Are Knowledge Spillovers Important for Productivity in Irish Firms?

Size: px
Start display at page:

Download "Are Knowledge Spillovers Important for Productivity in Irish Firms?"

Transcription

1 Are Knowledge Spillovers Important for Productivity in Irish Firms? Justin Doran * Department of Economics, University College Cork Ireland Abstract This paper analyses whether knowledge spillovers play an important role in determining the performance of Irish firms. This is accomplished through the use of a knowledge augmented production function. The Irish Innovation Panel, a firm level dataset, is utilised which contains information on firms performance and research and development efforts. Spillovers are weighted according to the geographic proximity of firms in order to allow for the decaying effectiveness of transmitting tacit knowledge across distance; with the unit of geographic measurement being the county level. The results indicate that knowledge spillovers play an important role in explaining the performance of Irish firms. Keywords: Spillovers, Productivity, Geographical Proximity, Knowledge Production Function JEL: O3, D2, R1 * justin.doran@ucc.ie Telephone number:

2 1. Introduction This paper analyses whether geographically bounded knowledge spillovers play an important role in explaining the performance of Irish firms. The data set used is the Irish Innovation Panel (IIP); consisting of three surveys covering the innovation activity of firms over the time period 1996 to In an Irish context, while there has been a number of studies estimating the importance of business linkages for innovation and productivity such as Roper (2001), Jordan and O Leary (2008) and Roper et al. (2008), no study to date has focused on empirically measuring the effect of pure knowledge spillovers on Irish firm performance. This paper utilises a measure of knowledge spillovers similar to that suggested by Griliches (1992) while also weighting the effectiveness of these spillovers by the spatial proximity of firms (Aiello and Cardamone 2009). The importance of knowledge spillovers for economic growth and development are highlighted in the endogenous growth models of Romer (1986) and Lucas (1988). These papers emphasise how increasing returns to scale can be achieved by internalising the research and development (R&D) of knowledge conducted by entities external to the business. Jaffe (1986) notes that these spillover effects can have direct benefits for firms productivity should they be able to internalise pertinent knowledge originating in other enterprises. However, Audretsch and Feldman (2004) note that the transmission of knowledge spillovers may be bounded geographically; with the effectiveness of the spillovers decaying as distance increases. This may result from an inability to transmit tacit knowledge over longer distances due to the inability to codify or formalise this form of knowledge (Jacobs 1969). While codified knowledge can be transmitted easily over large spatial areas, tacit knowledge is best transmitted over short distances through frequent face-to-face interaction (Von Hipple 1994; Nonaka et al. 2001). Therefore, as the distance between economic actors increases the effectiveness of the transfer of tacit knowledge from one to the other decreases. Thus, it is important to consider the diminishing effectiveness of firms to assimilate knowledge spillovers as the spatial distance between actors increases. This paper considers, for the Irish case, whether geographically bounded knowledge spillovers contribute to firm performance and productivity. A knowledge augmented production function is employed to estimate the effective return from internal knowledge generation and external knowledge spillovers (Griliches 1979, 1992; Jaffe 1986). The extent to which knowledge spillovers are geographically constrained is modelled using an approach similar to Aiello and Cardamone (2009), where spillovers are weighted by the spatial distance between economic actors. The IIP provides data on firms located in the 26 counties of the Republic of Ireland and it is possible to identify the county of origin of the firm, however, the exact location of the firm within the county is unknown. Therefore, the distance between firms is proxied for by the Euclidean distance between the capital cities of the relevant counties. The main contributions of this paper are threefold. Firstly, it uses Irish microlevel data at the county level; providing for a highly disaggregated level of geographic analysis. Secondly, it directly measures the effects of geographically 2

3 bounded knowledge spillovers on firm performance using both turnover and value added. Finally, it provides a number of important insights into the way in which the geographic limitations of knowledge spillovers may result in the concentration of economic activity in a small number of locations. The remainder of this article is structured as follows. The next section presents a discussion of the theoretically underpinnings of knowledge spillovers and the extent to which they may be geographically constrained. This is followed by an outline of the methodology employed by this paper. The key variables utilised from the Irish Innovation Panel are then described. Following this, the empirical results are presented and discussed. The final section concludes. 2. Literature Review 2.1 Knowledge Spillovers Knowledge spillovers have been identified by endogenous growth theory as being a key determinant of economic growth (Romer 1986; Lucas 1988). The availability of knowledge spillovers allows firms to achieve increasing returns to scale; thereby, providing an important mechanism through which firms can improve their performance. This can be facilitated through the internalization of knowledge generated by other enterprises outside of the business (Kafouros and Buckley 2008; Aiello and Cardamone 2005). Firms engage in research and development (R&D) in order to generate new economic knowledge which can be exploited in order to improve their performance (Griliches 1992). However, firms do not simply rely on their own knowledge generation, but also the knowledge generated by others. This is facilitated by the non-rivalrous nature of knowledge; which effectively means that the use of knowledge by one firm does not preclude the use of that knowledge by another (Bernstein and Nadiri 1988; Los and Verspagen 2000). This occurs due to the inability of firms to fully appropriate the knowledge which they generate (Griliches 1979). The impact of these spillovers on a business can be diverse. Through the internalisation of these spillovers firms may be able to generate higher levels of innovation output, reduce their costs or increase their productivity (Griliches 1979; Jaffe 1986; Bernstein and Nadiri 1988; Kafouros and Buckley 2008; Jacobs 1969). A distinction is required, however, between pure knowledge spillovers and rent spillovers which may occur (Los and Verspagen 2000). Rent spillovers are embodied in traded goods. For example, the sale of new technology from one industry to another (Griliches 1979). Pure knowledge spillovers are related to the aforementioned non-rivalrous nature of knowledge. Los and Verspagen (2000) identify the most important classification of a pure knowledge spillovers is that the knowledge is transferred from one firm to another without the receiver having to pay for the knowledge directly. This paper focuses on identifying the potential benefits of these pure knowledge spillovers. 3

