Princeton Election Consortium

Innovations in democracy since 2004

Outcome: Biden 306 EV (D+1.2% from toss-up), Senate 50 D (D+1.0%)
Nov 3 polls: Biden 342 EV (D+5.3%), Senate 50-55 D (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

InTrade inefficiencies

November 3rd, 2008, 11:06pm by Sam Wang

Electronic markets such as InTrade are driven by polls and arrive at similar probabilities. But a few exceptions offer opportunities. As a general rule, simple political contracts match the sign of the polling margin closely. For example, if a McCain contract is trading above 50, it is nearly certain that he leads in an average of polls, and vice versa. However, when polling margins are in the 3-10 point range, inefficiencies arise:

In general, electronic markets monitor the wisdom of crowds. One benchmark is Presidential races, where they folllow polls fairly well; see the current standings at They lag polls a bit and express much less certainty than polls, even on Election Eve. Both phenomena may account for why the current standings there are Obama by 37 EV, half the margin reported here.

As you can see in this graph of InTrade prices versus polling margins, even a 10-point margin isn’t enough to saturate bidder confidence:

InTrade prices vs state polls

InTrade prices vs state polls

A highly significant polling margin, statistically speaking, can have a price in the 20-80 range, which is an understatement of confidence by the market. This understatement is consistent with observations of individuals by Kahneman, Tversky, and others that we overestimate the probability of highly unlikely events, which I describe in Welcome To Your Brain.

The same is true for “aggregated” InTrade contracts, which are usually priced within a
factor of 2 of their actual value as calculated from polls.

Discrepancies in prices versus poll-based probabilities fall into three easily identified categories.

1) State-level contracts.

OHIO.DEM, “Democratic Party Nominee to win Ohio’s Electoral College Votes in 2008 Election,” is trading at 80, but Obama leads or ties McCain in all of the 10 polls taken over the last week. I would buy this contract.

ARIZONA.DEM is trading around 19, yet McCain leads in 8 polls. I would short sell this contract.

GEORGIA.DEM is trading around 22, yet McCain leads in 6 polls. I would short sell this contract.

2) The Electoral College. For most of October, the Meta-Analysis has fluctuated in the range of 350-368 EV for Barack Obama. Its dynamics are similar to what happened in 2004, where it fluctuated around the eventual result of Bush 286 EV, Kerry 252 EV. The top end of the 95% confidence
interval has stayed relatively constant, just below 380 EV. This is a structural problem that I have referred to as Obama’s red ceiling.

DEM.ELEC.COLLEGE.>=380 is at 20.5. Because its value is extremely low, I recommend short selling this contract.

3) The House of Representatives. Aggregated polls at currently indicate 257+/-3 seats for the Democrats, with a 68% CI of (254, 260) and a 95% CI of (251, 263).

08.HOUSE.DEM.261-270 is at 46.0, but is actually worth about 16. I would short sell this contract.

08.HOUSE.DEM.251-260 is at 35.0, but should be worth about 80. I would buy this contract.

The House trades carry substantial risk. I include them mainly because they are more fun. Of course, I would not recommend making both HOUSE.DEM trades since they are nearly equivalent.

Tags: 2008 Election

7 Comments so far ↓

  • BCC

    I’d buy Georgia, myself.

  • berb

    I’m an Intrade user myself and just the other day I sold off a bunch of shares in Obama @ 85 each (even though I am certain they’ll be worth 100 after tomorrow) simply so I could invest more into “08.HOUSE.DEM.251-260” because I bought in at 25 per share and that was just too good to pass up.

    If only I had some magical way of sending them a check overnight to add funding to my account, then I could have kept the Obama and still invested some into other stuff… oh well.

  • Nicholas J. Alcock

    Dear Sam.
    All buyers need information. They buy their information from 1)polls 2)reports
    3)inside knowledge 4)intuition.

    Now, not all buyers are as wise as each other.
    My guess(hunch) is that electronic markets are less accurate than 538, PEC, VEC et al.
    I think if you check the data I would be right.?
    Of course, the market makers would say they are better judges but ask Lehman Bros about that?

  • Marc in Asia

    Betting markets such as Intrade don’t do well for extreme probabilities (>90%, <10%) for a few reasons. One is the bias in assessing low probability events. A bigger one may be that participants loose interest:

    Some participants are mostly in it for fun or gambling. There is little fun or glory in making a small profit on an almost sure thing. Thus players will prefer to play a more exciting contract. Berb above is a great example of this.

    Other participants are in it for a serious profit. In liquid financial markets, they can take advantage of even small discrepancies to profit. However on Intrade, where there are substantial trading fees and no interest is paid on deposits, there is little profit to be had on correcting a market mispricing of only a few percent. This is why, for example, the Al Gore contract is still trading at 0.3 cents — no one cares enough to correct it.

    The best thing is to think of Intrade as accurate to within 5%. Once the likely probability goes above 85-90% or below 10-15%, Intrade is no longer a useful indicator.

  • Eddie

    I agree (of course) that the average citizen overestimate the probability of highly unlikely events (as has been shown with death data), but it looks like you’re saying that a McCain win is a highly unlikely event based solely on the data from these new combined-polling models, and doing this in a year in which the currently-winning presidential candidate is Black.

    It seems to me that the average citizen overestimates the probability of the less-likely event for this election (, but the combined-polling sites are overestimating (massively in some cases) that same probability. I think the distribution for this election is unknowable/unmodelable, but that 5-10% McCain from the markets on the last day is the guess that I would take.

    I don’t know for sure if this is the same thing, but don’t folks in the financial world get into trouble (‘blow up’) for relying too much on statistical models?

  • Nigel Eccles

    There is of course the possibility that bettors are pricing these extreme events more accurately. Beware of financial models projecting events as ‘extremely unlikely’ (like the current financial crisis).

  • F Lo

    to arbitrage the Gore contract would mean posting collateral of $100 in the hopes of winning 0.03 cents. Let’s say you do it today, so you are playing for a daily return of 0.0003% or an 0.11% annual rate. You are probably better off leaving that money in your savings account.

    (this assumes no transaction costs too)

Leave a Reply to BCC (Cancel)