3. Empirical Evidence
3.1 Data
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
3. EMPIRICAL EVIDENCE
In this section we use firm-level data to test more formally the empirical validity of our propositions. First, we examine the relationship between capital intensity and R&D, and show the clear correlation in Figure 1.1 by a simple regression model. Next, we expand the simple regression by controlling other factors that would reasonably be expected to affect R&D spending. Finally, we analyze the determinants of the R&D spending and show that R&D spending is significantly affected by both capital intensity and level of outsourcing competition.
3.1 Data
In order to evaluate the correlation between capital intensity and research behavior under incomplete contract, it is more convincing to test our hypothesis under firm-level data. Our empirical work draws upon survey data, Report on Foreign Investment Strate-gies of the Manufactures, collected in 2004 by Taiwan’s Ministry of Economic Affairs (MOEA). This survey includes 1712 foreign manufacturing firms owned by Taiwanese enterprisers. Furthermore, it contains the basic characteristics of firms, including their research spending, number of employees, industry type and incentive for overseas invest-ment. This model was built with companies under incomplete contract and we select specifically ODM and OEM businesses that are likely to suffer from incomplete contract problems. Among 1712 firms, 462 firms were engaged in ODM or OEM business, and all
‧
of them also made R&D spending in 2004. We use this cross-sectional data to verify our first two propositions.
Our dependant variable, R&D spending, is collected by MOEA in 2004. Some in-dependent and control variables are also taken from this survey. Capital intensity is not provided in this survey, so we use the data from another survey, ”Industry, Commerce and Service Census”, collected in 2006 by Taiwan’s Directorate-General of Budget, Ac-counting and Statistics (DGBAS). The capital intensity of each firm is measured as the ratio of total capital stock to total employment in the corresponding industry. We assume there is a high correlation between unobservable firm-level capital intensity in 2004 and industry-level capital intensity in 2006.
K
where vi is an error due to the fact that (K/L)ij and (K/L)j are not exactly correlated.
We also assume that, once (K/L)j is controlled, the expected value of (K/L)ij does not depend on other determinants of the R&D spending.
E
where X denotes other determinants of the R&D spending. Hence, we adopt the capital-labor ratio in industry-level as a proxy variable of the unavailable firm-level data. In figure 1.1, we can obtain a clear relationship between the capital intensity and R&D spending.
In order to examine our proposition 2, we must find an indicator which reflects the condition of business of each firm. In this survey data, it investigates the motivations which induce these firms to engage in overseas investment, such like local market devel-oping, cheaper raw materials and abundant labor pool. For these firms seeking cheaper labor poor, most of them are confronted with a fierce cost-down competition, so we use this incentive as a dummy variable LF, to capture fierce outsourcing competition. There is one thing to be noted that we can not distinguish such overseas investment between vertical and horizontal integration. If such foreign supplier invests for vertical
integra-‧
tion, we could infer that this supplier faces a potential competition; however, horizontal integration will not be a good indicator to capture the fierce competition. Moreover, we expect that capital intensity and fierce competition interact in R&D spending. Higher capital intensity induces foreign supplier to engage in R&D, and fierce competition would enhance its R&D spending.
Tab. 3.1: The mean data sorted by industry
Industry No of obs. Sales R&D spending Lobor
Food Manufacturing 2 1, 332, 772 3, 109 179
Textiles Mills 17 2, 776, 436 20, 752 2,282
Wearing Apparel & Clothing 13 1, 757, 114 20, 722 1,963
Leather & Fur 12 12, 628, 396 439, 485 12,005
Pulp & Paper 3 118, 217 3, 333 1,225
Chemical Material 4 6, 997, 958 26, 069 331
Chemical Products 8 992, 918 61, 235 321
Rubber Products 6 7, 670, 340 123, 788 3,285
Plastic Products 24 1, 570, 678 26, 441 1,106
Mineral Products 6 9, 095, 903 21, 224 2,197
Basic Metal 26 1, 508, 274 16, 562 704
Fabricated Metal Products 29 1, 422, 400 22, 937 575
Computers, Electronic & Optic 87 9, 689, 496 200, 939 1,734 Electronic Parts & Components 126 11, 163, 962 269, 816 1,876
Electrical Equipment 46 2, 296, 995 29, 931 1,339
Motor Vehicles & Parts 40 2, 698, 088 54, 992 508
Other Manufacturing 13 1, 635, 583 34, 126 726
Total 462 6, 415, 129 139, 839 1709
N otes : The value of sales and R&D are measured in thousands of New Taiwan Dollars (NTD).
