Bio-based and Applied Economics 7(3): 233-247, 2018 ISSN 2280-6180 (print) © Firenze University Press ISSN 2280-6172 (online) www.fupress.com/bae Full Research Article DOI: 10.13128/bae-7677 Corporate R&D and the performance of food-processing firms: Evidence from Europe, Japan and North America HeinricH Hockmann1, Pedro andres Garzon delvaux2,*, Peter voiGt3, Pavel ciaian2, serGio Gomez y Paloma2 1 Leibniz Institute of Agricultural Development in Central and Eastern Europe (IAMO), Germany 2 Joint Research Centre (JRC), European Commission, Spain 3 DG ECFIN, European Commission, Belgium Date of submission: xxxx xxth, xxxxxxxxxx; accepted xxxx xxrd, xxxxxxxx Abstract. This paper investigates the impact of corporate research and development (R&D) on firm performance in the food-processing industry. We apply Data Envel- opment Analysis (DEA) with two step bootstrapping using a corporate data for 307 food-processing firms from the EU, US, Canada and Japan for the period 1991-2009. The estimates suggest that R&D has a positive effect on the firms’ performance, with marginal gains decreasing in the R&D level as well as the performance differences are detected across regions and food sectors. R&D investments in food processing can deliver productivity gains, beyond the high-tech sectors generally favoured by innova- tion policy. Keywords. Corporate R&D, DEA, double bootstrapping, food-processing industry. JEL codes. O30, L66. 1. Introduction Both the theoretical and empirical literature established that R&D is critical for firm productivity growth. For example, the empirical literature has found that between 1% and 25% of variance in the actual productivity across firms can be explained by differences in R&D investment (Hall et al., 2010). However, there is considerably less agreement on the size of the R&D impact on the firm’s productivity (e.g. the size of marginal impact, dimin- ishing vs. increasing returns to R&D). Existing analysis of the implications of R&D mainly focus on knowledge-intensive businesses; there are less studies covering R&D and innovation in low- medium-tech sec- tors such as food-processing. The literature is highly scattered in the field of agro-food sector ranging from conceptual analysis, system-oriented approach analysis (e.g. Jongen and Meulenberg, 2005; OECD, 2012, 2013) to public R&D in agro-food sector (Alston, *Corresponding author: pedro.garzon-delvaux@ec.europa.eu 234 Heinrich Hockmann et alii 2010). Analyses on public R&D and its impact on primary agriculture production are more numerous given that the relevant data is more accessible. Conversely, much less effort has poured into the private R&D even though it probably represents the largest share of the overall sector’s R&D (e.g. 59% in Japan, 51% in US according to Alston et al. 2010). Furthermore, the firm level studies seldom focus on specific aspects of R&D (e.g. adoption, product variety). Most are case studies with a limited regional or sectorial cov- erage (e.g. one country, part of the sector). Broader quantitative analyses are limited by data measurement and availability constraints. The food-industry is usually considered to be a medium to low R&D intensity sec- tor representing around 0.27 % of the total output in the EU agro-food industry (Food- DrinkEurope, 2015) compared to other sectors such as the automobile (5.5%) or phar- maceutical (13.1%) industries (Hernández et al. 2015). This is understood, among others, to be related to the fact that the agro-food sector is dominated by SMEs which do little research, many innovations are often derived from other input sectors and thus are incor- porated in machinery, packaging and other manufacturing supplies (e.g. Menrad, 2004) as well as many food-products are rather easy to imitate with significant R&D spillovers which reduces firms’ incentive to invest in R&D (Gopinath & Vasavada, 1999). Although this general patterns may hold, the agro-food industry shows a high het- erogeneity in the R&D intensity (Avermaete et al., 2003; Winger and Wall, 2006; Feigl and Menrad, 2008; Capitanio et al., 2010). There is a strong geographic heterogeneity in the level of private R&D. Heterogeneity is also present in the type of innovation among firms: process, product, or organisational innovation. Finally, it is important to mention that firms also differ whether they invest in R&D externally or internally. The objective of this paper is to contribute to this literature by providing empirical evidence on the impact of private (corporate) R&D on productivity of food-processing firms. More specifically, we analyse the size of firm inefficiency and explore the determi- nants of the inefficiency against the frontier production function using a unique corporate data set of food-processing firms from the EU, US, Canada and Japan for the period 1991- 2009. To derive productivity parameters, we apply Data Envelopment Analysis (DEA) with two step bootstrapping which allows us to correct the bias in (in)efficiency and gen- erate unbiased estimates for (in)efficiencies. 