Saturday, August 11, 2012

PredictWise and The Signal: Betting Markets and the Election

The Wisdom Of Headlines ...
We continue our digital politics tag series with a look at PredictWise. PredictWise is the engine behind Yahoo's "politics portal" The Signal. It uses a "Wisdom of Crowds" approach by taking a set of Internet Betting markets and aggregating them to try to analyze predicted events.

The idea of betting on events is topical: the Department of Defense just killed its Terrorism Betting Market amid outcry about the appropriateness of casting lots on damage to America (I wonder if the market had odds on Congress shutting it down?).

The Site Itself
The Signal is a disappointment. As far as I can tell, it's just a series of articles culled from PredictWise and elsewhere with no discussion or feedback outside of the ever-present "follow me" buttons. PredictWise isn't much better: you can create an account and log in--but that seems to be it. Other than giving them some personal information I don't seem to have any special capabilities for creating an account.

Functionally, PredictWise aggregates several predictions but there's little analysis. It has an excellent state-map (that is apparently updated in real time--that's awesome) where you can mouse-over each state and see the individual odds for it.

It has a blog that's worth a look ... if not updated all that frequently.

What Does It Mean?
Well ... here's exhibit "A". This is taken from the most recent blog post and headline. It's from Aug 9th.
The Crowds ... They Seem To Have Lost Big ...
Exhibit 'B' is their July 8th blog PostA response to David Leonhardt's "When the Crowd Isn't Wise" The argument revolves around InTrade famously getting the SCOTUS decision on ObamaCare wrong (badly) and whether or not debiased (that's "with bias mathematically removed") polling ( is more accurate than debiased prediction markets (InTrade, etc.). The PredictWise paper concludes
In 2008, FiveThirtyEight, a debiased poll-based forecast, offered to the general
public a more accurate forecast than raw poll numbers or raw prediction market prices. But, the analysis here shows that were Intrade’s prices debiased, they would have provided a more accurate forecast and more valuable information than the best poll-based forecasts currently available, especially early in the cycle and in uncertain races. 
Okay ... but let's look at exhibit "A" again. Ryan isn't anywhere on that list. How come?

I don't know. Certainly Ryan's name floated around. There's no reason not to include him in the numbers. Is it just that PredictWise had a cut-off and had to cut someone? No--those three were in the "final running." There's no way that PredictWise just "cut them." Something else is going on: here's the InTrade:

So--we knew the "final three" for several days and everyone just got it wrong. The reason Ryan wasn't showing up on Exhibit A was probably that he wasn't statistically significant.

So ... What?
I think the lesson here is that betting markets and polling are now correlated. In other words, unlike guessing the weight of an ox (Wisdom of Crowds anecdote: the average of the crowd's guesses was dead on) prediction markets are now driven by polling. People are betting based on what they read on 538. Once that happens--well, I'm no statistician--but it seems that rather than using disparte sources of information (as happens when each person tries to guess the weight of the ox) the feedback cycles information (good and bad) back into the system and amplifies it.

It's like if the ox-guessers were betting money and a guy said "I work with oxen--I weigh them daily--that's a 1.3 ton ox!" and everyone went ... "Ohh--he sounds smart! Okay: I bet 1.3 tons too--here's my fifty bucks!"

It turns out the guy was working with the carnival and they all lost--but hey, he sounded good, right?

So I think betting markets are now part of the Internet Info-sphere and are no longer a separate variable as they might've been in 2008 (remember: Internet growth has exploded since then).

We'll see if PredictWise has anything to say!


  1. Thanks for the comments. We're already working to incorporate some of your suggestions into the PredictWise site (for example, there will be functionality available only to those that have an account in the very near future), as well as many other things that we think you'll like. Stay tuned on those...

    The prediction market data on the Map View updates every couple of minutes, so that's as "real time" as is available anywhere. We're glad you appreciate that feature!

    As for the polls versus prediction markets debate, we believe that one is more clear cut than you do: prediction markets dominate polls as a forecasting tool. We're working on proving that with some very convincing evidence.

    Anyways, thanks for checking out the site.

    -Andrew from PredictWise

    1. Fantastic--I'll be watching. I very much look forward to the evidence. InTrade was *very* accurate for 2008 so, hey, I'm paying attention.

      I believe the gold standard here of convergence that ALL the big-data players are looking for is mapping real-world events to real, significant shifts in the electorate--and doing that within the time-cycle of polling (and with greater precision, reach, breakdown by national vs. state, demographic, and so on).

      Betting Markets with real-time variance (the current 10% drop in Florida) have potential to illuminate that relationship between event (Ryan announcement) and sentiment.

      So my question is, big-data guys, when leaks started coming out last night what happened in your information models? When did the betting markets start showing the impact? After the press-release? When Twitter knew (significantly before?).

      This propagation is important since it shows how the electorate becomes informed and when that information turns to action (selling shares in Portman).

    2. As we speak, David is working on a blog post about that very topic (an event study of how and when the prediction markets responded to the Ryan news). Check for it on PredictWise tonight or tomorrow.