‧
To control for other potential determinants of R&D spending, we run the multiple regression including other firm and industry characteristics. First, we have to control firm size. The expected sign of size is ambiguous, depending on the characteristics of each industry (Veugelers and Cassiman, 1999). For example, Acs and Audretsch (1987) find that large firms tend to have a relative innovative advantage in a capital-intensive industry, while small firms do in a highly innovative industry. Moreover, as we show in table 3.1, R&D spending is higher in electronic parts industry and electronic equipment industry, so we take 16 industry dummy variables, to measure the characteristic of each industry.
Technology sourcing is also an important determinant, which might be either a sub-stitute for or a complement to a firm’s internal R&D spending (Veugelers and Cassiman, 1999; Piga and Vivarelli, 2004). We use a dummy variable PT, to capture the effect if such firm’s technology is sourced from purchase. Several papers have indicated that the larger export market provides higher returns to R&D investment Bustos (February 2011).
We take the variable ex, the percentage of export of total production to control larger market effect.
Tab. 3.2: Empirical variables
Variable Definition and measurement Expected sign
R&D Total R&D spending of the firm in thousands of New Taiwan Dollars
K/L The ratio of capital stock to employment in the corresponding industry + LF Dummy variable, whose value is 1 if the firm invests for seeking cheaper +
labor poor; 0, otherwise
L The amount of hired labor ?
PT Dummy variable, whose value is 1 if the firm sources technology from ? other firm; 0, otherwise
EX Percent of export +
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
3. Empirical Evidence 21
We show the descriptive statistics of our empirical variables in table 3.3.
Tab. 3.3: Descriptive statistics of empirical variables
Variable Mean Std. errors Maximum Minimum Median
ln(R&D) 4.22 0.88 7.10 0.60 4.29
ln(K/L) 3.26 0.25 4.36 2.76 3.18
LF 0.03 0.16 1.00 0.00 0.00
LF*ln(K/L) 0.09 0.54 3.93 0.00 0.00
ln(L) 2.67 0.65 4.95 1.00 2.60
PT 0.05 0.22 1.00 0.00 0.00
EX 66.72 37.89 100.00 0.00 83.50
In order to avoid correlation among our independent variables, table 3.4 displays the correlation coefficient matrix of our variables. From the row 1 of table 3.4, we observe that our independent variables, except for PT, move significantly with our dependent variable.
There is a serious multicollinearity problem between LF and LF*ln(K/L), and that will lead to an unclear or ambiguous result if we put these two variables into a regression model. To resolve such problem, we choose to only use LF to capture fierce competition in the multiple regression model. Except for this problem, the correlation coefficients between the other control variables are not significant, so the multicollinearity problem will not be a concern in our model.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
3. Empirical Evidence 22
Tab. 3.4: Correlation coefficient matrix
Variable ln(R&D) ln(K/L) LF LF*ln(K/L) ln(L) PT EX
ln(R&D) 1.00 0.24 0.12 0.13 0.54 -0.00 0.12
ln(K/L) 1.00 0.10 0.12 0.13 0.03 -0.08
LF 1.00 0.99 0.08 -0.03 -0.08
LF*ln(K/L) 1.00 0.09 -0.04 -0.08
ln(L) 1.00 -0.02 0.13
PT -0.07
EX 1.00