¶1Hello, hope you are doing well. So, today I I have a lot of questions about kind of how I think about setting up my AI systems for looking at, let's say, good prospects or finding good markets on This could be on prediction markets. It can be like on other markets, too. But, basically, the idea here is that we are going to go for some hunting for some expected value. And this is kind of what I've been thinking about lately because when I talk to like my friends and stuff like that, they are always kind of mostly interested in who is going to win the game.
¶2So, in about 3 days, there's a game Arsenal versus Manchester City. And let's say if they want to bet, they are mostly focused on who is going to win. Is it going to be Arsenal? Is it going to be City? And yeah, that's mostly what they talk about, who they're going to place their bet on.
¶3But, if you kind of think if you want to kind of win in the long run, if you want to stop losing in the long run, you got to kind of twist your mind and think about more expected value because this is where you want to be if you kind of want to be profitable over the long run. And this doesn't just apply to sports betting. Basically, everything you do, if you're going to do some automatic trading with AI, we just need that expected value. Other than that, it's just gambling, I guess, right? So, today I thought I'd just do like a simple showcase of how you can try to avoid losing to the to Polymarket, Kalshi, whatever betting platform if you do sports betting or something like that.
¶4I don't really do sports betting, but it's a bit too random for me and I really like the odds. But, you could find a deal there and they're probably just going to ban you if you keep exploiting that. But, let's just do an example here. So, let's say we have this match Arsenal versus Manchester City. Let's go to Polymarket and we just check out here, right?
¶5So, you can see Arsenal 37%, City 36, and a draw is 28. So, let's just type that down. So, Arsenal was 37. 37, whoa. 36.
¶6And draw was like was it 25? 28. So, in percentages on Polymarket, 37% chance to Arsenal to win the game, 36 for City, and we have 28% chance to draw. So, that's a really even game, right? Let's take a look at Calshi.
¶7So, if you look at Calshi, you can see it's 37 for Arsenal, 37, 35, 28. So, that looks a bit strange, right? Because Nah. Doesn't really add up, but that's just the way it is because there are some minor details here, but basically, this is exactly the same, right? So, they are really correlated, the market on Polymarket and the market on Calshi.
¶8They are really correlated. And it's really hard to find any good value between those. But what we can do is actually look at bookmakers, at least for like these type of events. We can check bookmakers, right? Because they are kind of doing the same, but in a bit of a different format.
¶9So, if you want to check like more European style, we have these 2.45, 3.6 for draw, 2.55 for City win. If you check like more American like Arsenal City, we have the plus 150, 235, 160. So, how can we kind of use this now to see if we can find any mispricings in the market? So, this is a very small sample. You kind of want to do this in like a big scale, but just for this video, I think this is really interesting.
¶10Okay, so let's just do some simple statistical math here. So, basically, if you look at Bet365, you can see we have these numbers here, and we can just do like one over all of this, right? So, we just do one over this, and then we end up with something like So, let's just do one over 2.45, and this will give us like let's just round it to 0.41. So, this is like one at 0.41, 0.28, and here we have three nine. Okay, so we have these three You can say this is like percentage.
¶11So, let's just add these together. And this kind of adds up to 108% right? That's really strange. Uh shouldn't it be 100? But this 8% over is kind of the margin for the bookie.
¶12So, we kind of need to normalize this. So, we kind of want to remove uh get this down to like 100%, and then of course we just do 41% divided by our margin here. So, that's 1 8. We end up with 38% on the home win, right? Uh 26 for draw, and 36% for away.
¶13So, these are kind of normalized, and now we can kind of look at the Bet365 bookie odds, and we can compare them here over to Polymarket, right? And then we can kind of see are there any mispricings here? You can see we have 2% uh difference in kind of the the draw here. So, Bet365 thinks that a draw is a little bit less uh possible than actually doing like uh instead of looking at Polymarket. So, that could be like a 2-cent edge there if you want to call it like that, 2% edge.
¶14But, these are like very small edges, but this is just an example of things you can think about when you're kind of looking for a good deal, right? So, let's say for example we found out that for some reason bet365, let's say DraftKings, and a bunch of other bookies or other measures you have for fair value gave this like a really high percent chance something like, I don't know, 45 45% chance of a home win. And that is kind of something you should be aware of. Now, Arsenal looks really cheap, right? Because if the all the bookies thinks that Arsenal will win 45% of each game, and the bookies or Polymarket Calshi only says 37, now we kind of have like an 8% edge here if we believe that the bookies are kind of correct.
¶15And this is a good bet, right? So, we definitely should take that. Of course, you got to think a bit about like the fees on Polymarket and the stuff like that, but just as a good rule of thumb, this is a really good bet now because you can see we have a difference here between win is like 8%. And if the bookies are more calibrated, then these prices are kind of mispriced if we believe that the bookies has the best calculator, best way to have find the right probability. So, that is basically kind of how I think about looking on Polymarket for expected value.
¶16I don't really care about who wins. It doesn't really matter. also in this case, this does not mean that Arsenal will win, right? Uh that doesn't really matter. The only thing we are thinking about is the price we kind of get uh for uh our bet, right?
¶17So, if the bookies think if the bookies thinks that this is like 45% and we can buy in for 37 cents per share, uh then we get a much better deal. And that is the things I'm looking for. And now we kind of come to I am building a bunch of systems that is kind of using um AI, right? To automate this. So, we can look for this autonomously.
¶18So, I thought I can just throw like a super simple system you can build that kind of looks for deals like this. And this can calculate and find this autonomous. So, let's just do a super simple example here on Codex. Uh I'm just going to say the prompt. I want to create a system to check for fair value on a sports game.