2. Methodology To estimate the impact of private R&D on firm productivity we adopt a two-step approach. First, we use DEA to estimate firm performance (inefficiencies). Second, we run regression to explain the determinants of firm inefficiencies on a set of explanatory vari- ables including private R&D. Different approaches have been applied in the literature to identify production fron- tiers using both parametric and non-parametric methods. Here we adopt a non-paramet- ric approach - DEA with two step bootstrapping (Simar and Wilson, 2007). The advantage of DEA is that it does not require imposing assumption on the functional form of the frontier, there are no restrictions regarding the number of parameters required, it is rela- tively easy to deal with a whole range of inputs and outputs, and inputs and outputs can have very different units. However, in general, some limitations remain in terms of con- 235Corporate R&D and the performance of food-processing firms sidering time series, sensitiveness to outliers, demanding to incorporate (nonparametric) statistical inference, etc. Methodologically, however, the assumption of a common frontier across countries and sectors is a sensitive issue potentially leading to biased results (Koop et al., 2000; Limam and Miller, 2004; Orea and Kumbhakar, 2004). This paper avoids assuming a com- mon technology across sectors by estimating at industry-specific technology level. A frontier production function, in general, defines the maximum output achievable, given the current production technology and available inputs. We estimate DEA model in the formulation of output distance function: δ̂ i = δ i(X,Y |T ) = max δ > 0 |δyi ≤ Yλ, xi ≤ Xλ, i'λ = 1{ } (1) where δi is inefficiency parameter of firm i, yi is output; δyi is maximum output achievable (frontier), xi and X are inputs; λ are weights used to construct the virtual producer (fron- tier). The main idea of DEA is to find virtual firm (combination of other firms) capable of producing more output for the given inputs. In the second stage, the inefficiency parameters are regressed on a set of explanatory variables, zi, to estimate the determinants of inefficiency: δ i = ziβ + ε í ≥1 (2) where β are parameters to be estiamted and εi is an independent and identically distrib- uted (i.i.d.) error term. For estimation, δ i has to be replaced by δ̂ i (the estimated efficiency scores from the first stage): δ̂ i = ziβ +ξí ≥1 (3) Usually a Tobit regression is applied to estimate the parameters of β. This procedure become necessary because the error term ei is truncated and not symmetrically distributed with mean zero. Examples of the z variables – and as such also used in this study – are R&D intensity, capital intensity, time, country dummies (capturing different institutional settings), etc. Simar and Wilson (2007) point to several problems with this approach and advo- cate for the use of a truncated regression, instead. The δ̂ i are serially correlated in an unknown way since each δ̂ i depends on all observation in T. Thus the δi are not inde- pendent of each other which induces biased estimates in the second step since the usual assumption regarding the error term does not hold. Moreover, since xi and yi are correlated with zi (otherwise the second step would make no sense), zi is correlated with ξi. The correlation disappears asymptotically, how- ever, at a very slow rate. As a solution to this bias they suggest a two-step bootstrap algorithm (Simar and Wil- son, 2007). First, we correct the bias in (in)efficiency (in DEA). Second, we get unbiased 236 Heinrich Hockmann et alii estimates for (in)efficiencies (in the truncated regression). That is, the bootstrap allows to bias-adjust coefficient estimates and also for calculating proper confidence intervals for the statistical inference. Bootstrapping tends to affect the structure of the data, potentially generating other forms of bias through an ‘over-manipulation’ of the data. A possible alternative is to devel- op an instrumental variable to control for the bias. However, this alternative was not seen as operational taking in consideration the available data. 3. Data and variables Considering strengths and limitations of several potential sources of data,1 Standard & Poor’s (S&P) COMPUSTAT data set (S&P, 2014) was favoured which contains data at firm level collected from companies’ audited annual/quarterly reports. The selection process of firms from the available population of companies entailed several steps. The first consisted in retrieving firms classified as belonging to agriculture (industry code: 0xxx) as well as those to the food-industry (industry code: 2xxx); cover- ing the period 1991-2009. Data had to cover revenue, sales, net income, capital and R&D expenditures (if any); number of employees and/or wage sum, industry code, and region/ country (i.e. info on the location of the company’s headquarter/where it is registered). However, as most companies from agriculture did not report R&D expenditures, they were dropped from the final sample. Labour input is critical when considering firm performance. In the case of missing ‘number of employees’ but available labour expenditures, the number of employees was approximated by using average wage levels taken from International Labour Organisation (ILO) and for values of labour costs vice versa. The dataset does not allow distinguishing whether R&D was conducted domestically or abroad. All companies’ R&D expenditure was assigned to the country where the com- pany is registered. The DEA approach applied in this paper is sensitive to outliers. Moreover, presuming a common production frontier for companies across countries implicitly assumes that all companies have access to the same technology and produce under virtually the same tech- nological restrictions. Hence, reducing the sample to a sub-sample comprising of rather homogeneous countries/companies appeared advisable in order to ensure widely unbiased empirical results. Outlier observations, however, still need to be excluded from the sample. After carrying out a final outlier check (checking for consistency and order of magni- tude across observations as well as along the time series) some further firms/observations had to be dropped. Thus, outliers were excluded based on the results of Grubbs’ tests cen- tred on the sectoral average growth rates of firms’ R&D stock intensity (K/revenue) over 1 For instance, the AMADEUS database may contain sufficient cross-section and time series firm level data, but provides information on R&D (if at all) only for very recent years. The presumed emergence of the food-pro- cessing sector as medium-tech, evolving from formerly low-tech, could not be investigated accordingly based on such data. Another possible source of data could be the EU Industrial R&D Scoreboard (released by EC Joint Research Centre). This database comprises of fully consolidated firm level data of top R&D investors in Europe and elsewhere (year of last audited report + 3 years back in time). However, among the listed companies, there are too few belonging to the food-industry. 