4 2.2 The Geography of Knowledge Spillovers While spillovers occur due to the inability of firms to fully appropriate knowledge, the extent to which they can cross geographic space may be constrained (Feldman 1999). Greunz (2004), Audretsch and Feldman (1996) and Jaffe et al. (1993) note that in the growth literature there is an assumption that knowledge flows take place more readily within a country than between countries. They develop this further by noting that this may hold for smaller geographic classifications also. Indeed, Breschi and Lissoni (2001) argues that Jaffe et al. (1993) geographical analyses, based on US State level data, remains too broad; suggesting that finer levels of geographic measurement are required. This is supported by Glaesar et al.(1992) assertion that ideas must first cross hallways before they can cross oceans or continents; suggesting that proximity facilitates the absorption of knowledge spillovers. Cincera (2005) and Feldman (1999) also highlight the importance of considering this geographical component of knowledge spillovers. Krugman (1991) identifies Marshall s (1920) triad as one of the key underpinning theories of why knowledge spillovers are geographically constrained. This triad is comprised of; economies of specialisation, labour market economies and knowledge spillovers. All of these are based on the assumption of geographic concentration in a given location, and that the benefits accrued by firms from these three factors are conditional on their collocation within the same geographical area as other firms. However, Breschi and Lissoni (2001) identify the first of these two to be rent spillovers with only the third classified as a pure knowledge spillovers as described by Griliches (1979, 1995). Marshall (1920) suggests that these pure spillovers occur as knowledge can flow more freely among agents located in close geographical proximity to one another due to the social bonds which develops between them, trusting relationships which can be fostered and the ability to frequently communicate with one another on a face-to-face basis. While knowledge with a low degree of tacitness can be easily codified and transmitted, it is likely that the type of knowledge most relevant for firms innovation production is highly specific and contextual and, therefore, more tacit in nature (Feldman 1999). This tacit knowledge cannot be easily codified and, as a result, is difficult to communicate. Nonaka et al. (2001) and Von Hipple (1994), in a similar way to Marshall (1920), suggest that this form of knowledge is best transferred using frequent face-to-face interaction while Adams and Jaffe (1996) suggest that the cost of transmitting this knowledge increases as distance increases. Therefore, the more tacit knowledge is the more important geographical proximity is for assimilating spillovers. 4

5 2.3 Evidence on Localised Knowledge Spillovers Numerous empirical studies have shown that the assumption of spatially decaying knowledge spillover hold. Jaffe et al. (1993) show that knowledge spillovers are geographically bounded through the use of data on patents and patent citations. They find that when registering new patents, firms are more likely to cite patents which originated in the same region as the business. This suggests that firms assimilate knowledge generated in their locality for their use in the development of further innovations. This finding is supported by Autant-Bernard (2001) who, for a sample of French regions, finds that knowledge spillovers are geographically bounded; with regions relying on the knowledge generated within their own boundaries more than that generated in neighbouring regions. Aiello and Cardamone (2009) also find that geographical proximity between actors is important in determining the extent of knowledge spillovers. Their analysis, for a sample of Italian firms, suggests that firms in close proximity to one another benefit from spillovers more than those located further apart. Similarly, Adams and Jaffe (1996) also find this type of effect for a sample of US firms. They suggest that knowledge generated by firms located in different regions and more than 100 miles apart has a lesser impact on firms than knowledge generated within the same state or less than 100 miles away. 2.4 The Irish Case In the Irish case, while a number of papers have studied the effect of linkages or formal cooperation between economic actors, to date, there has been no analysis of the effects of pure knowledge spillovers on firm level innovation or productivity. Roper (2001) provides an analysis of the effects of firm networking and internal R&D performance on the innovation performance of Irish firms. While he concludes that networking does provide an important source of knowledge for innovation, this sheds no light on the potential benefits of pure knowledge spillovers. Similarly, Jordan and O Leary (2008) provide a detailed analysis of various types of knowledge linkages, with diverse external agents, on the likelihood of Irish firms engaging in both product and process innovation. Again, while they control for R&D performance, they do not consider the potential benefits of R&D knowledge spillovers from other firms. Also, both of these papers focus exclusively on the innovation production function and do not consider the impact of knowledge on firms performance. Roper et al. (2008) provide an analysis of the effects of knowledge linkages on innovation and the subsequent impact of this innovation on firm level productivity. Their analysis focuses on the impact of internally generated innovation activity on firm level productivity. However, it is probable that external knowledge also factors into the production function of firms in the form of knowledge spillovers. This paper expands on these studies by generating a direct measure of knowledge spillovers and estimating the effects of these knowledge spillovers on firm level productivity. This provides an insight into the potential role knowledge generate in other enterprises may have on firm performance, and complements the existing literature on knowledge linkages, innovation and productivity outlined in this section. 5

