«Der Open-Access-Publikationsserver der ZBW – Leibniz-Informationszentrum Wirtschaft The Open Access Publication Server of the ZBW – Leibniz ...»
In order to account for the parametric shortcomings associated with the estimations of Tobit models, column 5 presents the results of a semiparametric model proposed by Blundell and Powell (2007). This approach accounts for censoring and endogenous regressors in a semiparametric setting (nonparametric ﬁrst stage). Interestingly, the impact is still present although its magnitude is clearly reduced. These results suggest that ﬁrm heterogeneity is a factor explaining ﬁrms’ strategic decisions and that ICT exert an important and independent impact in generating such heterogeneity.
5 Robustness Checks
This section presents some alternative speciﬁcations in order to check the robustness of the main results. In particular, two diﬀerent alternative speciﬁcation of the two stage procedure were investigated. First, the speciﬁcations presented in Table 5 were estimated for the whole available sample and not only for ﬁrms that reported ICT consulting activities in 2003, revealing the importance of ICT infrastructure in their operations. The results are presented in Table 6. Second, alternative measures of innovations were also taken into account under the same ﬁrst stage speciﬁcation. The alternative innovation measures correspond to the percentage of sales reported in 2006 that are derived from the product innovations introduced during the period 2004-2006, as well as the percentage of cost reductions achieved in 2006 from the introduction of process innovations during the period 2004-2006.
Consistent with the results reported in Table 5, Table 6 shows that the main results with respect to the impact of ICT induced heterogeneity on the ratio of R&D employees and total employees are maintained when the full sample is considered for the parametric approaches presented in columns 1-4. Although the endogeneity tests suggest the validity of the two stage approach, the overidentiﬁcation tests are inconclusive in the IV GMM and IV Tobit speciﬁcations presented in columns 2 and 4 of Table 6 (i.e. rejection is achieved only at 10%, while in Table 5 no rejection was possible). Within the analysis of this paper, rejection of the overidentiﬁcation test implies that the use of ICT does independently aﬀect R&D incentives, as well as ﬁrm heterogeneity. If rejection is achieved with high conﬁdence, which is not the case in the present analysis, the proposed two stage approach would be invalid.
In addition, instead of analyzing the impact of ICT induced heterogeneity on R&D incentives, the analysis also considers the impact on the percentage of sales reported in 2006 that are derived from the product innovations introduced during the period 2004-2006 as a measure of innovative output. As the analysis presented in Table 5, the sample used corresponds only to ﬁrms that reported ICT consulting activities in 2003 in order to consider the impact of ICT induced heterogeneity on ICT intensive ﬁrms. In this case, the coeﬃcients of the parametric speciﬁcations were signiﬁcant although the speciﬁcation test rejected the validity of the approach. This result is not surprising given the demand side factors (not considered in the analysis) that might aﬀect the relationship between ﬁrm heterogeneity and the beneﬁts of product innovation.
Similarly, the analysis also considers the case of a diﬀerent measure of innovative output.
In this case, the percentage of cost reductions achieved in 2006 from the introduction of process innovations during the period 2004-2006 was investigated. As before, the analysis was performed for ICT intensive ﬁrms. The parametric results are consistent with the main results of the paper reported in Table 5, including the speciﬁcation tests. This result suggests that the adoption of ICT is a useful tool to diﬀerentiate a given ﬁrm from its competitors (i.e. ﬁrst stage results are signiﬁcant) and this ICT induced heterogeneity has an independent impact on the ﬁrm innovative output (i.e. second stage results).
6 Conclusions This paper studies how the adoption of information and communication technologies (ICT) aﬀects ﬁrm heterogeneity and thereby contributes to the productivity driven selection mechanism that determines aggregate productivity growth within industries. Given the well documented existence of high and persistent productivity diﬀerences within industries where more productive ﬁrms grow faster, exhibit a higher probability of survival and displace low productivity ﬁrms, this paper attempts to speciﬁcally estimate the role of ICT on such ﬁrm heterogeneity. Moreover, to explore the role of ICT induced heterogeneity on the ”creative destruction” process, the results are related to observed ﬁrm strategies relevant to that process (i.e. R&D incentives and innovation outputs). The results show that ICT has a robust, positive impact on ﬁrm heterogeneity only when ICT is used intensively and jointly with speciﬁc ICT applications. In addition, the analysis showed that ICT induced heterogeneity is not innocuous: it has a signiﬁcant and positive, albeit small, impact on the incentives to innovate. These results suggest that ﬁrm heterogeneity is a factor explaining ﬁrms’ strategic decisions and that ICT exert an important and independent impact in generating such heterogeneity.
References Acemoglu, D., P. Aghion, F. Zilibotti, 2003, Distance to Frontier and Vertical Integration, Journal of the European Economic Association, 1, 630-38.
Acemoglu, D., P. Aghion, F. Zilibotti, 2006, Distance to Frontier, Selection, and Economic Growth, Journal of the European Economic Association, 4, 37-74.
Acemoglu D., Aghion P., Lelarge C., Van Reenen, J. and F. Zilibotti, 2007, Technology, Information, and the Decentralization of the Firm, Quarterly Journal of Economics, 122 (4), 1759-99.
Aghion, P., N. Bloom, R. Blundell, R. Griﬃth and P. Howitt, 2005, Competition and Innovation: An Inverted-U Relationship, Quarterly Journal of Economics, 120, 701Aghion P. and R. Griﬃth, 2005, Competition and Innovation: Reconciling Theory and Evidence. Zeuthen Lectures. MIT Press.
Bartelsman, E. and M. Doms, 2000, Understanding Productivity: Lessons from Longitudinal Microdata, Journal of Economic Literature, 38 (3), 569-94.