237Corporate R&D and the performance of food-processing firms the investigated period.2 Moreover, some further observations were dropped for reasons related to the computation of the R&D and capital stocks. In accordance with the literature (see Hulten, 1991; Jorgenson, 1990; Hall and Mairesse, 1995; Bönte, 2003; Parisi et al., 2006), stock indicators (rather than flows) were used as impact variables. It is thus implicitly assumed that a firm’s productivity is affected rather by the cumulated stocks of capital and R&D expenditures and not only by current or lagged flows.3 Accordingly, our main impact variable is a firm’s R&D stock (K) and the second impact variable is ‘capital expenditures’ (C) captured as capital stocks. By consid- ering the per capita values of these variables (i.e. per number of employees), it allows us both to standardise the data and to eliminate firms’ size effects (see, for example, Crépon et al. 1998). In this framework, knowledge (R&D) and physical capital stocks were com- puted using the perpetual inventory method based on the following formulas: Kt0 = R&Dt0 gs,c(K )+δ (4) Kt = Kt−1 ⋅(1−δ )+ R&Dt with t = 1991, … , 2009 (5) Ct0 = It0 gs,c(C)+φ j and (6) Ct = Ct−1(1−φ)+ It (7) where R&D is R&D expenditure and I is gross investment (capital expenditure). The OECD ANBERD and the OECD STAN databases were used to provide growth rates g(K) and g(C) for K and C, respectively. We computed the compounded average rates of change in R&D and fixed capital expenditures in the food-processing sector and per country (c). For some European countries the mentioned databases did not report or allowed calculating specific growth rates for R&D- and capital-stocks. The corresponding European averages were assumed in these cases instead. For the US, Canada, and Japan, however, the growth rates were taken from the literature.4 In general, different depreciation rates (δ) and (ϕ) for K and C should be assumed depending on whether the industry is high-, medium-high, medium-low/low-R&D inten- sity. In fact, more technologically-advanced sectors are characterised (on average) by short- 2 Grubbs’ test – also known as maximum normalised residual test – assumes normality (which is a desirable property anyway). Accordingly, we ran normality tests on the relevant variables (assumption was never rejected). 3 Using cumulated R&D and capital stocks – as in the previous relevant literature – overcomes a potential endo- geneity problem which can arise if flows are used. 4 For capital growth from OECD (Capital Services, total; mean percentage change 1985-2009; see: http://stats. oecd.org/Index.aspx) and for R&D growth rates the average over the period 1980-1998 was taken from (http:// www.ulb.ac.be/cours/solvay/vanpottelsberghe/resources/DGBVP_OES.pdf 238 Heinrich Hockmann et alii er product life cycles and by a faster technological progress which together accelerates the obsolescence of the current knowledge and physical capital. In this light, Ortega-Arquiles et al. (2009) suggested sectoral depreciation rates of 20%, 15% and 12% to the knowledge capi- tal and 8%, 6% and 4% to the physical capital respectively for the high, medium-high-, and medium-low/low-tech sectors, with the latter (δ=12%, ϕ=4%) to be applied here to the food- processing industry. These are similar to the 15% and 6% commonly used in the literature (Musgrave, 1986; Nadiri and Prucha, 1996; Pakes and Schankerman, 1986; Hall, 2007). All variables in monetary units were transformed into 2007 Euro using the end of year exchange rate. In cases where no direct exchange rate to Euro was provided by COM- PUSTAT, for a certain year, the corresponding currency was transferred into USD first and then into Euro. After processing the data, the sample used in this paper consists of 307 companies (2948 observations) for the period 1991-2009 registered in either of the following country groups: EU (557 observations), North America (USA and Canada, 1,050 observations), and Japan (1,341 observations), as shown in Table 1. Europe is less represented than Japanese and North-American counterparts. There is no information on Japanese firms prior to 1999 and most regions are less represented for this period. However, the period starting in 2000 is more balanced, including for Europe. To control for this data structure we use a dummy variables in our estimations to distin- guish these two periods. As shown in Table 1, there is observed significant heterogeneity among the 307 firms. The mean number of employees varies between 2211 in Japan to 15293 in the EU. Nev- Table 1 Descriptive statistics of main variables Variable Mean Std. Dev. Min Max Firms Obs. Total sample 307 2948 Revenue 2308.3 5192.3 0.4 51514 COGS-costs 1443.5 3295.6 0.4 47137 R&D expenditure 89.7 451.7 0 7290.3 Capital expend. 1286.4 2996.3 0 25846 Employees 10610 31443 2 486000 EU companies 85 557 Revenue 2705.8 6602.6 0.4 51514 COGS-costs 1561.2 3323.9 0.4 22873 R&D expenditure 175.6 926.9 0 7290.31 Capital expend. 1768.9 4020.4 0 25846 Employees 15292.7 36441.3 2 269000 US & Canada 79 1050 Revenue 3684.8 6607 1.7 50659 COGS-costs 2309.5 4578.3 1 47137 R&D expenditure 72.5 266.4 0 2476 Capital expend. 1839.9 3584.2 0 24759 Employees 18054 43375 2 486000 Japan* 143 1341 Revenue 1065.3 1983.5 5 15913 COGS-costs 716.6 1330.7 2 9785.7 R&D expenditure 67.5 181.5 0 1642.2 Capital expend. 