Indeed, it seemed to me even then,
well before I studied the subject in any depth, that people's responses to
opinion polls are very sensitive to the form and structure of the
questionnaire and that the validity of the findings are even more sensitive
to the sample of voters taken and whether those surveyed are likely to vote.
Meanwhile, decisions backed by hard-earned money are likely to be very
carefully considered and to weigh and discount relevant and irrelevant
information in a serious and appropriate fashion.
That’s the theory at least. The most
convincing point, however, was that the market had done so much better a job
of predicting the election result than the trusted pollster. It was several
years before I realized that what applied to the small rural backwater of
Brecon and Radnor applied equally well in a study of every U.S. presidential
election between 1868 and 1940. In only one year, 1916, did the candidate
clearly favored in the betting the month before the election end up losing,
when Charles Evans Hughes lost to the incumbent, Woodrow Wilson, in a tight
race. (See “Historical Presidential Betting Markets,” Paul W. Rhode and
Koleman S. Strumpf, The Journal of Economic Perspectives, Vol. 18,
No. 2, Spring 2004, pp. 127-141.)
The power of the betting markets in
assimilating the collective knowledge and wisdom of those willing to back
their judgment with money has only increased in recent years as the volume
of money wagered has risen dramatically. Indeed, by 2004 the
Intrade market model went
stratospheric in predictive accuracy as the market favorite won the
electoral votes of every single state in that year’s U.S. presidential
election. Meanwhile more than one respected pollster and analyst called the
race for John Kerry as late as election day itself.
The betting markets saw their best
triumph of 2004 in Florida. Even though a number of polls put Kerry ahead in
that state, or said the race was too close to call, the betting markets
consistently showed Bush would win Florida comfortably.
Indeed, if the Democrats had paid as
much attention to the markets as the polls, I am convinced that the election
result would have been different. They could have downsized their effort in
Florida and focused their efforts more on other swing states where betting
sites showed the race was much closer.
Intrade followed up in 2006 when the
market favorite won each and every Senate seat up for election. Moreover, in
large part the stronger the favorite, the bigger was the margin of victory.
Follow the Money
The assumption, or at least the hypothesis, must be that the accuracy of the
betting markets is created out of the information and intuition of many
people rather than the conclusions of a few. Those myriad people feed in the
best information and intuition they can because their own financial rewards
depend directly upon them. And it really is a case of “follow the money”
because those who know the most are likely to bet the most.
Moreover, the lower the transaction
costs (there is no tax on betting in the UK) and information costs (in never
more plentiful supply due to the Internet) the more efficient we might
expect betting markets to become in translating information today into
forecasts of tomorrow. The rise in the importance of person-to-person
betting exchanges like Betfair offers
another reason why the markets are becoming ever more efficient in
predicting the future. These exchanges differ from traditional betting
markets by eliminating the odds-setting bookmaker, instead providing the
technology to match up the best offers to back and lay an outcome on offer
from all the clients of the exchange. In so doing they ensure that the
margins implicit in the odds are lower than they have ever been. For all
these reasons, betting markets today are likely to provide better forecasts
than they have done at any time in history.
Buoyed by this enthusiasm for the
power of market forces I was sufficiently confident, when asked by The
Economist magazine, to call the winner and the seat majority in the 2005
British General Election over two weeks out. My prediction of a 60-seat
majority for the Labour Party, repeated in an interview on the BBC Today
program was challenged in a BBC World Service debate with the chairman of
the MORI polling organization. He wanted to bet me that his figure of a
Labour majority of over 100 was a better estimate. I declined the bet and
saved him some money. The Labour majority was a little over 60 seats.
In October of this year The
Economist ran a follow-up article on election betting, asking me the
likely outcome of the British General Election if Prime Minister Gordon
Brown, then actively considering his options, decided to ask the Queen for
the requisite dissolution of Parliament. After consulting the markets, I
declared his odds of a majority just slightly better than even. “So,” I
asked, “is the Prime Minister willing to risk his majority on the toss of a
coin?” The answer was no.
