Innovation And Export Behaviour At The Firm Level

Research Policy 26 Ž1998. 829–841

Innovation and export behaviour at the firm level Katharine Wakelin

)

MERIT, P.O. Box 616, 6200 MD Maastricht, Netherlands Received 28 February 1997; revised 20 October 1997; accepted 12 November 1997

Abstract This paper considers the role of innovation in determining export behaviour for a sample of UK firms including both innovating and non-innovating firms. Export behaviour is defined both as the probability of a firm exporting and the propensity to export of the exporting firms. An empirical model of the determinants of export behaviour is estimated, and the determinants are found to vary between innovating and non-innovating firms. Non-innovative firms are found to be more likely to export than innovative firms of the same size. However, the number of past innovations has a positive impact on the probability of an innovative firm exporting. The paper concludes that the capacity to innovate changes the behaviour of the firm relative to non-innovating firms. q 1998 Elsevier Science B.V. All rights reserved. Keywords: Innovation; Export behaviour; Firm level

1. Introduction There is considerable macroeconomic evidence that differences in innovation, in addition to relative prices, can influence export behaviour Žsee for instance Fagerberg, 1988, for the OECD countries and Greenhalgh, 1990, for the UK.. This paper aims to extend that analysis to the firm level, considering the role of innovation in determining export behaviour for a sample of UK firms. The impact of innovation on firm performance is treated in two ways: the direct impact of being an innovative firm; and the role of spillovers of innovations from other firms. In contrast to other microeconomic studies which have used R & D expenditure as a proxy for innovation,

)

Corresponding author. Tel.: q31-43-3883886; fax: q31-433216518; e-mail: [email protected].

this paper uses the innovation history of the firms taken from the SPRU innovation survey. The role of innovation in trade behaviour is of particular interest in the case of the UK. As a number of studies have pointed out, the UK’s trade performance in manufacturing, as shown by the UK share in world manufacturing trade, has been declining for much of the post-war period, although this trend may have been reversed in the 1980s. 1 It is frequently noted Žfor instance by Thirlwall, 1986. that it is in the area of non-price competitiveness that the UK economy is particularly weak, reflecting, among other factors, the poor skills of the work

1 See for instance Landesmann and Snell Ž1989. and Anderton Ž1992. for examinations of the UK’s changing trade patterns in the 1980s. This decline in trade share does not necessarily indicate a decline in ‘competitiveness’; Fagerberg Ž1988. gives a broad definition of competitiveness.

0048-7333r98r$19.00 q 1998 Elsevier Science B.V. All rights reserved. PII S 0 0 4 8 - 7 3 3 3 Ž 9 7 . 0 0 0 5 1 - 6

830

K. Wakelinr Research Policy 26 (1998) 829–841

force, poor product design and quality, after sales service, and reliability. There have been a number of studies investigating the impact of non-price competitiveness on UK trade at a sectoral level. Greenhalgh Ž1990., Buxton et al. Ž1991. and Greenhalgh et al. Ž1994., have found some evidence of a positive role for innovation in trade performance. In this paper the impact on export behaviour of being an innovating firm is estimated, while controlling for other factors which influence exports. In addition, the level of innovation in the sector, and the use of innovations in the sector are also considered as determinants of firm export behaviour. The paper is set out as follows. Section 2 outlines the characteristics of innovation at the firm level, and summarises other empirical work concerning the determinants of exports at the level of the firm. Section 3 presents the data set to be used and the empirical model to be tested; Section 4 gives some descriptive statistics. Section 5 considers the model specification, and discusses the results for the probability of a firm exporting, and for the propensity of firms to export. Section 6 concludes. 2. Firm behaviour It is at the firm level that decisions about the commitment of resources to innovation, and the innovative strategy of the firm are made. It is also principally at the level of the firm that the benefits of innovation are enjoyed: in terms of cost reductions, new markets and potential monopoly rents. This makes the firm a suitable unit of analysis when considering the role of technology in export behaviour. The sector, and more broadly the country in which the firm is located, provide the context for these decisions, and have a strong influence on them. Considering the relationship between innovation and exports for firms in a single country does not reflect on the larger issues of a country’s competitiveness Žhowever this is defined.. The exports of each firm give no indication of the imported content of the firm’s output, nor the level of import penetration within the industry in which the firm operates. The tacit and non-codifiable nature of technology, the importance of learning-by-doing and learningby-using in technological change and the potential to appropriate at least some of the benefits of innova-

tion may lead to the accumulation of innovations by individual firms. 2 As technology is not always codifiable, the transfer and diffusion of it can occur only slowly. Firms accumulate skills from using new technologies, learning via the production process, and from the implementation of innovations. Innovation becomes a cumulative process ŽRosenberg, 1976, 1982., which is at least partly specific to the firm. 3 One outcome from these firm specific innovation patterns is that asymmetries exist among firms in terms of their technological capabilities and their general economic performance. These asymmetries provide the motivation for examining differences in export behaviour between innovating and non-innovating firms. The majority of firm level studies considering innovation have concentrated on testing the Schumpeterian hypothesis of a positive relationship between firm size and innovation. However, a number has examined the relationship between innovation and exports. Kumar and Siddharthan Ž1994. analysed the relationship between R & D expenditure and exports for 640 Indian firms from 1988 to 1990 grouped according to industry: they found R & D expenditure to be an important factor in low and medium technology industries, and concluded that India does not have a competitive advantage in high technology sectors, but innovation positively influences her performance in other sectors. Willmore Ž1992. concentrated on the role of transnationals in Brazil’s trade, estimating both the determinants of exports and those of imports. He found no significant role for R & D expenditure as a determinant of exports, although R & D appeared to play a small negative role with respect to imports: technological effort led to increased domestic inputs and less reliance on imports. Hirsch and Bijaoui Ž1985. considered the relationship between R & D expenditure and export behaviour for Israel; a small country which experienced a rapid rise in exports in the 1970s. They tested the importance of innovation advantages for 111 Israeli firms, all of which had undertaken R & D expenditure and were thus classified as innovators.

