Australian PM John Howard has called an election for 9 October. I’ve discussed the political issues here, but CT readers will also be interested in the implications for the efficient markets hypothesis. Centrebet , which didn’t do brilliantly last time, has the (conservative) Coalition at $1.55 and Labor at $2.30. If I’ve done my arithmetic properly, and allowing for the bookies’ margin, I get the implied probabilities as 0.60 for the Coalition and 0.40 for Labor. The polls have Labor ahead, but looking at all the discussion, I’d say that the consensus view is that the election is a 50-50 proposition, and that’s also my subjective probability.
How good a test of the efficient markets hypothesis will this be? Bayesian decision theory provides an answer[1]. If our initial belief is that the EMH is equally likely to be true or false, and the Coalition wins, we should revise our probability for the EMH up to 0.55. If Labor wins, we should revise it down to 0.45.
fn1. The workings are easy for those who know Bayes’ theorem and accept the modern subjectivist interpretation , but they won’t make much sense to those who don’t.
{ 19 comments }
willchill 08.30.04 at 12:05 am
can you really apply efficent markets theory to a phenomenon where there is no actual market clearing going on? markets are all about matching the last seller with the last buyer, and i’m not sure how you can relate that to pure speculation about a future event based on a few opinon polls
Jason 08.30.04 at 1:15 am
If you really think it’s 50-50, why not place a bet on Labour? Are you really so risk averse that you can’t cover the cost of transaction without it being a negaitve deal for you?
John Quiggin 08.30.04 at 1:37 am
“f you really think it’s 50-50, why not place a bet on Labour?”
I discuss this point on my own blog
Jason 08.30.04 at 2:01 am
So it is the combination of risk aversion and transaction cost that are hindering you.
OK then, a slight more complicated, but less risky scheme. Get all those people in the consensus (the 50-50 people) and take out small bets on Coalition from each of them. Combine all the money and do a large bet on Labour. At the very least it would provide for a conversation starter, and your wife would have less to complain about if you make money either way.
(I know, transaction costs way too high, but I feel it’s too easy to just say you and everyone else feel it’s 50-50 when there is no action that supports the statement)
PS It’s an interesting thought experiment to estimate the probability of some variation of EMH by using past data in a Bayesian way. I don’t know how you would incorporate the changing probabilities over time though.
dsquared 08.30.04 at 3:43 am
But surely, John, if all the bettors are facing a similar loss function to you, then the Bayesian calculation is different?
[the guys who did the Hewlett Packard internal market experiment ended up confronting this problem and found it to be serious. Their final solution involved using an average price of the final 50% of trades made to generate their forecast, which obviously saves the village by destroying it; their predictive accuracy was only possible in the context of admitting that a simple moving average rule would generate positive returns)
John Quiggin 08.30.04 at 4:01 am
dsquared, I’m assuming that the people who actually bet are not subject to joint budget constraints like mine.
dsquared 08.30.04 at 4:05 am
speaking as one of them, we bloody well are :-)
dsquared 08.30.04 at 4:09 am
ahhh, here’s the working paper version of the work that the HP guys did on what to do when a market is made up of bettors like me and John.
James Surowiecki 08.30.04 at 4:43 am
Daniel, I assume you know this, but that’s not the HP internal-market paper. The HP internal market that was set up to forecast printer sales was organized by Kay-Yut Chen and Charles Plott, who’ve written a couple of working papers on the subject. The paper you’ve linked to is part of a series of experiments that Bernardo Huberman, Chen, and Leslie Fine have been doing to figure out whether it’s possible to arrive at better collective forecasts by taking into account things like individuals’ risk preferences and private vs. publicly shared information.
Huberman, Chen, and Fine are especially interested in applying these ideas to small groups, which often don’t have the characteristics — diversity and independence — that characterize larger groups and that are crucial, or at least so I would argue, to intelligent collective forecasts, market or otherwise.
dsquared 08.30.04 at 4:49 am
Yeh; I think I linked to the actual HP paper in my review of TWOC. Although it is interesting (for both of us) that the HP paper also uses a decision rule for producing the final forecasts which is not consistent with “market efficiency” in the Samuelson sense of properly anticipated prices fluctuating randomly (they use some form of averaging over quite a long time window of trading, which suggests to me that there is a technical rule on the HP market that would make money). IIRC, Chen is the HP employee.
James Surowiecki 08.30.04 at 4:51 am
Also, as far as testing the EMH (or the wisdom of crowds, for that matter), Centrebet’s current price doesn’t seem to me all that important. Howard called the election less than 24 hours ago, so Centrebet’s current price reflects mainly the judgment of its own oddsmakers with a few bettors tossed in. The interesting question is what the market’s verdict will be once a sizeable number of bettors have weighed in and presumably — at least if Centrebet is not trying to take one side of the bet or the other — the bookies have adjusted the odds to reflect the bettors’ judgment.
John Quiggin 08.30.04 at 5:28 am
James, I’m pretty sure the market opened some time ago. Uncertainty about the election date is just one of the factors that has to be taken into account in laying bets well in advance.
