It's interesting that this is explicitly for non-sports markets because I see no reason why this would be less applicable there. Sports betters have long talked about that the winning strategy is usually to bet the under (i.e. the no) on most bets. The over is generally a more exciting and fun outcome which causes it to attract more betters which in turns makes that side overpriced.
Like with this bot, I have no idea if that will lead to actual positive returns. This as this might just be a remnant from a time when these betting lines were set less intelligently. But all things being equal, it seems logical that "boring" bets would have a better return in the long run than "exciting" bets as long as some betters are primarily motivated by entertainment.
There's probably a lot of knowledge like this that sports betters have built up over decades that applies to these new forms of non-sports gambling.
The contrarian bet is fun but I wonder how it actually holds up. Prediction markets do tend to overprice dramatic outcomes, so "always bet no" isn't as dumb as it sounds. Would love to see real P&L over a few months, not just the thesis.
I've backtested this kind of strategy, and it had a good return (like 100% APR), but then I realized it was cheating by knowing when things are going to resolve. Often times it's not clear. Your return depends a lot on how quickly you can get your money out. I never got around to trying a strat that doesn't know the resolution time, which actually has to be manual cause it takes some judgement to pick things that you expect to resolve soon.
Also requires a lot of volume to be "predictable" obviously, since 1 loss sets you back 10-20 wins. It's surprisingly hard to find reasonable-liquidity markets after all your filtering. Scare quotes around "predictable" because you never know if others will use this strat or a lot of unlikely events will happen due to insiders.
Edit: Another thing, betting on the overdog in sports markets actually looked more appealing because there are plenty of those events with large volume, they're kinda homogenous, you know exactly when they resolve, and they're harder to rig. Just like the author, I was excluding sports at first. I simply never got around to putting real time or money into that.
I drew the same analogy. You put up $0.95, a YES gambler only puts $0.05 (ignoring spread); you're "lending" the money in case of a YES. Problem is, leaving the money in longer doesn't seem to mean a higher premium for the YES side.
Basically arbitraging human imagination. People love coming up with fantastical concepts because they get attention, but the more exciting a market is, the less likely it is to actually happen. Reality is usually boring.
My general observation is that people tend to underestimate the likelihood of black swan events (covid, financial crisis) even as it's pretty obvious they're happening. And then when they do accept it, they react too far the other way and assume it's never going to end.
I've had success playing the markets in these specific cases. I did fritter away a lot of my gains from the financial crisis thinking I was a genius market timer. But I learned my lesson and didn't waver once I jumped back in after covid.
In both cases I got out before a bulk of the crash and timed the bottom almost to the day. Lucky I know, but I had reasons for both. For the financial crisis it was when Bill Fleckenstein closed his bear fund and put it all in MSFT. For covid it was when it looked like the lockdown was working and NYC hospitals weren't going to completely fall over like Northern Italy or Wuhan.
For any non black-swan scenarios, I assume I'll never get one up on the masters of the universe and just leave everything in blended age-appropriate funds.
I'm very concerned about an AI crash and the future of white collar work in general. But it feels more like a slow death to me than a black swan. So I'm just hedging with bonds and cash and stocks that hopefully don't crash as hard in a recession.
I would also not want to take even a fair bet against black swan because the day that "S&P500 falls to lowest level since 2016 as Labubus collapse" is the headline is the exact day I least want to lose a big pile of money gambling. If it's the shareholder's money though.... I'm probably getting laid off in that scenario anyways...
This makes sense to me, but isn't there a risk of increasing the potential payoff high enough that someone is motivated to go out and make the yes side happen?
Consider this bot running on us military outcomes or something.
By design it's a game where people with inside knowledge or enough power to bend reality can steal money from people with gambling addiction. Automating your addiction might not be the best move.
This is what markets like Polymarket boil down to. Normies can't win. Some will, of course, but that's just chance and there's no way if ensuring it's you.
It's really no different than a casino: if you ever find yourself with more money than you walked in with, cash out and leave.
Best strategy for most people though is to simply not participate and you'll break even.
You're thinking like an engineer and making the laughable assumption that "prediction markets" are markets. It's totally unregulated with all sorts of grifts and cheats. One of the platforms was promoting a high-return bet against Rory at the Masters yesterday.
You can make money off of all sorts of stuff. You can "sell" the bets, so there's lots of live pump and dump.
We've gone full circle. The bookie with no neck that smelled like onions was more honest than these platforms.
