20+24+25+26+27+28+36+37 124
Food products, beverages, tobacco 15 + 16 36
Chemicals and chemical products 24 105
Basic metals and metal products 27+28+29 79
Electrical and optical equipment 30+31+32+33 172
Transport equipment 34+35 66
Real estate 72+73+74 218
50
3.3.2 Operationalisation of firm growth
Before starting with our analysis, an operationalisation of the term firm growth is necessary.
In the empirical literature, there exists a wide range of definitions of firm growth. Some definitions are based on the number of employees (e.g. Kirchhoff and Greene, 1998;
Schreyer, 2000; Garnsey, Stam, and Heffernan 2006; Hoelzl and Friesenbichler, 2008) whereas others are based on turnover (Daunfeldt, Elert, and Johansson 2010). More precise, Garnsey et al. (2006) shows that firm growth can be measured in terms of inputs (e.g.
employees), in terms of value (e.g. asset) or outputs (e.g. turnover, profit). In our analysis, we use turnover data which is one of the most commonly used measures for growth. In the regression model we apply the relative turnover growth indicator. The characteristics of this indicator are important for the choice of an adequate regression approach. From the literature it is well-known that the logarithm of the growth of firms (e.g., profit rates) is Subbotin (exponential power distribution) distributed. However, regressions with a Subbotin distributed error term are no standard tool. The use of standard regression approaches can be justified by the fact that our residuals as well as the independent variable are approximately normally distributed. There is no evidence for a deviation from a normal distribution in our data. We also do not find other problems, such as heteroscedasticity, for our regressions with the logarithm of relative growth as dependent variable. In addition, we use the variance inflation factor (VIF) to test for multicollinearity (see appendix 3.11). To avoid multicollinearity between the different items of R&D expenditure (i.e. R&Dexp05, R&Dexp04, R&Dexp03) and capex (i.e. capex05, capex04 and capex03) we set up different regressions, each time one average values (i.e. R&Dexp or Capex) and one of the single values (i.e. R&Dexp05, R&Dexp04, R&Dexp03 and capex05, capex04 capex03) is included in the model. This procedure removes all problems with multicollinearity. Hence, we use a standard regression approach, although we are aware of the fact that the real distribution of growth differs slightly from the assumptions on which this approach is based. We define our dependent variable by measuring turnover growth (TURN) as the change in the logarithms of the turnover from year t-1 (2005) to turnover to year t (2006). A study by Coad and Rao (2010) investigates that sales growth are followed by growth of R&D expenditures. We therefore suggest strong associations (‘positive feedback loops’) between R&D activities and turnover growth. Hence, we see turnover growth as an adequate growth measure for our study. The equation for the dependent variable follows as:
TURN, ln , ln ,
3.3.3 Independent variables
To test the hypotheses, we employ five general independent variables (including the control variables there are seven). These variables are the relative turnover growth in the periods 2003 to 2004 and 2004 to 2005, firm size, R&D expenditures - which we see as mainly intangible - and capital expenditures - which we see as mainly tangible. In addition we use industry affiliation and profit as control variables because some empirical studies claim that sales growth is related to certain industry affiliation and is associated with subsequent growth of profits. As discussed above (section 3.2) we basically deal with the relevance and importance of resource-based innovation indicators (Grupp 1998), such as R&D expenditures, and investments in tangible goods, such as capital expenditures. It is beyond the scope of this study to explore performance indicators such as market cap and cash flow. As such, this enables the analysis of inputs such as R&D expenditure and capex investment to be related to
51
outputs such as sales neglecting the performance indicators such as market capitalisation and cash flow. The independent variables are defined as follows:
Turnover (Turn03/04; Turn04/05)
We use the relative growth indicator in terms of sale to measure the firm growth from 2003 to 2004 and from 2004 to 2005. These independent variables testify whether firms that experience growth in any one year repeat this performance in the following year or two years later.
Size of firms (Size)
First, we classify the firms into three size classes: small- and medium-sized (SME) enterprises (less than 250 employees), sized enterprises (251 to 1000 employees) and very large-sized enterprises (more than 1000 employees). We conduct most regressions for these size classes separately. Second, in the other cases to control for firm size and to avoid endogeneity we use the log form of the employment number reported in year 2003. The frequency of the different size classes is presented in Table 3.2.
