Could Prediction Markets Help to Find MH370?

Like everyone else, I have become obsessed with the disappearance of Malaysian Airlines flight MH370. When I read that the flight lost contact on March 8, I assumed that it would be found crashed into the ocean in a matter of days if not hours. Nearly two weeks later, people are starting to wonder whether it will ever be found.

There is no shortage of theories about what happened to the flight. Pilot suicide seems to be the most likely answer, but there is scant evidence of motive. Terrorist hijacking is an obvious possibility, perhaps by the Taliban or Uighurs seeking to strike back at China, but no groups have claimed responsibility. Piracy is a possibility; the list price of a Boeing 777 is over $200 million. Pilot Chris Goodfellow claims that an electrical fire is the most likely cause, and many believe this explanation is the most likely, but it has difficulty explaining the several course and altitude changes made by the flight. Similarly, Australian Pilot Desmond Ross argues that the flight could have depressurized, explaining why the plane first rapidly descended. He then argues that errors induced by the depressurization could explain the plane’s other maneuvers. Any, or none, of these could be true.

As theories have proliferated and the official search area has widened to a significant portion of the Earth’s surface, I have started to wonder whether prediction markets might help to locate the missing flight. Prediction markets are similar to stock markets, but the traded contracts are predictions rather than shares of a corporation. Contracts in prediction markets have a payoff (say $1) if the associated prediction is correct, and no payoff if it is incorrect. Such markets have proven remarkably powerful in predicting the outcomes of certain types of events, such as political elections.

Thinking about MH370, I was reminded of this passage from James Surowiecki’s The Wisdom of Crowds:

In May 1968, the U.S. submarine Scorpion disappeared on its way back to Newport News after a tour of duty in the North Atlantic. Although the navy knew the sub’s last reported location, it had no idea what happened to the Scorpion, and only the vaguest sense of how far it might have traveled after it had last made radio contact. As a result, the area where the navy began searching for the Scorpion was a circle twenty miles wide and many thousands of feet deep. You could not imagine a more hopeless task(xx).

The search radius for MH370 is up to 4,400 kilometers, 275 times that of the submarine, but there are clear similarities between these searches. In the submarine’s case, James Craven, a naval officer, approached a group of experts and asked them to place bets on what likely happened to the submarine and where it would end up. He then combined these guesses into a prediction of where the sub would be found. This aggregate prediction was only 220 yards from the sub’s actual location, even though it was not a location that any individual bettor had predicted. As Surowiecki explains, “No one knew why the submarine sank, no one had any idea how fast it was traveling or how steeply it fell to the ocean floor. And yet even though no one in the group knew all of these things, the group as a whole knew them all”(xxi).

Search area represented in Google Earth. The white circle represents the plane’s maximum range from its last known location. The red circle indicates the possible locations of its last known satellite connection. (http://ogleearth.com/2014/03/flight-mh370-search-data-in-google-earth/)

Nobody knows what happened to MH370 either, though many people may think they do. But nearly all of these theories are based on some information. Perhaps what none of us know individually we may know together?

How would such a market work? I think it would go something like this:

  1. Divide the total search area, defined by MH370’s last confirmed location, its fuel supply, and ocean currents, into a grid.
  2. Each grid location represents a possible value for a prediction of where the plane will be found.
  3. Each grid location has a pair of yes/no contracts associated with it. A ‘yes’ contract pays $1 if the flight is located within that grid location and $0 otherwise. A ‘no’ contract pays $1 if the flight is not located within that grid location and $0 otherwise.
  4. To buy into the market, you purchase a pair of contracts for a grid location for $1. Since the contracts are mutually exclusive, when the flight is found the combined value of these contracts is assured to be $1.
  5. Traders can then sell or buy contract halves through a double-auction market similar to that of the stock market. Traders should buy ‘yes’ contracts for grids they think the plane is likely to be found in, and sell contracts for grids where it isn’t. Market prices could therefore be interpreted as predictions of where the flight will be found.

Since the market would be zero-sum, it would be self funding, though of course someone would have to organize and run it. It could even be run as a play money market, as such markets have also shown to be surprisingly accurate.

What are some potential challenges? Search area size could be one. The search area I have proposed is 61 million square kilometers. Even with 1000 square kilometer grids there would be 61,000 contract pairs. While the market as a whole might be popular and liquid, the market for any individual contract may not be. It might be possible to solve this problem with a clever user interface that allows users to buy contracts through a batch process, but this means that such a market couldn’t be easily integrated into existing prediction exchanges.

Another is that there might not be sufficient dispersed knowledge to effectively solve the problem. While prediction markets have been shown to be effective for many tasks, they are not universally accurate. For example, in his book Infotopia Cas Sunstein relates the failure of prediction markets to predict President Bush’s first Supreme Court nominee(134). Rather than predicting the successful candidate, John Roberts, markets essentially followed the predictions of pundits, nearly all of which turned out to be wrong. Similar problems might exist for MH370, as there is very little information but many news organizations and bloggers confidently offering opinions that may have little basis in fact. Were a prediction market operating today, for instance, we might be experiencing an unjustified “Chris Goodfellow bubble”.

On the other hand, one potential extra benefit of a prediction market for this case might be its anonymity. When the Challenger space shuttle disintegrated at launch in 1986, stock markets quickly pointed to solid rocket booster manufacturer Morton Thiokol as the likely culprit (Maloney and Mulherin 2003). In all likelihood, the market made the correct prediction because someone with insider knowledge leaked information. But if so, nobody knows who that was. Similarly, someone might have information about the plane’s location, either because they were involved in its disappearance or because they have access to information that has not been made available to the public or searchers. This hypothetical person could potentially indicate the plane’s location through market trades without compromising their anonymity. Yes, this potentially allows a terrorist to profit from their actions, but it would come with discovering the plane’s location—perhaps a price worth paying.

So while it’s certainly possible, perhaps even probable, that a prediction market wouldn’t help to locate MH370, it would be an extremely interesting experiment. It’s certainly beyond my ability to implement in any reasonable time, but it’d be great if someone else did.

5 Comments

  • Kevin Reply

    I’ve had the same thought this whole week. Prediction markets could allow us to separate the wheat from the chaff, and provide balance to all the wild speculation about MH370. Do you, perchance, know Robin Hansen’s work in this area?

    • Mike Thicke
      Mike Thicke Reply

      Yeah I’ve read several of his papers. He’s definitely one of the leaders in the field, though I think he’s somewhat overoptimistic about prediction markets. I discuss his proposal for prediction markets for science here and explain why I think they might not work as well as he thinks.

    • Mike Thicke
      Mike Thicke Reply

      Thanks, I will be sure to check those out!

Leave a Reply

Your email address will not be published. Required fields are marked *