15 Wins in a Row: Luck for a Real Opportunity, a Guarantee for a Trap

"Arbitrage" sounds sophisticated. Strip it down and it just means earning a spread: the same thing is cheap here and pricier there, so you buy low, sell high, and keep the bit in the middle.

Tech people have a soft spot for it. It doesn't rely on smooth talk or recruiting people under you. It looks like something you can calculate and verify: test with a little money first, scale up once it works. Flipping used graphics cards, moving goods between platforms, "stable yields" advertised at 20% a year — they all run on that same idea.

I keep a small automated trading script running on the side. Over the years, the moments that make me most uneasy are not when it loses. They're when it keeps winning. A winning streak makes you relax. It makes you feel the thing has already been verified.

A couple of days after the National Day holiday, I turned the question "what does a winning streak actually prove?" into twenty-odd lines of Python. When the first number came out I froze for a second, sure I'd written the formula backwards.

The setup is simple.

A real opportunity: each time you take it, there's a 97% chance you make 1% and a 3% chance the spread suddenly collapses and you lose 25%. Over the long run its expected value is positive, roughly +0.22% per round. It genuinely makes money.

A trap: it lets you win every single time, until you feel safe enough to place a big order, and then it closes the net. Whoever set it up is after that last trade. The small wins before it are just their customer-acquisition cost.

Now assume that out of ten opportunities that look about the same, two are traps.

You've just won 15 times in a row. The chance of a real opportunity doing that is 0.97 to the 15th power: 63.3%. The chance of a trap doing it is 100%. That's exactly what it was built to do.

Put the two side by side, and after 15 straight wins, the odds that what you're holding is a trap have gone up, from 20% at the start to 28.3%. The formula is 0.2 ÷ (0.2 + 0.8 × 0.633). Anyone can punch it into a calculator.

That's where I froze. We're used to treating a streak as evidence: the more you win, the safer you feel. But as long as "a real opportunity loses now and then, and a trap never lets you lose," a winning streak can only push the suspicion of a trap up, never down.

The cleaner the wins, the harder you should look.

The intuitive fix: watch it a while longer, verify more, and only then go big.

The trouble is that this road costs you at both ends.

Go in early, and you catch both the traps and the real ones, with the highest odds of getting scammed. Go in late, and the trap can wait just as well. It closes the net the moment you place your big order, whether you watched it 5 times or 50. The real opportunity, on the other hand, won't wait. The fattest part of the spread is usually eaten by someone else while you're still watching with a small position.

So extra verification does almost nothing against a trap, and against a real opportunity it is pure cost. There's no optimal point you can calculate between early and late. The people who set traps don't count on your greed. They count on the fact that at some point, you have to act.

Worse, a streak quietly changes your position size.

I ran another batch. Same real opportunity, no traps. One group always bets 30% of its capital. The other starts at 30%, adds 10% after every win, goes 2x leveraged after 10 straight wins, and drops back to 30% after any loss. 100,000 runs each, 60 rounds per run.

The group that sized up with its streak does look better on average: 1.24x versus 1.04x. But out of 100,000 runs, 34.7% ended down more than 30%. For the group that stayed at 30%, that figure was 0.2%.

The average looks pretty because the winners win big. The losing third got pushed into heavy positions, one step at a time, by the streak that came before.

The plainest version of the math: win 15 times at full size and your capital reaches 1.16x. Then hit one collapse at 2x leverage, lose half, and you're left with 0.58. Fifteen rounds of "experience," handed back in one go, plus another 40-odd percent on top.

Outside the simulation, there's a ready-made, full-size version of this.

In March 2021, a deposit protocol called Anchor launched in the crypto world. You deposited a stablecoin called UST (a digital currency that claimed to always be worth exactly 1 US dollar), and it paid about 19.5% a year.

That rate held for over a year, paid on time, and never made depositors lose. By the time things broke in May 2022, the UST sitting in Anchor made up nearly 75% of UST's entire market cap.

On May 9, 2022, UST clearly lost its peg and spent the next few days swinging between $0.95 and $0.25. Common estimates put the combined market value wiped out across UST and its sister token LUNA at around $40 billion.

Looking back, the painful part isn't that it collapsed. It's that for the fourteen months before the collapse, the "evidence" it produced every month looked exactly like a healthy high-yield product. There was not a single "loss" anywhere that would have told people to get off early.

A real opportunity usually makes you lose a little a few times. Those small losses are how it tells you roughly what its risk looks like and how big it is. A return that has never once made you lose hasn't gotten rid of its risk. It has just moved it to the very last trade.

Next time a spread hands you win after win, there's one small thing you can do first: go find its record of losses. If you can't find one, place your order as if it were a trap. Before you hit Enter, write down "if this whole trade goes to zero, how much do I have left?" If you can live with that number, then go ahead.