Determinants of football transfers

Abstract

The analysis of football transfers is hampered by selectivity bias. In most empirical estimations, simple regression is used and selectivity is ignored. In this article we propose an estimation method that corrects for sample selectivity and allows the use of more observations in a simple manner. The ordered probit estimates point in a similar direction as the estimates from commonly applied estimation techniques but the significance is higher.

Keywords:

• football transfers
• sample selectivity
• ordered probit model

JEL Classification:

• J44
• J41

Notes

¹  The precise way countries are ranked is given on http://www.uefa.com/memberassociations/uefarankings/country/

²  The UEFA also published an European club ranking (http://www.uefa.com/memberassociations/uefarankings/club/). We can not utilize this ranking because many clubs are not present on the list because they did not participate in one of the international UEFA competitions in the last 5 years.

³  The information on the number of points earned in a season is taken from www.vi.nl, the website of Voetbal International, the leading football magazine of the Netherlands. It contains information on all European first football leagues.

⁴  In the UEFA club ranking Fenerbahçe SK is listed as the number 51 and Galatassaray AṢ as the number 44. The UEFA club ranking contains fewer clubs than our ranking.

⁵  We also tried using only five categories (big improvement, improvement, neutral, decline, big decline), but this reduces the already limited number in especially the category improvement even further. The estimation results were very similar.

⁶  Of course we can immediately apply OLS on this equation. The estimation results show very marginal significance (p-value of the F-test on regression = 0.218), indicating that our precise ranking itself might not be very informative.

⁷  We experimented with different class sizes both in Equations 4 and 5. The estimation results remained remarkably constant as long as sufficient numbers of observations were left in each of the categories.

⁸  In Table 4, and . In Table 5, and .

⁹  We have a wealth of individual information available but we were unable to get significant results with respect to more sophisticated variables.

¹⁰  In Table 4 not finding a new club is put down as a large decline.

¹¹  Since we restrict our analysis to players who left a football team, the selectivity discussed by Carmichael et al. (1999) is ignored.

¹²  The p-value of this variable when added to the transfer fee equation is 0.122.