There is a market (several actually but that’s not important), but it is capped at 5,000 traders. On top of that there is a cap of $850 per trader per market so some of these traders are in a position where they can sell but not buy. Even with all that there are people are actively trading right now for various reasons. Finally, every once in a while someone sells off all their shares in a market and there’s a chance for a new person to enter.
Trading is quite active on PredictIt. The markets aren’t exactly closed, each contract has a trader cap (5000 traders). So when someone sells all their shares in a contract, someone else can buy in. I’ve been able to get into every market after a few tries except for the main Presidential market.
You can put in buy/sell orders with max/min prices and those orders can be accepted even if you’re offline.
When I bought my 1,000 shares in the PA, GA, AZ market I mentioned earlier today, I saw that it was moving around at $0.86 - $0.88. I put in a buy order for 1,000 at $0.85 and went to bed. In the morning that order had been filled in pieces by automatically matching me with several sellers when the price dipped at various times.
I’m still waiting for a couple other markets to settle. It will be over $500 all told from shares I bought well after the election. A nice little Xmas bonus. Thanks MAGA bettors!
With this narrow of a market, it doesn’t seem as though it would in any way necessarily correspond with reality. There simply isn’t enough of an ability for those with rational foresight to force to price down any further. The people left in the market at this are self-selected for irrationally supporting the ability for Trump to win, and pretty much nothing is going to change their mind. I might admit that the data says that these markets did a better job than polling aggregators this time around, but I would say that’s simply luck.
It doesn’t matter if you admit it or not. It’s simply a fact that a PredictIt derived probabilistic forecast outperformed 538’s probabilistic forecast in the months leading up to the election. It wasn’t even close.
Again, I’m not disputing that fact. But I am disputing that it’s a meaningful fact, simply because it’s possible that someone throwing darts could have done even better. The procedure behind how the forecast is determined is much more important than the actual forecast when it comes to who you should listen to in the future. Wasn’t there an octopus somewhere that successfully predicted so many Stanley Cup finals results?
Maybe your entire point is that betting markets did better this one time, with absolutely no implications for what that might mean for the future. But I don’t see what the point of saying that over and over is. By stating how limited the liquidity in the market is, I see that’s it’s not all that more sophisticated than throwing darts. A little, sure, but it’s simply not using the weight of nearly all of the available opinions on the matter, whereas polling aggregators are using as much information as they have available. Sometimes that information is inaccurate and the weight of opinions that happen to present themselves in the market happens to be well-aligned with reality.
The odds available right now shows why the betting markets are not particularly good metrics as a whole, because right now we know the probability of Trump winning is negligible, yet his shares are trading at vastly more than the minimum price. That they used to be a better indicator is no more useful than that octopus up until the time it randomly gets it wrong.
A paper I linked earlier in the thread showed that Intrade outperformed 538 in 2008. So in two out two times where a prediction market was compared head to head with 538 in a presidential election, the prediction market did better.
I suspected that PredictIt did better than 538 in 2016, but I didn’t collect data that year and there didn’t appear to be a way to get data after the fact. I rectified that by starting data collection in April of 2020 so that the two could be properly compared.
So we have some evidence that prediction markets are better than 538 at forecasting elections in the months leading up to the election, and zero evidence that they are not. Seems like a meaningful fact to me. Your mileage may vary.
Yes, their predictions before the elections seem to be pretty good. But I’d say it’s a major problem when their “predictions” after the election are utterly irrational. It suggests that the reason they happen to be right in the time before the election is just luck and small sample sizes. When your system predicts nonsense in cases that are not in question, there’s clearly something wrong.
Still, the sample size thing is puzzling. If that (or any other) PredictIt market had had not 5,000 participants, but at 50,000 would things be different?
I fully understand how a properly selected scientific sample set can accurately stand in for a comparatively vast population. I also understand how a unscientifically self-selecting sample set can grossly mislead about the corresponding full population.
Since PredictIt is self-selecting, that gives a higher bar on how big a sample is needed to “blend out” the self-selection biases.
All the above is talking pre-election. As you say, post-election we’re dealing with a weird market wrap-up effect. But also one that can be subject to capture by a small number of the clueless. Who can then be duly fleeced by small numbers of the clueful.
Such as yourself. Congratulations on a job well-done that needed doing!
Yeah, I’m wondering that, too. It seems pretty clear that the post-election market is overly limited, but raises the question if 5,000 is a big enough size either.
However, I’m not sure if the self-selection bias is worse or better than the selection bias between people who will versus won’t answer their phones from an unknown number, given the same sample size, which the prediction markets aren’t. It would be one thing if they simply were answering a poll about their own preferences, which costs nothing and absolutely would introduce self-selection bias toward the faithful and trolls and bots, but I’m not sure that people are so irrational that they won’t do any introspection about actual percentages before risking real money.
The only thing that looking at PredictIt right now tells us is that people exist who are buying the President’s bullshit. But we already knew that and have known that for years.
There were at least 20 different Presidential markets, including the state markets, where you could still bet after the election. Most of the markets had at least two contracts. The 5000 trader and $850 limit applies separately to each contract, and many contracts never hit the trader cap.
There are also more than 20 other Secret markets that expire in March where you can make pro and anti Trump bets.
All the pricing in these markets told the same story. It’s not like the main Presidential market was an outlier. I don’t think the 5000 trader cap made a difference in the market prices.
Article in Slate about the pro-Trump skew in the betting markets. (It’s about betting sites with lines set by bookies versus markets such as PredictIt, but most of the underlying issues are the same.)
One interesting point I had not thought of: Trump supporters are disproportionately likely to be the betting type.