THE IOWA ELECTRONIC MARKETS

The IEM are real money-prediction markets operated at the University of Iowa’s Henry B. Tippie College of Business. They have proven accurate in predicting election outcomes (Berg et al. Reference Berg, Forsythe, Nelson, Rietz, Plott and Smith2008) and generally forecast vote shares better than polls (Berg, Nelson, and Rietz Reference Berg, Forsythe, Nelson, Rietz, Plott and Smith2008). The relative accuracy of the IEM versus polls actually increases farther in advance of an election.

Because the IEM is described elsewhere, ^{Footnote 2} this article discusses only the IEM tied to the 2014 US House and Senate elections. Three markets predicted the control configuration of the House and Senate as a result of the 2014 US election: the House Control Market, the Senate Control Market, and the joint Congressional Control Market. We use the Congressional Control Market for three reasons. First, it shows the likelihood of various control combinations of the two chambers, which allows a simultaneous analysis of the combined effects of events on House and Senate control probabilities. Second, it represents whether parties have absolute control of each chamber. It does not have contracts associated with increasing or decreasing the degree of control (as in the other two markets) and neither does it effectively ignore third parties (which could play a pivotal role in a closely divided chamber). Third, it is more liquid and, therefore, more likely to price outcomes efficiently. ^{Footnote 3}

The Congressional Control Market prospectus is in the online appendix associated with this article. Table 1 lists the contracts traded in the market.

Table 1

Contracts Traded in the 2014 IEM Congressional Control Winner-Takes-All Market

Market participants trade contracts with payoffs based on the joint outcome of Senate and House control. Each contract pays $1 or $0 based on the composition of Congress following the 2014 US elections. The contracts represent possible combinations of Democratic and Republican House and Senate control, with an “Other” contract representing neither major party outright controlling one or both chambers because of independent and third-party seats.

The contract representing the actual election outcome pays $1. All other contracts expire worthless. Contracts are created when traders buy “unit portfolios” (i.e., one of each contract) from the IEM exchange. The IEM continuously stands ready to buy or sell unit portfolios for $1 each. This forces the risk-free rate of return to zero. In addition, there are always equal numbers of each contract at any time, which creates zero-aggregate market-level uncertainty. Together, these factors imply that contract prices should equal expected values. Thus, for each contract:

where $P_t^i$ is the price of contract $i$ on date $t$ , ${E_t}\left( {LV_T^i} \right)$ is the date-t expected liquidation (i.e., payoff) value of contract $i$ at the terminal date ( $T$ ), and $p{r_t}$ represents the probability given date-t information. ^{Footnote 4} Thus, the contract price equals the forecast probability that the contract’s event will occur: for example, the DH_DS14 price forecasts the probability that Democrats will control the House and the Senate, the DH_RS14 price forecasts the probability that Democrats will control the House and Republicans will control the Senate, and so forth. Price changes show how traders’ beliefs regarding the parties’ control chances evolve over time.

TIMELINES, EVENTS, AND PRICES

Figure 1 shows Congressional Control Market prices from September 1, 2013, through February 28, 2014. This period includes several significant events: the government shutdown, the rollout of ObamaCare and periodic enrollment reports, and the Senate nuclear option on filibusters. For most of this period, prices forecasted that the most likely election outcome would be continued Republican House control and Democratic Senate control. However, major price swings indicated significant shifts in the control prospects of the parties, leading to a Republican sweep (i.e., the actual outcome) as the most likely outcome by the end of the period. We asked whether significant swings appear related to major political events identified in the news.

Figure 1

Prices, Contract Volumes, and Events in the 2014 Congressional Control Market from September 1, 2013, through February 28, 2014

Instead of relying solely on judgment to identify significant price swings, we used statistical price-change measures related to χ²-tests to identify major and sustained price changes. We used these measures to tag significant events, as follows:

1. 1. We computed the one-day change measures for each day relative to the prior day, eliminating days that were not above the 90th percentile.

2. 2. We computed the two-day change measures across a day (prior to the next day), eliminating days that were not above the 90th percentile.

3. 3. We computed the two-day change measures leading up to a day (current versus two days prior), eliminating days that were not above the 90th percentile.

The first step identified significant price changes. The second and third steps assured that the changes neither result in a reversal the next day nor from the reversal of a prior significant change.

The measure we used mirrors the weighting scheme in a χ²-test of independence in a contingency table. Specifically, we defined the measure of price change from day t to day t+j as follows:

where prices are normalized and the sum is across contracts in the market. The numerators capture sizes of price changes of individual contracts; the denominators weight these by relative sizes of initial prices. Intuitively, a given price movement compared to high- and low-probability outcomes is more likely statistically significant than the same change relative to mid-range probability outcomes. ^{Footnote 5}

Dates in 2013 that were tagged as significant by passing all three measures are October 1; November 14, 15, 23, and 24; and December 18. These dates appear as vertical bars in figure 1 .

For comparison, we collected national polling data on overall congressional approval ratings from PollingReport.com. The polling organizations and the specific questions used for analysis are listed in table 2. For each poll, we defined the normalized net approval rating as follows:

Table 2

Congressional Approval-Rating Polls Selected for Analysis

where i indexes the polling organization, t indexes the date, %Approve is the percentage of respondents who approve of the job that Congress is doing, and %Disapprove is the percentage of those who disapprove. Figure 2 shows net approval ratings between September 1, 2013, and February 28, 2014, as well as IEM-tagged significant events. Overall, net approval ratings are negative. ^{Footnote 6}

Figure 2

Normalized Net Approval Ratings from Polls and IEM-Tagged Events between September 1, 2013, and February 28, 2014

To identify political events perceived as significant in the news, we conducted a Lexis-Nexis search of news articles using the keyword “Congress.” After identifying events with numerous articles, we narrowed the search to the New York Times to create the summaries listed in table 3. This is by no means a comprehensive list, but it allowed us to correlate major political events with the IEM-tagged events. ^{Footnote 7}

Table 3

Partial List of Political Events between September 1, 2013, and February 28, 2014

SUPPLEMENTARY MATERIAL

To view supplementary material for this article, please visit http://dx.doi.org/S1049096515000785.