What happens if you flood the market with a bunch of implausible bets like "sun won't rise tomorrow"? Sure, you might try to filter that out with some sort of "seasoning" period (ie. don't buy new markets), but then that means more time for arbitrageurs to correctly price the market, depriving you of any price advantage you might have had.
Except that the mere existence of the market with the question posed for people to consider, probably activates the availability heuristic[1], causing people to overestimate the likelihood.
That logic doesn't work because not every bet have even payouts. If there's a market for whether a dice rolls 1 or not, the odds might resolve to "no" 83% of the time, but if it only pays you $1.1 per dollar wagered on "no", you're still losing money.
I think we've collectively DDoSed it. I'm getting a 504 timeout.
The author [page](https://github.com/sterlingcrispin) is there on github, but I can't even find his full list of his repos to confirm it's still there (I also get a 504 on that).
Honest question: Why in all hell would you open source this?
I have been making money with a bot off a statistical anomaly in prediction markets lately. There is no way in hell I will open source it or tell you what that anomaly is because I have capacity back-tested it and there are so many players in the market; if all of HN and Github start downloading and use my code it WILL cease to work.
Put another way, your orders are helping move the market and price the market more efficiently; that's the market compensating you for pricing things better. If a thousand people run your strategy, prices will get moved to exactly the point where your strategy stops working. You effectively split that pie with a thousand people.
If this seems interesting for you remember that if you are putting $100 in a 99 to 1 bet you need to win 100 times to get $100 but only need to loose 1 time to loose $100.
And the chance of losing at least once in a 99% sure bet after 100 rounds is around 60%. Even if you reduce to 30 rounds it still is around 30%.
This may seem smart at first glance, but the math doesn't really checks out.
"A comprehensive dataset of 1.9 billion trading records from Polymarket, processed into multiple analysis-ready formats. Features cleaned data, unified token perspectives, and user-level transformations — ready for market research, behavioral studies, and quantitative analysis."
No, data situation is bad, at least for market making - you need to scrape the orderbook yourself to be able to do any realistic backtesting. And even then, it's hard to know whether other bids at the same price are ahead of you or behind you in the queue.
Disclaimer: I contribute work as a political advisor and don't participate in betting markets as a market participant.
Nevertheless, Polymarket is a very interesting marketplace of sentiments and information, and it can be a very strong leading indicator of huge price movements in "real" markets like the NYSE, in part because it directly measures one factor of sentiment, i.e. whatever the prediction is about. Market sentiment determines market prices on very large and deep markets, too.
In the run-up to the election, when Trump was running against Biden, a betting market was a leading predictor of NASDAQ (a very deep, very liquid index of stocks). I wrote up the findings here: https://medium.com/@rviragh/does-the-stock-market-react-posi...
This indicator was the best one anyone has ever shown for NASDAQ for any signal, period. The signal was so strong it trumped all other signals and variances of any kind. Traders trading with just this signal and no other signal of any kind could have made practically an unlimited amount of money as long as the signal was intact. (Basically, until Biden dropped out.)
I myself didn't place any bet due to my role as a political advisor at the time, but the size of the correlation is still the biggest and most surprising one I've ever seen.
It's interesting that this is explicitly for non-sports markets because I see no reason why this would be less applicable there. Sports betters have long talked about that the winning strategy is usually to bet the under (i.e. the no) on most bets. The over is generally a more exciting and fun outcome which causes it to attract more betters which in turns makes that side overpriced.
Like with this bot, I have no idea if that will lead to actual positive returns. This as this might just be a remnant from a time when these betting lines were set less intelligently. But all things being equal, it seems logical that "boring" bets would have a better return in the long run than "exciting" bets as long as some betters are primarily motivated by entertainment.
There's probably a lot of knowledge like this that sports betters have built up over decades that applies to these new forms of non-sports gambling.
https://x.com/sterlingcrispin/status/2043723823678382254
They admit no returns.
But it does seem like a fun project and nowhere does it say anything about returns or profits so not scammy imo just funny meme backed code
Yes exactly.
The bot has zero risk management and I have a strong disclaimer on the github it is essentially a meme.
73% of all polymarkets do resolve to No though.
There's a good dataset on huggingface if you wanted to do some data science
https://huggingface.co/datasets/SII-WANGZJ/Polymarket_data
The contrarian bet is fun but I wonder how it actually holds up. Prediction markets do tend to overprice dramatic outcomes, so "always bet no" isn't as dumb as it sounds. Would love to see real P&L over a few months, not just the thesis.
I've backtested this kind of strategy, and it had a good return (like 100% APR), but then I realized it was cheating by knowing when things are going to resolve. Often times it's not clear. Your return depends a lot on how quickly you can get your money out. I never got around to trying a strat that doesn't know the resolution time, which actually has to be manual cause it takes some judgement to pick things that you expect to resolve soon.