6 3. Empirical Methods In order to analyse whether knowledge spillovers play a role in explaining productivity in Irish firms this paper adopts the use of a knowledge augmented production function (Griliches 1979, 1995). This is specified as: Y = f(a, K, L) (1) Where Y is firm output, A is technology, K is a measure of capital and L is a measure of labour. Technological improvements, which can be derived from the performance of R&D, can increase the productivity of firms. However, as noted earlier, economic knowledge is non-rivalrous in nature and the returns can not be fully appropriated by the originating firm. The knowledge which is generated by one firm performing R&D may spill over to others, who may be able to internalise some, or all, of this knowledge for their own productivity gains (Aiello and Cardamone 2005; Audretsch and Vivarelli 1996; Griliches 1992). Therefore, new economic knowledge can be disaggregated into two components; that generated internally in the firm through the performance of its own R&D and that which spills over from other firms R&D efforts. Griliches (1979) and Jaffe (1986) suggests that a crude way to measure spillovers from one firm to another is the summation of the R&D performed by all other firms. This can be represented as: K = R & (2) i D j j j i Which suggests that the spillovers, K i, received by firm i are equal to the sum of the R&D performed by all other firms. This is postulated as the research performed by other firms may be appropriated by firm i assuming that the knowledge which is generated by other businesses in non-rivalrous in nature. However, the assumption that all knowledge generated by other firms will be readily available to firm i is unrealistic. Therefore, as suggested in the literature, there is a need to weight spillovers by some mechanism in order to capture the limited ability of firms to internalise knowledge generated by other businesses (Jaffe 1986; Griliches 1979; Aiello and Cardamone 2005). As discussed earlier one of the key concepts surrounding knowledge spillovers is that they may be geographically bounded (Autant-Bernard 2001; Krugman 1991; Almeida and Kogut 1997). Therefore, one possible weighting mechanism for knowledge spillovers is the distance between firm i and j. Firms which are closer to one another are more likely to engage in face-to-face interaction and as a result transfer tacit knowledge more easily (Nonaka et al. 2001; Glaesar et al. 1992). Thus, following from Aiello and Cardamone (2009) and Beise and Stahl (1999), this paper uses the geographic distance between firms to weight knowledge spillovers. The Irish Innovation Panel (IIP) provides information on the county in which each firm is located but does not provide the precise location of the firm. As the 6

7 location of the firm is only identifiable at the county level, the distance from a firm in one county to another is proxied for by the distance between the capital city/town of the counties. Once the distance between firms is computed, an index of geographical proximity is derived as: g ij d ij = 1 (3) max[ d ] ij Where g ij is the index of the distance between two firms, d ij is the distance between the two firms i and j and max[d ij ] is the maximum distance which can exist between two firms (Aiello and Cardamone 2005). Using this formula a spatially weighted matrix is generated with the following properties: w ij g M = M gi 1,1,1 L O L L O L g1 j M M g ij (4) Where the distance between each firm and every other firm in the sample is included in the matrix w ij. Due to the manner in which the index is calculated the value for firms located within the same region is unity, given by d ij =0, and zero when far apart, given as d ij = max[d ij ]. This is then included in the spillover equation (2) as a weight; resulting in (5): S = w R & D (5) i j j i ij j This specification of spillovers provides a greater weighting to the knowledge spillovers generated by firms in close geographical proximity to firm i. This is consistent with the theory discussed earlier which indicates that knowledge spillovers are spatially bounded, with firms being able to internalise knowledge from close geographical neighbours more easily than from more distant neighbours (Jaffe et al. 1993). As a result of this distinction between internal R&D generation (A) and spillovers from neighbouring firms (S), it is possible to rewrite the production function for empirical estimation as: Y it α β β β λ β εi = e K it Lit Ait Sit Zit e (6) Where all variables are defined as above and in addition α 0 is an intercept term, ε i is an error term, Z it represents a vector of control variables which may impact on a firms productivity and the subscripts i and t respectively indicate firm and time. Taking the natural logarithms of equation (6) results in the logarithmic equation (7): ln Y α Z + ε (7) it= 0+ β1 ln K it+ β2 ln Lit+ β3 ln Ait+ λ1 ln Sit+ β4 it i 7

