D S E Dipartimento Scienze Economiche Working Paper Department of Economics Ca’ Foscari University of Venice Fabrizio Castellucci Giovanni Pica Mario Padula The age-productivity gradient: evidence from a sample of F1 drivers ISSN: 1827/336X No. 16/WP/2009 Working Papers Department of Economics Ca’ Foscari University of Venice No. 16/WP/2009 ISSN 1827-3580 The age-productivity gradient: evidence from a sample of F1 drivers Fabrizio Castellucci Università Bocconi Giovanni Pica University of Salerno and CSEF Mario Padula Ca’ Foscari University of Venice and CSEF July 2009 Abstract Estimating the effect of aging on productivity is a daunting task. First, it requires clean measures of productivity. Second, unobserved heterogeneity at workers, firms and workers/firms level challenges the identification of the age-productivity gradient in cross- sectional data. Finally, the study of the age-productivity link requires to partial out the role of experience and to account for the selection bias that arises if less able people drop out faster than more able ones. We tackle these issues by focussing on a panel of Gran Prix Formula One drivers and show that the age-productivity link has an inverted U-shape profile, with a peak at around the age of 30-32. We would like to thank seminar participants at the University Ca' Foscari of Venice, the University of Milan, and conference participants at the Brucchi-Luchino conference, Bologna, December, 11-12, 2008 and the Third ICEE Conference, Ancona January 30-31, 2009. The usual disclaimer applies. Keywords: Aging, individual effects, firm effects, match effects, Formula One JEL Codes: J24,C23, L83 Address for correspondence: Mario Padula Department of Economics Ca’ Foscari University of Venice Cannaregio 873, Fondamenta S.Giobbe 30121 Venezia - Italy Phone: (++39) 041 2349184 Fax: (++39) 041 2349176 [email protected] This Working Paper is published under the auspices of the Department of Economics of the Ca’ Foscari University of Venice. Opinions expressed herein are those of the authors and not those of the Department. The Working Paper series is designed to divulge preliminary or incomplete work, circulated to favour discussion and comments. Citation of this paper should consider its provisional character. The Working Paper Series Department of Economics is availble only on line Ca’ Foscari University of Venice (www.dse.unive.it/pubblicazioni) Cannaregio 873, Fondamenta San Giobbe For editorial correspondence, please contact: 30121 Venice Italy [email protected] Fax: ++39 041 2349210 1 Introduction Aging is a global phenomenon. If older individuals are less produc- tive, an aging working population can lower aggregate productivity, eco- nomic growth and fiscal sustainability. Therefore, understanding the age- productivity gradient is key in a aging society. However, estimating the effect of aging on productivity is a daunting task. First, it requires clean measures of productivity. Wages are not such measures to the extent that they reward not only productivity but also other workers’ attributes. Sec- ond, unobserved heterogeneity at workers, firms and workers/firms level challenges the identification of the age-productivity gradient in cross- sectional data. Longitudinal data attenuate some identification issues, but give rise to the problem of partialling out the effect of aging from the pure effect of time. Thereby, the study of the age-productivity link in a panel framework requires investigating and partialling out the role of experience and seniority. Third, selective entry and exit may take place as more productive workers may enter the labor force at younger ages and exit at older ages. Fourth, jobs differ with respect to the skills they require and different skills may evolve very differently over different working careers. The literature on micro data has tackled some of these issues, but some remain unresolved. The available evidence seems to indicate that the elderly do suffer a drop in productivity. Medoff and Abraham (1980, 1981), and Waldman and Avolio (1986) use supervisors’ rating to mea- sure productivity and show that older workers are less productive than younger ones. These early studies have been important attempts to tackle the above mentioned issues, but still suffer from severe shortcomings. First, being most of these studies based on cross-sectional data, they are unable to disentangle the effect of age from the effect of tenure, and are unable to control for the fact that workers may self-select into firms according to their productivity. Second, supervisors may tend to over- reward senior workers for loyalty and past achievements and therefore supervisors’ rating might be only an imperfect proxy for individual pro- ductivity. 