¶19Uh the first step for you is to use the web socket on Polymarket and Kalshi. These are free. To get the live prices for an event the user will give you. Then uh when you have found the events, the user will report a series of bookmaker odds. Here you can just place in like an API, right?
¶20Because I don't want to do that for this example, but that would be like the fully autonomous. I'm just going to put in a couple of numbers autonomous. This will uh need to be recalculated and normalized into percentages because we want to compare it to Polymarket like we did and Kalshi. Uh the goal is not to look for prices that the goal is to not look. The goal is to look for prices that are outliers, mispriced that may contain expected value.
¶21So, I'm just going to send this now. And now um our Codex uh agent here is going to check out the web sockets and we're going to give it an event. We're just going to use the the same event we had here. And it's going to try to calculate that as far as based on the bookmaker odds we put in, right? Okay, so now you can see we have kind of set this up, so it's going to ask me I'm going to say the event is Arsenal versus Manchester City Community Shield game money line with home draw and away.
¶22Find the live prices from each of these uh prediction market platforms. Just to get a baseline, and we can go grab some bookmaker odds, but this should should have been like I I based or something like that if I wanted this fully autonomous. Okay, so you can see we kind of grabbed uh home Arsenal. Um you can see we have both here. Calci bid and ask.
¶23Cheapest buy. 37 28 35. That's kind of what we calculated ourselves. And here is kind of the yeah, you can see there are some margins here 101 on Polymarket. That's a bit strange, but uh let's just go grab some um some bookmaker now.
¶24This is kind of very related to sports betting, but I think it's just a good example. This could be stocks or whatever you want. So, let's just grab this from I just found a sports book FanDuel. I don't know so much about that. Uh okay, and let's grab from DraftKings.
¶25Okay, and let's just do an image from three bet 365 or something like that. >> [snorts] >> Um just going to paste the image. So, I'm just going to feed that in. This should have been autonomous autonomous if you really wanted to set this up on these type of markets. We can just do fully like a fully pipeline, and this can keep monitoring this at all times.
¶26Uh we can just double check. Yeah, this looks pretty good. We got a couple of American. We got a decimal. Doesn't really matter because uh Claude code uh now Codex pipe GPT 5.6 is going to recalculate this for us.
¶27And it's going to check like is there any mispricing here? Is there anything that looks like it have some expected value? Because the more data points you can kind of feed this, the more you can kind of infer. Of course, finding the true fair value is not going to happen because there is no true fair value, but you just going to have to trust Do you trust the bookmakers to have the best fair value? Or do you trust the prediction markets?
¶28You probably not want to trust like some guy on Reddit that says, "Yeah, I give it like 60% chance that City will win." Uh because he doesn't have any stake in the game, so I kind of try to look for places, predictions where there are stake in the game. You can also try to build your own models. That is kind of what I'm doing, uh but not so much on sports, but I'm trying to look for markets where I can build hopefully a better fair value model to kind of look at what I think is the fair value, then compare it with the price to try to find expected value. So, this is what I'm using AI for. I'm trying to build a bunch of different systems that autonomous monitor these markets uh to see if there pops up something that is interesting and that is of expected value.
¶29And eventually it's just going to place that trade autonomously based on some parameters we have set up. So, that is kind of what I'm looking for. Uh let's see here now. So, you can see we uh we cleaned up the data. Uh let's see now.
¶30And we kind of got the the results here. You can see conclusion, Man City on Chelsea is the only gross outlier, but it's not actionable plus expected value as a taker. Okay? Because we can kind of see uh 35 and we find like 1.5 percentage points difference. but the verdict is no executable expected value at asks.
¶31Man City on Calci it's worth monitoring, especially at 34c or lower. So, if you go back to the Calci platform and let's say we suddenly see uh Manchester City here kind of dips down to 34, then we kind of get into something interesting here because then the spread is kind of widening a bit more and it could be interesting to place something on Man City at 34 cents. But, basically, this has nothing to do with their the the calculations from this. Uh basically, what I wanted just to talk a bit about today is that it's not really about uh trying to figure out who wins this game. Doesn't really matter too much if you're going to keep playing.
¶32If this is the only time you ever going to place a bet or a prediction, then of course it matters, but if you're going to keep doing this over time to actually try to have like a uh profitable strategy using AI, automatic trading, algo trading systems like that, then you only need to focus focus on the price and what uh the price is looking uh versus what you think is the fair price and you can find some good edges uh with expected value over time. So, yeah, just a bit of a different video. Uh I have some systems in work and I'm going to keep coming back to these in future videos, but this is just like a common misunderstanding I see a lot of people talk about when it comes to all types of platforms. This could even be on like uh platforms like HyperLiquid and stuff like that. Uh sometimes the price uh what you think is the price versus what is the price in the market, if there's a dislocation there and you think there's like 5 cents, $5 uh discrepancy and expected value, then you should probably just take it if you trust your fair value model.
¶33But uh yeah, doesn't really matter only at uh Polymarket, prediction markets, sports betting. It kind of relates to everything you do where you try to predict something that is unpredictable. Uh at least in our case, you can't really 100% predict who's going to win this game. That is impossible, right? Because no one knows until the game's game is finished.
¶34But uh yeah, hope this gave you some more information, and maybe you got a bit more interested in trying to build out systems using Codex, Claude Code, Open Code, LLMs. You can build some really advanced system to try to make this autonomous. That's what I'm doing. I'm going to share more of that. So, if you like this kind of content, leave a like, subscribe, and hopefully I'll see you again very soon.