Something quite interesting has been happening on prediction markets over the past month, especially in comparison with the forecasts emerging from statistical models based on opinion polls.
Wouldn't this too limit the wisdom of the crowd effect? Arbitrage allows information from one market flow to another, essentially combining their crowds.
That depends. If one market is dominated by a few traders and they have very large sums in play they can have a disproportionate effect on prices in other much smaller markets. I think we are in this environment at the moment.
Fantastic work Rajiv, thanks! Would you be willing to make comparisons with other sources, possibly by exchanging data with us? I'd love to see a comparison with bias-corrected estimates based on X polls, e.g., of the % of support for candidates [1], or our upcoming estimate of the probability of victory. Based on my personal observations, prediction markets exaggerate the advantage of leading candidates in comparison to the % of public support. For instance, bias-corrected estimates based on X polls show temporal trends resembling prediction markets, but less pronounced. Both sources showed a drop in support for Trump after the debate, increase at the beginning of October, and a drop between October/November [1, 2]. Perhaps there is a rational explanation for that.
> insulated from arbitrage by design
Wouldn't this too limit the wisdom of the crowd effect? Arbitrage allows information from one market flow to another, essentially combining their crowds.
That depends. If one market is dominated by a few traders and they have very large sums in play they can have a disproportionate effect on prices in other much smaller markets. I think we are in this environment at the moment.
Fantastic work Rajiv, thanks! Would you be willing to make comparisons with other sources, possibly by exchanging data with us? I'd love to see a comparison with bias-corrected estimates based on X polls, e.g., of the % of support for candidates [1], or our upcoming estimate of the probability of victory. Based on my personal observations, prediction markets exaggerate the advantage of leading candidates in comparison to the % of public support. For instance, bias-corrected estimates based on X polls show temporal trends resembling prediction markets, but less pronounced. Both sources showed a drop in support for Trump after the debate, increase at the beginning of October, and a drop between October/November [1, 2]. Perhaps there is a rational explanation for that.
[1] https://socialpolls.org/#/timechart
[2] https://www.threads.net/@przemyslslaw/post/DBjhL0koRmP
Yes definitely, let's talk after the election and exchange data. If you have daily prediction updates I can definitely use and evaluate them.