A review of Islam, E. and J. Zein, 2020, “Inventor CEOs,” Journal of Financial Economics 135, 505-527.
I. Research question and its importance
Drawing on the literatures of learning by doing, this paper investigates whether having an inventor CEO leads to better firm innovation outcomes. There are three reasons the topic is important. First, CEO hands-on experience is an unexplored dimension in analysing the effect of CEO characteristics. Second, as firm’s innovation trajectory and the selection of CEO is endogenous, a randomized experiment is desirable yet impossible. Hence, making inference is especially challenging because there are numerous alternative explanations. The authors take lengthy paragraphs to deal with each of them. Lastly, this topic is also of interest for human capital practitioner because to what extent an inventor CEO is worthy its salary premium vis-à-vis non-inventor CEO remains unanswered empirically.
II. Method, finding, and limitation
The hand-collected data consists of U.S. high tech firms from 1992 to 2008. OLS with fixed effects (firm-year and firm-tech class) was utilized in the baseline model, while difference-in-difference model was employed to make causal inference with the data. More precisely, the authors exploit an exogenous CEO turnover (e.g., sudden death or health shocks, etc.) as an identification strategy. Consistent with their conjectures, CEO inventor leads to higher quality of innovation, such as more patents, citations, as well as economic values. However, since they exploit only the departure of inventor CEO (and a non-inventor CEO replacement), their causal claim tells only half of the story. In other words, it’s more interesting to see if hiring an inventor CEO leads to better innovation outcome as postulated.
III. Improvement
The difference-in-difference method has several assumptions (Bertrand et al. 2004) but the paper only covers part of them. While the authors test for parallel trend assumption and cluster standard error at firm level, they do not test for serial correlation, neither offering a matching method that is more generally available (e.g., PSM or CEM) to replicating the research in another context. Moreover, the treatment group consisting of only 17 observations puts the results less robust, albeit it was due to data unavailability. Finally, the period selection of DiD specification (three years before and after the event) does not reflect the average tenure (e.g., 6 to 7 years) reported by the authors. In brief, the temporal impact of CEO inventor on innovation is always the query for practitioners because it helps in devising compensation contract with CEO.
IV. Future research
Future work may replicate the design to assess to what extent university research outcome (measured by patent, grant, or awards) is affected by the experience of its Principal (measured by impact score and fields of research). Exogenous shock such as universities merger can be exploited for this research.
V. Reference
Bertrand, M., Duflo, E., Mullainathan, S., 2004. How much should we trust differences-in-differences estimates? The Quarterly journal of economics 119, 249-275