1 These shortcomings are later addressed in the literature. Stephan and Levin (1998) study researchers in the fields of Physics, Geology, Physiol- ogy and Biochemistry. The number of publications and the standard of the journals they appear in are found to be negatively associated with the researchers’ age. Similar evidence is found in the field of economics where Oster and Hamermesh (1998) conclude that older economists publish less than younger ones in leading journals, and that the rate of decline is the same among top researchers as among others.1 The productivity of in- dividuals doing “creative” jobs, such as authors and artists, is measured by the quantity and sometimes the quality of their output. The evidence seems to indicate that the elderly are less productive. Kanazawa (2003) shows that age-genius curve of scientists bends down around between 20 and 30 years. Similar curves are also found for jazz musicians and painters. Fair (2007) focuses on record bests by age for athletes and chess players and shows that the decline starts around age 35. These papers share a common feature, the use of piece-rate samples, which provide a clean measure of productivity. However, given the cross-sectional nature of these studies, they cannot easily deal with unobserved heterogeneity and with selection bias problems. Those settings also do not allow to investigate the effect of the worker-firm match on productivity and to separately identify the effect of age from the effect of experience. A different approach consists in analyzing the age-productivity rela- tionship exploiting employer-employee matched data-sets. The evidence based on such data-sets, where individual productivity is measured as the workers’ marginal impact on the company’s value-added, finds an in- verted U-shaped work performance profile.2 Individuals in their 30s and 40s have the highest productivity levels. Employees above the age of 50 are found to have lower productivity than younger individuals, in spite of their higher wage levels. These papers basically estimate the effect of aging on productivity by comparing output (or value added) per worker in plants (or firms) with a different age composition of the workforce. 1The same pattern applies to Nobel economists (Dalen, 1999) 2See Andersson et al. (2002), Cr´epon et al. (2002), Ilmakunnas et al. (2004), Halti- wanger et al. (1999), Hægeland and Klette (1999). 2 Again, a problem with most of those studies is that they are based on cross-sectional evidence (with the notable exception of Dostie (2006)). Thus, reverse causality may be at work: for example, successful firms generally increase the number of new employees and this mechanically leads to a younger age structure. Thus, a younger workforce could be the effect rather than the cause of firms good performances. In order to overcome these problems, this paper casts the age-productivity test within the worker-firm pair in a panel context.3 We rely on a unique data-set that records the race performances for all Gran Prix Formula One (F1) drivers from 1991 to 1999. Thus, our data-set shares the advan- tages of both piece rate samples and of longitudinal employer-employee matched data-sets. It provides a clean measure of productivity and has enough information to identify the age-productivity profile after con- trolling for a host of workers and firms characteristics. First, the panel nature of the data allows to include drivers, firms and match effects, thus controlling for the different sources of unobserved heterogeneity at the driver/firm/driver-firm level. The presence of drivers fixed effects also accounts for the possibility of drivers entering and exiting the sample in a non-random fashion, as long as drivers’ participation decisions are driven by fixed drivers’ attributes.4 Second, but equally importantly, our data allow to account for experience, as measured by the age of entry in the F1 industry, and still identify the age-productivity gradient. Gen- erally, the age of entry cannot be separately identified from the driver effect. However, our measure of productivity records for each driver his relative performance, which allows to restrict to zero the sum of drivers’ effect and therefore provides the extra condition to identify the effect of the age of entry on productivity. 3A recent paper by Van Ours (2009) focuses on runner, economists and manufac- turers data to provide evidence on the age-productivity gradient. The data on runners and economists are longitudinal, but do not allow to cast the age-productivity gradi- ent in the employer-employee pair framework. Manufacturers data allows to control for the employer-employee match: to account for the changing composition of the work force the estimates are obtained using an IV approach. 4As a robustness check, we use propensity score weights to deal with the additional non-random attrition coming from exit decisions driven by time-varying characteris- tics. 3 Our estimates show that productivity peaks at the age of 30-32 and then decreases. In accordance with the findings of Abowd et al. (1999), we also show that workers effects are more important than firms and match effects in determining productivity, as they account respectively for 25, 12, and 2 percent of the explained variability.