ABSTRACT
The difference in the price of the stock between the beginning and the end of the event window is proportionate to the impact of the policy change on the price of the company’s stock. However, policy changes rarely appear suddenly. Instead, they are created over many months or years. In order to judge the full impact of the policy change on the company’s stock, the researcher must judge the prior probability of the event at the beginning of the window. We call the chosen prior probability π. Since the posterior probability of the event is always 1, the full effect E of the policy change on the company’s stock is given by:2
, (3.1)
where st is the price of the company’s stock at time t. It should be clear that if the researcher chooses a different start time t, end time t′, or prior probability π, the event study can produce vastly different results in terms of the size and even sign of the estimated effect E. In particular, the longer the event window, the more likely it is that other, unrelated events will occur which will bias the results. Prediction markets can mitigate these problems. While prediction markets may take many forms, here we focus on the market for a contract that pays $1 if a certain event, such as a policy change, happens and zero otherwise.3 The price at any given time thus represents the market estimate of the probability of that event happening.4 Prediction markets give the researcher an accurate measure of π, the prior probability of an event happening, and may also help to identify an appropriate event window. Moreover, changes in the company’s stock price that are unrelated to changes in the probability of the policy change will show no correlation with changes in the price of the prediction market contract. Thus, the event window does not need to be carefully chosen to exclude other events. Finally, as the probability of an event may change many times in response to political events, each probability change can be analyzed as a separate event, where the change in probability of the policy change is accurately measured. By separating a single event window into many small sub-windows and then taking an appropriately weighted average effect across the sub-windows, prediction markets allow for more precise estimates than traditional event studies. This chapter consists of three examples from our previous research (Snowberg et al., 2007a,b).5 Each example illustrates a particular problem with traditional event studies and shows how the inclusion of prediction markets produces a more accurate estimate of the economic effect of political events. It should be noted that the three issues with traditional event studies are all inter-related – so each example will contain some elements of the other issues. Where possible, we show the difference between our results and research using traditional event studies or misusing prediction markets. The examples in this chapter provide insight into several questions, as well as demonstrating the methodological usefulness of prediction markets. First, we show that in the 2004 US presidential election, candidate convergence did not
occur, as predicted by Downs (1957) and many other models. Specifically, the stock market rose 2 percent in value on news of a Bush victory (over Kerry). Second, we show this difference of 2 percent between Republicans and Democrats has been remarkably consistent over time, appearing in an analysis of all elections between 1880 and 2004. This suggests that whatever the changes in party structure and policy issues over that period, Republicans have consistently been the party of capital, and Democrats the party of labor. Finally, we show that the stock market declined in response to the news of a Democrat victory in the Senate (and House) in 2006, suggesting that, contrary to conventional wisdom, markets do not prefer divided control of the legislature and executive to unified control of both branches.