Also requires a lot of volume to be "predictable" obviously, since 1 loss sets you back 10-20 wins. It's surprisingly hard to find reasonable-liquidity markets after all your filtering. Scare quotes around "predictable" because you never know if others will use this strat or a lot of unlikely events will happen due to insiders.
Edit: Another thing, betting on the overdog in sports markets actually looked more appealing because there are plenty of those events with large volume, they're kinda homogenous, you know exactly when they resolve, and they're harder to rig. Just like the author, I was excluding sports at first. I simply never got around to putting real time or money into that.
> One loss sets you back 10-20 wins.
didn't look at the numbers, but this one sentence reminds me of selling options for 'passive income' (don't do that)
I drew the same analogy. You put up $0.95, a YES gambler only puts $0.05 (ignoring spread); you're "lending" the money in case of a YES. Problem is, leaving the money in longer doesn't seem to mean a higher premium for the YES side.
Quintessential hustler logic: inability to compare the gains from wins to inevitable losses.
Don’t be gullible enough to fall for this bad math.
Say 70% of the time it resolves to ‘no’, you still don’t make money by blindly choosing ‘no’.
Guess why?
Hint: This strategy is also described with the macabre analogy: picking up pennies in front of a steamroller.
Do you want to pick up pennies in front of a steamroller?
Related: https://www.jbecker.dev/research/prediction-market-microstru... (previous discussion: https://news.ycombinator.com/item?id=46680515)
Basically arbitraging human imagination. People love coming up with fantastical concepts because they get attention, but the more exciting a market is, the less likely it is to actually happen. Reality is usually boring.
My general observation is that people tend to underestimate the likelihood of black swan events (covid, financial crisis) even as it's pretty obvious they're happening. And then when they do accept it, they react too far the other way and assume it's never going to end.
I've had success playing the markets in these specific cases. I did fritter away a lot of my gains from the financial crisis thinking I was a genius market timer. But I learned my lesson and didn't waver once I jumped back in after covid.
In both cases I got out before a bulk of the crash and timed the bottom almost to the day. Lucky I know, but I had reasons for both. For the financial crisis it was when Bill Fleckenstein closed his bear fund and put it all in MSFT. For covid it was when it looked like the lockdown was working and NYC hospitals weren't going to completely fall over like Northern Italy or Wuhan.
For any non black-swan scenarios, I assume I'll never get one up on the masters of the universe and just leave everything in blended age-appropriate funds.
I'm very concerned about an AI crash and the future of white collar work in general. But it feels more like a slow death to me than a black swan. So I'm just hedging with bonds and cash and stocks that hopefully don't crash as hard in a recession.
I would also not want to take even a fair bet against black swan because the day that "S&P500 falls to lowest level since 2016 as Labubus collapse" is the headline is the exact day I least want to lose a big pile of money gambling. If it's the shareholder's money though.... I'm probably getting laid off in that scenario anyways...
This makes sense to me, but isn't there a risk of increasing the potential payoff high enough that someone is motivated to go out and make the yes side happen?
Consider this bot running on us military outcomes or something.
By design it's a game where people with inside knowledge or enough power to bend reality can steal money from people with gambling addiction. Automating your addiction might not be the best move.
This is what markets like Polymarket boil down to. Normies can't win. Some will, of course, but that's just chance and there's no way if ensuring it's you.
It's really no different than a casino: if you ever find yourself with more money than you walked in with, cash out and leave.
Best strategy for most people though is to simply not participate and you'll break even.
You say that like it's bad thing, but really it's great!
It gives us normies a way to see what the powerful are thinking.
You're thinking like an engineer and making the laughable assumption that "prediction markets" are markets. It's totally unregulated with all sorts of grifts and cheats. One of the platforms was promoting a high-return bet against Rory at the Masters yesterday.
You can make money off of all sorts of stuff. You can "sell" the bets, so there's lots of live pump and dump.
We've gone full circle. The bookie with no neck that smelled like onions was more honest than these platforms.
Well, that's why things aren't priced uniformly, isn't it?
Basically, realized vol is lower than implied vol over time. Yes.
also people overpay for skew protection and you can make consistent money selling skews (until that one time it blows up on you)
Already priced in.
