Advances in Business and Management Forecasting Forecasting the 2008 U.S. presidential election using options data Christopher M. Keller, Downloaded by Australian Catholic University At 05:13 25 October 2017 (PT) Article information: To cite this document: Christopher M. Keller, "Forecasting the 2008 U.S. presidential election using options data" In Advances in Business and Management Forecasting. Published online: 08 Mar 2015; 173-182. Permanent link to this document: https://doi.org/10.1108/S1477-4070(2010)0000007015 Downloaded on: 25 October 2017, At: 05:13 (PT) References: this document contains references to 14 other documents. To copy this document: email@example.com The fulltext of this document has been downloaded 106 times since 2015* Users who downloaded this article also downloaded: (2010),"Forecasting using fuzzy multiple objective linear programming", Advances in Business and Management Forecasting, Vol. 7 pp. 149-156 <a href="https:// doi.org/10.1108/S1477-4070(2010)0000007013">https://doi.org/10.1108/ S1477-4070(2010)0000007013</a> (2010),"Forecasting performance measures – what are their practical meaning?", Advances in Business and Management Forecasting, Vol. 7 pp. 137-147 <a href="https://doi.org/10.1108/S1477-4070(2010)0000007012">https://doi.org/10.1108/ S1477-4070(2010)0000007012</a> Access to this document was granted through an Emerald subscription provided by emerald-srm:401304  For Authors If you would like to write for this, or any other Emerald publication, then please use our Emerald for Authors service information about how to choose which publication to write for and submission guidelines are available for all. Please visit www.emeraldinsight.com/authors for more information. About Emerald www.emeraldinsight.com Emerald is a global publisher linking research and practice to the benefit of society. The company manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as well as providing an extensive range of online products and additional customer resources and services. Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive preservation. Downloaded by Australian Catholic University At 05:13 25 October 2017 (PT) *Related content and download information correct at time of download. Downloaded by Australian Catholic University At 05:13 25 October 2017 (PT) FORECASTING THE 2008 U.S. PRESIDENTIAL ELECTION USING OPTIONS DATA Christopher M. Keller ABSTRACT The 2008 U.S. presidential election was of great interest nationally and internationally. Interest in the 2008 election was sufﬁcient to drive a $2.8 million options market by a U.K.-based company INTRADE. The options in this market are priced as European style ﬁxed return options (FRO). In 2008, the Security and Exchanges Commission approved, and both the American Stock Exchange and the Chicago Board Options Exchange began to trade FROs. Little research is available on trading in FROS because these markets are very new. This chapter uses the INTRADE options market data to construct exponential smoothing forecasts, which are then compared under a hypothetical trading strategy. The trading returns indicate that this market is relatively efﬁcient at least in the short term but that because of the all or nothing payout structure of a FRO, there may exist small arbitrage opportunities. Advances in Business and Management Forecasting, Volume 7, 173–182 Copyright r 2010 by Emerald Group Publishing Limited All rights of reproduction in any form reserved ISSN: 1477-4070/doi:10.1108/S1477-4070(2010)0000007015 173 174 CHRISTOPHER M. KELLER Downloaded by Australian Catholic University At 05:13 25 October 2017 (PT) INTRODUCTION AND BACKGROUND The INTRADE market is a speculative prediction market. Prediction markets are also known as information markets, idea futures, event derivatives, or virtual markets and exist for the purpose of making predictions (Spann & Skiera, 2003). There are many prediction markets publicly available: IOWA ELECTRONIC MARKETS – generally economic or political issues; TRADESPORTS – sporting events; SIMEXCHANGE – video games; HOLLYWOOD STOCK EXCHANGE – ﬁlms and ﬁlm-related people; and INTRADE – from business to politics to art and wine and even weather, but excluding sports, which is covered by BETFAIR. Raban and Geifman (2009) provide a useful wiki page. The promise and potential of prediction markets is also discussed in Arrow et al. (2008) and Wolfers and Zitzewitz (2004, 2008). Most evidence on prediction market efﬁciency compares ﬁnal pre-election forecasts with actual outcomes but does not analyze the efﬁcacy of ongoing forecasts throughout the entire market time. The results of ﬁnal election market forecasts is mixed: Erikson and Wlezien (2008) claim that election prediction markets are inferior to polling estimates, whereas Jones (2008) claims that candidate futures provided the most accurate popular-vote forecasts. Lee and Moretti (2009) consider a Bayesian