2

See for instance Dosi Ž1984, 1988. and Freeman Ž1982.. Blundell et al. Ž1995. found evidence of ‘history dependence’ in patterns of firm innovation. 3

K. Wakelinr Research Policy 26 (1998) 829–841

Initially, the authors contrasted the propensity to export of their innovating firms with the average propensity to export in each sector, and found that the innovating firms, grouped into sectors, had a higher propensity to export than the sector average. In the model which followed, they found lagged R & D expenditure to be significant in explaining the rate of change of exports in a cross-section. The size of the firm, measured by sales, and the change in firm sales, taken as an indicator of ‘other firm characteristics’ were also significant. The authors concluded that innovation is an important factor in explaining export performance; they also noted that while a minimum size is probably required to export, beyond that firm size is not a major factor. The majority of firm level studies have used R & D expenditure as an indicator for innovation, and as a basis for the classification for firms as innovators. Taking information from an innovation survey, as in this paper, has the benefit of not excluding those firms too small to have a separate R & D department, or even a R & D budget, yet which nevertheless innovate ŽPavitt et al., 1987.. The distribution of such firms may be concentrated in some sectors, such as the engineering and instrumentation sectors, in which many innovations are produced as part of the production process rather than through R & D. By using the actual number of innovations produced to classify firms as innovators, the size bias of R & D expenditure as an innovation proxy can be avoided. 4

3. The data set and empirical model The empirical analysis presented here aims to assess the importance of different determinants of trade behaviour, in particular the role of innovation. The differences between innovating and non-innovating firms are treated in two different ways. First, a dummy variable is included which takes the value of one for firms which have innovated; this is designed 4 There have been some studies combining this innovation survey with other firm data to investigate a range of issues, see for instance the work of Blundell et al. Ž1995. on the historical dependence of innovation, and Geroski and Machin Ž1993. for the relationship between innovation and firm profitability.

831

to test if there is a fixed effect as a result of being an innovator. In addition, the number of innovations a firm has had in the past is also included. Second, the estimates are made separately for the two groups of firms, innovators and non-innovators, to see if the determinants of trade behaviour vary across these two groups; this separation is tested relative to a model with all firms pooled together. The latter approach indicates that being an innovator fundamentally changes the nature of the firm, so that innovating firms need to be treated differently from non-innovating firms. The data set used in this paper is a microeconomic data set of UK firms which covers 320 firms for a period of 5 years from 1988–1992 and accounts for over half of total UK manufacturing output over the 5 years. 5 It consists of two sub-samples. The first sample of innovating firms is chosen from the firms covered by the SPRU innovation survey. The definition used for the inclusion of an innovation in the survey was ‘the successful commercial introduction of new or improved products, processes or materials’. Only the firms in manufacturing sectors were then chosen. As Pavitt et al. Ž1987. point out, 90% of the innovations were exploited by firms with their principal activity in manufacturing, so this selection covers the majority of innovations. The second sample consists of firms chosen randomly from the population of UK firms; the innovating firms, which are defined as firms included in the SPRU survey, are then removed from the sample to leave a sample of non-innovating firms. 6 Only firms in the manufacturing sector were then chosen from this sample. Approximately one third of the firms are quoted on the UK stock market, while two thirds are not: the latter are expected to be generally much smaller in size. In order to separate differences between sectors from differences between firms, firms were randomly selected so that there are equal numbers of

5 For details of the data sources, see Appendix A and for more details on the selection and choice of firms, see the work of Wakelin Ž1997.. 6 The firms are picked randomly from two data sources; Datastream for the quoted firms, and ICC for non-quoted firms. Datastream covers all quoted firms, while ICC has data on more than half a million non-quoted firms.

832

K. Wakelinr Research Policy 26 (1998) 829–841

innovating and non-innovating firms in each of the ten sectors Žsee Appendix A for details of the sector definitions.. The proportion between quoted and non-quoted firms was maintained. Without this selection procedure the sample of innovative firms came from more innovative sectors on average than the non-innovative firms. Some sectors, such as mechanical engineering, have a high level of technical opportunity as a result of the production process; others, such as footwear and clothing provide fewer opportunities. The data for each firm cover 5 years from 1988 to 1992 inclusive although data are missing for some firms in some years lowering the number of total observations. This period is considerably after that covered by the innovation survey Žwhich covers 1945–1983. and relies on the longevity of the impact of the innovations, or rather the continuing status of these firms as innovators Žand of the other firms as non-innovators.. Some firms may have innovated in the intervening period and thus be misclassified as non-innovators. As the survey has not been updated this risk is unavoidable. The criteria for being included in the survey are strict however, indicating that this problem is likely to apply to only a small number of firms in the sample at most; as a result we assume that the problem is not serious. 7 There is increasing consensus in the literature on international trade that no single factor can neatly account for the trade patterns of developed countries. As a result, the empirical model considered here includes a number of different explanations for export behaviour. All the variables considered are available for both groups of firms, and each is available annually for the period 1988 to 1992. The definitions of the main variables available in the data set are given below starting with the dependent variable: propensity to export: PX f s X frTS f ; average capital intensity 8 : KSA f s TK frTS f ;