652.6 1497.7 0 13127 Employees 2211 4203 16 36554 *(1999-2009 period only) Table 2 Sample composition - observations per subsector Subsector Codes No. observations Beverages, including alcohol 2080-2087 561 Mixed/generalist 2000 490 Prepared foods 2090-2099 491 Meat and poultry packing 2010-2015 272 Sugar and confectionery 2060-2068 252 Canned fruits and vegetables 2030-2038 225 Grain 2040-2048 226 Bakery 2050-2053 197 Dairy 2020-2026 18 Oils 2070-2079 116 Total 2948 239Corporate R&D and the performance of food-processing firms ertheless, in each macro region, apparently, there are also a number of small and even micro-companies. It has to be stressed that the final sample gathers rather large compa- nies, inherent with stock listed company data. This entails that results cannot be easily generalised as rather small private companies operating in the food-processing sector are not captured, but should be considered pertinent to large firms which, in fact, are inclined to be more active in terms of R&D. Also, this kind of “pick the winner” effect might be particularly severe in medium and low-tech sectors (like food-processing), where the overall company population tends to be dominated by smaller firms which scarcely engage in R&D investment (Becker and Pain, 2002). The sample mean of R&D-intensity (R&D/sales) is above 1% in all macro-regions with the EU reporting the highest rate (~6%). This would allow classifying the companies/ sector as medium-tech (even medium-high), according to the commonly applied classifi- cation (Hatzichronoglou, 1997). Considering the median R&D-intensity rather than the mean, the R&D/sales ratios do not change significantly in magnitude in Europe and the US/Can, but they drop below 1% in Japan. However, in the EU and the US/Can only a few firms perform R&D at all (but those which do, however, have significant spending), while in Japan most companies are engaged in R&D activities but modestly at individual level. In general, the companies active in the food-processing sector in the EU and in the US/ Can seem to be fairly similar: EU companies are, in average, a little smaller in terms of rev- enue (sales) and number of employees but have almost exactly the same ratio of net income/ revenue as those from US/Can and also comparable figures in terms of spending on R&D and capital (including their accumulated stocks). In contrast, Japanese firms appear smaller and less profitable, more inclined to do corporate R&D, but, in average, at a lower financial (Table 1). These differences between macro regions need to be taken in consideration when interpreting the estimated results and performing cross-country comparisons. In terms of sub-sector representation, observations from beverages companies are the most present followed by mixed-activity or generalist food-processing firm and prepared foods, accounting for 53% of the total sample. The remaining subsectors account individu- ally between 4% and 9% the dairy sub-sector which is marginally present in the sample (Table 2). Table 1 Descriptive statistics of main variables Variable Mean Std. Dev. Min Max Firms Obs. Total sample 307 2948 Revenue 2308.3 5192.3 0.4 51514 COGS-costs 1443.5 3295.6 0.4 47137 R&D expenditure 89.7 451.7 0 7290.3 Capital expend. 1286.4 2996.3 0 25846 Employees 10610 31443 2 486000 EU companies 85 557 Revenue 2705.8 6602.6 0.4 51514 COGS-costs 1561.2 3323.9 0.4 22873 R&D expenditure 175.6 926.9 0 7290.31 Capital expend. 1768.9 4020.4 0 25846 Employees 15292.7 36441.3 2 269000 US & Canada 79 1050 Revenue 3684.8 6607 1.7 50659 COGS-costs 2309.5 4578.3 1 47137 R&D expenditure 72.5 266.4 0 2476 Capital expend. 1839.9 3584.2 0 24759 Employees 18054 43375 2 486000 Japan* 143 1341 Revenue 1065.3 1983.5 5 15913 COGS-costs 716.6 1330.7 2 9785.7 R&D expenditure 67.5 181.5 0 1642.2 Capital expend. 652.6 1497.7 0 13127 Employees 2211 4203 16 36554 *(1999-2009 period only) Table 2 Sample composition - observations per subsector Subsector Codes No. observations Beverages, including alcohol 2080-2087 561 Mixed/generalist 2000 490 Prepared foods 2090-2099 491 Meat and poultry packing 2010-2015 272 Sugar and confectionery 2060-2068 252 Canned fruits and vegetables 2030-2038 225 Grain 2040-2048 226 Bakery 2050-2053 197 Dairy 2020-2026 18 Oils 2070-2079 116 Total 2948 240 Heinrich Hockmann et alii 4. Results 4.1 The size of inefficiency An output-oriented efficiency model (variable returns to scale-VRS) was run with a simple specification made of one output and three inputs. Inputs consist of capital stock (C), labour (number of employees, E) and total cost of goods sold (COSD). The out- put was measured as the value of total revenues assumed to be total food related sales, although firms may have sales revenue from other lines of activity and streams of income such as asset management (Fuglie et al., 2011). The distribution of efficiency scores by frequency is displayed in Figure 1. In general, the figure shows that the inefficiency distribution is skewed to the left indicating that most of the companies operate relatively close to their frontier (panels b and c). Very high inef- ficiencies could only be found for a few companies. Moreover, panel (a) shows an estimate of the bias of the inefficiency estimate. The distribution reveals that the bias is consider- able. Thus conducting an analysis without bootstrapping would have led to largely biased estimated parameters in the second step. Panel (b) gives an example of the inefficiencies calculated with the adjusted technology T*. Finally, panel (c) give the unbiased estimator (distribution) of the inefficiency. 