I lost 122 times on Polymarket so you don't have to.

I expected the numbers to be inconclusive

340 trades settled between 21 July and 12 August. 218 won, 122 lost.

When I saw the 340 trades, I thought they would still be inconclusive, maybe at best break even. But it turns out that I actually have a win rate of 64.1%, and I win more often than just flipping a coin. I still can’t mathematically verify that I’m reliably making money.

Both halves of that came out of the same 340 trades.

What happens to your money in a five minute window

The way these binary options work is that, let’s say, you put in 1 USD. You buy at 55%, which tells you, or the market tells you, that your odds of winning sit at 55%. If it goes up, you roughly double your money. If you lose, you lose whatever you put in, so in that case it would be 1 USD.

Obviously, I noticed every time I was on the wrong side, which is quite painful because whatever you invested goes straight down to zero. Later I also observed that I crawled back to the winning side and managed to offset it over a longer time period, but every loss still stings a bit.

The fees move the line before you start

Polymarket publishes its fee formula. The fee is the number of shares you buy, times a rate, times the price, times one minus the price.

Crypto markets carry a rate of 0.07, and only the taker pays it. Whoever leaves a resting order pays nothing.

Work it through on a single dollar. At 55 cents that dollar buys about 1.8 shares, and the fee comes to 3.15 cents. So the price you actually paid is 56.7 cents, not 55. That moves your breakeven from 55% to 56.7% before you have made a single call.

Two tests, two answers

The first test asks whether I beat a coin.

So if you flip a coin and it’s a fair coin, on average you end up on either side about 50% of the flips. In my case, if I flip my coin 340 times and I get 218 heads, it means that the coin is overweighting the heads side. Statistically speaking, it is not just a good run anymore. We’re looking at a biased coin.

That test scores 5.2. Anything past about 2 is treated as decisive. Not luck.

The second test asks whether I make money, and it runs on the same 340 trades. It scores 1.02. It also needs about 2. So it says nothing at all.

The coin flip thing is just an analogy, and you can tell fairly fast whether a coin is fair or not. To say for sure whether this makes money, we need more trades to settle that question.

The two disagree because a win and a loss are not the same shape. My typical winning trade returns 32.5%. Every losing trade returns exactly -100%, because the stake goes to zero. A typical win doesn’t cover a loss, so the profit rests on the rare big winners. That is what makes the money test slow.

The single best trade returned many times its stake. I thought that it’s for sure an outlier, and I didn’t trust it. So far it turns out that it was pretty much an outlier when I was on the right side, which makes me happy.

The breakeven point could be at 70. In other words, you would still be losing money.

Congratulations on the 64%. Now where is your breakeven?

I would congratulate you on the win rate, because that’s a feat in itself. However, I would ask a follow-up question. The question would be regarding where the breakeven point lies, because the breakeven point could be at 70, which, in other words, means you would still be losing money.

My average loss is 1.5 times my average win, which puts my breakeven at 60.1%. I’m at 64.1%.

The rest of the scoreboard: up 7.1% on the money staked, and the worst drawdown along the way ran to 78% of the final profit.

What I am not publishing, and why

The reason why I’m publishing the metrics behind the things is that I have this yardstick to measure my success in a statistically rigorous way. Without that knowledge, it would be hard to tell whether this entire endeavour was profitable or not.

The actual parameters are proprietary and confidential.

One more caveat, named rather than hidden: those 340 trades pool several strategy variants. The 64.1% describes the mix, not any single one of them.

The takeaways

1. Find your breakeven before you celebrate your win rate. Once we outlined the stats behind it, we could easily work out where the breakeven point is. Be aware of that point, then try to get the win rate above it, which in other words would mean the expected value is positive.

2. A win rate and a profit are two separate questions. The same trades answered one of them decisively and left the other wide open.

3. Count your trades before you trust your number. 340 settled the coin question. It takes roughly 1,300 to settle the money question, and I’m not sure we manage to have enough trades by 31 August.

4. Publish the yardstick, not the machine. The maths has to be public for the result to mean anything. Nothing else does.

What comes next

I still don’t know whether this is luck or actually a profitable strategy. This is why I want to have more trades and a statistical backbone behind them, so the outcome is statistically significant and tells me whether it’s profitable or not.

There is a way to make any strategy look profitable. Banks sell it as QIS.

Which of your wins was actually pure luck?