Low-Volume and Play-Money Exchanges
This is not to say that all betting markets are always right. The
Iowa Electronic Markets have
been allowing selective trading for modest amounts of money on their
exchange since 1988, with some success. Nevertheless, their odds were far
off the mark in the 2000 election, predicting that Mr. Bush would win a
larger share of the popular vote than Mr. Gore (meanwhile, the real-money
“spread betting markets,” available to significant trading volume in the UK,
had the electoral winner priced up as a dead-heat). Moreover, the Iowa
markets failed to predict the Republican victory in the Senate in the
mid-term elections in 2002 and have struggled to maintain their foothold at
the top end of predictive accuracy since the rise of the high-volume
exchanges. Perhaps this is because Iowa's markets keep bets to a small size,
which puts less pressure on traders to get their predictions right. Perhaps
it’s a blip. Time will tell and these low-volume markets, as well as
play-money exchanges such as
Newsfutures, continue to command respect.
Forecasting elections is not simply a
matter of comparing the accuracy of polls with betting markets, however.
There is a sizeable literature on the value of econometric models, which
take account of such factors as employment, inflation, interest rates,
incumbency, in predicting election outcomes. Other methodologies include
various ways of weighting the opinions of political scientists, political
journalists, political pundits and experts of other stripes, in forecasting
the winners and losers.
In my book, Information Efficiency
in Financial and Betting Markets, published in 2005 by Cambridge
University Press, I employed a blind “moment-in-time” case study, conducted
exactly a month before the 2004 U.S. presidential election, to compare and
contrast the forecasts implicit in each method. I was left with least
confidence in econometric models and most in betting markets, which is what
I expected.
One interesting angle, though, is
whether a combination of one or more of these methodologies, weighted in one
way or another, can do better than any individual forecasting methodology.
Just as interesting is the extent to which the accuracy of forecasts derived
from election betting markets is affected by altering the structure of the
market. For example, is the accuracy of these forecasts affected by the
nature of rewards within these markets -- whether
participants use their own money or “play money” to make trades in the
prediction market; is it for example affected by whether artificial limits
are placed on the amount of money used by market participants to trade? More
generally, to what extent is the accuracy of forecasts derived from election
prediction markets affected by altering the structure of the prediction
market?
The bottom line, therefore, is not to
assume that betting markets provide all the answers, though sometimes it
seems like they do. We must always be looking to improve our predictions.
And when the markets do fail, we must not be afraid to ask why. When these
prediction markets succeed, we must also ask why, and whether they could do
even better.
These questions and more are at the
forefront of two new journals, published by the University of Buckingham
Press, namely The Journal
of Prediction Markets and The
Journal of Gambling Business and Economics.
On recent evidence, then, some markets
have seemed so powerful in predicting election outcomes that one might be
forgiven for bothering to turn out to vote at all. Of course, that makes no
sense. What does make sense, however, is that political operatives need to
pay more attention to the markets in planning their election strategy. If
those close to Mr. Kerry had listened to this advice in 2004, the senator
from Massachusetts would in my convinced opinion now be President Kerry. Any
sophisticated campaign needs to make sure that it doesn’t repeat that
mistake.
For those who want to know more, the
Betting
Research Unit at Nottingham Business School, now linked to the
newly-formed
Political Forecasting Unit at the same institution, is just an e-mail
(leighton.vaughan-williams@ntu.ac.uk) away.
In conclusion, recent years have
witnessed groundbreaking shifts in the way in which betting is taxed,
regulated and perceived by economic theorists. This means that betting
markets will become more than just a major part of our future. Properly
utilized they will be able to tell us what that future is likely to be! We
seem therefore to have created, almost by accident, a “high-tech” crystal
ball that taps into the accumulated expertise of mankind and makes it
available to all. The challenge now is to make the best use of it.
