Catching up after one of the busiest weeks in recent memory... I've wanted to post regarding Barry Ritholtz' most recent sage comments on prediction markets but haven't had enough cycles until now... Ah, Saturday mornings in the summer... cool breeze on the back porch, a big hot steamin' cuppa Joe... Here goes:
First, I find it somewhat encouraging that the dialogue around this tool is starting to sound just a little more like well, a dialogue - attracting intellectually honest examination that doesn't automatically bounce to the extremes of rejecting prediction markets out of hand or assuming that they will create eternal prosperity, world peace and a libertarian paradise on earth. I'm also encouraged by a growing recognition that prediction markets are a tool - and only that. They are not a religion, nor do they need to be about political philosophy, even as they may help to calibrate political phenomena, (as well or slightly better than other tools for doing so, e.g. polls.) As Mr. Ritholtz points out, they will always rely, (at least in part), on those other flawed tools for 'fuel'... and that's OK.
Second, I find myself in almost total agreement with Mr. Ritholtz, given my interest in prediction markets as a potentially powerful component of a much larger, contextualized scenario-based approach to thinking about and planning for different ways the future may evolve. I don't know if that makes me a card-carrying member of the prediction markets 'crowd' or puts me right in the middle of a dangerous prediction markets 'DMZ'. Draw your own conclusions while we review the bidding (pun intended):
Ritholtz #1: "The inherently unknowable is, by definition (duh) inherently unknowable. The further into the future one peers, the less accuracy one can expect on forecasting singular events. Call this the rule of the unknown future."
Yep. As I like to say, nobody has actually been to the future - God and a handful of prophets excepted; but we won't get into that here. :)
Ritholtz #2: "...the crowd cannot create wisdom simply by pooling their collective ignorance. Indeed, the less an event is dependent upon the actual behavior of crowds, the less likely it is that prediction markets will forecast a particular outcome correctly."
I have previously put this another way, (...a contract on the actions of a small group is inherently more risky and less liable to be correct than one for a general election in which actions and actors are widely distributed), but we're basically in agreement. To use an overly simple analogy: I cannot easily predict what my neighbor will do; I may be able to predict what my wife will do; I can usually predict what I will do. The prediction market failures Ritholtz cites: the Michael Jackson verdict, the Purcell resignation from Morgan Stanley, and the Howard Dean implosion in Iowa all involve closely held decisions. The principals (e.g., the MJ jury, Mr. Purcell, Mr. Dean) and their inner circles were probably not participating in the markets in any significant way, if only because the risk of reputational loss or legal liability is high relative to the money to be made.
Ritholtz #3: "The collective behavior of market participants can often provide insight into future market behavior. In other words, the present behavior of investors can sometimes be used to predict the future behavior of these same investors. The key is to use markets that are liquid, widely followed, with significant money at stake."
Ritholtz #4: "Where futures market do particularly well is where they track the behavior, beliefs and perspectives of their collective participants."
These seem like two sides of the same idea, and a clarification of #2. I.e., if a 'prediction' market contract is about the collective actions of market participants themselves (and their close proxies, e.g., other voters, other movie-goers, other purchasers of a particular product, other influenza sufferers, etc.), then it's more likely to 'predict' the actual outcome. Or back to my simple analogy: I'm more likely to predict the behavior of myself than I am the behavior of my wife, and certainly more likely to do that than to predict the behavior of my neighbor. That said, I am more likely to predict the actions of my neighbors collectively (e.g., whether >50% of them will have brought in their morning newspaper off the stoop by 7:00AM on a particular day of the week), than I am to predict whether one particular neighbor will have done so, (assuming I haven't been spying on him, making myself an insider and thus changing the nature of the problem.) That's an assertion. I welcome other views.
I do take issue with another part of Ritholtz' point #4 however: "I do not believe political futures predict anything -- instead, they are more like an alternative polling method that consolidates and synthesizes all the pre-existing tracking polls." In the larger philosophical sense, he's right; nothing can be 'predicted'. Too many folks lend to this whole subject what I can only describe as a visceral yearning for the absolute, the prescient and the prophetic.
It is human nature to desire to know what is going to happen - to seek certainty. Not only around prediction markets, but around any attempt at 'seeing' the future (e.g., scenario planning, science fiction, expert opinion, visionary leadership, computer modeling, opinion polls, authoritative columnists, and name-brand media outlets), it is human nature to want to feel that someone really has been to the future and that we can rely on their prediction and not expend energy worrying about thousands of wild outlier possibilities. The real world, unfortunately, is different.
In the real world small things happen that make any predictive method prone to failure some of the time, (aka, the butterfly effect). Howard Dean gets a little over-excited on mike. Mr. Purcell has some second thoughts after a late-night conversation with an old friend. (I'm making this up.) Someone on the MJ jury makes a statement that tips the scales of another juror's thinking, and that changes the whole deliberation, etc. Recognizing all of this is a healthy start to any discussion in this vein.
So yes, prediction markets do synthesize pre-existing tracking polls, but that is not necessarily all that they synthesize. I would not put a derogatory 'spin' on that feature. Synthesis can be an incredibly valuable function. It is especially so when the synthesis 'engine' (i.e., a market) is open to sources that may not be obvious to all. Yes, markets will swing wildly on public news. That's just what they do. But they also can incorporate news that is not-so-public (back to the neighbor analogies), and do so much more quickly than other methods. If nothing else, the speed of synthesis is an important benefit in some applications. As I like to say in a scenario context: knowing you're wrong before everyone else knows that you're wrong.
One key to sorting out the dueling graphs of prediction market swings I think, is establishing in advance, a norming function to smooth out market noise leading up to a singular event. I did that on the morning of election day last year - smoothing out Tradesports data snapshots I'd taken over the previous four days. The result was perfect on a state-by-state basis and carried an R-squared value of 0.77 for how the magnitude of state votes related to the magnitude of prediction market values. That's just one event and one interpretation of it, but enough for me to remain encouraged that prediction markets have some important applications and features not fully addressed by any other method.
We'd all be well served by spending our collective energies defining those applications and features more tightly, (including what kinds of tasks prediction markets are poorly suited to), rather than arguing about their universal usefulness versus their depravity and foolishness.