7

There is no empirical evidence available quantifying to the seriousness of this problem. 8 Capital intensity is defined relative to sales and not to labour as the capital to labour ratio was found to be collinear with the average remuneration variable, making it difficult to separate the effects of the two variables.

average wages: AS f s TR frSIZE f ; unit labour costs: ULC f s TR frTS f ; where X stands for exports, TS for total sales, TK for total capital, SIZE for the number of employees, TR for total remuneration of employees, the subscript f is for the firm. The explanatory variables are scaled either by an indicator of firm size, such as the number of employees, or total firm output. As Deardorff Ž1984. points out, scaling is important in reducing heteroscedasticity. Firm characteristics are used to give an indication of the attributes of the firm, and the general level of firm competitiveness. The capital variable indicates the level of firm assets, including machinery and buildings, which embody past innovations, as well as influencing the marginal cost of the output of the firm. As a result a positive relationship between capital intensity and exports is expected. The two labour cost variables defined above, unit labour costs and average wages, are included to indicate labour costs and skill levels. Unit labour costs are likely to have a negative impact on exports in cost sensitive export markets. For the average wage variable, a high average wage may indicate a firm with a large degree of accumulated human capital; this implies that the coefficient on the average salary variable might be positive. There is considerable evidence for the importance of skills in determining the export performance of developed countries at a more aggregate level Žfor instance, Oulton, 1996 for the UK and Germany.. Fagerberg Ž1988., however, found only a small role for relative unit labour costs in influencing the export market shares of 15 industrialised countries; he found investment and technology to play a more important role. Verspagen and Wakelin Ž1997. on the other hand, found relative wage costs to be an important factor in explaining bilateral trade flows between OECD countries in some sectors. It appears that the evidence relating to the importance of wage levels in affecting trade at the sector level is mixed. Firm size is expected to have a positive relationship to exports as larger firms have more resources with which to enter foreign markets. This may be particularly the case if there are fixed costs to exporting such as gathering information or covering the uncertainty of a foreign market. There may also be economies of production and marketing which bene-

K. Wakelinr Research Policy 26 (1998) 829–841

fit large firms. In addition, the square of firm size is used to allow for non-linearities in the relationship between size and exports. It is possible that a minimum size is required to overcome the additional costs of exporting, beyond which increases in size have no impact on export behaviour. There are a number of studies which include such non-linearities; both Kumar and Siddharthan Ž1994. and Willmore Ž1992. find a negative relationship for the quadratic term of size in the context of developing countries. Although size is an advantage in exporting, this may not apply to very large firms which can be more orientated towards the domestic market due to, for example, a domestic monopoly giving them no incentive to export. A firm faced with a domestic monopoly can exploit domestic demand; a foreign market would typically have a higher elasticity of demand, and would involve the domestic firm becoming a price taker. Following this hypothesis the quadratic term of size is expected to have a negative coefficient. Finally, innovation variables at a sectoral level showing the innovation potential of the sector, and the use of innovations in the sector, are expected to have a positive relationship with exports. The different innovation variables available are: User, the number of innovations used in the sector from 1979 to 1983, taken from the SPRU survey, scaled by the number of enterprises in the sector; Prod, the number of innovations produced in the sector for 1979–1983, from the survey, excluding

833

each firm’s individual innovations, scaled by the number of enterprises in the sector; R & D, the expenditure on R & D in the sector scaled by the number of enterprises in the sector. The innovations produced in each sector are an indication of the technological opportunity of the sector, while innovations used show the innovation spillovers from other firms in the economy. At the two-digit sector level, 67% of innovations are used in a different sector from the sector that produced them ŽRobson et al., 1988.; the use of innovations thus indicates spillovers of innovations both within and between sectors. A positive coefficient would indicate the importance of the innovative environment in influencing the export patterns of firms. Firm level R & D expenditure is available only for the quoted firms in the sample; instead the number of firm innovations, FIRM, taken from the SPRU innovation survey, is used as an indicator of a firm’s past technological capabilities. In the first model, being an innovator is also included as a firm characteristic by including a dummy variable Ž DUM . for innovating firms.

4. Descriptive statistics Some descriptive statistics for the variables are given below in Table 1 for the four separate classifications of innovating exporters and non-exporters, and non-innovating exporters and non-exporters.

Table 1 Descriptive statistics: means Žstandard deviations. Innovators Exporters Propensity to export Average capital intensity Unit labour costs Average remuneration Ž£. Number of innovations Number of employees Innovations produced by sector Innovations used by sector R & D in the sector Ž£ million. Number of observations Ž N . Proportion exporting

0.43 Ž0.26. 0.49 Ž0.27. 0.22 Ž0.09. 13,578 Ž3575. 3.9 Ž9.6. 14,311 Ž25, 637. 546 Ž425. 305 Ž176. 721 Ž704. 355 0.64

Non-innovators Non-exporters 0.39 Ž0.34. 0.20 Ž0.11. 13,805 Ž5748. 1.7 Ž2.5. 1867 Ž6382. 526 Ž465. 278 Ž196. 483 Ž610. 200 0.64

Exporters

Non-exporters

0.38 Ž0.31. 0.51 Ž0.49. 0.23 Ž0.10. 12,485 Ž3947.

0.43 Ž0.65. 0.26 Ž0.16. 10,742 Ž3802.