4.2 The determinants of inefficiency The basic hypothesis of the second stage is that R&D has a positive impact on firm performance. In general, the determinants of inefficiency will be captured by the knowl- edge base of a company which depends on (a) on own R&D and (b) knowledge created elsewhere (universities, research institutes, companies) and diffuses to the public domain. The main objective of this paper is to capture the effect of the first type of knowl- edge. As a result, we include the variable own (private) R&D expenditure of companies Figure 1. Illustration of inefficiency estimates and estimated bias, frequencies. (a) (b) (c) Inefficiency units Inefficiency units Inefficiency units BIÂS(δ̂ i ) = BIAS(δ̂ i )+ vi δ̂ i * = δ (x i ,yi |T*) ˆ̂δ i = δ̂ i − BIÂS(δ̂ i ) 241Corporate R&D and the performance of food-processing firms (without distinguishing whether it is internal or external R&D) in the set of explanatory variables (z) considered in the second stage estimations. Usually the information is avail- able when the companies are required to publish their investments. Although it can be safely assumed that large companies in all countries have some R&D, however, they have no spontaneous incentive to report it since this would reveal information about the firm’s strategy and threaten the firm’s competitive position. This lack of data may bias the result. However, no information on R&D is less severe than expected. Given the basic hypotheses, the impact of R&D on performance might be less significant since firms which do not report but conduct research should be more effi- cient than expected. Regarding the knowledge created elsewhere (technological opportunities), firm R&D impacts not only the revenues directly but in addition also affects the technological opportunities of the firm. The firm’s technological opportunities consist of two parts: the knowledge external to the sector (universities, public research institutes) and the exist- ing knowledge at the competitors which diffuses to some extent into the public domain (Cohen and Levinthal, 1989). The degree of openness depends on the institutional reg- ulations regarding the protection of firm specific knowledge but also from the type of technology. The use of public knowledge depends on the absorption potential. This absorption depends on the height of the R&D expenditure as well the characteristics of the scientific and technological foundations. In addition it is determined by the ease how this knowl- edge can be absorbed. In order to account for differences in the knowledge and research infrastructure we consider regional dummy variables in the estimation. We expect that the US and Japan have a favourable knowledge base to conduct R&D and this knowledge base also finds its expression in better firm performance. Some indication of this can be seen Table 1 which shows that Japan and the US have the highest research expenditures related to outputs. The same effect can be expected for the old EU Member States (“EU15”). Similar to Japan and the US, they belong to the group of countries with a highly developed research infrastructure. Given the structural difficulties of EU New Member States (“NMS”) from Eastern Europe in particular related with their past history of planned economy, the research systems in these countries are likely less developed thus attaining lower pro- ductivity levels. The reference region for these regional dummy variables is Canada. Note that, some studies find that Canada reports lower performance of food-processing firms than their peers from other developed countries such as US (Chan-Kang et al. 1999; in Fuglie et al. 2011) To further control for the knowledge and research infrastructure beyond the regional dummies, the contemporaneous general public R&D investments per capita is also intro- duced in the regression (GERD of government sector, Euros equivalent, 2007 constant prices). The time lags and dynamic effects (e.g. see Andersen & Song, 2013) are not controlled for in the analysis, given that the availability of data in the sample for different years varies strongly across firms and regions. However, to account for the differences in the sample structure over time, dummy variables are used for the 1990s period and the period 2004- 2009 with the 2000-2004 period serving as reference. 242 Heinrich Hockmann et alii Ta bl e 3. T ru nc at ed re gr es si on e st im at es o f t he d et er m in an ts o f e ffi ci en cy . In de pe nd en t v ar ia bl es 1A (b ia se d) 1A 1B 2A 2B 3A 3B 4A 4B 5A 5B C on st an t 2, 28 80 5 ,7 32 1* 5, 74 37 * 4, 99 21 * 5, 02 46 * 5, 91 01 * 5, 77 05 * 5, 70 44 * 5, 83 31 * 5, 98 44 * 6, 07 73 * R& D , p er pe tu al in ve nt or y -0 ,8 24 3 -0 ,7 93 1* -1 ,0 60 6* -0 ,8 68 4* -1 ,1 70 3* -0 ,6 63 9* -0 ,9 15 7* -1 0, 76 39 * -1 0, 89 21 * -0 ,9 76 7* -1 ,1 72 9* (R & D , p er pe tu al in ve nt or y) ² 0, 04 47 * 0, 05 32 * 0, 03 86 * 0, 11 76 * 0, 08 10 * G ER D , g ov . s ec to r/ c ap ita -0 ,0 02 3 -0 ,0 02 6* -0 ,0 02 8* -0 ,0 04 0* -0 ,0 04 1* -0 ,0 06 2* -0 ,0 05 4* -0 ,0 05 2* -0 ,0 05 7* -0 ,0 06 3* -0 ,0 06 6* Ja pa n -0 ,7 09 8 -0 ,9 46 9* -0 ,9 24 8* -1 ,2 18 1* -1 ,1 46 5* -1 ,1 87 8* -1 ,2 03 3* -1 ,2 02 0* -1 ,2 32 8* U SA -0 ,7 37 8 -1 ,0 09 0* -1 ,0 08 2* -1 ,0 15 7* -1 ,0 20 4* -1 ,0 54 2* -1 ,0 57 9* -1 ,1 01 3* -1 ,1 26 3* EU 12 , N M S 0, 96 66 1, 65 72 * 1, 64 79 * 1, 48 37 * 1, 51 42 * 1, 46 66 * 1, 48 20 * 1, 52 67 * 1, 63 51 * EU 15 -0 ,2 25 6 -0 ,2 82 3* -0 ,3 05 2 -0 ,5 03 5* -0 ,4 02 2* -0 ,4 32 1* -0 ,3 46 5* -0 ,4 83 2* -0 ,5 17 3* 19 90 s d um . 0, 11 87 0, 19 38 * 0, 18 52 * 0, 27 08 * 0, 28 27 * 0, 09 04 0, 10 29 0, 09 04 0, 07 68 0, 08 61 0, 07 90 A fte r 2 00 4 du m . 