Yet more elections

Thinking about the German election outcome, it struck me that this would be an ideal test for betting markets. I’ve always thought that, if there’s a bias in such markets it would be towards the right, so the toughest test for them would be predicting a left-wing upset win like this one (I’m calling a win on the basis that left parties got a majority of the vote, not making a prediction about what goverment might emerge). A quick Google reveals that there is such a market, called Wahlstreet but my German isn’t good enough to deal with their site, which has lots of graphs bouncing around without an obvious control. Hopefully someone will be able to help me.

Anyway, if there’s a contract allowing a bet on the share of votes for the three left parties (SPD, Greens, Left party) and if, two weeks in advance, that market was predicting a vote share of more than 50 per cent (as actually happened), I’ll concede that the case for the superiority of betting markets over polls has been established, at least as a reasonable presumption. [I didn’t follow the polls closely but I had the impression that most of them were predicting a CDU/CSU win until the last days of the campaign].

Reader Andrew Reynolds has also invited, a couple of times, my thoughts on the Japanese election. My reticence in this case is due to the fact that I have a strong antipathy towards Koizumi (the Japanese PM) that I can’t really explain. There are plenty of objective grounds I could cite, like his visits to shrines for war criminals, but my feeling is based more on a gut reaction that the guy is something of a fraud, and this colours my view of his policies. For example, I think corporatisation of the Post Office (this was the current proposal actually being debated in the recent election with privatisation a future option) is a good policy, given its history as an LDP porkbarrel, but this is scarcely an innovative or radical idea, and I see no evidence that Koizumi can do anything more than drag up warmed-over neoliberal policies which (in Australian terms) date back to the 1980s.

In addition, there’s something clearly wrong with a situation where the same party always wins even when its performance is consistently lousy for 15 years. If Koizumi were really the innovator he’s presented as, he would have split from the LDP and joined forces with the opposition.

As I say, you should take all this with a grain of salt, given the biases I’ve confessed above, but those are my thoughts for what they are worth.

17 thoughts on “Yet more elections

  1. Steve, I said “split with”. Koizumi purged his party but it’s still basically the same old LDP as far as I can see.

    GDP, Andrew Leigh and I have already debated this, and I’ve given reasons for finding this evidence unconvincing. See here for example.

  2. Hello John,

    you are right about the forecasts: no one predicted such an outcome of the election, so both parties and voters were quite surprised last sunday evening.
    I had a look at the “Wahlstreet” and it’s more a kind of site where “shares” of the parties could be traded, like a stock market. It looks like they tried to get a better prediction from that but i don’t think it has anything to do with betting.

    Cheers from Germany,

  3. Thanks – I thought your position would be interesting (in the non Yes Minister sense). I would tend to agree with you. I think it a pity that the DPJ, with its more explicit reform agenda did not do better, but at least there is now a case for reform being argued by the leaders of both of the major parties and that there is a hope that something like a two party system will emerge. Koizumi also seems to be a little bit more reluctant to visit Yasukuni – he said he would not do so when asked to during the campaign.
    Regarding the betting – I think your criteria need to be revisited. Surely all that would be needed would be for the betting odds to be closer to the eventual outcome than the polls, not an over 50% prediction. In any event, this is from five days out from the election – not two weeks.

    the betting market Wahlstreet (ouch) has the SPD on 34% and the CDU just under 40%, with Greens on 8.5%, Left on 7.5% and FDP on 7.5%. This would put the Red-Red-Green buggered imagination option in the box seat with exactly 50%, the CDU/FDP on 47.5%…and the Ampelkoalition over the finishing line with an impressive 50%.

  4. JQ —

    Is there some theoretical research which justifies the widely-held belief that betting markets predict the probability of the event of interest? Or even that they tend to predict this probability, as some appropriate variable, such as time or the number of betters, tends to infinity?

    It is not at all obvious to me, a math. statistician by training, that the mean of multiple independent bets would be an unbiased estimator of anything, much less the variable of interest. It strikes me that people may have conflated one [0,1] variable (the probability of an event happening) with another [0,1] variable (the average normalized amount bet), and assumed that one of these variables estimates the other. Why should this be so?

  5. Peter, the best theoretical model is that there are some “insiders” who have private information that gives them a better ability to predict the result. They have an incentive to bet on this info, whereas uninformed voters do not. I’ve never been convinced by this and remain unconvinced.

  6. Andrew will be presenting a paper (joint with Justin) at ACE (Monday morning, 11h45) on this topic. See you there.

  7. Unfortunately, I’m going to arrive too late for this. Bad planning as I will also miss Sutton who I would very much like to see.

  8. Thanks, John.

    I’m not questioning whether or not the betting is accurate, but what precisely the betting is estimating. It’s not obvious to me that the average of multiple, independent bets = a prediction. This equation seems to be taken for granted by economists, so perhaps there’s something I’m missing.

    There’s an obvious problem with the inside knowledge argument, since for many domains (eg, national elections) there is no inside knowledge for anyone to have. Since many citizens only decide whether and how to vote on their way to the polling station, there is no fact-of-the-matter prior to the vote taking place itself.

  9. John, there is a market on the German election at Intrade, based on number of Bundestag seats won. It should be noted that many of the traders active in these markets don’t actually have a view on the result at all. They are actively buying and selling before expiry or arbitraging within or between markets. Your test of a market forecasting an upset win seems a bit steep, not least because if markets were forecasting it, it would not be an upset at all. I suspect if these markets were radically out of line with polls or some other information, people would say they were irrational and just lucky when they got things right.

    On Koizumi, he did come close to splitting with the LDP on several occasions and the LDP and Koizumi barely tolerate one another. Read Aurelia George Mulgan’s book on “Koizumi’s Failed Revolution.” It is very good.

  10. What makes prediction markets interesting is that they help coordinate and reveal information that might not be available otherwise in the absence of a market. Benchmarking their performance against alternative models (as Andrew does in his various papers) is the more relevant test of their usefulness, rather than John’s truth judgement view of them.

  11. My reticence in this case is due to the fact that I have a strong antipathy towards Koizumi (the Japanese PM) that I can’t really explain.

    Is it his hair-do? Its seems a little too artfully tousled.

Comments are closed.