The Dark Side (Prediction Markets, Manipulation and Self-Fulfilling Prophecies)
Upon reading letters in which her husband described his encounter with the three seer-witches, Lady Macbeth effused that they "have transported me beyond this ignorant present, and I feel now the future in the instant." For the three witches had predicted that Macbeth, then a regional lord, would ultimately become King of all Scotland. Feeling anointed by fate thereafter, Macbeth went on to murder the King and assume the throne, thus lending a self-fulfilling aspect to the witches' prophecy.
What powers lie in self-fulfilling prophecies? Can we rigorously delineate their dangers as they relate to the future of prediction markets?
Koleman Strumpf presented his paper, authored with Paul Rhode, "Manipulating Political Stock Markets" at the Mini-Conference on Information And Prediction Markets in London earlier this week. Part of their study traces the development of the seldom-remembered political betting markets of the late 19th and early 20th centuries, which peaked in volume in 1916. Here is a particularly fascinating passage:
Contrary to these sanguine assessments were the frequent assertions that active partisan involvement, especially by Tammany Hall, systematically distorted the betting odds and, in selected instances, speculative attacks and manipulation sought to change the momentum of the races and influence voter turnout. [...] As an example, in 1916, Democrats charged "the money was being sent to Wall street to force the betting odds to Wilson's disadvantage" [...] "Already," one prominent Democrat said, "we are hearing that many up-State farmers are struggling between their conscience and fear that Hughes will be elected and it might be found out that they voted for Wilson." (my emphasis)Perhaps it was no coincidence that political betting fell off by the next Presidential election of 1920 in the wake of the Black Sox Scandal of 1919. On the other side of the world, this intervening period saw the Russian Revolution of 1917, which brings to mind Karl Popper's and George Soros' idea that Marx sought to influence the course of history by claiming to predict it. The specter of self-fulfilling prophecies will always haunt political prediction markets — and since there is a good amount of prediction concerning these markets, it's best for their proponents to meet these concerns head-on.
The Strumpf/Rhode paper corroborates previous work by Robin Hanson, Ryan Oprea, Colin Camerer and others concluding that attempts to manipulate liquid markets will be largely unsuccessful. In short, the Hanson/Oprea paper argues that a manipulator's price bias can be treated as "noise," and variance in that bias can actually increase price accuracy, by enticing more market participation and liquidity. As with all models, there are many assumptions here that may not hold in the real-world. For instance, what if the market contains a preponderance of momentum/feedback traders — those for whom the manipulator's "noise" acts as new information? In actual markets, very often participants act on such endogenous "clues," and traders may have no idea that manipulation is even occurring. Also note that the cited works on manipulation specifically address the influence of price on price, whereas when we are discussing self-fulfilling prophesy, we are more precisely referring to the effect of price on fundamentals. Likewise, some studies have stressed the role of ethylene gas or hallucinogens on the Delphic Oracle's behavior and reception, but here we are less interested in divergences between reality and perception/representation than we are in cases where the latter actually becomes the former.
By "fundamentals" all that is meant here is some external influence on the market price. In self-fulfilling cases, often this exogenous factor is related to collateral or some other source of funding or liquidity. "On The Fundamentals of Self-Fulfilling Speculative Attacks" examines currency crises and concludes that while self-fulfilling dynamics may trigger the timing of major market moves, they cannot unfold without certain fundamental circumstances already in place. The fundamentals determine the "whether" of the move, while the sentiment and technicals merely determine the "when." In this 1997 interview, Paul Krugman basically agrees with this prognosis, adding that self-reinforcing dynamics usually exaggerate the magnitude of the eventual jump:
John R. MacIntyre: You would say then that there is nothing arbitrary about these crises and that it is the market responding the policy inconsistencies of the government?Self-fulfilling dynamics may move events forward in time or climactically intensify them, but they are not a free-floating power with respect to fundamental circumstances, at least in liquid markets with reasonably good and up-to-date information dissemination.
Krugman: I would say that is ninety percent true. Currency crises can, in principle, arise more or less arbitrarily. One way is what is termed a self-fulfilling crisis. It is like a bank run. We all know that a bank that is fundamentally sound can nonetheless be broken if all depositors panic and try to cash their deposits all at once. [...] In practice, there are reasons why a currency crisis is most often worse than actually deserved by a country. But in no case that I can think of has there been a crisis completely unjustified by the economic fundamentals [...]