Bartelsman, E., J. Haskel and R. Martin, 2006, Distance to Which Frontier? Evidence on Productivity Convergence from International Firm-Level Data, Working Paper, University of Amsterdam.
Bloom, N., R. Sadun and J. Van Reenen, 2008, Americans do IT Better: US Multinational and the Productivity Miracle, Working Paper, Stanford University.
Blundell R. and J. Powell, 2007, Censored Regression Quantiles with Endogenous Regressors, Journal of Econometrics, 141, 65-83.
Bresnahan, T., E. Brynjolfsson and L. Hitt, 2002, Information Technology, Workplace Organization and the Demand for Skilled Labor: Firm-Level Evidence, Quarterly Journal of Economics, 117, 1, 339-76.
Brynjolfsson, E. and L. Hitt, 2003, Computing Productivity: Firm Level Evidence, Review of Economics and Statistics, 85, 4, 793-808.
Chun H., J. Kim, R. Morck, B. Yeung, 2008, Creative Destruction and Firm-speciﬁc Performance Heterogeneity, Journal of Financial Economics, 89, 109-35.
Crepon, B., E. Duguet and J. Mairesse, 1998, Research, Innovation and Productivity: An Econometric Analysis at the Firm Level, Economics of Innovation and New Technology, 7 (2), 115-58.
Draca, M., R. Sadun and J. Van Reenen, 2007, Productivity and ICT: A Review of the Evidence, In Mansell, R., C. Avgerou, D. Quah and R. Silverstone (Eds.), The Oxford Handbook of Information and Communication Technologies, Oxford.
Foster, L., J. Haltiwanger and C. Syverson, 2008, Reallocation, Firm Turnover, and Eﬃciency: Selection on Productivity or Proﬁtability?, American Economic Review, 98 (1), 394-425.
Griliches, Z., 1995, R&D and Productivity: Econometric Results and Measurement Issues, in Stoneman, P. (Ed.), Handbook of the Economics of Innovation and Technological Change, Oxford, 52-89.
Hall, B., F. Lotti and J. Mairesse, 2008, Innovation and Productivity in SMES: Empirical Evidence for Italy, NBER Working Paper No. 14594.
Jorgenson, D., M. Ho and K. Stiroh, 2008, A Retrospective Look at the U.S Productivity Growth Resurgence, Journal of Economic Perspectives. 22, 1, 3-24.
Li, Q. and J. S. Racine, 2007, Nonparametric Econometrics: Theory and Practice, Princeton University Press, Princeton.
Syverson, C., 2004, Product Substitutability and Productivity Dispersion, Review of Economics and Statistics, 86 (4), 534-550.
Van Ark, B., M. Timmer and M. O’Mahoney, 2008, The Productivity Gap Between Europe and the United States: Trends and Causes, Journal of Economic Perspectives, 22, 1, 25-44.
Appendix A: Blundell and Powell (2007) The previous analysis included a semiparametric estimator proposed by Blundell and Powell (2007). This estimator takes simultaneously into account the presence of censoring in the dependent variable, as well as the existence of endogenous variables. The proposed estimator
follows a two-stage approach, which estimates the censored model:
The variable xi contains a set of regressors including an endogenous regressor, as well as an scalar error term ui. The endogenous regressor can be represented as a possibly nonparametric
function of its instruments zi and its corresponding error term vi :
The estimation procedure follows in two stages. First, the conditional quantile qi ≡ Qi [yi |xi, zi ] of the dependent variables is estimated used a nonparametric quantile regression (See Appendix B). In the same stage, a control variable vi corresponding to the error of the nonparae) metric estimation of the endogenous regressor vi = xi − π(zi ) is derived (See Appendix B).
Subsequently, in a second stage the coeﬃcients for βi are derived in a weighted least-squares
regression following a ”pairwise diﬀerencing” argument. The estimator is given by:
The variable Kv is the kernel function chosen. In this paper, an Epanechikov kernel was employed. The term hn corresponds to the optimized bandwidth derived by a cross-validation method and ti represents a trimming term ensuring that censoring is taken into account.
Standard Errors are derived by bootstrapping with 500 iterations.
Appendix B: Multinominal Kernel regression
Estimation of the nonparametric kernel regression and nonparametric quantile regression are described in Li and Racine (2007). Both procedures used kernel estimations with mixed data continuous and categorical data to allow the use of discrete regressors. Importantly, bandwidths for the smoothing parameters are derived by data driven cross-validation methods.
Nonparametric Kernel regression Li and Racine (2007) propose a method for nonparametric regression that includes both, continuous and discrete regressors. They apply a kernel method based on work of Aitchison
and Aitkens (1976). The nonparametric function consider is:
The vector Xi = (Xc, Xd ) contains continuous, as well as discrete variables. The modiﬁed kernel estimator Kh,ix = (Wh,ix Lλ,ix ) contains an estimator for each, continuous variables Wh,ix and discrete variables Lλ,ix. The resulting kernel function for both variables is deﬁned
with h and λε[0, c − 1/c] being non-stochastic smoothing parameter that are derived by a cross-validation bandwidth estimation. Wh0 is a kernel function for continuous variables and L(Xid, xd, λ) for discrete variables. The continuous kernel can contain any usual kernel available for continuous data (i.e. Epanechikov or Gaussian), while for discrete variables is Aitchisons and Aitkens (1976).
Nonparametric quantile estimation Li and Racine (2007) provide an estimator to derive conditional cumulative distribution function (CDF) nonparametrically by a kernel estimator and to determine the conditional quantile of a function of continuous and dichotomous regressors. Estimation uses the same kernel function as in the nonparametric kernel regression also provided by Li and Racine (2007).
In a ﬁrst step, the conditional cumulative distribution function (CDF) of Y is estimated: