- The Colts entered the game with a 68% chance of winning.
- The Bears winning the coin toss made them 0.5% more likely to win.
- Devin Hester’s kick return for a touchdown made the Bears 10.25% more likely to win.
- Thomas Jones’ 52 yard run to the Colts’ 5 yard line had the biggest impact of any offensive play. (10.75%)
- When the Bears took an 8 point lead, the market still viewed the Colts as the favorite to win.
- Kelvin Hayden’s interception was the biggest impact play of the game. It increased the Colts chances of winning by 18% to 94.5%.
- The market was fully convinced of a Colts victory with 5:00 left in the game.
- MVP Peyton Manning ranks only 6th on the list of top impact players of the game for the Colts.
- Rex Grossman’s poor play contributed 36.5% to the Colts’ chance of winning, more than twice as much as the top performing Colt.
Question to Keith Jacks Gamble. Have you taken into account the delay?
- TradeSports imposes a 4-second order-entry delay on some sports and live entertainment contracts to compensate for the varying delays with which members see an event (radio vs. tv; cable vs. satellite; etc.).
(I suppose the answer is “yes”.)
Also, what do people think of this delay? Does BetFair impose this delay too? Is it justified? Fair? Should TradeSports-InTrade get rid of it?
I’m curious about what people think of it.
—
Additional question to Keith Jacks Gamble. Would you prefer the term “prediction market analysis” or “prediction market event study” for what you have done. What’s the better term?
See:
Prediction Market Event Studies – (Eric Zitzewitz’s work on prediction market analysis of the financial markets)
http://www.midasoracle.org/2006/12/15/prediction-market-event-studies/
Prediction Markets for the CFO – by Eric Zitzewitz – 2006-12-13
http://www.midasoracle.org/2006/12/13/prediction-markets-for-the-cfo/
“When the Bears took an 8 point lead, the market still viewed the Colts as the favorite to win.”
For how much time? —just curious.
Yes, I did take into account the delay Tradesports imposes on orders. I recorded the bid and ask prices at least 8 seconds after a play and before the next play began. It’s easy to tell when orders have come in after a play. Traders who have orders near the current price cancel them at the start of the play to avoid getting executed at a bad price if a big move happens on that play. So before the orders come in, the bid ask spread is very wide. In cases where I observed more than one price after a play, I always used the one with the smaller bid-ask spread (though there was never a meaningful difference.) The average bid-ask spread for my data is 1.58.
I think my analysis is better described as a prediction market event study.
I have posted the Excel file I used so that those who are interested can see the effect of each play.
Data.
Thanks.
(((Still a mystery to me why the phrase “prediction market analysis” would not fit. Anyway.)))
I think “prediction market analysis” fits too. I suggested that “event study” is better because I think it says more specifically what type of analysis was used. Perhaps I’m not seeing something you see about its meaning. You have been very careful about using terms carefully and consistently when others have not, which I really appreciate.
With the Colts starting at their 16 yard line after the kickoff, the Bears became slim favorites for the first time. This probability lead lasted only a few plays before Benson fumbled.
Thanks for your comment. (We will have to wait till Monday for other Midas Oracle commenters to pop up here. Saturdays and Sundays have low web traffic —I mean, besides Marginal Revolution linking to this blog post!!)
And thanks for the posting of the Excel file (XLS file). (I suppose that people with Open Office, which is an open-source software, can open it.)
And what about “prediction market event analysis”?????
Or “prediction market and event study”???? (Note the “and”, which says that the study focuses on both the prediction market history and the event history.
And what about “prediction market history analysis”???? Isn’t that better? The idea is to convey efficiently a message to the guy in the street. “History” rings a bell —as opposed to “event study”.
If the idea is to convey efficiently a message to the guy in the street, then I think “prediction market event analysis” or simply “prediction market analysis” is just fine. “Event study” means something specific to an academic, but I don’t think it conveys the idea best to the guy in the street.
Midas Oracle trying to appeal to both demographics (normal people and academic people), I’ll go with both “prediction market analysis” and “prediction market event study”.
Excerpts of the paper:
An Analysis of the 2007 Super Bowl Using Price Changes on TradeSports
http://www.midasoracle.org/2007/02/09/an-analysis-of-the-2007-superbowl-using-price-changes-on-tradesports/
I re-watched the Super Bowl to verify the play-by-play data and to identify clearly dropped balls by receivers (which in my last draft were credited to the quarterback and not the receiver). This resulted in a couple small changes, and my paper has been updated. Peyton Manning’s performance now ranks 6th on the list of impact players for the Colts.
OK. Is there a need to update the chart posted in this blog post? If yes, contact me privately if you need technical help to substitute the old chart with the new one.
There’s no change to the chart since the only changes were to attributions, not prices. The play-by-play data doesn’t attribute dropped balls to the receiver just the quarterback, so I re-watched the game to fix this. Two of Peyton’s incompletions were clear drops (one by Utecht and one by Moorehead.) Taking these drops away from Peyton’s performance measure improves him to 6th on the list of impact players for the Colts. This is the only fact that needs to be edited on previous posts.
Gamble on Gambling
by Michael Abramowicz
http://www.concurringopinions.com/archives/2007/02/gamble_on_gambl.html
The methodology of prediction market event study — Multiple Causes Edition
http://www.midasoracle.org/2007/02/13/the-methodology-of-prediction-market-event-study-multiple-causes-edition/
Rex Grossman for MVP
http://statastic.com/2007/02/13/rex-grossman-for-mvp/
My reply to “Gamble on Gambling”, which includes a warning for those who are using historical data only to estimate these probabilities.
Super Bowl Analysis Highlights — MR Comments
http://www.midasoracle.org/2007/02/15/super-bowl-analysis-highlights-mr-comments/
My reply to the MR comments is now posted on the MR page.
Super Bowl Analysis Highlights — Keith Jacks Gamble’s Second Turn
http://www.midasoracle.org/2007/02/15/super-bowl-analysis-highlights-%e2%80%94-keith-jacks-gambles-second-turn/