2883 Ž9318. 529 Ž431. 296 Ž193. 765 Ž741. 350 0.66

134 Ž240. 461 Ž455. 250 Ž187. 489 Ž631. 180 0.66

834

K. Wakelinr Research Policy 26 (1998) 829–841

They are taken as averages across the 5 years for each of the variables. The most interesting difference between the classes is the variation in average firm size as shown by the number of employees. On average, smaller non-innovative firms export than innovative firms, indicating that being an innovative firm does not necessarily provide an incentive to export. On the contrary, relative large innovative firms do not enter export markets: the average size of the innovative non-exporters is 1867 employees, while the equivalent for the non-innovators is 134 employees. However, innovative firms that do export have a higher propensity to export than non-innovators 9. Overall, the proportion of observations having positive exports is similar across the two groups despite the much large average size of the innovating firms. 10 This result indicates that small innovative firms are less likely to enter export markets than the equivalent non-innovating firms. Those innovative firms which do export are much larger, and have a higher proportion of innovations Žalmost 4 on average against 1.7 for the non-exporters.. Smaller innovative firms may concentrate on the domestic market due to their favourable position in that market. Past innovations may have given them good links with other firms, or greater power in the market place thus limiting the incentive to search out new markets abroad. Firm size shown by the number of employees, and the number of innovations are both highly positively skewed. There is a small number of very large firms, and a small number of firms with a very large number of innovations. In general, many of the smaller firms have one or two innovations, while there are some firms which have over 20. Due to this variation, the number of innovations ŽFIRM. is also included in the later model along with the status of the firm as an innovator or non-innovator, to capture the importance of the extent of innovations.

9

Hirsch and Bijaoui Ž1985. also found a higher propensity to export for Israeli firms, but they did not look at the probability of exporting. 10 This is not the same as the number of firms exporting as there are five observations per firm and some firms export in one year and not in another.

The use of sample means to summarise the data may disguise the complexity of the relationship between innovation and firm size. Pavitt et al. Ž1987. found a U-shaped relationship between innovation and firm size, based on the innovation survey data used here. Small specialised firms were found to be very active in innovation, while large firms exploited the possibilities of R & D based diversification into other product markets. Nevertheless, the descriptive evidence points to an increasing relationship between firm size and innovation, with innovating firms being characterised by higher than average size when contrasted with non-innovating firms. While the sector distribution between innovators and non-innovators is the same due to the selection of the sample, the sector distribution within each group can be analysed. The exporting firms appear to be generally located in more innovative sectors within each group, especially when R & D expenditure is considered. On the whole however, the difference between sectors does not seem large enough to explain export behaviour in each group. Section 5 will examine if being an innovator does lower the probability of exporting using an econometric model for export behaviour. 5. The results 5.1. The choice of specification In the data set there are a number of firms which have no exports. The dependent variable, the propensity to export, which varies between 0 and 1 by definition, therefore frequently takes a value of zero. As a result OLS regression may not be the most suitable estimation procedure. 11 In order to find the best model specification two alternative models are tested against each other, following Cragg Ž1971.. The first specification estimates a single censored Tobit model. This uses all the available information from the explanatory variables, but includes both the decision of whether or not to export, and the level of exports, in one model Žsee Lin and Schmidt, 1984 for details.. The alternative specification separates 11 As it can give estimates, which imply predictions of the propensity to export outside its possible range, i.e., higher than one and lower than zero.

K. Wakelinr Research Policy 26 (1998) 829–841

835

Table 2 The choice of specification

a KSA AS SIZE SIZE 2 USER ULC DUM FIRM N

Probit Y s Y

Truncated Y s PX

0.15 Ž0.18. 0.18 Ž0.09. ) ) 0.02 Ž0.009. ) ) 0.78 10y4 Ž0.11 10y4 . ) ) ) y0.56 10y9 Ž0.11 10y9 . ) ) ) 1.66 Ž0.92. ) y0.15 Ž0.33. y0.59 Ž0.09. ) ) ) 0.04 Ž0.02. ) ) 1040

y0.30 Ž0.12. 0.16 Ž0.04. ) ) ) 0.03 Ž0.005. ) ) ) 0.15 10y4 Ž0.02 10y4 . ) ) ) y0.11y9 10 Ž0.25 10y10 . ) ) ) y0.76 Ž0.40. ) 0.11 Ž0.20. y0.008 Ž0.04. y0.0005 Ž0.003. 694

)

Significant 10%. Significant at 5%. ))) Significant at 1%. s Truncateds 0.36 Ž0.02.. Loglikelihood: probits y590.0; truncated 46.7. ))

the decision of whether or not to export from the decision of how much to export. The first stage uses the whole data set and considers the decision of whether or not to export using a probit model. 12 The model assumes an underlying Y ) which cannot be seen. Instead a variable Y can be observed which takes a value 1 when Y ) is greater than 0, and 0 when it is equal or less than zero. For the second stage only the subset of firms which export are considered. A truncated estimation procedure is used as the dependent variable is observed only if it is greater than zero. This double specification can be tested as the unrestricted model against a Tobit model as the restricted model. The unrestricted model is given below in Eqs. Ž1. and Ž2.: Y s a q b 1 KSA q b 2 AS q b 3 SIZE q b4 SIZE 2 qb5 FIRMq b6 INN q b 7 ULC q b 8 DUMq e

Ž 1. where Y s 1 if x ) 0, x is exports and Y s 0 if x s 0. For the firms for which Y s 1: PX s d q g 1 KSA q g 2 AS q g 3 SIZE q g4 SIZE 2 qg 5 FIRMq g6 INN q g 7 ULC q g 8 DUMq m

one of the choice of innovation variables, which is the number of innovations used ŽUSER. in the results given below. For the restricted Tobit model Eq. Ž2. is estimated for the whole sample. Taking the Tobit model as the restricted model, and the probit and truncated together as the unrestricted model, the Tobit model was rejected at 99% probability using a x 2-squared likelihood ratio test Žthe test statistic is 50.4.. 13 The results are given below in Table 2, with the 5 years of data pooled together; the standard errors are given in brackets. 14 The first column of the table shows the results for the probit model for the probability of exporting; the second column gives the truncated model for the propensity to export for the sub-set of exporting firms. There appear to be important differences between the influence of the innovation variables in the two models. The number of innovations used at the sector level is positively and significantly related to the probability of exporting, and is negative and significant for the propensity to export of the exporting firms. 15 In addition, the dummy variable for innovating firms is negatively significant in the model

Ž 2. where all the variables are as defined earlier. INN is 12 The models are estimated using Newton’s method of estimation for maximum likelihood estimation, taking the OLS estimates as the starting values.