John Quiggin 08.30.04 at 7:31 am
Reading today’s Australian Fin Review, I see that not only is there a continuous market, but they were paying 3 to 1 for a Labor win in January, when Labor was already ahead in the polls. Things have been up and down since then, but nothing has fundamentally changed. A couple of observations
(1) There seems to be excess volatility here, as in other financial markets
(2) If I’d been paying attention, I would have been keen to put some money on at those odds
Andrew Boucher 08.30.04 at 11:09 am
This is almost surely (!) not a test of the EMH.
(1) At most then this is merely a test whether centrebet is an efficient market, and I doubt it is (anymore than the Iowa market appears to be).
(2) At best it is a test if the election is held today, and not Oct. 9.
(3) And in any case what version of the EMH is this a test of? Off the top of my head, I can’t even think of one.
John Quiggin 08.30.04 at 12:33 pm
Andrew:
(1) If you rule out all markets for which EMH does not hold as not efficient, then EMH is vacuously true
(2) Think about it, and you’ll see this is wrong. What kinds of contingent claims do you think are traded in futures and stock markets? Claims that pay off the same day?
(3) Either strong or semi-strong
Andrew Boucher 08.30.04 at 2:17 pm
(1) I create a market in my house for the 2004 U.S. election. Anyone can bet, but in reality the only participants are my wife and me. What does that say about the EMH? Anything?
I can’t access the centrebets page (I use Safari), but my guess is that it has a small number of participants who do not trade often and are biased to certain sections of the population.
(2) Well I thought about it, and it’s still not wrong. You’re using an event on Oct 9 to determine whether the prediction of a market today is correct. But there will be new information between now and Oct. 9.
That is, suppose a market predicts today that A will beat B on Oct. 9. Suppose there’s a major event Oct. 1 – e.g. A has been found committing a major crime. The election swings to B. Will you use the outcome of the election to say that the market made an incorrect prediction?
Now suppose a market predicts today that A will beat B on Oct. 9. B wins the election.
How will you determine that there *wasn’t* such an event between now and Oct 9?
(3) Nah. First, if the participants in centrebets are biased to a certain small segment of the population, as I think they probably are, then the market only reveals the information accessible to that segment. Secondly, I presume that the amount of gain is fairly limited – so participants have little incentive (unlike the NYSE stock market) to improve the information they have. Thirdly, the market may correctly have summarized all information, but still be incorrect about the election.
Jack 08.30.04 at 3:42 pm
It is interesting how poor Bayes theorem is at dispute resolution — the stronger your initial convictions, the less convincing you find the alleged contrary evidence.
Andrew has a point about the time gap — not that a future’s contract is not the same sort of thing but that the actual outcome is the best proxy for the true best assessment now. I think this is one of the problems of the EMH, the lack of a competing hypothesis. What is Not EMH? There doesn’t seem to be a universal one that provides a best guess at a price. What Warren Buffet would pay?
Is there a market for the truth of the EMH? Investment company research budgets perhaps? It seems reassuring that the smaller the size of the research budget the less plausible the truth of the EMH becomes and conversely the more money is spent on it not being true the more plausible it becomes. This suggests to me a natural rate of inefficiency (not a permanent one just the possibility of there existing an equilibrium). It should perhaps be possible to work out the market price but I’m not sure how to take into account the disinformation and corporate finance related roles of many equity analysts. Should these efforts be counted as further expenditure on the falsity of the EMH or on the opposite?
John Quiggin 08.30.04 at 9:03 pm
Jack and Andrew, I’ve offered a competing hypothesis, the judgement that it’s 50-50 based on polls and expert judgements (I could use a formal estimate based on polls if you like). New information arising between now and the election is new in both cases.
Jack’s point about priors is right, and explains why disputes between people with strongly held views tend not to be resolved quickly.
Jack 08.30.04 at 11:02 pm
I guess the question is what happens if the market price changes between now and the election, say to suggesting that the chances are 50/50? On the one hand that would be some kind of vindication of the expert view. On the other there is no knowledge of the Platonic best guess at the starting point, the closeness to which is the issue.
Expert opinion is an interesting candidate for an alternative hypothesis but clearly difficult to apply in some situations, market forecasting for example. On the other hand in law and medicine nothing could be more usual. Unfortunately it it doesn’t reliably come up with a simple competing thesis. It of course has a hall of shame of its own and shares some of the same events.
I’m rather wary of the interpretation of this in that the Bayesian reasoning feels like a way of sidestepping caveats against overinterpretation provided by measures of statistical significance.
To be fair it seems to me that the uncertainty of market prices is rarely taken into account in discussions of the EMH and admitting it might undermine some of the stronger claims about it.
One particular implication might be that there is little discussion of risk preferences so there is no scope for somebody to choose a non-market solution that might under some circumstances be sub-optimal but which might be more robust and keep working under more circumstances — in exchange rate issues for example.
Comments on this entry are closed.