Endogenous product versus process innovation and a firm's propensity to export
Summary (3 min read)
1 Introduction
- Research on the role of innovation on economic outcome has for long been at the heart of three different fields of the profession: macro-economics, international economics, and industrial economics.
- To a large extent, product and/or process characteristics and the corresponding modes of innovation are typically viewed to be beyond a firm’s choice.
- The latter is, however, largely at odds with both economic intuition and stylized facts.
- The next section provides an overview of earlier theoretical and empirical work on innovation to motivate determinants of innovations and derive hypotheses about their consequences for productivity and export propensity.
- Section 4 summarizes the main features of their survey data.
2.1 Economic theory on innovation
- There is a sizeable body of theoretical work that elaborates on the determinants of innovation and their consequences for productivity and economic growth and, to a lesser extent, for exports.
- As indicated before, innovation is endogenous itself and firms innovate more likely in large economies (where fixed costs can be covered more easily), if the productivity in research labs is high, product markets are competitive, and if consumers value a large variety and/or a high quality of available products (see Grossman and Helpman, 1991, chapters 3 and 4).
- While both process and product innovation spur aggregate income, product innovation is preferable by avoiding the adverse effects of technological unemployment.
- There, investment in firm-specific assets (to be associated with product innovation, see Spence, 1984) and a high corresponding outcome (i.e., a high total factor productivity) are the key determinants of a firm’s export propensity.
- When assuming that the aggregate efficiency can be measured by the (inverse of) average production costs, then, Boone’s (2000) analysis suggests that a higher level of competitive pressure cannot increase product and process innovation at the same time.
2.2 Empirical work on the determinants and effects of
- Two exceptions in the latter regard are Cassiman and Mart́ınezRos (2004) and Lachenmaier and Wößmann (2006).
- Both studies exploit information from panel data.
- Cassiman and Mart́ınez-Ros (2004) focus on innovations as such and treat them as predetermined variables (hence, they use once-lagged instead of contemporaneous innovations in the export regressions).
- One of their major findings is that innovations are indeed endogenous and their exogenous treatment leads to largely downward-biased estimates of the impact of innovations on firm-level exports.
2.3 Contribution of this paper
- This paper departs from the strategy adopted in previous micro-econometric work on the innovation-driven exports hypothesis in two important ways.
- 1A smaller number of studies that employed the less preferable R&D expenditures as an indirect measure of innovations lacked to find such a positive impact (see Cassiman and Mart́ınez-Ros, 2004, for a survey).
- 5 First, it explicitly distinguishes between product and process innovations in the analysis and, second, it accounts for their endogeneity by allowing for an endogenous selection of firms into product and process innovations.
- In contrast to earlier work, the authors use matching techniques for multiple binary treatments – in their case, new product and/or process innovations versus no innovations at all – to account for self-selection of firms into either type of innovation.
3 Empirical framework
- In the subsequent analysis the authors assume that, after controlling for a set of observable variables, treatment participation does not depend on treatment outcome.
- Since their data set allows us to disentangle product innovation from process innovation – hence, there are two treatment indicators at the firm level –, the authors have to depart from the strategy typically applied in models with a simple binary treatment variable.
- Neither of these studies considers the impact of these two modes of endogenous innovations on exports.
- Estimates of the average treatment effects can be obtained as follows.
4 Data
- The authors data are based on the Ifo Innovation Survey that is conducted annually by the Ifo Institute, covering more than 1,000 firms in Germany per year.
- The survey asks about the structure of innovations at the firm level.
- In particular, it collects information about process versus product innovation activities and about export status.
- Furthermore, the survey explicitly covers questions relating to exogenous innovation impulses and obstacles as well as other firm-level characteristics.
- Beyond that, there is an industry indicator that allows us to link industry characteristics to the micro-level data.
4.1 Dependent variables
- Regarding the dependent variables, the database provides information on whether a firm has exported and applied new product innovations or process innovations over the last six months or not.
- In the year t the authors have introduced (or started but not yet finished) new product innovations.
- First, 80.00 percent of the firms in their sample conduct exports.
4.2 Independent variables
- Beyond the information for the dependent variables in their analysis, the survey asks about a set of incentives/impulses and obstables/impediments to innovation.
- Of those, in their empirical model, only the following four impediments exert a significant impact on a firm’s probability to innovate: lacking own capital; lacking external capital; long amortization period; imperfect opportunities to cooperate with public or academic institutions.