R&D expenditure (R&Dexp)
In the European Scoreboard, firms report the R&D expenditure (which we see as mainly intangibles) for each year. We measure the ratio between their R&D expenditure and their total turnover. First, we use the average of this R&D expenditure ratio for the observed years (2003-2006). Second, we use this R&D expenditure ratio for each year separately (R&Dexp03, R&Dexp04 and R&Dexp05).
Capital expenditure (Capex)
Furthermore, the European Industrial R&D Investment Scoreboard database presents additions to tangible fixed assets such as capital expenditure. Capital expenditures are investments used by a company to acquire or upgrade physical assets. Again, we measure the average capital expenditure as the ratio of capital expenditure to total turnover (2003-2006).
Additionally, we use the capital expenditure ratio for each year separately (Capex03, Capex04 and Capex05).
Profit (Profit)
From a micro-economic perspective, we control for profit efficiency. To avoid endogeneity, we use the operating profit for the year 2003. We hence suggest that turnover growth might be associated with (subsequent) growth of profit (see Coad 2009). We control for this
‘feedback loop’ to investigate whether firms behave superior as profit has grown previously.
Coad (2009 p: 73) shows that “if turnover growth have grown recently, aim to keep R&D levels at a roughly constant ratio with respect to the firm size”.
Industry classes (IndDummy)
As mentioned above we use industry affiliation as a control variable. For this purpose we aggregated the NACE-2-digit industries classification. We use 11 different types of industry which are presented in Table 3.9 (in the appendix). For each sector, except one, a dummy is included in the regression analyses.
3.3.4 Regression approach
We set up a regression approach with a simple linear model (see Equation (1)). In the course of the regression analysis we find that the variables R&Dexp03, R&Dexp04 and R&Dexp05
52
are highly correlated with each other. The same holds for the variables capex03, capex04 and capex05. To avoid multicollinearity (see correlation matrix 3.11 in the appendix) between these explanatory variables, we set up different regression models. The comparison of the results provides information about the time structure of the impact of R&D on firm growth.
(1) GROWTHj a0a1R&Dja2capexj a3profitj
R&D stands for the various measures of R&D expenditures and capex stands for the various measures of capital expenditures as discussed above. In regressions that are done for all firms of any size together, the log value of the size of the firms is included in the regression as an independent variable. Similarly, in regressions that consider firms from all sectors dummies are included as independent variables, which reflect each of the industries as described in Table 3.2. Hence, if all firms are analysed together, the following model is used:
(2)
j j j j j
j a a R D a capex a profit a size a IndDummy
GROWTH 0 1 & 2 3 4 log( ) 5 14 1 10,
The time structure of the impact of R&D on growth might interfere with the time structure of the growth process itself. If growth in one year depends on growth in previous years and growth in previous years depends on R&D activities in previous years or R&D activities further in the past, a relationship between current growth and previous R&D activities might be a direct effect or an indirect effect. To disentangle this structure we conduct each regression one time including measures of R&D and capital expenditures as independent variable but without considering growth in the past (see Equations (1) and (2)), one time without any R&D activity considered but including growth in the past (Equation (3)), and one time with past growth rates and R&D and capital expenditures as independent variables at the same time (Equation (4)).
(3)
j
j
j j
j j
IndDummy a
size a
profit a
growth a
growth a
a GROWTH
, 10 1 14
5 4
3 2
1 0
) log(
05 / 04 04
/
03
(4)
j
j j
j j
j j
j
IndDummy a
size a
profit a
capex a
D R a growth
a growth
a a GROWTH
, 10 1 16
7 6
5
4 3
2 1
0
) log(
&
05 / 04 04
/ 03
Furthermore, we conduct one regression set for all firms together and then three regression sets for each firm size separately. Through this, we are able to analyse whether our findings depend on the size of firms. In order to analyse the differences between industries, we also study a number of firm subsamples that are defined by their industry classification as discussed above.