8 This equation measures the effects of spillovers on the levels of output in a firm. To assess the impact of spillovers on firm productivity we subtract the natural logarithm of labour from equation (7) resulting in: ln( Y it / Lit ) (8) = 0+ β1 ln( K / Lit ) it+ β2 ln( Ait / Lit ) + λ1 ln Sit+ β4 α Z + ν Where all variables are defined as before. The vector of control variables, Z it, contains a series of binary variables indicating the sector in which the firm operates and the age of the firm. These factors have previously been shown to also have an impact on the productivity of firms (Pavitt 1984; Cohen and Klepper 1996). As the data set contains observations for firms over three waves of the IIP survey panel estimation techniques are used to estimate equations (7) and (8). The estimation techniques utilised by this paper are fixed effects and random effects models. Both of these estimation techniques take into account the panel nature of the IIP and allow for firm level heterogeneity and variation across time periods to be controlled for. In order to determine the most efficient model for interpretation a Hausman specification test is applied (Greene 2008; Hausman 1978). The Hausman test indicates that the random effects estimator can be used 2. it it 4. The Irish Innovation Panel Data 4.1 The Irish Innovation Panel The data set utilised by this study is the Irish Innovation Panel (IIP). The IIP provides information on the innovation, technology adoption, networking and performance of manufacturing and tradable services plants in Ireland. This paper uses data from waves three to five of the IIP containing information on the innovation performance of firms over the time period 1996 to Each wave of the survey covers the innovation activities of manufacturing plants with 10 or more employees over a three year period. For manufacturing the IIP is a highly unbalanced panel reflecting firms non-response but also the closure and opening of manufacturing units over the nine year period covered by the panel used here. While the IIP contains data on firms located in both Northern Ireland and the Irish Republic, data on location is only available to the author for firms located in the Republic. Therefore, this study focuses solely on these firms (Roper and Hewitt- Dundas 2003; Roper and Anderson 2000; Roper and Hewitt-Dundas 2006). When deriving the sample of firms to be analysed using the IIP a number of factors were considered. Firstly, while the IIP contains five waves; for consistency in the measures of productivity, R&D and geographic scope only the latter three surveys can be exploited in this study; thus determining the time period 1996 to Further to this, it is necessary to restrict the sample to those firms which have performed R&D during the reference periods (Aiello and 2 The results of the Hausman tests for each estimation are presented in the relevant tables displaying the results of the estimations of equations (7) and (8). The Chi 2 and p values are presented, with the null hypothesis stating that the difference in the coefficients of the random effects and fixed effects models is not systematic. 8

9 Cardamone 2005). The exclusion of firms with no R&D expenditure is necessary, as firms which do not engage in R&D activities must be treated as not generating any new knowledge. This results due to a lack of information in traditional innovation surveys, such as the IIP and the European wide Community Innovation Survey, on other forms of internal knowledge generation apart from formal expenditure on R&D. After imposing these restrictions the data assembled is comprised of an unbalanced panel of firms engaged in R&D activity for each of the three waves of the IIP giving a total sample size of 454 firms 3. Table 1 displays the descriptive statistics for the variables used in this study. Two measures of firm performance are considered; turnover and value added. Value added is calculated as turnover minus the total material costs. Both values are adjusted to real prices using GDP deflators at 2000 level prices. The average total turnover of the firms considered is approximately 20 million with a standard deviation of approximately 70 million while the average turnover per employee is 133,240 with a standard deviation of 306,091. The average firm value added is approximately 12 million with a standard deviation of approximately 57 million. In terms of value added per employee, the average is 3,706 with a standard deviation of 1,032. [insert Table 1 around here] Given that this paper focuses on the geography of knowledge spillovers it is important to note that when considering the geographical distribution of average turnover and value added per worker, there is substantial variation across counties. This can be observed in Figure 1a and 1b. Figure 1a displays the average turnover per employee for firms based in each county with the state average equaling 100. Similarly, Figure 1b displays the average value added per employee for firms based in each county, again with the state average equaling 100. The unequal distribution of firm performance across counties raises the possibility that firms in high-performing regions may be exploiting resources not accessible to firms in other regions. [insert Figure 1 around here] Indeed, while Table 1 indicates that the average expenditure by firms per employee on R&D is 3,325 it can be observed in Figure 2 that the average R&D expenditure of firms per employee varies substantially by county. There is also a degree of concordance with Figure 1, which displays firms performance; with R&D intensive counties possessing higher levels of both turnover and value added (see appendix 1 for scatter plots of variables). This may suggest that knowledge spillovers are present and accruing to firms operating in regions where substantial new economic knowledge is being generated; contributing to the observed heterogeneity in firm performance across counties. [insert Figure 2 around here] Internal knowledge generation, and the pool of knowledge which can spillover from one firm to another, is proxied in this paper by R&D expenditure. In the IIP, 3 When estimating the value added equations this falls to 443 due to a number of firms providing no information on the material costs associated with production. 9