- In addition to these firm-level determinants the authors use characteristics that vary across NACE 2-digit industries published by EUROSTAT (NewCronos Database).
- By way of contrast, the higher the weighted foreign wage costs are relative to foreign output, the lower the authors expect the competitive pressure for German producers to be ceteris paribus.
- Table 2 summarizes mean and standard deviation of all covariates.
5 Estimation results
- Table 3 presents the results of a multinomial logit model (assuming a logistic cumulative density function, respectively) determining a representative firm’s choice of product and/or process innovation.
- This type of matching requires that the matched control units exhibit a propensity score that differs by not more than the radius from the propensity score of the treated unit they are matched onto.
- In the table, the authors report estimates of all three treatment effects, θm,l, αm,l, and γm,l for all treatment pairs m and l and their standard errors.
- The authors results indicate that the analytical standard errors are slightly more conservative (i.e., smaller) than the bootstrapped ones.
- Firms that conduct new product and process innovations (the treated – T in the first table column – receive (d, c)) exhibit a significantly higher export propensity than ones that neither do product nor process innovations (the matched controls – C in the second table column – receive (0, 0)).
6 Sensitivity analysis and discussion
- The authors undertake several robustness checks to assess the sensitivity of their findings.
- The latter ensures that the authors estimate the impact of innovation on export propensity from a comparison of treated firms with untreated ones where the export status in the past was the same between the treated and the untreated.
- Across the board, neither changing the radius nor the matching estimator (nearest neighbor or alternative kernel matching estimators with different bandwidths instead of radius matching) affects their conclusions from above, neither in qualitative nor in quantitative terms.
- T denotes the treatment, C the control group.
7 Conclusions
- The authors goal in this paper was to provide novel empirical insights in the role of product versus process innovation on export propensity at the firm level.
- Either of these modes of innovation has been hypothesized to affect firm-level productivity in previous theoretical work.
- Economic theory suggests that firms do not undertake innovations at random, neither product nor process innovations.
- Viewing innovations as a ’treatment’, this lends support to an endogenous treatment approach to innovations and export propensity.
- The largest estimated self-selection bias in the data amounted to more than 200 percent, depending on the mode of innovations (product and/or process innovation).
Did you find this useful? Give us your feedback
Citations
3,920 citations
499 citations
302 citations
Additional excerpts
...Prob(Expt = 1jExpt 1 = 0) = f(log TFPt 1; log kt 1; log lt 1; logNoPt 1; time) (5)...
[...]
287 citations
277 citations
References
28,298 citations
9,036 citations
6,911 citations
"Endogenous product versus process i..." refers background in this paper
...…likely in large economies (where fixed costs can be covered more easily), if the (exogenous) productivity in research labs is high, product markets are competitive, and if consumers value a large variety and/or a high quality of available products (see Grossman and Helpman, 1991, chapters 3 and 4)....
[...]
...International economic theory spots the role of product innovation for trade in open economy growth models (Dollar, 1986; Jensen and Thursby, 1987; Grossman and Helpman, 1989, 1990, 1991, chapters 9-11; Segerstrom, Anant, and Dinopoulos, 1990)....
[...]
5,633 citations
5,186 citations
Related Papers (5)
Frequently Asked Questions (6)
Q2. what is the expected average effect of treatment m relative to treatment l?
(1)The expected average effect of treatment m relative to treatment l for a firm randomly selected from the group of firms participating in either m or l is defined asαm,l = E(Y m − Y l|S = m, l) = E(Y m|S = m, l) − E(Y l|S = m, l), (2)where S is the assignment indicator, defining whether a firm receives treatment m or l.
Q3. What is the expected average effect of treatment m relative to treatment l?
The expected average effect of treatment m relative to treatment l for a firm drawn randomly from the population is defined asγm,l = E(Y m − Y l) = E(Y m) − E(Y l).
Q4. What is the pseudo-R2 of the effect before matching?
For instance, for the effect (d, c) versus (0, 0), the pseudo-R2 before matching is 0.354, i.e., the covariates are relevant predictors in the overall sample.
Q5. What is the main difference between the two treatment indicators?
Since their data set allows us to disentangle product innovation from process innovation – hence, there are two treatment indicators at the firm level –, the authors have to depart from the strategy typically applied in models with a simple binary treatment variable.
Q6. What are the main exceptions to the theory of innovation-driven exports?
Cassiman and Mart́ınez-Ros (2004) focus on innovations as such and treat them as predetermined variables (hence, they use once-lagged instead of contemporaneous innovations in the export regressions).