Senate and Recession Odds

Experimenting with AI Helping Me Write Substack Posts By Hand

I have seen many posts about AI potential to replace human writing. This post is an experiment in trying to work with AI in a way I find fun & liberating. I first asked Claude a research question, then wrote up my thoughts on the evidence by hand, then had Claude take my handwritten prose and type it up. That's the meta-context; the object-level post is what I think about prediction market pricing of senate and recession odds.

The Tails Are Too Fat

Whoever is market making Senate elections on Interactive Brokers & Polymarket appears to be modeling the elections with HEAVY FAT tails. On Polymarket, the democrats have an 8% chance of winning Idaho. Non-Partisan polls are non-existent right now but we can use recent electoral history as a guide. Republicans win in Idaho by 30-40 points routinely in federal elections. That kind of shift in margin is literally unprecedented in the modern era but Polymarket traders assign an 8% chance. Mississippi and South Carolina also look interesting to me as mispriced tails but that’s because there are fewer undecided voters and the electorate is highly racially polarized. Dems would need to win truly conservative white people to win.

Once we start getting polling closer to the election date, these Claude generated charts of historic polling error relative to the average since the 1990s are helpful. 90% of senate races are within a 10% error of final polls. Thanks to 538 for putting the data online so Claude could go wild.

Kalshi Recession Odds: Too Damn High

Kalshi prices the odds of a recession this year at 9% and the odds of a recession next year at roughly 41%. Recessions occur in roughly 18% of years and growth is presently strong in mid Q3.

I have shown the orderbooks for buying "NO Recession" in both years below from my phone. Claude on my phone had trouble getting order books without api credentials. I do not think I’d buy at these prices given expected returns of stuff I’m not sharing but I do like the no side at these prices better than saying buying an index fund.

Conclusion

In the politics space there are many markets with tails that feel too fat on Polymarket and interactive brokers (if the one user of interactive brokers forecast x reads this blog). Idaho seems the most extreme to me but there are dozens of examples like this. In economics on kalshi, there is often a pervasive doomerism that sharp bettors fade. The recession market is the prime example but there’s a whole set of markets like this. Be careful, do your own research, beware of fees, this is not investment advice, etc, etc!

As for the meta experiment, Claude is nearly infinitely intelligent but my handwriting is harder than an erdos problem it seems. Editing all the times it marked my handwriting as illegible was humbling.

Here’s the raw data.