0, 23 99 0, 28 15 * 0, 27 63 * 0, 38 97 * 0, 39 90 * 0, 16 97 * 0, 18 43 * 0, 17 81 * 0, 18 08 * 0, 19 50 * 0, 19 65 * D ai ry 0, 30 69 * 0, 32 73 * -0 ,0 63 6 -0 ,0 55 6 -0 ,0 90 1 -0 ,1 27 0 0, 18 76 0, 20 57 C an ne d 0, 11 05 0, 10 26 0, 29 52 * 0, 25 45 * 0, 25 93 * 0, 23 13 * 0, 25 00 * 0, 22 80 Be ve ra ge s -0 ,5 80 4* -0 ,5 82 8* -0 ,7 58 5* -0 ,7 19 0* -0 ,7 06 9* -0 ,7 59 8* -0 ,7 69 8* -0 ,7 93 6* G en er al 0, 17 60 0, 15 83 0, 15 51 0, 14 61 0, 16 76 * 0, 12 92 0, 11 08 0, 18 26 M ea ts 0, 31 46 * 0, 28 54 * 0, 42 65 * 0, 38 32 * 0, 40 36 * 0, 34 87 * 0, 62 28 * 0, 62 48 * O ils -0 ,5 02 0* -0 ,5 39 6* -0 ,3 13 0* -0 ,3 25 1* -0 ,2 88 7* -0 ,3 19 8* -0 ,0 02 2 -0 ,0 19 9 Ba ke ry 0, 30 42 * 0, 29 58 * 0, 46 44 * 0, 41 24 * 0, 39 15 * 0, 35 27 * 0, 46 86 * 0, 46 96 * Pr ep ar ed fo od s 0, 38 93 * 0, 39 19 * 0, 42 36 * 0, 40 49 * 0, 38 70 * 0, 37 86 * 0, 34 61 * 0, 36 03 * Su ga r -0 ,1 58 7 -0 ,1 69 6 -0 ,0 46 6 -0 ,0 95 2 -0 ,0 75 8 -0 ,0 79 2 0, 01 55 -0 ,0 03 0 Ja pa n x R& D 10 ,2 00 2* 9, 98 97 * U SA x R & D 10 ,1 01 5* 9, 83 61 * (E U 12 , N M S) x R & D -0 ,2 31 7 -0 ,5 67 5 EU 15 x R & D 9, 99 56 * 8, 70 63 * D ai ry x R & D -1 ,5 26 5* -1 ,5 85 1* C an ne d x R& D -0 ,1 39 0 -0 ,0 88 3 Be ve ra ge s x R & D 0, 21 83 0, 19 48 G en er al x R & D 0, 18 30 -0 ,4 62 4 M ea ts x R & D -9 ,0 02 5* -9 ,1 33 2* O ils x R & D -4 ,4 68 9* -4 ,5 71 5* Ba ke ry x R & D -0 ,8 57 3 -0 ,8 75 4 Pr ep ar ed fo od s x R & D 0, 56 44 * 0, 47 38 * Su ga r x R & D -1 ,2 18 6* -1 ,1 68 2* N M S: E U N ew M em be r S ta te s. * in di ca te s st at . s ig ni fic an ce a t 5 % . S ou rc e: o w n ca lc ul at io ns o n R v2 .1 4 w ith F EA R pa ck ag e. 243Corporate R&D and the performance of food-processing firms The estimated results of the second stage pooled truncated regression are reported in Table 3. We have estimated several alternative and complementary model specifications to avoid potential collinearity between explanatory variables. Model 1A starts with a sim- ple specification of the estimated equation which includes private R&D (perpetual inven- tory), public R&D (GERD/per capita), time dummies, and regional dummies (US, Japan, EU, etc.) with Canada serving as the reference country. For comparison purposes, we also report the results obtained with the biased estimators for the first model (1A biased). The remaining models are only presented with their unbiased estimators. The extended first model (1B) also considers squared value of private R&D with the aim to capture the change in marginal gains from additional investment in private R&D. The second set of models (2A, 2B) considers sectoral dummies instead of regional dummies with firms specialised in grain processing being used as the reference sub-group. Model 2B expands 2A with adding squared value of private R&D. The third set of models (3A and 3B) add both regional and sectoral dummies in the estimated equation. Again, model 3B expands 3A with adding squared value of private R&D. The remaining model sets (4 and 5) consider interaction variables between private R&D and regional and sectoral dummy variables, alongside the variables considered in the first three model sets, in order to capture whether the impact of private R&D vary across regions or sectorial circumstances, respectively. That is, the fourth set of models (4A, 4B) includes interaction variables between private R&D and regional dummies, while the fifth set of models (5A, 5B) interacts private R&D and sectoral dummies. The estimates largely confirm the hypothesis that private R&D has a positive effect on performance (i.e. it reduces inefficiency) of the food-processing firms (Table 3). How- ever, the variable controlling for marginal gain of additional investment does systemati- cally capture decreasing marginal returns of R&D investments on performance at firm level. Public R&D has also statistically significant contribution to performance, in line with country specific studies such as for the Spanish food sector by Acosta et al (2015). However, the relationship is complex as hinted by Maietta et al (2017) whose analysis of the R&D sector in Europe over the 2007-2009 period suggest a displacement effect on intra-muros (internal) R&D by government R&D. These results are consistent across all estimated models. Private R&D investing seems to more positively affect performance in Canada (the reference country) than in the USA, Japan or EU15 countries (4A and 4B). The estimated coefficient for new EU member states is not significant in both models where the inter- action variables between private R&D and regional dummies are considered (i.e. 4A and 4B). These results suggest that additional R&D investment in Canada and NMS would produce greater firm efficiency gains than in in the USA, Japan or EU15. With regards to sub-sectorial sensitivity to R&D investment on firm performance (5A and 5B), some sub-sectors (dairy, meat processing, oils and sugar) seem to be more responsive to R&D investment and statistically significant compared to the reference sector (grain). In con- trast, processed food sectors are less sensitive to R&D investment, while the remaining sub-sectors were found to be statistically insignificant relative to the reference sector. The performance of food-processing firms during the period after 2004 is significant- ly lower compared to the 1990s and especially compared to the reference period (2000- 2004). In terms of regional variation of firm performance, the estimates suggest that Japa- 244 Heinrich Hockmann et alii nese, US, and EU15 firms are more efficient that Canadian firms which corroborates with previous studies comparing US and Canadian firms (Chan-Kang et al. 1999, Fuglie et al. 2011). The food-processing firms from the NMS tend to underperform the Canadian peers, and hence the firms from other countries. Firm operating as generalist of the food-processing sector tend not to indicate a sta- tistically significant difference with the reference group (grains). In most models, this is also the case for dairy and sugar-related firms, while for oil and canned producers the results are mixed in terms of statistical significance. However, firms specialised in meats, bakery and prepared foods tend to be less efficient than those involved in grains; these resulte are statistically significant across all models. 