In 2016, on Tuesday, looking at data rolling in, I told the then-global head of rates at JPM that Trump would win. It was not, shall we say, a consensus call at the time. So I want to talk a little bit about what I'm seeing right now, since the timeline is going crazy. On a trading desk, it matters more to be right than to be popular, and it matters more to be right than to predict what you want to happen. I have the curse of being a mathematician by training. So I still have the muscles, even if I don't use them as much. Right now I would say this: 1 - Anyone who thinks they know anything with high confidence you should immediately discount in this election cycle. 2 - The error bars on races, especially the swing races in the Senate, are likely so large that polling may not be particularly predictive. 3 - There are conflicting cross factors that really are making a mess out of some of the traditional polling which is causing this. Namely, what I'm seeing is a few things that gives me major heartburn. First, non-response rates to polls are now so high I think it's impossible to know if we have a representative sample of any subgroup. We can 100% get the aggregate weighting correct to voter turnout, but when you model your R sample at 10% crossover votes and get 4%, or your I sample at mildly D leaning at +2% but instead get +10%, it means all your re-weighting was an exercise in false specificity. Second, we have major conflicts in the data. People are claiming they did not vote for Trump in 2024 when we are pretty sure they did, and we are getting R samples in some swing states where if the sample is correct about the sides voting in 2024, Harris would be president (she's not, if you've been in a coma). At the same time, we have young people looking maybe way more like they did in 2022 than 2026, and we know we are missing a large chunk of that sample because they never fucking answer the phone, so the net result is... ... ...? Yeah, nobody knows. Third, mutually combative forces. Anti-incumbency is a thing in mid-term elections, it's also a thing in American politics where "fuck the guy in charge I hate it here" seems to be stronger than partisanship in an increasing number of cases. This means that helps Dems in red states but hurts Dems in blue states? The reverse would be true for Republicans. These kinds of forces produce compression. Fourth, social media and Twitter are not the real world, but the distortionary forces here are influencing the polling landscape, where I would imagine the most media-connected people are also the ones taking polls, which leads to the invisible sample also being the ones in a different discourse bubble than the ones on here. The punch line is simple: take any error bar and triple it. Quadruple if you're in Maine or something. We don't know anything, and any confident prediction about Dems taking the Senate or Republicans holding the House are almost certainly wrong. Not in their specific outcome, but rather, in good process. All pigs are fish therefore 2+2 = 4 type thinking. Right now, depending on how you want to torture the data, you can see any outcome from R/R to D/D and any middle state. We've had large polling misses meaning we should be favoring Dems (NJ governor, anyone?), Rs being underweighted in the past meaning we should favor them (IA, anyone?), and just internal polling that was way off (El-Sayed way up in the primary, anyone?). Anyone giving you a confident hot take here, in short, is full of it. Even the early voting is not uniform, and while the information content there is not always high, there is usually not zero, and if you're looking at data from GA vs. OH, you aren't getting the exact same vibes, much less some other places. In short: national politics is getting weirder, polling is getting harder, the parties are fragmenting internally, and nobody knows shit yet. Good news, though: they'll all claim they were right after the fact. What you should do is track both predictions and process to see who is really worth listening to over time.