The Impact of R&D Expenditures on Firms’ Productivity in the Formal Private Manufacturing Sector in the Eastern Europe and Central Asia Economies and in the Middle East and North Africa Region

Abstract

This paper examines - through the lens of theory – the impact of R&D spending on productivity in small and medium enterprises SME’s in the Middle East and North Africa & Eastern Europe and Central Asia nations, using the BEEPS 2013 database. The aim is to contribute to the existing empirical work and drawing the conclusions from the firm level ECA and MENA dataset about the impact of R&D on labour productivity particularly in the manufacturing sector in more than 4000 firms across 28 nations in the ECA region and 3275 firms throughout 9 countries in the MENA region.

To better allow for firm heterogeneity in the analysis, the study adopted two types of matching analysis. The propensity score matching (PSM) and the Mahalanobis distance matching (MDM). The findings of this paper show that there is statistically significant impact of R&D spending (Treatment) on firms performance proxied by output per worker as the (Outcome) variable. 

The novelty of this research is inspired  by the argument of (King and Nielsen, 2016). This is where they suggested that the propensity score matching techniques, could approximate a low-standard experimental design, and could ignore much of the potentially useful information without efficient use, leaving us with higher imbalance, model dependence, and ultimately bias. Thus, these recent developments in the matching methods suggested that the conclusions drawn from PSM analysis are best supported by a second estimator, such as MDM, which has the property of double robustness, and reduces imbalance, model dependence, and bias.

To a great extent, the above (King and Nielsen, 2016) argument is proved to be true. This is where the impact of R&D expenditures on firms’ productivity was found to be statistically significant – as mentioned above – in the outcomes resulted from using both the PSM and MDM, but thedifference is that the imbalance in the confounding covariates, appeared to be extremely lower, and the bias reduction reached 100% in some covariates in the MDM outcomes compared to the PSM results.