5. Conclusions This paper confirms the hypothesis that R&D investment influences firm perfor- mance: food-processing firms which invest in R&D tend to be closer to the efficiency frontier compared to those that do not invest in R&D (i.e. private R&D has a negative effect on inefficiency). Estimates of this paper also point to decreasing marginal returns in reducing (increasing) inefficiency (efficiency) by private R&D as well as that that the gen- eral public R&D has a positive effect on efficiency of food-processing firms. When looking at the drivers of firm performance, country/region dummies do cap- ture differences and similarities in knowledge systems and nature of the sector. Similari- ties can be detected in the US and Japanese contexts. Further, as expected, less favourable eastern European (NMS) context is indentified in the estimated results as compared to the performance of firms from old EU Member States. However, the results suggest that gains from additional investment in R&D could be greater in NMS than old EU Member States or the US. The findings of this paper have to be considered, however, with some caution on the account of the data limitations. The persistent lack of reporting R&D in certain countries in the EU may create biases in the estimated effects. Further, the sample contains rather larger firms from the food-processing industry (a key factor determining R&D, as illus- trated by Acosta et al (2015) for the Spanish food sector), while small firms are under- represented. This data limitation does not allow to fully extrapolate the results obtained in this paper to the whole food-processing industry. Overall, the results of this paper show that R&D in food-processing industry is asso- ciated with higher firm performance. At the same time, the sample used in this paper includes medium-high-tech (and larger) food-processing firms, questioning the generally held view on the sector as being rather low-tech. By prioritising high-tech sectors, emerg- ing technologies, knowledge-based services, etc., the current backbone of the European economy, mainly constituted by industries that are often rather medium- and even low- tech, tend to be somewhat marginalised from the policy attention perspective (Hanse and Winther, 2011). Results of this paper show that growth opportunities could also be expected and encouraged from this type of non-high-tech innovative sectors. Further, the results of this paper suggest heterogeneity in R&D effects across EU Member States, hence innovation policies may have different implications across EU regions. 245Corporate R&D and the performance of food-processing firms 6. Acknowledgements The analytical framework and analyses of R&D effects on the performance of food- processing firms presented in this paper are based on Garzón Delvaux et al. (2018) developed within the IMPRESA project. We would like to thank our IMPRESA consor- tium partners for their comments received during the development of the project. Spe- cial thanks go to Davide Viaggi and Michele Vollaro for the review of a previous version of the paper. Also, our thanks go to the following JRC colleagues: Mark Boden, Andrea Conte and Pietro Moncada Paternò Castello, for having first described and then provided access to the COMPUSTAT database relevant to our analysis. We would like to express our appreciation to our former colleagues Marianne Lefebvre and Sébastien Mary who initiated IMPRESA at JRC. Finally, we would like to acknowledge the funding received for this project from the European Union’s Seventh Framework Programme for research, technological development and demonstration under the grant agreement No 609448. 7. Disclaimer The authors are solely responsible for the content of the paper. The views expressed are purely those of the authors and may not in any circumstances be regarded as stating an official position of the European Commission. 8. References Acosta, M., Coronado, D., & Romero, C. (2015). Linking public support, R&D, innova- tion and productivity: New evidence from the Spanish food industry. Food Policy 57: 50-61. Alston, J. M. (2010). The benefits from agricultural research and development, innova- tion, and productivity growth. OECD Food, Agriculture & Fisheries Papers, No. 31, OECD Publishing. Andersen, M.A., Song, W. (2013). The Economic impact of public agricultural research and development in the United States. Agricultural Economics 44: 287-295. Avermaete, T., Viaene, J., Morgan, E.J., Pitts, E., Crawford, N., Mahon, D. (2004). Determi- nants of product and process innovation in small food manufacturing firms. Trends in Food Science & Technology 15(10): 474-483. Becker, B. and Pain, N. (2002). What determines industrial R&D expenditure in the UK? National Institute of Economic and Social Research, London. Bönte, W. (2003). R&D and productivity: Internal vs. external R&D – Evidence from West German manufacturing industries. Economics of Innovation and New Technology 12: 343-360. Capitanio, F., Coppola, A., Pascucci, S. (2010). Product and process innovation in the Ital- ian food industry. Agribusiness 26(4): 503-518. Coelli, T.J., Rao, D.S.P., O’Donnell, C.J., Battese, G.E. (1998). An Introduction to Efficiency and Productivity Analysis. 2nd ed. (2005), XVII, 350 p. ISBN 978-0-387-25895-9 Cohen, W. M., & Levinthal, D. A. (1989). Innovation and learning: the two faces of R & D. The economic journal 99(397): 569-596. 