I would never use the word ‘dishonest’ in this forum. I was giving CarnalK the benefit of the doubt: I assumed he stopped reading after the conclusion’s first sentence agreed with his, well, bias.
But here he doubles down. Is it illicit for archaeologists to calibrate their C[sup]14[/sup] dates? Are mechanical engineers not supposed to consider frictions in their analyses??
BTW, from Rothschild’s paper I see the “long-shot bias” appears to be much greater than I realized. Here are some uncorrected estimates, and the result after de-biasing with the formula he borrows from Leigh:
0.500 -> 0.500
0.400 -> 0.339
0.300 -> 0.195
0.200 -> 0.084
0.100 -> 0.018
0.053 -> 0.004
0.030 -> 0.001
Here’s the conversation:
“Betting markets aren’t that useful”
“They’re very accurate if you debias them”
“Maybe, but you never use debiased numbers”
“So what??”
This might be the conversation you think you’re having, but no one else, as far as I can tell is posting the things you are claiming here.
Betting markets are useful even if you don’t debias them. Debiasing is only necessary for examining the tails and even then its not necessary in every circumstance. And in my last post before this one I posted that I do, in fact, debias betting markets for some of my side projects.
Here’s the actual facts in the thread so far and not an imaginary conversation:
Prediction markets are, in general, slightly more accurate than aggregated polling.
Debiased aggregated polling (like 538) can outperform prediction markets.
Debiased prediction markets can perform as well, and in some case better, than even debiased aggregated polling.
The debiasing transformation that Rothschild uses, for favorite-longshot bias, is p’ = ϕ(1.64 * ϕ[sup]-1/sup) where ϕ(x) is the cumulative distribution function of the standard normal distribution and the constant 1.64 was determined empirically by other economists in a different paper using out of sample data (Leigh, et. al. 2007). I wouldn’t want to calculate ϕ(x) by hand, but it, and its inverse, are standard statistical functions that should be built in to whatever you’re using to do this type of analysis.
There’s no physical switch to flip, but one can either choose to do it or not depending on the project requirements. In my Python code it consists of a single line of code that I can toggle on and off by adding/removing a single character to comment/uncomment the line.
I’m saying that contrary to what the post I replied to suggested, I don’t believe that there’s some simple bias in betting markets like “betting markets overestimate people who are in the news a lot”, because such a bias would be easily exploited for money.
I have no doubt that people do make money betting against unrealistic bets. Which will tend to bring betting markets toward better predictions.
Is it illicit for archaeologists to calibrate their C[sup]14[/sup] dates? Are mechanical engineers not supposed to consider frictions in their analyses??