Predicting the presidential election cycle in US stock prices: guinea pigs versus the pros

Abstract

The notion that US stock prices follow a pattern that is synchronized with presidential elections has been discussed among financial investors for a long time. Academic work exists that supports this idea, quantifies the pattern and has demonstrated its robustness over several decades and across parties in power. This article takes the existence and robustness of this presidential election cycle for granted and asks whether individuals exploit it when they predict stock prices. It considers and contrasts two types of such forecasts: Those made by professionals included in the Livingston survey and those made by students in a laboratory experiment. A key result is that neither group fares particularly well, though participants in the experiment outperformed the professionals.

Notes

¹Examples of academic work in this area of research are the papers by Umstead (1977), Allivine and O'Neill (1980), Huang (1985), Gärtner and Wellershoff (1995), Hensel and Ziemba (1995), Booth and Booth (2003) and Wong and McAleer (2009).

²A classic in this area of research is Dokko and Edelstein (1989). A more recent effort is Söderlind (2009).

³See Haruvy et al. (2007).

⁴During the Bush presidency, examined in isolation, the pattern proved significant only when the stock market crash of October 1987 was neutralized by means of a dummy variable.

⁵While this is the information provided by the Federal Reserve Bank of Philadelphia in early documentation after it took over conducting and administering the Livingston surveys, it later felt compelled to caution users, stating that ‘that description was a bit too general’. For details see the online documentation provided at http://www.philadelphiafed.org/files/liv/NewFilesJun04/SPI/Old_Stock_Price_Web_Doc.pdf

⁶Others who opted for this approach include Dokko and Edelstein (1989) and Söderlind (2009).

⁷Even here we should note, though, that individuals observe the stock market even when they do not participate in the Livingston survey. So their real-world experience reaches far beyond the sample of predictions each one contributes to the survey.

⁸ One should note, however, that some data points were lost with this new specification, bringing the number of observations down to 25.