246 Heinrich Hockmann et alii Crépon, B., Duguet, E. and Mairesse, J. (1998). Research, Innovation and Productivity: an Econometric Analysis at Firm Level. Economics of Innovation and New Technology 7: 115-158. Feigl, S., Menrad, K. (2008). Innovation activities in the food industry in selected Euro- pean countries. Project report “Traditional United Food Europe” (TRUEFOOD), WP 7.5: Traditional products and the economic impact of innovation. University of Applied Sciences of Weihenstephan (ed.). Straubing / Germany. FoodDrinkEurope (2015). Data & Trends 2014-2015. Based on OECD STAN and Eurostat data. http://www.fooddrinkeurope.eu/uploads/publications_documents/Data_and_ Trends_2014-20151.pdf Fuglie, K.O., Heisey, P.W., King, J.L., Day-Rubenstein, K., Schimmelpfennig, D., Wang S.L., Pray, C. E. and Karmarkar-Deshmukh, R. (2011). Research investments and market structure in the food processing, agricultural input, and biofuel industries world- wide. USDA-ERS Economic Research Report 130. Garzon Delvaux, P.A., Hockmann, H., Voigt, P. Ciaian, P. Gomez y Paloma, S. (2018). The impact of private R&D on the performance of food-processing firms: Evidence from Europe, Japan and North America. EUR 28249 EN, Publications Office of the Euro- pean Union, Luxembourg. Gopinath, M., Vasavada, U. (1999). Patents, R&D, and market structure in the US food processing industry. Journal of Agricultural and Resource Economics VOL?: 127-139. Hall, B.H. (2007). Measuring the returns to R&D: The depreciation problem. NBER Work- ing Paper No. 13473, National Bureau of Economic Research, Cambridge, MA. Hall, B.H., Mairesse, J. (1995). Exploring the relationship between R&D and productivity in French manufacturing firms. Journal of Econometrics 65: 263-293. Hall, B.H., Rosenberg, N. (2010): Handbook of the Economics of Innovation, Volume 1 & 2. Elsevier. Hansen, T., Winther, L. (2011). Innovation, regional development and relations between high- and low-tech industries. European Urban and Regional Studies 18(3): 321- 339. Hatzichronoglou, T. (1997). Revision of the high-technology sector and product classifi- cation, OECD Science, Technology and Industry Working Papers, 1997/02, OECD Publishing. Hernández, H. H., Soriano, F. H., Tüebke, A., Vezzani, A., Dosso, M., Amoroso, S., Gras- sano, N., Coad, A. & Gkotsis, P. (2015). The 2015 EU Industrial R&D Investment Scoreboard (No. JRC98287). Institute for Prospective and Technological Studies, Joint Research Centre. Hulten, C.R. (1991). The measurement of capital, E. R. Berndt and J. E. Triplett (eds.): Fif- ty years of economic management. University of Chicago Press, Chicago. Jongen, W.M.F., Meulenberg, M.T.G. (2005): Innovation in agro-food systems. Wagenin- gen Academic Publishers/The Netherlands. Jorgenson, D.W. (1990). Productivity and economic growth. In: E. R. Berndt and J. E. Tri- plett (eds.): Fifty years of economic growth, 19-118, Chicago University Press, Chi- cago. Koop, G., Osiewalski, J. and Steel, M. F. (2000). Modelling the sources of output growth in a panel of countries. Journal of Business and Economic Statistics 18(3): 284-299. 247Corporate R&D and the performance of food-processing firms Limam, Y. R., Miller, S. M. (2004): Explaining economic growth: Factor accumulation, total factor productivity growth and production efficiency improvement. Depart- ment of Economics Working Paper Series, 2004-20, Connecticut University. Maietta, O. W., Barra, C., & Zotti, R. (2017). Innovation and University‐Firm R&D Col- laboration in the European Food and Drink Industry. Journal of Agricultural Eco- nomics 68(3): 749-780. Menrad, K. (2004). Innovations in the food industry in Germany. Research Policy 33: 845- 878. Musgrave, J.C. (1986). Fixed Reproducible Tangible Wealth Series in the United States, 1925-91. Survey of Current Business 66: 51-75. Nadiri, M.I. and Prucha, I.R. (1996). Estimation of the Depreciation Rate of Physical and R&D Capital in the U.S. Total Manufacturing Sector. Economic Inquiry 34: 43-56. OECD (2012). Improving Agricultural Knowledge and Innovation Systems, OECD Confer- ence Proceedings. OECD Publishing. http://dx.doi.org/10.1787/9789264167445-en OECD (2013). Agricultural Innovation Systems: A framework for analysing the role of the government, OECD Publishing. http://dx.doi.org/10.1787/9789264200593-en Orea, L., Kumbhakar, S.C. (2004). Efficiency measurement using a Latent Class Stochastic Frontier Model. Empirical Economics 29(1): 169-183. Ortega-Argiles, R., Potters, L., Voigt, P. (2009). R&D-intensive SMEs in Europe: What do we know about them? JRC-IPTS Working Papers on Corporate R&D and Innova- tion 2009-15, Institute of Prospective Technological Studies, Joint Research Centre. Pakes, A. and Schankerman, M. (1986). Estimates of the Value of Patent Rights in Euro- pean Countries During the Post-1950 Period. Economic Journal 96: 1052-1076. Parisi, M., Schiantarelli, F., Sembenelli, A. (2006). Productivity, innovation creation and absorption and R&D – Microevidence for Italy. European Economic Review 8: 733- 751. Simar, L., Wilson, P.W. (2007). Estimation and inference in two-stage, semi-parametric models of production processes. Journal of Econometrics 136(1): 31-64. S&P (2014).COMPUSTAT. http://marketintelligence.spglobal.com/about-us/about-us.html (now, S&P Global Market Intelligence). Winger R, Wall G (2006). Food product innovation: A background paper. Agricultural and food engineering working document 2. FAO (ed.), Rome. Positive Mathematical Programming and Risk Analysis Quirino Paris The hedonic contents of italian super premium extra-virgin olive oils Luca Cacchiarelli1,*, Anna Carbone2, Tiziana Laureti1, Alessandro Sorrentino1 Corporate R&D and the performance of food-processing firms: Evidence from Europe, Japan and North America Heinrich Hockmann1, Pedro Andres Garzon Delvaux2,*, Peter Voigt3, Pavel Ciaian2, Sergio Gomez y Paloma2 Can menu labeling affect away-from-home-dietary choices? Elena Castellari1,*, Stéphan Marette2, Daniele Moro3, Paolo Sckokai1 A preliminary test on risk and ambiguity attitudes, and time preferences in decisions under uncertainty: towards a better explanation of participation in crop insurance schemes Attilio Coletta1, Elisa Giampietri2, Fabio Gaetano Santeramo3,*, Simone Severini1, Samuele Trestini2