Resolves based on the number of times out of 100 that the new 538 Model predicts Harris would win the presidential election.
🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ8,751 | |
| 2 | Ṁ7,993 | |
| 3 | Ṁ7,642 | |
| 4 | Ṁ5,823 | |
| 5 | Ṁ3,915 |
As it turns out the prediction market could not really tell you, the right answer was around 15% and the distribution was pretty spread.
Ya, to be fair though, I have no idea what people were doing on this market. It was so mispriced! And I had already bet so heavily on it for months that I didn't want to correct it.
This was somewhat complicated that they likely changed the model. They might have changed the polls:fundamentals weighting (as @Conflux recently wrote). From what I understand of their methodology it's not clear why there would be such a huge shift in weighting given the amount of polling is similar to pre-dropout.
Though they could have changed more than just the polls:fundamentals weighting function. They have Harris +4 in polls and +3 in fundamentals now. These are way better poll numbers than back in April when they had Biden at 62% odds. It's still not clear to me why they have a lower number for Harris now without methodology changes. But maybe details of the model are just (overly)complicated and it's actually the same.
Either way, I agree with @benshindel that the market was mispriced, particularly with the weighting on 51-55. I lost some Mana thinking the market was more mispriced on even higher than 60, but I think feel ok that it even in hindsight still seems like a positive-EV bet.
Basically, approximately a discrete distribution with a continuous is just for summarizing the market. It doesn’t provide any advice on how to bet. Same way Fourier Analysis can reconstruct a signal perfectly* based on finitely many samples.
*minus high frequency components.
Actually it might be fair to say you had a normal distribution (let’s say 56% +/- 3%) that differed significantly from the markets normal distribution (let’s say 52.5% +/- 2.5%) and made a profit since your model was better (gave a 7.5 higher likelihood of the result in this scenario)
Interesting, why would it be bimodal (i.e. how is it possible that there are two peaks and the mean of them is less likely than either peak)? You suspect there’s 2 different models GEM was choosing from? There’s a binary variable in the model (eg Silver had a kamala_mode flag so incumbency?)
Doesn’t seem that reasonable in this context.
Anyways the idea of summarizing a discrete market as a normal distribution (2 parameters) is just the 1st order approximation which does its job for this market. It can easily be extended to a Gaussian Mixture Model if the market doesn’t appear normal and there’s enough granularity in the market data (analogous to the Discrete Fourier Transform of a signal with N samples having up to N different frequencies)
He is off the rocker, maybe taking same meds as the SBF, the Trump or Roger Stone? Very opinionating and thinking he cannot be wrong, he is most right in thinking mentality. I would not be trusting him for any of the predicting at this moment. Thinking he is very right, it is getting inside of his head.
I was wrong about Nate. He's gained some common sense since leaving 538. This is 100% what was going on.