Path to Ruin or Reward? Analysing Martingale Tendencies on Polymarket
There's a couple things you might want to know to better understand the content of this dashboard.
This dashboard only analyses gamblers who started trading on Polymarket between 3 months ago and 2 weeks ago. ⁽¹⁾ 📅
This dashboard only analyses a subset of those gamblers with at least 5 and at most 2000 trades on Polymarket. Accounts with over 2000 trades over the last 3 months are considered market makers and are not relevant to our analysis. ❌
PnL charts and ROI figures are derived from all-time realised PnL data. It does not track unrealised PnL (gains or losses that the gambler has not materialised/sold yet). ⁽²⁾ 📈
The theory that inspired this dashboard is called the Martingale betting system, which I shortly describe below, and I highly recommend not to skip it if you want to make the most out of this dash! 🎰
⁽¹⁾ Accounts less than 2 weeks old are considered too new to have a real strategy, and still at the discovery stage. My code might have accidentally included accounts that aren't actually new, and were already active over 6 months ago, then stopped and came back during the last 3 months.
⁽²⁾ The PnL of a gambler is calculated as the sum of balance changes on share-buying and -selling events, and incoming payout events upon market resolve.
"I ain't reading all that!" sure, but you'll miss out. Go straight to the Wojak map below to understand what the quadrants of the heatmaps correspond to.
I'm not sure if anyone has tried this approach to studying Martingale tendencies before, so I hope a few of you at least will understand my gibberish (lol). This section was made possible by the new heatmap chart type from Flipside, so shout out to them for that.
Below you'll find 6 heatmaps, 2 for each gambler profile. The ones on the left are ROI heatmaps (the values they contain are the average ROIs for each square or (x,y) bin), and the ones on the right are gambler concentration heatmaps (the values are the number of gamblers for each (x,y) bin).
Why is there a need for 2 heatmaps for each profile? To gauge the significance of the average ROI values in the left heatmap, and also for the sake of curiosity. An average ROI from a sample of 5 gamblers will be much less significant than an average ROI from a sample of 300 gamblers. That way, we know which parts of the ROI heatmap we must take with a grain of salt, ya dig?
Now onto the x and y axes, what do they represent?
The x axis is the correlation between a gambler's PnL and the size of their bets, when the PnL goes up.
The y axis is also the correlation between a gambler's PnL and the size of their bets, but when the PnL goes down.
So why split the data series that way? Because remember, a gambler shows Martingale tendencies when they increase their bet size (as a % of their portfolio) when they win, and don't cut it or even increase it more when they lose. So we need to isolate the cases where they win and lose, and study them separately.
Here are a few examples:
x = 1 and y = 1: the gambler bets more when they win and less when they lose, every time.
x = 1 and y = 0: the gambler bets more when they win every time, and keeps betting the same when they lose.
x = 1 and y = -1: the gambler bets more when they win and more they lose, every time (the absolute degen).
x = -0.4 and y = 0.3: the gambler tends to bet less when they win, and also less when they lose: a more conservative and wise gambler.
So the takeaway is that 1 and -1 are extreme values showing a systematic behaviour. Between the two, it's more of a tendency to behave like so (not every time).