13

The probit model correctly predicts 68% of the outcomes from the mode; s is the standard error of the residual. 14 Dummies for each year were also included but were not significant. 15 These results are the same substituting PROD or sector RandD for USER in the model.

836

K. Wakelinr Research Policy 26 (1998) 829–841

for the probability of exporting, but insignificant in the model for the propensity to export. However, the number of innovations FIRM has a positive impact on the probability of exporting Žand no relationship to the propensity to export.. 16 Firms with a large number of innovations are more likely to export, indicating heterogeneity even within the group of innovating firms. These results confirm the pattern found in the descriptive statistics in Table 1. The number of firm innovations and the level of innovation in the sector increases the probability of exporting, while the status of being an innovator decreases the probability. Small innovative firms appear to be more orientated towards the domestic market. Given their size this is unlikely to be a preference for direct production abroad rather than exports as multinational firms are generally large in size. The most likely explanation appears to be that exporting is something that small firms are forced to do in order to reach new markets, and in the case of innovative small firms they feel less pressure than non-innovative firms to search for new markets as they are relatively secure in their domestic markets. Section 5.2 will investigate the relationship separately for the two groups of innovating and non-innovating firms. 5.2. The probability of exporting Taking the probit model for the probability of exporting, a model separating the firms into innovating and non-innovating firms was tested to see if the probability of exporting varies over the two groups. The restricted probit model is that given in column two of Table 2 which already has separate intercepts for the innovating and non-innovating firms; the unrestricted model separates the explanatory variables for the two different groups as well. A Chisquared test was then made for the unrestricted model relative to the restricted model, following the test outlined above. The restriction was rejected at 99% confidence level, indicating that the innovating firms should be estimated separately from the noninnovating firms.

Thus, it appears that the two sets of firms—innovators and non-innovators—behave differently. This may be due directly to the accumulated economic benefits of the individual innovations themselves, or that innovating firms have some generic differences from non-innovators, such as better management. Size is certainly one factor, as we have seen from the descriptive statistics the innovating firms are significantly larger than the non-innovating firms on average. However, the role of both size and the quadratic term seems to be similar for each group; size is positively significant while the quadratic term is negatively significant. So it appears that innovating firms have different determinants for the probability of exporting which cannot just be explained by their difference in size. This is consistent with the results of Geroski and Machin Ž1993.; they found that the determinants of profits varies between innovating and non-innovating firms. The calculations were repeated with two alternative innovation variables: PROD—the number of innovations produced in the sector, and R & D—the level of R & D expenditure in the sector. For each of these innovation variables, the restricted model was rejected relative to the model separating the non-innovating from the innovating firms. The results are given in Table 3; the first half of the table gives the results for the innovating firms, the second half for the non-innovating firms. The results for many of the variables are as expected. Size clearly plays an important role in export behaviour, and a U-shaped relationship is confirmed. Firm innovations raise the probability of exporting, as shown earlier. The capital intensity variable has a positive effect as expected but this effect is not significant for either group of firms. There are three main differences between the results for the innovating and non-innovating firms and they refer to the two labour variables—average salary and unit labour costs—and the sector innovation variables. The differences are outlined below. 17 Ž1. The sign on the average wage variable is different for the two groups: positive and significant

16

Geroksi Ž1991., using the same innovation survey, found that lags of innovation put to eleven years old had a significant affect on productivity growth. Thus, the most recent innovations in the survey may still be positively influencing firm performance.

17 The models each have the same percentage of correct predictions at 67%; each model has a similar explanatory power: no one sector technology variable appears to perform better than another.

Table 3 The unrestricted probit models, dependent variablesY AS

SIZE

SIZE 2

INN

ULC

FIRM

InnoÕators N s 536 USER y0.21 Ž0.27. PROD y0.31 Ž0.29. R&D y0.16 Ž0.28.

0.23 Ž0.19. 0.27 Ž0.20. 0.21 Ž0.20.

y0.005 Ž0.01. y0.006 Ž0.01. y0.006 Ž0.01.

0.68 10y4 ) ) ) Ž0.11 10y4 . 0.68 10y4 ) ) ) Ž0.11 10y4 . 0.68 10y4 ) ) ) Ž0.11 10y4 .

y0.47 10y9 ) ) ) Ž0.11 10y9 . y0.48 10y9 ) ) ) Ž0.11 10y9 . y0.48 10y9 ) ) ) Ž0.11 10y9 .

1.62 Ž1.25. 1.92 ) Ž1.04. 0.43 Ž0.31.

1.01) Ž0.58. 0.95 ) Ž0.58. 0.95 ) Ž0.58.

0.03 ) Ž0.02. 0.03 ) Ž0.02. 0.03 ) Ž0.02.

Non-innoÕators N s 504 USER y0.49 ) Ž0.28. PROD y0.57 ) ) Ž0.28. R&D y0.50 ) ) Ž0.27.