10 R&D is measured as a flow variable and captures the expenditure by a firm on R&D activities over a three year period; for example, the 2005 IIP survey asked firms to report their expenditure on R&D during the period While ideally the stock of R&D capital would be used in this analysis, this variable is not present in the IIP data set, nor are there a sufficient number of observations across time with which to calculate a stock measure 4. However, the use of R&D expenditure is common in the literature with Autant-Bernard (2001) noting that R&D expenditure can be satisfactorily used as a proxy for firms knowledge stock. R&D expenditure by firms is also converted to real 2000 level values using GDP deflators. Given the knowledge production functions specified in Section 3 capital and labour measures are also required. Capital is proxied for by the investment by the firm on plant, machinery and fixed assets in the year the survey is conducted. Again, while this variable is measured as a flow rather than a stock, it still provides an accurate representation of the capital intensity of the firm and its use is consistent with that of Roper et al. (2008). The average capital expenditure by firms is 1,079,492 with a standard deviation of 6,968,322 while the mean capital per employee is 7,672 with a standard deviation of 27,320. Labour is measured as the total number of employee in the firm. It can be noted that the average number of employees is 115 with a standard deviation of 200. A variable representing the age of the firm is also included to control for the differences which may exist between the productivity levels of incumbent and start up firms. The average age of firms in the IIP is 27 with a standard deviation of 28. The sector in which the firm operates in is also included to control for differences in R&D activity and firm performance across different industries Calculating the distance Between Points The exact location of the individual firms is not available, with data only indicating the county in which the firm is based. Therefore, the distance between points is calculated as the Euclidean distance between the capital city/town of each county. The Euclidean distance provides an as the crow flies distance between the points considered (Deza and Deza 2009). It is derived from the latitude and longitude coordinates of the capital cities/towns of the regions. As such, the distance between firm i and j is given as: d ( i j ) 2 + ( i j ) 2 = (9) ( ij) lat lat lon lon Where d (ij) is the distance between firm i and firm j based on the location of the capital city/town in the county they are based, lat indicates the latitude of the respective firm and lon indicates the longitude of the respective firm. 4 This is due to data only being available for three IIP surveys covering the time epriod 1996 to The sectors controlled for in all regression models are; (i) Food, Drink and Tobacco, (ii) Textiles and Clothing, (iii) Wood and Wood Products, (iv) Paper and Printing, (v) Chemicals, (vi) Metals and Metal Fabrication, (vii) Mechanical Engineering, (viii) Electrical and Optical Equipment, (ix) Transport Equipment and (x) Other Manufacturing 10

11 5. Empirical Results 5.1 Results of Equations (7) and (8) The first column of Table 2 presents the results for the estimation of equation (7) using turnover as the dependent variable. Firstly, it can be observed that the intensity with which firms perform internal R&D has a significantly positive effect. This suggests that firms internal knowledge generation efforts have a positive effect on turnover. This is as anticipated, as firms would be expected to be able to exploit new knowledge generated within their own firm to improve their performance (Griliches 1979). However, there is also evidence to suggest that knowledge spillovers are important in explaining firm turnover. The positive R&D spillover coefficient suggests that the more knowledge firms have access to from outside their enterprise the higher their turnover. This suggests that firms can benefit from knowledge generated, not only internally in the firm, but also generated by other businesses outside their enterprise. It is important to note that as this coefficient is weighted by the spatial distance between economic agents the highest benefit to a firm s turnover arises from spillovers which are spatially proximate. [insert Table 2 around here] As both the dependent and independent variables are measured as logs, the elasticity of internal R&D expenditure and R&D spillovers is given by the coefficients presented in Table 2. The elasticity of internal R&D expenditure is while the elasticity of R&D spillovers is This suggests that firms benefit more from knowledge generated by other firms than they do their own internal R&D performance; pointing to the importance of R&D spillovers for turnover. This result is consistent with Bernstein and Nadiri (1988). It can also be noted that firms with higher levels of capital investment are more productive. Similarly, firms with higher levels of employment have higher levels of turnover. These results are as expected. More capital intensive firms and larger firms would be expected to possess higher levels of turnover relative to less capital intensive or smaller firms. It can also be noted that older firms also have higher levels of turnover. The second column in Table 2 presents the results of the analysis when value added is used as the dependent variable. Again, both internal R&D expenditure and external knowledge spillovers are found to be statistically significant and positive; suggesting that both play an important role in explaining the performance of firms. However, while the elasticities vary slightly from that observed in the turnover equation, knowledge spillovers remain more important for firm performance than internal R&D. Also, as expected, capital intensity, employment size and firm age exert a positive effect on firm performance. Overall, the results presented in Table 2 suggest that firms which appropriate knowledge from other businesses have higher levels of turnover and value added. However, it is also interesting to assess whether spillovers exert a direct effect on 11

