A minor point from afar on the debates I see about the fundamentals after yesterday's Q2 numbers: the problem isn't just that the economic data is out of sample, it's that there's no reason to assume that 'fundamentals models' are actual forecasting models

The 'fundamentals models,' particularly those emerging from academia, are just intended to explains variance in past election results. No one serious interprets them as literal forecasts. To the extent they have predictive power, it's only under a series of assumptions

To take one obvious assumption: relative peace and prosperity. Some fundamentals models are even explicit in recognizing that there might be different rules under different circumstances, like the 'bread and peace' model which controls for Vietnam and Korea

There are other obvious examples, even if there wasn't an election that year: does anyone think George W. Bush would have lost a hypothetical November 2001 election because of the state of the economy?

The more dangerous reason we might be out of sample right now isn't because a handful of economic indicators are out of whack, it's because of coronavirus: this may not be usual moment of peace and prosperity that allows sentiment to simply track the course of economic growth

To some extent, forecasters this cycle are quite lucky! The direction of the economy *happens* to line up with the course of coronavirus in the US.
That was not inevitable: other world leaders, despite terrible growth, are quite popular. A different president could have been too

That's not to say that the economy exerts no influence on the outcome. But let's not pretend that the economy represents the obvious basis for our prior right now

On a personal note, one thing that @ForecasterEnten and I have in common is that we were both into meteorology before elections, and I don't think it's a coincidence that we're less enthusiastic about social science modeling than the other data types

Weather forecasting models are dynamic: they assimilate the data into a picture of the world as it is, and they simulate forward in accordance with the laws of physics.

Weather forecasts are incredible, but they're still panned by most of you despite the incredible advantage of having exact measurement and the laws of physics on their side

The election forecasts are empirical. They don't really simulate the future and, if you're accustomed to weather forecasting or something, they're barely forecasts at all

A poll-based model is almost the weather equivalent of saying 'well, it was 85 degrees today, it'll probably be around 85 degrees tomorrow, plus or minus x' (and you can do your best to dredge through the records to add the right margin of error).

With this kind of model, you'll basically never really anticipate a surprise. The candidate who leads the polls will basically always be favored, but sometimes they'll lose because sometimes cold fronts do come through

You certainly have no way of anticipating crazy things that might not have any precedent in your observed experience, like a hurricane or something.

The addition of the 'fundamentals' are generally helpful. In weather, maybe adding the time of year to our 'today's temperature-based' model is a nice example. If it's 50 in January, maybe you shouldn't simply bet on 50 +/- 10 tomorrow. But this isn't magic or something

That said, we have quite a bit more reason to be confident about the 'seasons' as our fundamentals in the weather than we do about the economy, even if we used it in the same way

Anyway, this isn't to say the empirical models are useless. Not at all! But they're disappointing if you entered the field having previously thought about forecasting in a scientific settings, like physics or weather. They might not even feel like forecasts at all