0.05 Ž0.12. 0.05 Ž0.12. 0.06 Ž0.12.

0.06 ) ) ) 0.06 ) ) ) 0.06 ) ) )

0.14 10y2 ) ) ) 0.14 10y2 ) ) ) 0.14 10y2 ) ) )

y0.21 10y7 ) ) ) y0.21 10y7 ) ) ) y0.21 10y7 ) ) )

1.58 Ž1.60. 2.47 ) ) Ž1.20. 1.02 ) ) Ž0.13.

y1.37 ) ) y1.36 ) ) y1.26 ) )

Ž0.02. Ž0.02. Ž0.02.

Ž0.24 10y3 . Ž0.24 10y3 . Ž0.24 10y3 .

Ž0.39 10y8 . Ž0.40 10y8 . Ž0.39 10y8 .

Ž0.56. Ž0.56. Ž0.56.

)

Significant at 10%. Significant at 5%. ))) Significant at 1%. McFadden’s R 2 : USER s 0.10; PRODs 0.12; R&Ds 0.12. Loglikelihood: USER sy228.5; PRODsy226.8; R&Ds 289.3. ))

Table 4 The unrestricted truncated models, dependent variables PX

a

KSA

InnoÕators N s 348 USER y0.27 ) ) ) Ž0.10. 0.27 ) ) ) PROD y0.27 ) ) ) Ž0.10. 0.26 ) ) ) R&D y0.27 ) ) ) Ž0.10. 0.27 ) ) )

AS Ž0.06. 0.02 ) ) ) Ž0.06. 0.02 ) ) ) Ž0.06. 0.02 ) ) )

Non-innoÕators N s 346 USER y0.54 ) ) Ž0.26. 0.16 ) Ž0.09. PROD y0.72 ) ) Ž0.29. 0.17 ) Ž0.09. R&D y0.46 ) ) Ž0.23. 0.14 ) Ž0.08.

0.05 ) ) ) 0.05 ) ) ) 0.05 ) ) )

SIZE 2

INN

Ž0.005. 0.97 10y5) ) ) Ž0.005. 0.94 10y5) ) ) Ž0.005. 0.92 10y5) ) )

Ž0.17 10y5 . y0.71 10y10 ) ) ) Ž0.16 10y5 . y0.71 10y10 ) ) ) Ž0.16 10y5 . y0.66 10y10 ) ) )

Ž0.17 10y10 . y0.46 Ž0.38. Ž0.17 10y10 . y0.25 Ž0.31. Ž0.17 10y10 . y0.19 ) ) Ž0.09.

Ž0.01. 0.37 10y4 ) ) ) Ž0.01. 0.38 10y4 ) ) ) Ž0.01. 0.31 10y4 ) ) )

Ž0.13 10y4 . y0.38 10y9 ) Ž0.23 10y9 . Ž0.14 10y4 . y0.44 10y9 ) ) Ž0.24 10y9 . Ž0.13 10y4 . y0.26 10y9 Ž0.23 10y9 .

SIZE

ULC 0.58 ) ) ) 0.62 ) ) ) 0.62 ) ) )

FIRM

K. Wakelinr Research Policy 26 (1998) 829–841

KSA

a

Ž0.20. y0.001 Ž0.002. Ž0.20. y0.001 Ž0.002. Ž0.20. y0.001 Ž0.002.

y1.76 ) Ž0.99. y0.83 ) Ž0.46. 0.32 Ž0.85. y0.66 Ž0.46. y0.80 ) ) ) Ž0.27. y0.85 ) ) Ž0.43.

Innovators: s USER s 0.47Ž0.05.; s PRODs 0.48Ž0.05.; s R&Ds 0.47Ž0.05.. Loglikelihood: USER s 26.7; PRODs 25.0; R&Ds 30.5. Non-innovators: s USER s 0.27Ž0.01.; s PRODs 0.27Ž0.01.; s R&Ds 0.27Ž0.01.. Loglikelihood: USER s 46.6; PRODs 46.1; R&Ds 47.8.

837

838

K. Wakelinr Research Policy 26 (1998) 829–841

for the non-innovators and negative but not significant for the innovators. The positive relationship for the non-innovators indicates a possible skill effect for those firms: firms with higher average salaries are characterised by higher skills which in turn increases the probability of them exporting. The skill effect is not found for the innovating firms. Ž2. Unit labour costs are negatively related to the probability of exporting for the non-innovating firms, indicating that cost considerations play some role in their export performance with higher cost firms less likely to export. For the innovating firms, unit labour costs are positively related to the probability of exporting. This positive sign may indicate that innovating firms export higher quality goods which are less price sensitive. One explanation for these surprising results for the innovating firms is that many smaller innovative firms are not exporting when similar sized non-innovating firms are. These non-exporting innovative firms have neither higher unit labour costs than the exporters, nor lower average salaries. It is possible that many innovative firms with high average salaries Žand thus high skills. and low unit labour costs are choosing not to enter export markets, leading to the surprising signs on the labour variables. Ž3. The results for the sector-level innovation variables also show some variation over the two groups. As a result of the selection method, the two groups of firms come from the same sectors. Any differences between the groups are thus due to differences in the importance of those sector characteristics to the firms, rather than to the characteristics of different sectors. The production of innovations in the sector has a positive impact on the export probability of both groups of firms, while the use of innovation has no significant role. The R & D expenditure of other firms in the sector has a positive impact on non-innovators, and a much smaller effect which is not significant on innovators. This is despite the balancing of the sample into the same sectors. The R & D expenditure of other firms has potentially two different roles: R & D reflects greater competition in the sector as firms spend R & D to improve products and lower costs; or the R & D of other firms in the sector may spill over to firms providing a positive externality. The non-innovative firms appear to experience the positive spillovers from R & D