model of adaptive investor learning in the 2008 U.S. presidential futures market but considers only the ﬁnal outcome not the state-by-state electoral results. Chen, Wang, Yang, and Yen (2009) suggest that the minimum number of market participants in a futures market may be quite small (75). In the INTRADE market, the options are priced as European style FROs (FROs). A European style FRO pays out a ﬁxed amount at a ﬁxed time in the future. In the case of the INTRADE market, the pay out for each option is $10, and the time is ﬁxed by the U.S. presidential election. The option payouts are tied to whether a Republican candidate or a Democratic candidate wins the electoral votes of a particular state. INTRADE MARKET DATA This INTRADE market was composed of 102 different assets, one for each of the two primary parties within each of 51 voting districts (including the District of Columbia), and modeled accurately the electoral process of the U.S. presidential election. The INTRADE market is a more accurate election model than the perhaps more well-known Iowa Electronic Market Downloaded by Australian Catholic University At 05:13 25 October 2017 (PT) Forecasting the 2008 U.S. Presidential Election Using Options Data Fig. 1. 175 Daily Closing Prices for Pennsylvania. since the Iowa market is modeled on total popular vote that is not the method used in the actual election of the U.S. president, and the Iowa Electronic Markets are limited to positions of only $500. This chapter analyzes all daily trading data and is composed of more than 5,400 individual daily trades over the 721 calendar days after the ﬁrst trade. The data show variable forecast trends within the trading period. The state average number of Democratic trading days is 57.8 and of Republican trading days in 50.4. The state average of Democratic trading volume is $30,284 and of Republican trading volume is $26,035. Fig. 1 shows an example of the daily closing prices for the state of Pennsylvania, which was the most actively traded Democratic option. FORECASTING METHOD This chapter uses Holt–Winters method for forecasting as speciﬁed below (Winters, 1960). Let pi represent the option price on trading day i. With the standard parameters a, b 2 ð0; 1Þ, for iW1, recursively calculate the level Li ¼ api þ (1 a) (Li 1 þ Ti 1), Ti ¼ b(Li Li 1) þ (1 b)Ti 1 from the arbitrary initial speciﬁcations that L1 ¼ p1 and T1 ¼ p2 p1. For any forecast period k, then a forecast can be generated as Ft þ k ¼ Lt þ kTt. Three notes are speciﬁed regarding this chapter’s implementation of the Downloaded by Australian Catholic University At 05:13 25 October 2017 (PT) 176 CHRISTOPHER M. KELLER general method. One, since the initial conditions may be variously speciﬁed, this chapter chose to use the least extensive data for initiating the process. The initial level estimate is simply the ﬁrst data point observed. The initial estimate of the slope is simply the ﬁrst possible slope estimate from the data. Two, because the initial estimates are based on such limited data, it is necessary to allow the process to ‘‘burn-in’’ over an initial period so that the arbitrary initial estimates do not bias the overall forecasting effort. This chapter uses a burn-in period of 20 trading days. Three, the trading days in this model are not consecutive. That is, the most common applications of this forecasting method are for complete data with equal and identically spaced intervals. However, since this chapter is interested in the trading effects, the intervals here are not equal and identical calendar-wise, but are equal and identical in whether or not a trade occurs, that is trading days. In this regard, this model is reasonable since there is no necessary assumption of precluded or pent-up demand that is transacted only intermittently as in a service stock environment example like Johnston and Boylan (1996), which may use, for example, exponentially weighted moving averages (Johnston, 1993). For any forecast period k, the optimal values of a and b are determined for Republican options and for Democratic options by minimizing the total sum of squared errors across the 50 states. Table 1 shows the optimal values for a and b for four different forecast periods. The optimal value of b for all forecast periods is relatively constant at a value of about 10%. The optimal value of a however changes dramatically for the shortest forecast period of k ¼ 1 trading day from a value of about 80% to a value of about 20% for long forecast periods of k beyond 35 trading days. This relationship is more extensively illustrated in Fig. 2. Table 1. Optimal Values of Forecasting Parameters for Select Forecast Periods. Optimal values k Democratic options 1 15 35 50 Republican options a b a b 0.81 0.71 0.22 0.18 0.13 0.10 0.07 0.07 0.80 0.29 0.22 0.20 0.14 0.10 0.10 0.10 Downloaded by Australian Catholic University At 05:13 25 October 2017 (PT) Forecasting the 2008 U.S. Presidential Election Using Options Data Fig. 2. 