Given that new political poll releases have been a bit sparse over the Xmas/New Year’s period I thought I’d branch out and look at some of the stock on iPredict.

iPredict is a New Zealand-based prediction market, and they are currently running stocks based on the result of the first Roy Morgan political poll for New Zealand to be released in 2010. In fact they have two bundles of stocks for the results of the poll for both National and for Labour. The prices of the stocks earlier this afternoon are shown in the screen captures below.

Screen capture of NAT.JAN10 stocks for first Roy Morgan poll of 2010 on iPredict taken on 2010/01/12.

Screen capture of LAB.JAN10 stocks for first Roy Morgan poll of 2010 on iPredict taken on 2010/01/12.

Each stock will only pay out if the polling result falls into a predefined range; for example, if the poll shows support for National of between 53.5% and 55.5% inclusive then the NAT.JAN10.HIGH stock will be judged at $1 and the other four stocks in the bundle will be judged at $0. The values listed next to “Probability” give the implied probabilities for each stock to pay out based on the most recent trade. Unfortunately the stocks don’t give an expected polling result, but it is possible to figure out roughly what the market is pricing the results at by calculating a weighted average of the stocks in each bundle. A very naive average of the stocks for both National and Labour gives polling expectations of approximately 52.8% +- 2.4% for National, and of 30.3% +- 2.2% for Labour’s polling

It is interesting to compare these results with the most recent Kiwi Poll Guy polling averages from the 18th December 2009 which had National polling at 55.0% +- 1.4%, and Labour polling at 29.7% +- 1.1%. Based on these numbers the iPredict stocks appear to be priced very appropriately. Any differences between the most recent polling averages and the current stock price averages lie within a reasonable margin of error, and the standard deviation also seems feasible given that there have been no political polls for a month or so, and opinion may well have changed over the Xmas/NY period. Additionally, the slightly larger implied variation on the stocks seems reasonable given that the poll will most likely be based on about 1000 respondents, introducing an additional increase in the standard deviation of the results of approximately +- 1% on top of the margin of error in the polling averages.

To analyse the results further, and try and look for a few possible bargains, I’ve tried to determine the payout probabilities for each of the five stocks in the LAB.JAN10 and NAT.JAN10 bundles as determined by a MC Simulation of the January 2010 Roy Morgan poll based on the most recent polling averages. The simulation takes pollster bias into account by comparing the results of previous Roy Morgan polls with those of other pollsters. It also takes into account the statistical error caused by the limited number of respondents, assumed here to be distributed according to a normal distribution with mean and standard deviation of 1300 and 100 respondents, respectively. The simulation does not take into account any changes in support for any of the eight major or minor NZ political parties since the most recent polling averages were calculated in December, and should therefore be taken with an appropriate amount of caution. The results are shown in the table below.

Table showing the payout probabilities for each of the five stocks in the LAB.JAN10 bundle (left) and the NAT.JAN10 bundle (right) on iPredict, as determined by MC Simulation of the January 2010 Roy Morgan poll based on recent polling averages.

When comparing these results with the current stock prices shown above there is again fairly good agreement, with the possible exception of the National VLOW stock and the Labour HIGH and VHIGH stocks, which appear to be slightly overpriced, and the Labour LOW stock, which appears to be slightly underpriced. There doesn’t, however, appear to be a lot of money to be made here by making single one-off trades, either by trading and waiting for the market to correct to cover your position, or by holding until judging takes place.

In order to make any profits here I think it is necessary to take a more in-depth look at the numbers and try to find some kind of correlation between the stocks in the National and Labour bundles. The probabilities for each of the 25 different possible permutations are shown in the following table.

Table showing the payout probabilities for each of the different permutations of the five stocks in the LAB.JAN10 and NAT.JAN10 bundles on iPredict, as determined by MC Simulation of the January 2010 Roy Morgan poll based on recent polling averages.

Here the five rows correspond to the five possible outcomes for the LAB.JAN10 stock (VLOW, LOW, MID, HIGH, VHIGH), and the five columns correspond to the five possible outcomes for the NAT.JAN10 stock (also VLOW, LOW, MID, HIGH, VHIGH.) You can see a string of more-likely outcomes running along the lower-left to upper-right diagonal, with outcomes in the upper-left and lower-right corners of the table predicted to be much less likely. This agrees with expectations, as the support for the National and Labour parties are known to be highly negatively correlated; a high level of support for the National Party implies a low level of support for the Labour Party, and vice versa.

If you compare the probabilities in this table with the prices of the stocks shown above you may start to see opportunities to make some fairly risk-free profit. For example, consider buying the stock NAT.JAN10.VHIGH at 11c and selling LAB.JAN10.HIGH at 26c, for an instant profit of 15c per pair of stocks traded. The only way this strategy can go wrong is if the result is HIGH for Labour and anything other than VHIGH for National, a probability of only about 6% according to simulation results shown in the table. In fact the actual calculations give an expected profit of 19c with a risk of 39c for each pair of stock traded, which isn’t too bad a return for a couple of weeks. The even better news is that because of the negative correlation between the polling results for Labour and National mentioned above this result is likely to hold even if the simulation predictions are slightly off. This is the main advantage of trading one stock from each of these bundles together as a pair over trading the stocks individually. Another pair worth looking at is NAT.JAN10.VLOW and LAB.JAN10.HIGH.

These trades aren’t arbitrage in the technical sense, as they are not entirely risk-free, but they’re the next best thing.

In other news, I’m wondering if perhaps I could find some way to include these numbers in the Kiwi Poll Guy polling averages. At the moment my gut feeling is that I probably shouldn’t, mainly because the stock prices seem to react more to the results of new polls after they are released rather than moving based on other non-polling factors, so I’m not quite convinced of their predicative power. I’d be interested to hear any ideas on this though. I look forward to seeing some of the stats on the trading from the iPredict team after these contracts close in a couple of weeks.

Read Full Post »