What about election markets where the "fundamentals" are the intentions of voters? The main Tradesports Bush contract apparently underwent a series of manipulative episodes in August through October of 2004, in one case plunging to 10% before sharply rebounding. In the proceeding months, as in 1916, there was plenty of speculation, especially from Don Luskin, on the source of those "speculative attacks." In any case, these manipulation attempts dissipated at the level of the market as the literature predicts, but let's assume that future attempts in similar markets are more successful for whatever reason, probably through a prevalent degree of momentum/feedback/herd trading. Under which circumstances could there then be a danger that the prediction will be self-fulfilled in the election? In order for a market to have an effect on the actual outcome it would have to be very well-known and -regarded, so well-regarded in fact that it would not be compelled to close any divergences from standard polls on its own. However, if this were the case, this market should be exceptionally liquid and manipulation should be all the more difficult in the first place.
Nonetheless, assume that these two conditions occur together: 1) the market is very deep, well-known and -regarded, and despite this, 2) a successful manipulation attempt occurs. Leaving aside the possible urge of voters to be on the "winning side" regardless of rational political considerations, in a two-party system with secret balloting, a prediction market should have little influence on who voters ultimately vote for. Their votes will still be based on appraisals of the party platforms and reliabilities with respect to their personal utilities or political ideologies. The fact that the market price may be bandied about by the media, even if it creates an aura of defeat for one candidate in accordance with the manipulator's wishes, will not affect voter decisions as they might if those decisions were to buy or sell some asset. However, a manipulated market price may very well influence which voters vote. Anthony Downs' great 1957 book An Economic Theory of Democracy argues that in a two-party system, party platforms will tend to converge in what amounts to political Hotelling. This point seems increasingly prescient each major election year in the US, as popular and electorate votes uncannily approach a 50/50 split. Focusing only on the popular vote for simplicity, when an election market trades near 50, apart from the fact that many voters may be indifferent to the outcome, there is a stronger impetus for voters to participate than when the projection is wide and their votes have less chance of making a difference. The effect on voters in manipulating a 50/50 race to 55/45, for instance, is actually unclear with respect to the manipulator's intentions. Will the supporters of the "favorite" become incrementally complacent, while the party of the "underdog" is made incrementally more resolute? Perhaps manipulators should actually buy the candidate they do not support — especially since this would constitute a hedge of sorts! In any case, even if semi-contradictory pre-requisites are met, the potential for self-fulfilling prophecies in two-party election markets is exceedingly tenuous. (In multi-party elections or nominations, where party/candidate differentiation and tactical voting come into play, it may be stronger.)
The research suggests two relatively assuring results then: 1) Manipulating markets tends to be difficult and ineffective, and 2) Self-fulfilling dynamics may quicken or intensify an event, but they should not cause one ex nihilo. Furthermore, political election markets do not seem susceptible to self-fulfilling manipulation, at least not in the two-party case. Now, there is another idea which people often conflate with these others, and this is a source of considerable angst and controversy regarding prediction markets. This logically separate issue exists where an unknown individual or a small group of actors can have a material impact on the final outcome of a market. This was a basic problem with part of DARPA's Policy Analysis Market: the set of potential "insiders" (terrorists, for example) was too diffuse and uncontrollable. (There are still ways to mitigate this; they were mentioned in the link above.) An open group of insiders is particularly untenable when matters of life and death are directly in play. Keen to avoid excessive controversy, Tradesports was wise to instate rules 1.6 and potentially 1.9 to all political election markets. (If a candidate were to drop-out for any reason, the contract would have been voided and all open trades reversed.) The contract for Bush to win Texas, when it traded at only 96% in early August 2004, was therefore mainly a demonstration of the longshot effect, and didn't indicate the presence of other possibly suspected embedded options. For reasons similar to insurance companies' "insurable interest" logic, this contract probably would not have been acceptable without the special rules.
The mere fact that someone may find a contract distasteful is not an argument against its existence if the contract will not reasonably encourage a negative outcome through its "insider" structure, especially if it serves a legitimate insurance purpose, or provides an informational social good. If there is a "dark side" to prediction markets, it mainly relates to those contracts that may encourage negative outcomes, and not to the realm of self-fulfilling prophecies. In the preface to the 1994 edition of The Alchemy of Finance, the great trader George Soros concludes that "in most situations," reflexivity "is so feeble that it can be safely ignored." The academic literature on manipulation suggests that the feebleness of self-fulfilling manipulative attacks in prediction markets is thereby squared. Of course, the special cases are the interesting ones.