12 the productivity, i.e. the output per worker, of firms. Therefore, Table 3 presents the results of the random effects estimation of equation (8). [insert Table 3 around here] For turnover per employee, it can be observed that internal R&D expenditure and knowledge spillovers exert a positive effect on productivity. Firms which generate new economic knowledge internally or appropriate it from other firms, have higher levels of turnover per worker. Interestingly, the elasticity of knowledge spillovers is again greater than that for internal knowledge generation. This suggests that, as with levels of turnover and value added, knowledge spillovers play a more important role in explaining firm productivity relative to internal R&D. When considering value added per employee the results are similar. Both internal R&D and knowledge spillovers play an important role in explaining firm productivity with both being significant and positive. Finally, for both estimations, firms with higher levels of capital intensity per employee are more productive; with higher levels of turnover and value added per employee. Also, older firms have higher levels of productivity, perhaps representing the advantage incumbent and well established enterprises have over newer start up companies. 5.2 Comparison of Results with Other Studies The results presented in this paper for Ireland are consistent with those from the international literature. Audretsch and Vivarelli (1996), using patents as a measure of firm innovation performance, show that knowledge spillovers have a significant positive effect on the level of firm innovation while Autant-Bernard (2001) conclude that spillovers have a positive effect on the innovation output of French regions. Bernstein and Nadiri (1988, 1989) find that knowledge spillovers can reduce firm costs; thereby improving performance. While Aiello and Cardamone (2009, 2005) also find that spillovers are important in explaining the performance of Italian firms. Finally, Jaffe (1986) shows that firms whose neighbours are R&D intensive have higher levels of profitability than firms whose neighbours conduct relatively little R&D. Regarding the differences between the social and private returns of R&D Bernstein and Nadiri (1989) provide evidence to suggest that knoweldge spillovers play a greater role than internal knoweldge generation in expalining firm performance. Similarily, Los and Verspagen (2000) also find that knowledge spillovers have a higher elasticity relative to other forms of knowledge. These results are consistent with the findings presented in this paper which suggests that, for Ireland, the returns to the firm arising from spillovers are greater than the returns arising from private R&D expenditure. 6. Conclusion and Discussion This paper analyses whether knowledge spillovers have an effect on the performance of Irish firms. The results indicate that knowledge spillovers have a significant positive effect on firms performance regardless of whether firm 12

13 performance is measured as turnover or value added in levels or per employee; suggesting that new knowledge generated by firms is not fully appropriated, and that it spills over to other businesses (Griliches 1979). This is consistent with the economic theory proposed by Romer (1986), Lucas (1988) and Krugman (1991) and suggests that firms, through exploiting the knowledge generated by others, can increase their own performance. Interestingly, this paper provides evidence which suggests that the elasticity of knowledge spillovers is greater than that of internal knowledge generation. This suggests that firms benefit more from an increase in the R&D effort of other firms than they do from an increase in their own (Bernstein and Nadiri 1989). However, when discussing this factor it must be born in mind that as spillovers are weighted geographically, the benefit accruing to a firm from the spillovers of others is also determined by the proximity of the entities. Therefore, while knowledge spillovers may have a higher elasticity than internal knowledge generation, depending on where the knowledge spillover originates the firm may receive the full benefit, if it is geographically proximate, or practically none of the benefit, if it is geographically distant. Given the potential that firms can benefit more from the R&D efforts of others than their own R&D expenditure, this raises the question as to why firms engage in internal R&D and a free rider effect is not observed; with firms solely exploiting the knowledge of others (Bernstein and Nadiri 1989). Cohen and Levinthal (1989) propose that the reason a free rider effect is not observed is due to the need for firms to develop their internal absorptive capacity in order to be able to exploit knowledge spillovers. Therefore, firms which do not invest in their own R&D will be unable to fully assimilate knowledge from external sources outside the business and exploit it for performance gains. This proposition raises an interesting point worthy of future research; namely, whether geographical proximate and distant knowledge spillovers are constrained by the internal absorptive capacity of firms. This may entail testing for complementarities between internal knowledge generation and knowledge spillovers. However, this is beyond the scope of this paper, which sets out to establish whether spillovers exist in the Irish context. As was noted earlier in this paper the Irish context for geographically bounded spillovers is particularly interesting given the uneven distribution of firm productivity and R&D expenditure displayed in Figures 1 and 2. With some counties exhibiting both high performance and R&D expenditure while some possess low levels of both. Given the finding that knowledge spillovers from R&D investment are important for firm performance it is important to consider the implications of this for the economic duality observed. As knowledge spillovers are a function of other firms expenditure on R&D, the greatest of these spillovers will occur in regions where R&D expenditure is high. Given that this expenditure is concentrated in a relative small number of counties, and that the efficiency of spillovers deteriorates as distance increases, only firms within these counties or neighbouring them will benefit substantially. Also, firms located in low R&D intensive counties will receive relatively low levels of spillovers. This suggests that the most productive regions will continue to benefit 13

