Will Manifold arbitrage be automated by July?

111Ṁ1433

resolved Jun 29

Resolved

YES

ALL

There are lots of arbitrage opportunities on Manifold---duplicate markets (like https://manifold.markets/AVS/will-russia-officially-declare-war and https://manifold.markets/dglid/will-russia-formally-declare-war-on-fae0c436ac66), or just closely related markets, or markets on manifold that match markets on metaculus, and so on. Right now, people are doing this arbitrage work manually, whenever they notice that two probabilities are far enough out of sync to justify the investment. There's some profit to be made from automating the process. This will get easier when the official API supports trading, which is like to happen sometime in June: https://manifold.markets/LarsTrieloff/will-manifold-markets-offer-a-publi-64ff5aaccc48. This question will resolve YES if at any point from now through June 30, it becomes public knowledge that somebody has automated the arbitrage process in at least one case. I've initialized the probability to 25% to give someone an opportunity to do some insider trading.

Get

1,000

to start trading!

🏅 Top traders

#	Name	Total profit
1		Ṁ297
2		Ṁ52

19 Comments

8 Holders

15 Trades

Sort by:

predictedYES 2y

Here we go: https://ineffectivetheory.com/automating-arbitrage-manifold/

predictedYES 2y

Nothing yet. Clearly, I'm going to have to do this myself!

I'd gladly implement something like this once the graphql api is available, it seems to much of a hassle without

predictedYES 2y

@JoyVoid The annoying part is identifying similar markets by hand. Took me way longer than I expected to find a good, simple example. How did you intend to use the graphql api? I got the sense from your previous comment that you were thinking of using the time-series of probability. I have trouble believing that works well, partly because it's a measure of how well the markets are *already* being arbitraged! (And partly because there just isn't enough data on a lot of these markets.)

The broader idea I have is that you could draw your own graph of belief and change it in real time. The graph propagates probability across the network, and you could either enter your own probability for some markets, or draw from market probability. Then it buys/sell in function of your probability compared to that of the market's. Seems big, so I do not know when I will work on this. But on a local scale, I'd probably just have a program that just implements limit order or something vaguely similar.

I think this should resolve Yes: https://manifold.markets/ManifoldMarketsUser/will-this-market-close-at-50

predictedYES 2y

@EnopoletusHarding Looks like it. Will resolve later today if nobody objects.

predictedYES 2y

Upon reflection, I'm not sure that counts as arbitrage. It's just regular automated trading, right? I'll hold out for a more clear-cut example.

predictedYES 2y

I won't be counting it towards positive resolution, but I believe @MetaculusBot may be automating the process of making M$ bets based on metaculus probabilities. Oh brave new world, that has such bots in it...

Seems more likely, now that the appropriate API is in place.

predictedNO 2y

@ScottLawrence The betting API is live now?

predictedYES 2y

@IsaacKing I've at least tested the market-creating API. The same docs (https://docs.manifold.markets/api) indicate a /v0/bet endpoint, but I haven't tested it myself.

How will you know if two markets have been arbitraged automatically?

predictedYES 2y

@MathiasFoster I won't, necessarily. This question only resolves yes if "it becomes public knowledge". (Of course, if I do it myself, then I'll know. I have some YES shares, so I'll also be incentivized to make it public knowledge!)

How would this even be possible? Questions almost never use the exact the same wording, and modern NLP isn't good enough to figure out whether two questions are asking the same thing. Not to mention that most arbitragable questions aren't exactly the same, there are usually some minor differences in the resolution criteria or resolution date.

@IsaacKing The question resolves yes if the arbitrage process is automated in *at least one case*---not all cases. Meaning, that a human can identify two markets and say "these should always be within 5%" (for example), and then the automation is in charge of trading with that goal in mind.

@IsaacKing I mean, you could probably use a bayesnet to quantify how much each markets are dependent from each other. You input the graph and the program takes care of the rest

predictedNO 2y

@ScottLawrence How are you going to verify whether a given automated system is able to obtain a net positive profit via arbitrage?

predictedYES 2y

@IsaacKing Presumably someone will write up a blog post with reasonably convincing details (like "here's the log of trades made, here's the profit after a week of running"). If there's some reason to believe that the person is lying, then I'll use my discretion.

🏅 Top traders

Related questions

Related questions