The author also noted:
> yes this has to buy below 0.73 long term, the bot has a configurable ceiling set at 0.65 and checks for new markets buying closer to .5
https://x.com/sterlingcrispin/status/2043685362812461436
What happens if you flood the market with a bunch of implausible bets like "sun won't rise tomorrow"? Sure, you might try to filter that out with some sort of "seasoning" period (ie. don't buy new markets), but then that means more time for arbitrageurs to correctly price the market, depriving you of any price advantage you might have had.
> What happens if you flood the market with a bunch of implausible bets like "sun won't rise tomorrow"?
Who does "you" refer to in this sentence? Polymarket itself?
I'm pretty sure if Polymarket itself decides it wants to screw you, you're gonna lose no natter what your strategy is.
jane street is always hiring!
For this question I'm working on https://polygains.com
What other question would you like to be backtested? This one is fairly easy
Except that the mere existence of the market with the question posed for people to consider, probably activates the availability heuristic[1], causing people to overestimate the likelihood.
[1] https://philopedia.org/topics/availability-heuristic/
null hypothesis bot
> Heroku Workflow The shell helpers use either an explicit app name argument or HEROKU_APP_NAME.
nice to see heroku still alive...
I love heroku, death to vercel
any stats on your returns so far?
Turkey reported high winrate until Thanksgiving
Falling victim to the classic fallacy. So sad
We call it a "black turkey event", nobody saw it coming.
Not my project, but author said on X:
> Why predict the future when 73.4% of all Polymarkets resolve as No?
https://x.com/sterlingcrispin/status/2043398710013595857
That logic doesn't work because not every bet have even payouts. If there's a market for whether a dice rolls 1 or not, the odds might resolve to "no" 83% of the time, but if it only pays you $1.1 per dollar wagered on "no", you're still losing money.
I think we've collectively DDoSed it. I'm getting a 504 timeout.
The author [page](https://github.com/sterlingcrispin) is there on github, but I can't even find his full list of his repos to confirm it's still there (I also get a 504 on that).
Back up
GitHub is down yet again. Guess they forgot to tell their AI “make no mistakes” while vibecoding.
I was about to say “first Microsoft service to reach zero 9s of uptime”, but then I realized, it’s Microsoft… GitHub is definitely not the first
Honest question: Why in all hell would you open source this?
I have been making money with a bot off a statistical anomaly in prediction markets lately. There is no way in hell I will open source it or tell you what that anomaly is because I have capacity back-tested it and there are so many players in the market; if all of HN and Github start downloading and use my code it WILL cease to work.
Put another way, your orders are helping move the market and price the market more efficiently; that's the market compensating you for pricing things better. If a thousand people run your strategy, prices will get moved to exactly the point where your strategy stops working. You effectively split that pie with a thousand people.
Because they are doing it for fun?
If this seems interesting for you remember that if you are putting $100 in a 99 to 1 bet you need to win 100 times to get $100 but only need to loose 1 time to loose $100.
And the chance of losing at least once in a 99% sure bet after 100 rounds is around 60%. Even if you reduce to 30 rounds it still is around 30%.
This may seem smart at first glance, but the math doesn't really checks out.
What's the data situation like if you wanted to backtest a model like this? Is it easily accessible?
this is a good dataset
https://huggingface.co/datasets/SII-WANGZJ/Polymarket_data
"A comprehensive dataset of 1.9 billion trading records from Polymarket, processed into multiple analysis-ready formats. Features cleaned data, unified token perspectives, and user-level transformations — ready for market research, behavioral studies, and quantitative analysis."
No, data situation is bad, at least for market making - you need to scrape the orderbook yourself to be able to do any realistic backtesting. And even then, it's hard to know whether other bids at the same price are ahead of you or behind you in the queue.
Disclaimer: I contribute work as a political advisor and don't participate in betting markets as a market participant.
Nevertheless, Polymarket is a very interesting marketplace of sentiments and information, and it can be a very strong leading indicator of huge price movements in "real" markets like the NYSE, in part because it directly measures one factor of sentiment, i.e. whatever the prediction is about. Market sentiment determines market prices on very large and deep markets, too.
In the run-up to the election, when Trump was running against Biden, a betting market was a leading predictor of NASDAQ (a very deep, very liquid index of stocks). I wrote up the findings here: https://medium.com/@rviragh/does-the-stock-market-react-posi...
This indicator was the best one anyone has ever shown for NASDAQ for any signal, period. The signal was so strong it trumped all other signals and variances of any kind. Traders trading with just this signal and no other signal of any kind could have made practically an unlimited amount of money as long as the signal was intact. (Basically, until Biden dropped out.)
I myself didn't place any bet due to my role as a political advisor at the time, but the size of the correlation is still the biggest and most surprising one I've ever seen.