The value of celebrity endorsements: A stock market perspective

It is important to note that celebrity endorsements come at a significant cost to companies. The fees in our sample range from fairly modest (£0.75 million paid by Avon to the Williams sisters or $3 million paid by Hershey Co. to Jessica and Ashlee Simpson) to very significant ($25 million paid by GM to Tiger Woods or £30 million paid by Sony to Dale Earnhardt Jr.). Endorsement fees are not the only expense of the endorsement campaign. Other outlays (management expenses, TV ad costs, etc.) have to be considered as well. Our goal, however, is not to explicitly analyze endorsement fees and costs, but rather concentrate on the net present value of a celebrity endorsement, which accounts for all cash flows associated with it, and should be manifested in announcement-day stock returns.

A somewhat related strand of literature analyzes the impact of sponsorship contracts on stock returns. The results are generally mixed. Mishra et al. (1997) analyze 76 announcements of sponsorship events. They document positive abnormal returns on announcement-day, but insignificant returns for longer event windows. Farrell and Frame (1997) document negative announcement-day returns for 26 sponsorship announcements at the 1996 Atlanta Olympic Games. However, Miyazaki and Morgan (2001) document positive abnormal returns (albeit for a very specific event window) of 27 Olympic sponsorship announcements. Samitas et al. (2008) find marginally positive returns for Athens Olympic Games sponsorship announcements. Clark et al. (2009) analyze the impact of sponsorship announcements of tennis and golf tournaments, auto racing, and college football on stock returns in an event study setting. In general, the impact is not statistically significant. However, congruence between sport and a sponsor is significantly related to perceived sponsorship success.

On December 12, 2009, Gillette announced that they would stop all advertising involving Tiger Woods. On December 13, Accenture has severed its ties with Woods.

Such as CNN, Reuters, and BBC.

These are Deutsche Telecom, Siemens and Sony.

Seven endorsement extensions are identified. As a robustness check, we include these observations back into the sample and find no significant impact on our results.

Pritamani and Singal (2001) use three alternative thresholds to identify significant price changes—5% and 10% and three standard deviations in returns. Their results are generally robust to the choice of a cut-off. We select the most conservative threshold.

Because in daily return data, periods of low volatility alternate with periods of high volatility.

We choose the same event window as Agrawal and Kamakura (1995).

The matched sample is available from the corresponding author upon request.

Actual sample mean (median) estimates of α and β equal 0.00051 (0.00054) and 0.85 (0.86), respectively.

We do not report these results to save space. They are available from the corresponding author upon request.

Following normal practice, the dummy variable takes value 1 to indicate the presence of a certain characteristic, and otherwise it takes value zero. When testing for the match-up hypothesis however, the dummy takes one of three values: one when congruence is assumed, minus one when incongruence is assumed, and zero otherwise.

To save space, we do not report the detailed results. These are available from the corresponding author upon request. The idea behind this approach is that an early leak of news would generate an early increase in trading volumes. Our results confirm the presence of such an effect for the day before the announcement date. We find that the average abnormal trading volumes are significantly positive one day before the event day according to both the Patell test and the Rank test. Abnormal trading volumes are also significantly positive at the event day. Moreover, they are also positive on the first day after the event day, which may be an indication that not all news is fully digested at the announcement date. The reason may be that event announcements reported in newspapers typically lag corporate press releases by 1 day. However, this elevated trading effect on the first day after the announcement is not significant. Interestingly, abnormal trading volumes are not significantly different from zero any other day within the first week before or after the announcement date. The trading volume effect thus is short-lived in our sample.

To save space we do not report the estimates of the constants \( \gamma_j^A \) and \( \gamma_j^{CAR} \). They are all insignificant at the 5% level.

As a robustness test, we ran regressions on standardized (instead of raw) cumulative abnormal returns as well. The results are qualitatively very similar: The estimate of the match-up dummy remains significant at the 10% level, and all other parameter estimates remain insignificant. As a further robustness test, a Kolmogorov–Smirnov test was conducted. It shows that normality of the return distribution can not be rejected.

Louie et al. (2001) document that the endorsed firm witnesses negative stock returns when blameworthy endorsers are involved in undesirable events.