expenditure, indicating that non-innovative firms in high R & D spending sectors are more likely to export. Section 5.3 investigates the propensity to export of the firms instead of their probability of exporting. 5.3. The propensity to export The second model, for the propensity to export of the exporting firms alone, was then estimated using each of the three innovation variables. The restricted model including both the innovating and non-innovating firms together was rejected relative to the unrestricted models in all three cases; confirming the differences between the two groups of firms. The results for the three unrestricted models are given in Table 4. These results show important differences in the capital intensity, unit labour costs, SIZE 2 and innovation variables for the two groups of innovating and non-innovating firms. Ž1. The capital intensity variable, KSA, is positively and significantly related to exports for all the firms, but the size of the coefficient is larger for the innovating firms. As technical change is embedded in capital equipment, so that capital embodies past innovations, we would expect it to increase the propensity to export of the innovating firms more than the non-innovators. Ž2. The quadratic term of the size variable is negative for both the groups but not always significant for the non-innovators and with a lower coefficient. Unlike the probit model already discussed, the negative relationship between increasing size and the level of exports affects innovating firms more than non-innovators. Overall, however, the inverted Ushaped relationship between exports and firm size can still be seen. Ž3. While the average wages variable is positively related to exports for both groups of firms, there is a noticeable difference in the unit labour cost variable. The latter is positive and significant for the innovating firms, and negative and significant for the noninnovators, as with the probability of exporting. The explanation advanced for the probability of exporting Žthat there are innovative firms choosing not to export. can no longer apply as we are considering only the sub-group of firms which export. It appears

K. Wakelinr Research Policy 26 (1998) 829–841

that innovative firms’ exports are less-sensitive to cost than those of non-innovative firms. One explanation for this result is that innovating firms export higher quality goods and are thus less likely to be adversely affected by high unit labour costs. The skill-effect for the average salary variable is now found for both groups of firms, indicating that the negative relationship found for innovating firms in Table 3 is due to the firms which do not export. These firms do not appear to be not exporting due to lack of skills, or innovations, or excessively high unit labour costs; rather their decision to stay in domestic markets may be a question of preference. For the larger innovative firms which do export, high unit labour costs are clearly no deterrence to success in export markets, perhaps as these large firms compete on characteristics other than price. Unlike the probit model the choice of innovation variable affects the results. The production of innovations ŽPROD. is not significant for either group. The production of innovations appears to increase the probability of exporting but does not affect the propensity to export. Sector R & D expenditure, on the other hand, is negative and significant for all firms. The positive spillover effects found when considering the probability of exporting are no longer present. On the contrary, the R & D expenditure of other firms seems to indicate rivalry between firms in terms of competition. The use of innovations is also negatively related to the propensity to export for the non-innovating firms. There is no evidence of positive spillovers of innovation on export performance. The sensitivity of the results to the innovation variable used indicates that the results of studies which rely exclusively on R & D data Žsuch as those of Hirsch and Bijaoui, 1985; Kumar and Siddharthan, 1994. may differ from those presented here and in other papers using actual innovation counts. To summarise, the models give some consistent results. Innovating firms should be studied separately from the non-innovating firms, as the determinants of the probability of exporting and the propensity to export vary significantly for these firms. Unit labour costs have a negative effect on non-innovating firms’ export behaviour, but have a positive one on innovating firms’ exports, indicating that the former are more price sensitive than the latter. There seems to be a positive role for skills in affecting the export

839

behaviour for all firms, and firm size also plays an important role in exports.

6. Conclusions This paper has analysed the role of firm specific characteristics in influencing trade behaviour at the firm level. Particular emphasis has been placed on the innovation characteristics of the sector, and the firm, and the impact this has on firm behaviour. By considering the relationship between trade and innovation at the firm level, the paper has a number of advantages over more aggregate studies. It is only at the firm level that the influence of firm characteristics can be separated from those of the sector. By balancing the two groups of firms by their sectors of origin, the analysis abstracts from differences in sectors and concentrates on the role of firm characteristics. Examining trade behaviour at the firm level allows an assessment of the diversity among firms, and particularly between innovating and non-innovating firms. As the results show, there are considerable differences in the reactions of innovating and non-innovating firms, and in the determinants of their export behaviour. One of the main results to emerge from the analysis is that innovating and non-innovating firms behave differently both in terms of the probability of exporting and the level of exports. This implies that the capacity to innovate fundamentally changes the behaviour of the firm. Given their size, innovating firms are less likely to enter export markets than non-innovating firms, as shown by the descriptive statistics and the sign on the dummy variable for being an innovator. Large innovative firms are likely to export, and the more innovations they have had, the higher the probability that they will enter export markets. However, smaller innovative firms with one or two innovations are less likely to export, and more likely to service the domestic market alone than the equivalent non-innovative firms. The most likely explanation for this is that the cost of entering export markets is higher for smaller firms, which thus prefer to stay in domestic markets. In the case of innovative firms they may have advantages in that domestic market which reduces the necessity to search out new markets. As

840

K. Wakelinr Research Policy 26 (1998) 829–841

the same relationship is not found for larger more innovative firms this may relate to the fixed costs of entering foreign markets which are higher for smaller firms. The implications of the results for government policy are considerable. First, it appears that the production of innovations at the sector level improves the probability of all firms exporting, both innovative and non-innovative. The innovative environment can thus act as an encouragement to export. This relationship is not confirmed for the propensity to export, where no positive spillovers were found. Second, low unit labour costs appear to play little role in export behaviour at the firm level for innovating firms. Innovating firms with higher unit labour costs are both more likely to export and have a higher propensity to export. In addition, higher average wages have a positive impact on export behaviour for both innovating and non-innovating firms, possibly reflecting the importance of skills in export behaviour.