177 Convergence of Forecasting Parameters, a and b, as Forecast Period k Increases. For the shortest forecast period, k ¼ 1 trading day, the optimal a value for both Democratic and for Republican options is very high at about 80%. This value can be understood to indicate that over such a short forecast horizon, the current market price is substantially the best estimate of tomorrow’s market price. In other words, the prediction market is at least relatively efﬁcient over a very short forecast horizon. The forecast period of k ¼ 15 trading days is the point at which there is the most divergence between the optimal values for the Republican options and for the Democratic options. For forecasts of this length, the optimal value of a for the Democratic candidate remains relatively high at about 70%, again indicated relative market efﬁciency for this forecast period. On the contrary, the optimal value of a for the Republican candidate has dropped dramatically to a value of about 29%. This much smaller value of a can be understood to indicate that for this longer period for this candidate the error-minimizing forecast is much more stable from past values. Colloquially, the support for this candidate over this period of time has stabilized or converged to a relatively constant underlying set. Unfortunately for the Republican candidate this rapid stabilization or supportive cohesion was stabilized on an ultimately losing subset. For forecasts beyond k ¼ 35 trading days, the optimal forecasting parameters for both candidates stabilized at about the same level of about Downloaded by Australian Catholic University At 05:13 25 October 2017 (PT) 178 Fig. 3. CHRISTOPHER M. KELLER Contour Plot of a and b Parameter Optimization Sensitivity for Democratic. a ¼ 20%. As would be expected the overall error for each of the forecasting methods for each of the respective candidates increases dramatically as the forecast horizon k increases. The Sum-of-Squared-Error (SSE) minimization is most greatly affected by the value of the b parameter. Sample contour plots of the optimal solution and surrounding values are shown below for a forecast period of k ¼ 15 in Figs. 3 and 4. TRADING EVALUATION An overall assessment of the forecasting process is this prediction market is applied by retroactively assessing a simple trading strategy and petting in prediction markets in general is discussed in Fang, Stinchcombe, and Downloaded by Australian Catholic University At 05:13 25 October 2017 (PT) Forecasting the 2008 U.S. Presidential Election Using Options Data Fig. 4. 179 Contour Plot of a and b Parameter Optimization Sensitivity for Republican. Whiston (2007). Since the options in question are all-or-nothing options, one of the simplest investment strategies is a buy-winner strategy that is a variant of buy-and-hold that uses a single trigger price as the ‘‘winner’’ of the buy-winner strategy. That is, if the forecast at any time exceeds a speciﬁed trigger price, then the entire volume of the market is purchased at the respective closing price. The results of this analysis suggest that there is pricing behavior in the market that warrants further research. Since the Democratic candidate in fact won the election, virtually any evaluation of a buy-winner strategy for Democratic options would be proﬁtable. This chapter instead analyzes the buy-winner trading strategy as applied only to the Republican candidate and thus is an a fortiori analysis in attempting to mitigate any post hoc in-sample error problem. Trades in these stocks also incur a $0.05 cost per option to trade and a $0.10 expiry Downloaded by Australian Catholic University At 05:13 25 October 2017 (PT) 180 CHRISTOPHER M. KELLER cost if the option is in the money. These costs are incorporated in the trading strategy returns discussed later in text. The buy-winner strategy can be applied even with no forecast. That is, this strategy could be affected by setting a trigger price, so that anytime the market price is above this trigger price, all options in the market at that time are purchased. Basically, this strategy assumes that there is a ‘‘tipping point’’ (Gladwell, 2000), at which the current price indicates a future winner. This strategy yields a positive return on investment for every trigger price above $6.60. When the k ¼ 1 forecast is applied, the buy-winner strategy yields a positive return on investment for every trigger price above $6.80. The evaluation of the buy-winner strategy gets more interesting as the forecast period increases. At the point of maximal candidate optimal parameter difference, k ¼ 15, this strategy shown a dramatic positive return on investment for every trigger price above $8.80. For forecasts beyond k ¼ 15 periods, the buy-winner strategy shows negative returns on investment. The three positive trading return strategy models current market price, k ¼ 1 