How I Use Prediction Markets
- news
- interpreting some new event’s importance (“did [X] move the market? if so, by how much? if not, does it actually matter?”)
- tracking events as they happen (election night)
- get a highlevel sense of something i have little background info nor domain expertise on (taiwan invasion risk)
- broadly, find people who know what they’re talking about: if an individual is making a lot of money on prediction markets, they probably know quite a bit and are very well calibrated on the topics about which they’re trading. (there are some false positives, but not many; and many many false negatives.)
- then, read their Substack, blog, Twitter, etc. (exs)
- cause myself to feel more financially invested in seeking out the truth of some matter by committing to invest some small some (e.g. $100) in some market i want to learn more about.
- in the extremely rare case where i (a) think that i know something that the rest of the market doesn’t, and (b) have a compelling story for why someone who’s a better trader than i would decide to take the opposite side of a trade with me: trade, and (hopefully, in expectation) make a tiny bit of money.
Things I do not use prediction markets for:
- 100x my money, get rich quick, make $30k TONITE [hyperlink to big filthy]
- {track, bet on} shitcoins
Disclaimers:
-
Sometimes prediction markets are not accurate. I’m ripping off Scott here, but:
As the amount of money you can make from correcting a market goes to infinity, the accuracy (relative to other sources) approaches perfect. But if you can’t make much money on a market, it won’t necessarily be very accurate.
Tracking the total volume of a market will give you a good sense of how much financial pressure there is for traders to correct the market. If a market has $1,000 in total volume, I wouldn’t trust it that much — but $10mm, I’d have a lot more trust.
[insert graph of “how much you can trust the market” to “market volume”]
Some other reasons prediction markets might not be ideal:
They have transaction costs - eg fees or taxes [addition from Saul: or operational difficulties, like transferring funds between five crypto accounts that require lots of legal work to figure out what taxes need to be paid] - that make it not worth correcting small mispricings
There are easier ways to make money than to correct their mispricings. For example, on average you can make 5% per year in the stock market. If there’s a prediction market question about an event that will happen a year from now, and it’s mispriced by 4%, then it’s not worth buying shares in the prediction [market] because you could make more profits by putting your money in the stock market. This is especially problematic for questions about events many years in the future, for example “will there be AGI in 2050?”
People don’t trust the prediction market [platform]. If there is a 10% mispricing, but a 15% chance that the prediction market [platform] will go bust, steal your money, or wrongly resolve the question against you - then on average you lose 5% by correcting the mispricing!
I’ll add some more to Scott’s list:
- unclear resolution criteria, or resolution criteria that don’t match the title of the market
- I expect the most well-informed, best traders on the market to be substantially less-informed on the topic than I am (this is extremely rare!)
- filtering out some relevant traders (e.g. only legally accessible in one country, or by “authorized investors”)
-
COIs: I own a small amount of stock in Manifold (<1/4 of what 1y of salary would’ve gotten me), enough that I often forget I have any at all. I’ve worked as a contractor for Manifold, and projects I’ve led have had the following companies as sponsors: Kalshi, Polymarket, Metaculus, Futuur, Futuresearch, [I might’ve forgotten one or two — sorry!].