As you can see on the left heatmap, the conservative (top left) and emotional (top right) fellas have the lowest ROIs.
The biggest cluster of gamblers is at the bottom left of the gambler heatmap, meaning there is a tendency among weekly gamblers to bet less when they win, and systematically more when they lose, akin to revenge trading.
The daily gambler ROI heatmap looks similar to that of weekly gamblers. Both quadrants (conservative and emotional) share one common trait: they can be unsure of their betting decisions (feeling too confident or too unsure about gains, and too stressed about losses) hinting at the absence of a real strategy, which could be the reason behind these red clusters. Once again it's good to remember that some of the red squares represent a very small number of gamblers.
On the right we notice two clusters, with similar positions to the weekly gambler clusters.
The HF gamblers heatmaps do look a bit different from the others.
On the left we have only one red cluster in the emotional quadrant but not at the extremes, and a faint green cluster (if we can even call that a cluster) in the contrarian quadrant slightly off the centre, which is not very significant.
The overwhelming majority of HF gamblers trade in the emotional quadrant, but away from the extremes. We notice that the bottom left cluster present in the two other groups' heatmaps has completely disappeared, suggesting that our HF bettooors keep their head cool when their strategy isn't working out. The cluster starts at the centre and stretches out half-way to the top right corner, showing that the vast majority of them either stick to a strict %, or know when to scale up/down their bet size according to the situation. They're overall sharper betters.
Across the 3 profiles that we studied, we did not find clusters in the "Martingale zone" (high x, and y in the middle/lower part). But, gamblers with high x and moderate-to-low y mostly showed negative ROIs, or at most ~0.
How should you trade on Polymarket based on these results? Unfortunately we did not find significant green clusters, but now you know what kind of behaviour to avoid based on the red clusters.
Stay in the centre of the map - keep betting roughly the same % of your portfolio, and don't feel too confident when winning: this dashboard proved it to you with on-chain facts.
After filtering the gamblers following the criteria described in the intro, we break them down into 4 categories:
one-off visitors (between 5 and 17 trades): on one hand, they have the lowest average ROI of all, but a similar median ROI to the others. This means there's a higher % of ruined or almost-ruined gamblers than in other categories, which is dragging the average down. On the other hand, they have a relatively high % of profitable gamblers, so overall their ROI standard deviation must be higher than the other groups'.
weekly gamblers (between 18 and 92 trades): the group with the smallest % of profitable gamblers. They trade more often than the one-off visitors, but don't seem to have a more profitable strategy overall.
daily gamblers (between 93 and 460 trades): despite trading on a daily basis and thus following prediction markets more closely than the majority of gamblers, they still have less profitable gamblers (%) than one-off visitors. Their ROI distribution is the closest of all to a normal distro (average and median almost equal to each other).
high-frequency (HF) gamblers (between 461 and 2000 trades): despite having one of the worst average ROIs, they have the highest proportion of profitable gamblers. So while the majority of them has been losing, a subset of them has been consistently profitable with an actual +EV strategy (we can infer that since they trade a lot - the law of large numbers says that your PnL converges towards the EV of your strategy the more you trade).
From here on, we will only study the weekly, daily and HF gamblers.
Notes:
Overall, the ROI distributions are highly similar, with a few profitable weekly gamblers scoring a very high ROI which we do not see in the other two groups.
The x axis stops at 10 but there are a few outliers in every graph with higher ROIs (negligible). Notice that in all 3 graphs, the vast majority of profitable gamblers have an ROI below 1, more so below 0.75, which means they have not even doubled their money yet.
The y axis has a log scale.
ROI = 0 means the gambler has not made or lost any money down the line (PnL = $0). ROI = -1 means the gambler is ruined (they lost everything they bet). ROI = 1 means they doubled their money, and so on.
A quick look at the total number of markets traded by account shows that 6 is the most popular number of markets to trade. The craziest gambler of all has traded over 800 markets within the time span of this analysis (not shown on this graph).
This figure is very similar to popular estimates that you'll find of profitable gamblers on Polymarket, indicating that our code is thankfully doing the right thing.

Meet Wojak, a humble degen who has just discovered Polymarket and can't wait to start making money on it. Wojak is confident that he can succeed trading prediction markets because of a certain strategy that has already earned him a good amount trading memecoins. If you're a real degen, you're certainly going to relate to Wojak's situation on the right ➡️
Unfortunately for Wojak, this time on Polymarket, his strategy did not work out as intended and he ended up broke. What went wrong? Wojak kept aggressively betting a higher % of his portfolio every time he won, and did not cut that % enough when he lost. So even though Wojak had an above-average win rate of 60% (3 wins, 2 losses), he still lost all his money.
In this dashboard, we are going to figure out whether or not Polymarket gamblers (or a subset of them) tend to show that kind of behaviour, which could be the reason why so many of them have been losing money.

One day, I was scrolling on Twitter and stumbled upon a Polymarket infographic saying that almost 90% of Polymarket gamblers had been losing money - an overwhelming number of them had lost a small amount only. After seeing that, I started asking myself many questions because while it showed an important figure about Polymarket gamblers, I felt like this infographic alone was far from enough to get a good understanding of how gamblers bet over there, and why so few are profitable.
