What did we learn from the 2020 election? The SSG blog invited some contributions from other traders (read it here) and I wasn’t able to come up with much at the time of asking. But now with the benefit of a few more full nights of sleep, I’m ready to tentatively sketch out some lessons to take from 2020:
1. The polls were wrong but not unpredictably wrong
If there is a silver lining in this election for MAGA bettors, it’s that at least they can claim that they were right about the polls underestimating support for Trump. It’s unclear exactly why this happened (I am intrigued by the nonresponse bias hypothesis, perhaps exacerbated by the pandemic polarizing stay-at-home-by-the-phone behavior on partisan lines), and even why the extent to which it happened varied by state. Preliminarily, states with lots of educated white college-aged voters were polled well, while those with high populations of white non-college voters were not (see: MN vs WI). Although, district-level polling of suburban educated districts also showed down-ballot Democrats doing a lot better than they ended up doing! So perhaps not the entire story.
There were some hints of reality in the polling. Biden’s softening support with Hispanic voters? It was there, but did not show up as dramatically as it manifested. And the polls did accurately capture the partisan splits in vote by modality (they nailed vote-by-mail preferences), but probably missed on turnout by subgroup. From a bettor’s perspective, what can we learn to minimize our exposure to polling error in the future?
My takeaway is that you have to always be vigilant in challenging assumptions. Did I consider that Biden’s support among Hispanics was soft and getting softer? Yes. Did I ask “what would happen if the floor really fell out for him in this group?” No. Did I take ABC’s Biden +17 in WI poll seriously? No. Did I ask “what would happen if there’s systematic partisan nonresponse among white noncollege voters?” Also no. Did I even fathom that polls of down-ballot Democrats would be so incredibly bad? No, but I mean, lol.
One final point on polling: the polls “felt” a lot worse than they will end up being, and a lot of that has to do with the order in which votes were counted. At the same time, they told a story (“Dems will have a great night”) that did not come to pass, particularly in the House and Senate. And any time the narrative is up-ended, people are going to be mad at the polls. (Note that in 2018, the polls also got a lot of stuff wrong, but people weren’t mad because the overall narrative was correct: Democrats gained a bunch in the House).
2. The models were both wrong and good
In terms of spitting out a win probability, both 538 and The Economist put out models that did a reasonable job given the inputs they had. The inputs were just kind of shit. Was Biden really 90+ to win? He was not. Polling error was not evenly distributed around the mean; it was always more likely that polls favored Biden than that they favored Trump. This shows up clearly when you look at forecasted margins: the prominent modelers had mediocre performances. (Notably, one modeler, Plural Vote, stands out in the crowd; they also had Joe only at 66% to win overall).
3. The markets were right for the wrong reasons
Guess what! I think the market’s price for Joe’s odds of winning was closer to reality than what the models had. Fair value? Joe was worth 60-65c, well less than the 95+ that I had him at going into the election. The tipping point state margin was 0.6 points – and at the end of the day that’s just a close election. Certainly not a 90/10 election or a 95/5 election.
Do I think the market arrived at that price by carefully integrating information available to it? No. It got lucky that the combination of “smart” money and stupid money ended up putting things about where they should be. And if you’re of the mind that this proves markets work, please do go and look at prices right now. Yesterday (a week after the election! with most of the votes counted!) people were paying 10c that Trump would win the popular vote, when none of the lawsuits he’s filed makes that remotely possible. You can literally get paid for predicting what has already happened.
It’s also worth asking what the market’s price would have been if polls had zero polling error whatsoever. I think any universe where Trump is polling within a point in the tipping point states, +3 in Florida, etc, is a universe where the market prices him a 65c-70c favorite to win re-election. In other words, the market’s belief was that the polls *always* are wrong by some fixed amount, not that the market knew what the exact margins in each state would be (indeed, pricing in various states suggests the markets were not aware of the true odds).
4. Don’t rely on PredictIt
Those that follow this blog or me on twitter know that PI has a long history of crashing when many traders are trying to trade at the same time. I expected it. I just didn’t think it would crash so soon (nor last so long!) – and as a result I ate a lot of losses that I shouldn’t have had I been able to cancel orders when Miami-Dade nuked everything. My fault, really, and of course I was able to trade my way out of the hole relatively easily.
But the experience has led me to another (blindingly obvious) conclusion, maybe only relevant to people like me that are trading semi-professionally: I really need to learn how to use offshore sportsbooks and crypto exchanges. I would have absolutely crushed 10c Biden-in-Michigan lines had I been able to. There’s no excuse for this, and I think it only makes sense that to be a true professional at political betting (there are very few people that can claim to be) you simply have to know how and where to get your money down.
5. Which electoral shifts persist and which reverse in the coming cycles?
There are a lot of very interesting questions left over from this election, none of which I have any great answers to:
Is the lurch right among Hispanics reversible? Is it Trump-related? An artifact of the COVID economy? Concentrated among men but not latinas? Bigger in rural areas than urban? Are Hispanics becoming white, the same way that whiteness swallowed up immigrant ethnic enclaves in the past? How big a factor was disinformation?
What about white non-college voters, particularly in the north? If they are now culturally polarized, how much more margin can they give up? What about their turnout – can any Republican nominee other than Trump get that level of turnout in this group?
Will white suburban college-educated voters keep shifting left? Or does their higher rate of ticket-splitting suggest that they’ll be comfortable coming home to the Republican party – with a not-too-crazy nominee?
Will future Republicans continue to make up ground with Black men? Or was this, again, a function of Trump?
A lot to chew on and think about as we approach the next round of special elections, midterms, and 2024. See you all then.