14 from intensive knowledge spillovers while the lower productive regions will continue to stagnate due to an inability to fully realise the potential benefits from the spillovers generated in more R&D intensive regions. 7. Acknowledgments The author would like to thank Professor Stephen Roper of Warwick Business School and Dr Nola Hewitt-Dundas of Queens University Management School for access to the Irish Innovation Panel data. 14

15 References Adams, J., & Jaffe, A. (1996). The effects of r&d: An investigation using matched establishmentfirm data. The RAND Journal of Economics, 27(4), Aiello, F., & Cardamone, P. (2005). R&d spillovers and productivity growth: Evidence from italian manufacturing microdata. Applied Economic Letters, 12(10), Aiello, F., & Cardamone, P. (Eds.). (2009). R&d spillovers and firms' performance in italy: Evidence from a flexable production function (Spatial econometrics: Methods and applications). Germany: Physica-Verlag. Almeida, P., & Kogut, B. (1997). The exploration of technological diversity and the geographic localization of innovation. Small Business Economics, 9(1), Audretsch, D., & Feldman, M. (1996). R&d spillvoers and the geography of innovation and rroduction. The American Economic Review, 86(3), Audretsch, D., & Feldman, M. (Eds.). (2004). Knowledge spillvoers and the geography of innovation (Vol. Volume 4, Handbook of regional and urban economics): Elsevier Audretsch, D., & Vivarelli, M. (1996). Frim size and r&d spillvoers: Evidence from italy. Small Business Economics, 8(3), Autant-Bernard, C. (2001). The geography of knowledge spillovers and technological proximity. Economics of Innovation and New Technology, 10(4), Beise, M., & Stahl, H. (1999). Public research and industrial innvoations in germany. Research Ploicy, 28(4), Bernstein, J., & Nadiri, I. (1988). Interindustry r&d spillovers, rates of return and production in high-tech industries. American Economic Review, 78(2), Bernstein, J., & Nadiri, I. (1989). Research and development and intra-industry spillovers: An empirical application of dynamic duality. Review of Economic Studies, 56(2), Breschi, S., & Lissoni, F. (2001). Knowledge spillovers and local innovation systems: A critical survey. Industrial & Corporate Change, 10(4), Cincera, M. (2005). Firms' productivity growth and r&d spillovers: An analysis of alternative technological proximity measures. Economics of Innovation and New Technology, 14(8), Cohen, W., & Klepper, S. (1996). A reprise of size and r&d. Economic Journal, 106(437), Cohen, W., & Levinthal, D. (1989). Innovation and learning: The two faces of r&d. The Economic Journal, 99(397), Deza, E., & Deza, M. (2009). Encyclopedia of distances. Germany: Springer. Feldman, M. (1999). The new economics of innovation, spillovers and agglomeration: A review of empirical studies. The Economics of Innovation and New Technology, 8(1&2), Glaesar, E. L., Kallal, H. D., Scheinkman, J. A., & Shleifer, A. (1992). Growth in cities. Journal of Political Economy, 100(6), Greene, W. (2008). Econometric analysis. United States: Pearson-Prentice Hall. Greunz, L. (2004). Industrial structure and innovation: Evidence from european regions. Journal of Evolutionary Economics, 14(5), Griliches, Z. (1979). Issues in assessing the contribution of research and development to productivity growth. Bell Journal of Economics, 10(1), Griliches, Z. (1992). The search for r&d spillovers. The Scandinavian Journal of Economics, 94(Supplement), Griliches, Z. (Ed.). (1995). R&d and productivity: Econometric results and measurement issues (Handbook of economics of innovation and technological change). Oxford: Blackwell. Hausman, J. (1978). Specification tests in econometrics. Econometrica, 46 (6), Jacobs, J. (1969). The economy of cities. New York: Random House. Jaffe, A. (1986). Technological opportunity and spillovers of r&d: Evidence from firms' patents, profits and market value. The American Economic Review, 76(5), Jaffe, A., Trajtenberg, M., & Henderson, R. (1993). Geographic localization of knowledge spillovers as evidenced by patent citations. The Quarterly Journal of Economics, 108(3), Jordan, D., & O Leary, E. (2008). Is irish innovation policy working? Evidence from irish hightechnology businesses. Journal of the Statistical and Social Inquiry Society of Ireland, 37, Kafouros, M., & Buckley, P. (2008). Under what conditions do firms benefit from the research efforts of other organizations? Research Policy, 37(2), Krugman, P. (1991). Geography and trade. Cambridge; MA: MIT Press. Los, B., & Verspagen, B. (2000). R&d spillovers and productivity: Evidence from u.s. Manufacturing microdata. Empirical Economics, 25(1),

