Thursday, November 5, 2020

Election Betting's Wild Ride & Notes

 Another year and another election is (almost) in the books. Like the previous two, I'll jot down a few of my thoughts for those interested (which this year seems more than usual).  As is custom, a disclaimer this is my opinion, and as we see with percentages changing two days later, opinion is all we have. 

I wrote in 2016 how confusing things are in modern U.S. election betting, and 2020 (as we'd expect 2020 to be) was even more so. It's vital for us - who are using the same data as a decision desk - to be able to portend what we're seeing in real time. That is, what happens in Viggo County IN or Jessamine KY, will likely tell us a lot, and we can extrapolate and model as other results come in. That, this year, was almost impossible; there were clues Trump was going to do better than estimated but they were just clues that were difficult to act on. You simply could not use this like you could in 2016. 

The Florida Domino

The big tell of the evening, that likely cost a lot of people a lot of money, was Florida. It's the first state that really doesn't have its head up its ass in reporting results. And those results were good for Trump. If we knew on Monday that Trump was going to run ahead of 2016 in Florida, we'd portend Georgia would be a 5 point or more win (and hammer at -180), we'd surmise he should be chalk, and we'd bet accordingly. That's exactly what the market did and it acted perfectly rationally. 

When the numbers that followed in the midwest - Ohio looking near exactly the same as 2016 (if you lost money on the fake-out with early results showing a 10 point Biden lead you surely will remember for next time), and Iowa looking +9 - the markets in Michigan and Pennsylvania followed. Again, as perfectly expected. When Wisconsin, which counts fast, showed the same results as last time, it was another arrow in the -400 Trump quiver. 

This election was looking like 2016, and because of recency bias, the markets probably overreacted, but to say they were 'wrong' is missing the big picture on what we do each election. 

Perhaps we can point to softer vote in Arizona as a big red flag, but think about it, if NV came in faster than AZ, Trump would've been even lower. Georgia, which *looked* softer than Florida, maybe was another caution point. I was losing confidence in some of the county data in the midwest (it was softer where Trump won razor thin margins) and acted accordingly (a good move), but it was not a big position. 

This year, Florida along with knowing what happened last time, drove the bus. 

The Case of no 100%'s

The phenomenon of this election was the case of no 100%'s. Early voting (that comes in late in some states, with the results in others, or early in others) generally assured we'd get no 100% reported counties and that was pretty deadly. Even decision desks and needles, which can be no help at all but built real time models with a team for many months, had trouble with modeling turnout, and what's left in said county from where. I was checking Indianapolis and area at 9:30 for clues to bet MI and WI and PA, but it wasn't even 70% in. 100's are our best friends, and with mail-in voting and otherwise, we were toast. 

Biden Value

For everything presented above, the value this time was with Biden, no question. Not because we were sure he'd win but because in 2016 the markets were anchored to the polls and this time they were anchored to Florida (and to a lesser extent Ohio and later Wisconsin, where Trump in the RCP average was down 7%, and it looked even). This, truth be told, was again pretty difficult to act on. Gennessee County in MI showed Trump overperformance with most of the vote in. Kenosha in WI similar. You'd want to pull the trigger, but you'd get close to 100% data that threw you for a loop. 

In PA, where I tweeted out Trump felt softer, again it was tough, but with Trump at -250 your value was with Biden, in my view. It was the only bet you could make. But if you want to bet 4 figures on that, you have a stronger stomach than I. Good on ya, you made a nice bet. 

Notes:

Polling

I read a bunch of books on the 2016 election. In each of them I was struck by how damn smart Jared Kushner was - with data, with fundraising via the web and with his polling. I think I tweeted it out once and got dive-bombed by the left wing luftwaffe, but it doesn't make it less of a fact. I had read a week or so ago that whispers about internal polling from him and his team showed Iowa Trump plus 9 and Ohio Trump plus 8. I didn't act on it as much as I should. But with those numbers, we could forsee a Trump win, so maybe it's good I didn't. 

I like most of you, relied on the polls from the smart people (OH Biden plus 2, Iowa about tied), and those smart people in many places did not do well. This was my anchor at times and because I bet 50X more as the vote comes in than in positions before the vote, it might've been much ado about nothing. But I can't believe I had a blind spot. I'm sitting here giving more weight to Quinnipiac who had been dreadful since forever with Trump, and don't give weight to a guy I know is smart with this stuff, who co-created Cadre at like age 27. 

More money was perhaps to be made in the Senate and House markets (Collins RCP average minus 7, she won by 4, for example), but anchoring to the polls this time, like last, was deadly for your brain, and bankroll. 

With a normal candidate with less of a broad coalition the polls will be better no doubt, but wow, in some areas they were complete trash. 

My other blind spot this election was the size of the city vote, and just how big Biden's lead was with mail-in. It was difficult to model, but there were clues early this would be bigger than usual. How do I not up positions in MI, or PA on that? I simply missed it. 

Is there Meat Left on the Bone?

What struck me most perhaps on this election, is the normalcy of US sports betting and the number of people betting the election this time, rather than last time. When I was posting thoughts in 2012 I think most people blocked me because they weren't betting and I was going on about some obscure county that looked good for Obama. This time, like everyone was betting, and they are some really, really smart people. I'm glad these people weren't around in 2004 or 08 or 12, because the chances for profit would be a lot smaller. 

Back in 2008 I was playing the Missouri Primary at Betfair between Hillary and Barack Obama. I had precinct level data via the Missouri gov website in some sort of ASCI format. The vote came in and it was massive for Clinton - but it was virtually all suburbs and rural, where we knew she was super strong. Some networks were assuming it a win for her with these big numbers. 

But St. Louis had not come in. With some back of the napkin calculations, it appeared to me Obama was at the very least a coin flip. Betfair had Clinton trading at 1.01, yes, people just unloading positions. I took all I could - which was pretty sizeable at that time. About an hour later Obama had won, by a not insignificant 12,000 votes. 

In 2020 - that never happens of course. People are smart, they're engaged, and like in any game of skill, the implied probability will be near actual, not farther away. 

Having said that, if this was your first big betting election, I can stress - it's not usually this hard. One would expect states to be more like Florida in their vote (much of what we see with flash-quick results in Canada), and we won't have COVID issues changing just about everything that comes in. I am fairly sure the data holes will be plugged, which will make it easier to pick a winner, but maybe with no change in lesser value. 

Thanks for reading everyone, and have a nice Thursday. 



Summary of my bets (probably on my twitter feed, but since I am not selling anything, it's not like I am searching for them):

Trump (pre-COVID) long, January and February -105 average (for too much freaking money, but seeing the results this was one of the better bets I probably have ever made)

Iowa Trump -250 and later in the night -300. 

Biden +180 in October in GA (looked like a loss, but maybe not)

Biden -800 election night NH after some city results reported

Biden WI small as results came in

Trump Arizona, fairly large at I think -150, when the FL hispanic numbers came in. 

Biden 26-27 states at I think +375 or something

There are probably a couple others, but I have not gone through them all yet. I'll wait until it's over. 



Most Trafficked, Last 12 Months

Similar

Carryovers Provide Big Reach and an Immediate Return

Sinking marketing money directly into the horseplayer by seeding pools is effective, in both theory and practice In Ontario and elsewher...