Bet the Process - Nate Silver New Book

Podcast Release Date:
Podcast Link:
hagrinBPR Review Date: 2024-09-14 07:35:52
Rating: 7.00

Notes:
Bias - Going in, I'm sort of feeling overexposed to Nate at the moment.

3:20 - Politics talk - the only thing I found remotely interesting is that he says they changed very little code to account for the switching of the candidate & basically just added a "don't use polls before x date" logic.

6:40 - Nate retains the IP of the original 538 model so now it's the Silver Bulletin model at Substack.

7:45 - Book talk starts. Apparently there's a chapter on Jeff and a chapter on Rufus? 8:10 - Lessons Nate learned betting - 1) how much of a grind something like the NBA season is (welcome to the party pal) 2) model provided value in early post-player movement events (offseason/start of season, trade deadline). 1.8 mill in bets, $5k net win. 9:55 - Apparently was betting $2k a pop on DK openers.

Some talk about what makes good VCs. A lot of words that can basically be distilled down to long time horizons/perspective + asymmetric risk/reward (Jeff does a good job summarizing). 17:00 - Nate mentions that "a fair criticism of the book is that it focuses on the most successful people" and while there are a bunch of failure stories in the book there does seem to be a lack of non S tier VCs as Jeff mentions. Apparently the 2nd half of the book is about SBF? I'm already trending out on this.

18:30 - Was going to talk to a bunch of Wall St/hedge funds but didn't because it was "too close to sports betting, like sports betting but for stocks". Oh no.

19:10 - Rufus asks a great question - "What does risk taking look like in the next 20 years?" Nate says its bifurcated. Calls free speech "risky". Rufus mentions SBF and the Kelly Criterion and at 21:20 and there's literally a famous thread of SBF vs Dan Robinson talking about this exact topic, we don't have to speculate on what SBF thought. Nate brings that thread up 1 minute later (good on Nate!).

27:30 - Chooses Bill Gurley as the VC he would probably back himself if he had to choose one person from the book. Nate part ends at 28:30.

30:30 - Rufus mentions that Polymarket seems more efficient (while also larger). I think Polymarket has a very clear bias in it although as we get closer to the election that bias is starting to get bet out of the market and this is almost entirely due to the market participants.

32:10 - Jeff rightly calls out Nate and his prior position on prediction markets & his now seeming shift towards being pro predictive markets now that he's developed ties to Poly. 34:15 - Rufus mentions that "this is the Seth Burn argument that for a model to have any value you need to be able to bet fully Kelly on it". Is this what Seth is even actually arguing? Maybe, Seth can be pretty extreme, but whether he tracked the models with quarter or half kelly isn't going to change the fact that you just can't say "all models have value" (something Rufus has said) and then not be able to actually explain why said model, that loses to the market, has value and then be able to do so in a way that you accurately identify the value in the model away from just the variance built in to the subject being modeled. The person trying to hold these modelers accountable is not the bad guy here, I don't know why there's this natural instinct or reaction to question the tracker.

36:00 - Jeff brings up the idea of early stage VCs and how data driven you can possibly be with early stage VCing and I think that's a very interesting topic. Very funnily Jeff is trying to explain Rufus' theoretical approach to early stage VCing to Rufus better than I think Rufus understands the hypothetical.

43:45 - "I know people who built SGPs (pricing) for one book and then were betting them at another." Nice juicy nugget there.

44:30 - Jeff asks Rufus about his football betting plans for this season. One interesting tidbit is that they were going to do some betting near post that they expect to be breakeven or even slightly lose to help with account longevity. 46:10 - Going to use some qualitative opinions that can say "hey your model is going to be weak here" to stay off some games. This is basically my point above - if a naive scoring of a model loses vs the market, then the requirement for proving that the model actually has value is this ability to be able to explain where the value actually is and that's where the burden of proof lays, not with the tracker. Jeff then brings up NCAAF 2H and seems like they both agree that's a very interesting market. Rufus brings up the Cris live flip and how "fragile" the market really is (which is a great point and one that I also noted in the BTP Trench review about what actually happened during the undersea cable maintenance affecting Costa Rica).

50:15 - Some talk about automation for live betting. It's a good discussion - it's a very hard problem, there are many edge cases, spotters and messaging tech are a real advantage here, it can sort of be done, but it's very difficult.