Acknowledgements I would like to thank Nigel Pain, Martin Weale and two anonymous referees for their helpful comments.

The sector level R & D expenditure came from First Release, CSO Number 188, December 1993, ‘Business Enterprise Research and Development 1992’. Manufacturing was defined as between 2100 and 4992 of the 1968 SIC classification. Sector Ž1980 SIC classification.

Number of firms 18

Metal manufacturing and goods Ž22 and 31. Non-metallic manufacturing 12 Ž24. Chemical and man-made 32 products Ž25 and 26. Mechanical engineering Ž32. 100 Office and data machinery Ž33. 10 Electrical and electronic 54 machinery Ž34. Transport Ž35 and 36. 22 Instruments Ž37. 12 Food, textiles, leather, footwear, 40 timber, paper and printing Ž41–47. Rubber, plastics, other manufacturing 20 9Ž48 and 49. Total 320

References Appendix A. Data sources Data on innovations come from ‘Innovation in the UK since 1945’ Science Policy Research Unit, University of Sussex, data obtained from the ESRC archive, Essex. Additional firm balance sheet data, total sales, exports, total capital, average remuneration, the number of employees came from two sources: Datastream provided the data for the quoted firms Žincluding firm level R & D expenditure., and ICC the data on non-quoted firms via Datastream. The sector level data, total UK manufacturing output, sector output, the number of enterprises per sector, the sector concentration ratio, all came from the Central Statistical Office, Report on the Census of Production 1991 Summary Volume, PA 1002, from Business Monitor.

Anderton, R., 1992. UK exports of manufactures: testing for the effects of non-price competitiveness using stochastic trends and profitability measures. The Manchester School LX Ž1., 23–40. Blundell, R., Griffith, R., van Reenen, J., 1995. Dynamic count data models of technological innovation. Econ. J. 105, 333– 344. Buxton, T., Mayes, D., Murfin, A., 1991. UK trade performance and R&D. Econ. Innovation New Technol. 1, 243–256. Cragg, J., 1971. Some statistical models for limited dependent variables with application to the demand for durable goods. Econometrica 39, 829–844. Deardorff, A., Ž1984.. Testing trade theories and predicting trade flows, In: Jones, R., Keen, P. ŽEds.., Handbook of International Economics, Vol. 1. Dosi, G., 1984. Technical Change and Industrial Transformation, London, Macmillan, London. Dosi, G., 1988. Sources, procedures and microeconomic effects of innovation. J. Econ. Lit. XXVI, 1120–1171. Fagerberg, J., 1988. International competitiveness. Econ. J. 98, 355–374.

K. Wakelinr Research Policy 26 (1998) 829–841 Freeman, C., 1982. The Economics of Industrial Innovation, 2nd edn. Francis Pinter, London. Geroksi, P., 1991. Innovation and the sectoral sources of UK productivity growth. Econ. J. 101, 1438–1451. Geroski, P., Machin, S., 1993. Innovation, profitability and growth over the business cycle. Empirica 20, 35–50. Greenhalgh, C., 1990. Innovation and trade performance in the UK. Econ. J. 100, 105–118. Greenhalgh, C., Taylor, P., Wilson, R., 1994. Innovation and export volumes and prices, a disaggregated study. Oxford Econ. Papers 46, 102–134. Hirsch, S., Bijaoui, I., 1985. R&D intensity and export performance: a micro view. Weltwirtschaftliches Archiv. 121, 138– 251. Kumar, N., Siddharthan, N.S., 1994. Technology, firm size and export behaviour in developing countries: the case of Indian enterprise. J. Dev. Studies 32 Ž2., 288–309. Landesmann, M., Snell, A., 1989. The consequences of Mrs. Thatcher for UK manufacturing exports. Econ. J. 99, 1–27. Lin, T., Schmidt, P., 1984. A test of the Tobit specification against an alternative suggested by Cragg. Rev. Econ. Stat. 66, 174–177. Oulton, N., 1996. Workforce skills and export competitiveness: an

841

Anglo–German comparison. In: Booth, A.L., Snower, D.J. ŽEds.., Acquiring Skills: Market Failures, Their Symptoms and Policy Responses. Cambridge Univ. Press, Cambridge. Pavitt, K., Robson, M., Townsend, J., 1987. The size distribution of innovating firms in the UK: 1945–1983. J. Ind. Econ. XXXV Ž3.. Robson, M., Townsend, J., Pavitt, K., 1988. Sectoral patterns of production and use of innovations in the UK: 1945–1983. Res. Pol. 17, 1–14. Rosenberg, N., 1976. Perspectives on Technology. Cambridge Univ. Press, Cambridge. Rosenberg, N., 1982. Inside the Black Box. Cambridge Univ. Press, Cambridge. Thirlwall, A., 1986. Balance of Payments Theory and the UK Experience, 3rd edn. Macmillan, London. Verspagen, B., Wakelin, K., 1997. Trade and technology from a Schumpeterian perspective. Int. Rev. Appl. Econ. 11 Ž2., 181–194. Wakelin, K., 1997. Trade and Innovation: Theory and Evidence. Edward Elgar, Aldershot. Willmore, L., 1992. Transnationals and foreign trade: evidence from Brazil. J. Dev. Studies 28 Ž2., 314–335.