Causality and a firm-level innovation scoreboard

Abstract

In this article, we construct an innovation scoreboard based on a conceptually acceptable endogenous indicator which is shown to be the output caused by the innovation generating process. A balanced panel of data on 552 UK firms over the period 1994–2005 is used to test for causality using general methods of moments (GMM) estimation and other techniques. With innovation defined as the successful exploitation of new ideas, it is proposed that exploitation be measured by total factor productivity; success be measured by the return on capital employed and their multiple be taken as the indicator of innovation. Results for the sample firms are presented and discussed.

Keywords:

• innovation
• scoreboards
• exploitation

JEL Classification :

• O3

Acknowledgements

We would especially like to thank two anonymous referees for their comments, participants in the ‘Knowledge for growth: role and dynamics of corporate R&D’ conference organized by the JRS-IPTS of the European Union, Seville, Spain, 8–9 October 2007, in the Royal Economic Society annual conference, Warwick University, 13–17 March 2008, and in the workshop ‘Measuring innovation: The economic issues,’ Nottingham University Business School, 4 February 2009. Of course any errors remaining in the article are our responsibility alone.

Notes

The former DTI is now labelled the Department for Business Enterprise and Regulatory Reform (BERR) together with the Department for Innovation, Universities and Skills (DIUS).

However the method proposed at that time (which is very different from that proposed here) was not considered suitable for need and the report has not been released into the public domain nor have the recommendations of that study been introduced.

Several studies specify R&D in terms of a stock rather than a flow measure (e.g. Bernstein 1996, Shah 1994), and it is important to be aware of the difference between stock and flow when examining empirical studies as the stock will be much greater than the flow. However, when the equation used to estimate the impact of R&D on productivity is specified in logarithms, as it usually is, then the difference in results is not so clear (Hall and Van Reenen 2000). Moreover, the stock of R&D is generally calculated using the perpetual inventory method, but unfortunately there is little information upon which to determine the initial condition for modelling the knowledge depreciation rate. For these reasons we confined ourselves to the flow measure of R&D sourced from the DTI's UK R&D Scoreboard and Datastream.

For the purposes of the exercise it is the consolidated group accounts of the UK parent company that are used.

.

It has been pointed out to us that this approach rules out from our sample those companies that do not undertake R&D. We accept this criticism but counter that our causality approach needs a measure of R&D and this determines our sample, but application of the scoreboard once we have established the principles could be extended to samples of firms not undertaking R&D.

Numbers of employees may alternatively be used if simplicity of application is the key requirement.

It is worth pointing out that the firm-level TFP used here does not measure productivity in the traditional sense. Measuring productivity using output deflated by industry price captures both productivity improvement and increased mark-ups. Deflating by firm-specific deflators would miss some of the firm's performance enhancement such as the part that goes into higher prices via quality improvements or lower prices, via cost minimization. Therefore, our measure of TFP (which is intended to pertain to output) is actually a measure of revenue productivity, which is an appropriate measure for firm performance.

The influence of outliers has been tested and found to be relatively insignificant, especially when working with logarithms.

For more details on the GMM-SYS estimator see Blundell and Bond (1998). We use the estimated variance matrix incorporating the small-sample correction of Windmeijer (2000).

Although it is not clear why this feedback exists.

However, it is worth pointing out that levels are not appropriate if comparisons across different currency units are used. In such cases, (logarithmic) transformations would be necessary.

It has been pointed out to us that if ROCE is negative then higher TFP will place a firm lower down the scoreboard. This is in fact not illogical. If a firm is making a negative return on its investments (in exploitation), then it will increase profit performance by doing less and not more.

ROCE is also predictable from R&D.