forecast, and k ¼ 15 forecast are illustrated in Fig. 5. Fig. 5 establishes that both the market price basis and the k ¼ 1 forecast basis yield positive returns in the range of 5%–10% when implemented with any trigger price above approximately $6.50. Furthermore, this rate of Fig. 5. Percentage Return on Investment for Buy-Winner Strategy Based on ‘‘Winner’’ Determination Using Three Different Data Sets. Downloaded by Australian Catholic University At 05:13 25 October 2017 (PT) Forecasting the 2008 U.S. Presidential Election Using Options Data 181 return increases as the strategy trigger price increases to approximately $8.25, and then the rate of return begins to decrease. A simple least-squares approximation is added to the graph that suggests that overall the k ¼ 1 forecast very slightly outperforms the simple market price forecast. Fig. 5 also establishes that the k ¼ 15 forecast is not proﬁtable until the trigger price is at the relatively high value of $8.60, but thereafter the rate of return increases dramatically as the trigger price exceeds this value. It should be noted that as the trigger price increases, the volume, and hence the total proﬁt realized, must simultaneously decrease. This again is just a cautionary note of care that the analysis is not in any way suggesting some fool’s gold of a perfect arbitrage opportunity. Rather, this analysis suggests conservatively that further research into the behavior of pricing past a ‘‘tipping point’’ for FROs is imperative. At the least, one would expect that once prices or forecast of prices exceed some ﬁxed level, that trading at any level short of the all-or-nothing payout should virtually cease otherwise there appears to be a valuable arbitrage opportunity. DISCUSSION AND CONCLUSION These INTRADE futures are priced as European style FRO. In 2008, the Security and Exchanges Commission approved and both the American Stock Exchange and the Chicago Board Options Exchange began to trade FROs. Little research is available on trading in FROs because the markets are very new. This chapter provides illustrative information on this new market by examining the INTRADE trading data. In particular, the results from simulated simple trading strategies on the forecasted data indicate pricing anomalies that may be both potentially proﬁtable and inferentially generate better forecasts. Although many valid and justiﬁed qualiﬁers may be well applied to any simulated trading results, in any case, this chapter provides useful and informative forecasting analysis of a rich set of data with implications to a newly opened type of market. REFERENCES Arrow, K. J., Forsythe, R., Gorham, M., Hahn, R., Hanson, R., Ledyard, J. O., Levmore, S., Litan, R., Milgrom, P., Nelson, F. D., Neumann, G. R., Ottaviani, M., Schelling, T. C., Shiller, R. J., Smith, V. L., Snowberg, E., Sunstein, C. R., Tetlock, P. C., Tetlcok, P. E., Varian, H. R., Wolfers, J., & Zitzewitz, E. (2008). The promise of prediction markets. Science, 320, 977, May 16 2008. Downloaded by Australian Catholic University At 05:13 25 October 2017 (PT) 182 CHRISTOPHER M. KELLER Chen, A., Wang, J., Yang, S., & Yen, D. C. (2009). The forecasting ability of Internet-based virtual futures market. Expert Systems with Applications, 36(10), 12578–12584. Erikson, R. S., & Wlezien, C. (2008). Are political markets really superior to polls as election predictors? Public Opinion Quarterly, 72(2), 190–215. Fang, F., Stinchcombe, M., & Whiston, A. (2007). Put your money where your mouth is – A betting platform for better prediction. Review of Network Economics. Gladwell, M. (2000). The tipping point. New York: Little Brown and Company. Johnston, F. R. (1993). Exponentially weighted moving average (EWMA) with irregular updating periods. Journal of the Operational Research Society, 44(7), 711–716. Johnston, F. R., & Boylan, J. E. (1996). Forecasting for items with intermittent demand. Journal of the Operational Research Society, 47, 113–121. Jones, R. J., Jr. (2008). The state of presidential election forecasting: The 2004 experience. International Journal of Forecasting, 24(2), 310–321. Lee, D. S., & Moretti, E. (2009). Bayesian learning and the pricing of new information: Evidence from prediction markets. American Economic Review: Papers and Proceedings, 99(2), 330–336. Raban and Geifman (2009). Prediction markets wiki. Available at http://pm.haifa.ac.il Spann, M., & Skiera, B. (2003). Internet-based virtual stock markets for business forecasting. Management Science, 49(10). Winters, P. R. (1960). Forecasting sales by exponentially weighted moving averages. Management Science, 6, 324–342. Wolfers, J., & Zitzewitz, E. (2004). Prediction markets. Journal of Economic Perspectives, 18(2), 107–126. Wolfers, J., & Zitzewitz, E. (2008). Prediction markets in theory and practice. In: L. Blume & S. Durlauf (Eds), The new Palgrave dictionary of economics (2nd ed.). London: Palgrave Macmillan.