16 Lucas, R. (1988). On the mechanics of economic development. Journal of Monetary Economics, 22(1), Marshall, A. (1920). Principles of economics (8th Edition ed.). London: MacMillan. Nonaka, I., Toyama, R., & Konno, N. (Eds.). (2001). Seci, ba and leadership: A unified model of dynamic knowledge creation (Managing industrial knowledge: Creation, transfer and utilisation). Thousand Oaks, Ca: Sage. Pavitt, K. (1984). Sectoral patterns of technical change: Towards a taxonomy and theory. Research Policy, 13(6), Romer, P. (1986). Increasing returns and long run growth. Journal of Political Economy, 94(5), Roper, S. (2001). Innovation, networks and plant location: Some evidence from ireland. Regional Studies, 35(3), Roper, S., & Anderson, J. (2000). Innovation and e-commerce: A cross-border comparison of irish manufacturing plants. Northern Ireland Economic Research Centre - Report Series. Roper, S., Du, J., & Love, J. (2008). Modeling the innovation value chain. Research Policy, 37(6-7), Roper, S., & Hewitt-Dundas, N. (2003). Innovation, best practice adoption and innovation networks: A comparison of northern ireland and the republic of ireland. IIP Research Report. Roper, S., & Hewitt-Dundas, N. (2006). Irish innovation panel wave 5 survey report. IIP Research Report. Von Hipple, E. (1994). Stickey information and the locus of problem solving: Implications for innovation. Managment Science, 40(4),

17 Table 1: Descriptive Statistics Mean Std. dv. Firm Performance Turnover ( ) 19,938,150 70,266,400 Value Added ( ) 11,706,370 57,460,540 Turnover per Employee ( ) 133, ,091 Value Added per Employee ( ) 3,706 1,032 Physical Capital Capital ( ) 1,079,492 6,968,322 Capital per Employee ( ) 7,672 27,320 R&D R&D Expenditure ( ) 475,944 2,584,065 R&D Expenditure per Employee ( ) 3,329 10,927 Labour Firm Age Source: the Irish Innovation Panel 17

18 Table 2: Estimates of Equation (7) Turnover VA Constant (0.9854) (1.0170) Internal R&D *** *** (0.0276) (0.0271) R&D Spillovers ** * (0.0848) (0.0878) Physical Capital *** *** (0.0289) (0.0285) Employment *** *** (0.0518) (0.0503) Firm Age *** *** (0.0487) (0.0470) Hausman Chi P > Chi Obs Groups Wald Chi Prob > Chi Note a: *** significant at the 1% level, ** significant at the 5% level, * significant at the 10% level. b: Random effects estimator used as indicated by the Hausman test statistic. c: All variables expressed in natural logarithms d: Dummy variables are included in the model to control for sector specific effects. These are not presented in the table. 18

19 Table 3: Estimation of Equation (8) Turnover per Worker VA per Worker Constant (0.9789) (1.0148) Internal R&D *** *** (0.0274) (0.0269) R&D Spillovers ** * (0.0850) (0.0883) Physical Capital *** *** (0.0289) (0.0285) Firm Age *** *** (0.0481) (0.0465) Hausman Chi P > Chi Obs Groups Wald Chi Prob > Chi Note a: *** significant at the 1% level, ** significant at the 5% level, * significant at the 10% level. b: Random effects estimator used as indicated by the Hausman test statistic. c: All variables expressed in natural logarithms d: Variables expressed in per employee terms. e: Dummy variables are included in the model to control for sector specific effects. These are not presented in the table. 19

20 Fig. 1 Turnover and Value Added per Employee by County for 2005 Fig 1a Turnover per Employee (National Average = 100) Fig 1b Value Added per Employee (National Average = 100) > 118% National Average 83% - 117% National Average 59% - 82% National Average < 58% National Average > 130% National Average 88% - 129% National Average 66% - 87% National Average < 65% National Average 20

21 Fig. 2 Average R&D Expenditure per Employee by Firms (National Average = 100) > 133% National Average 56% - 133% National Average 34% - 55% National Average < 33% National Average 21

22 Appendix 1 Fig. 3 Scatterplot with Fited Vales of Performance against R&D Fig. 3a Turnover per Employee and R&D Expenditure per Employee (State Average = 100) Fig. 3b Value Added per Employee and R&D Expenditure per Employee (State Average = 100) Turnover per Employee Value Added per Employee R&D Expenditure per Employee R&D Expenditure per Employee Turnover per Employee FV Value Added per Employee FV 22