Types of R&D investment and firm productivity: UK evidence on heterogeneity and complementarity in rates of return

ABSTRACT

Existing evidence on the impact of R&D on productivity is heterogenous and does not address the question of whether different types of R&D are complements or substitutes. The aim of this research is to open the R&D black box by providing fresh insights about how different R&D types affect productivity in different industrial and technological contexts in the UK. The model adopted allows for non-linearities between R&D and productivity and interactions between R&D types. The analysis makes use of micro data from the Office of National Statistics, comprising 8284 firms from 1998 to 2012. The results show evidence of diminishing marginal returns to total R&D. This concave relationship also holds for intramural R&D, applied/experimental R&D and private R&D. These findings suggest that studies which do not allow for non-linear relationships between R&D and productivity could suffer from specification bias. The results also indicate complementarity between intramural and extramural R&D and between basic and applied/experimental research. Returns to publicly funded R&D are insignificant and there is neither complementarity nor substitution between publicly and privately funded R&D. The findings strengthen the case for modelling the sources of heterogeneity explicitly by taking into account non-linearities and interactions between the different R&D types and productivity.

KEYWORDS:

• Firm productivity
• R&D
• Pavitt class
• basic research
• applied research
• experimental research
• in-house research
• extramural research
• privately funded R&D
• publicly funded R&D

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Lokshin, Belderbos, and Carree (2008) specified changes in knowledge capital stock as a function of Internal and External R&D only. I expand this to include other types of R&D.

2 The first-differenced GMM estimator of Arellano and Bover (1991) is constructed by taking the first differences of all variables and instrumenting them using the lagged level values of the series. However, the first differenced estimator suffers from bias in short persistent panel because the lagged values become poor instruments of the first differenced series (Bond, Hoeffler, and Temple 2001). The system GMM estimator was created to overcome this problem. The estimator augments the first differenced series in the first differenced estimator with the original level series instrumented with the lagged first differences of the series. It is the validity of these extra instruments (lagged first differences) that the third condition tests.

3 The standard disclaimer applies: the use of these data does not imply the endorsement of the data owner or the UK Data Service in the interpretation or analysis of the data.

4 Unfortunately, I was not able to extend the data set due to funding limitations and the prohibitive time cost involved in drawing data from the two sources. In addition, access to the database through the Secure Data Service was disrupted between March and April 2020 which was the period when the paper was being revised. Note that there are several cases in the literature where there is a time lag between the year of publication and the end year of the data. E.g. the sample period of Grillitsch, Schubert, and Srholec (2019) was from 2004 to 2011.

5 The depreciation rate of 9.3% is the average depreciation rate of plant and machinery (6%), buildings (2%) and motor vehicles (20%).

6 To control for double counting, R&D capital expenditure on land and building, and plant & machinery were deducted from net capital expenditure.

7 See ONS (2015b) for more precise definitions.

8 There is a fifth Pavitt class consisting of unclassified industries which is not report, but is available upon request.

Additional information

Funding

This research has been funded by the ESRC [grant number ES/K004824/1].