Open interest jumping with aggressive flow as a signal

The idea: a sharp jump in open interest together with a price move shows money entering a small coin, so you should follow it in. Tested across several different months, both sides of the trade came out near zero after costs.

Why it sounded right

Open interest is the sum of positions still open. Unlike turnover, which is half just passing coins from hand to hand, open interest cannot be inflated: if it went up, there are physically more positions. Somebody brought money and left it in the market. It is the most honest trace available from outside.

Now lay the price direction on top. Open interest rising with price — new positions are going long. Rising while price falls — they are going short. Add the imbalance between aggressive buys and aggressive sells and you get a picture showing not only that money arrived but which side it took. On a small coin this should work especially well: there the size of open positions is comparable to the coin itself, so a large player entering has to move price.

Then a noise filter goes on top. One signal can be chance, so you wait for a second on the same coin within half an hour, and you look at open interest relative to the coin's own size — taking only cases where positions are genuinely large against its scale. Plus you throw out situations where holding the position is expensive. Put together it looks like a careful system with a clear cause, not candle-reading. And the first rough checks backed it completely.

How we tested it

We downloaded the full archive of signals over several months, parsed the alerts into fields, and counted the result on both sides — buying after a rise and selling after a fall — exactly the way the author of the signals proposed, including waiting for the second signal and all the filters.

The key move was to count month by month instead of in one pile. The reference result had been produced on a few days inside a single month and looked flawless: it even passed a check that it was not just the market moving. So the question was put this way: does it repeat in other months? The same set of rules was run over several consecutive months, each with its own market character, and the months were compared against each other, not against the average. We also checked whether correcting for the market's overall move rescues anything.

How it ended

Neither side survived. Selling after a fall gives slightly more than half the trades up and a small typical gain — which is break-even after costs, and it is equally dull in every month. Buying after a rise is worse: the typical trade is negative in each month. Its average is sometimes positive, but only when one rare but huge case lands in the sample — and in a falling month the lottery simply loses. Correcting for the market's overall move changes nothing.

The loud numbers the whole thing was built on turned out to be a property of one lucky window. That is the classic trap: a rule fitted to a recent favourable stretch looks perfect on that stretch. And we managed to fool ourselves twice. First on a hand review of a small batch of cases — two conclusions came out of it, one about the buy/sell imbalance and one about the cost of holding, and both flipped on the full sample. Then on a short measurement that looked flawless and even passed an extra check — but did not repeat in the neighbouring months.

What remains is a possible weak lean: after signals where price was falling, the coin keeps sliding slightly more often. But that is right at the cost of trading and is not a strategy on its own. Plus, where these signals fire, trading is cramped anyway — the coins are thin.

What this teaches

• Test across several months of different character before you believe anything. One good month proves nothing, especially on small coins.
• A hand review of a couple of dozen cases will confidently hand you conclusions that flip on the full sample. It is good for a hypothesis, not a decision.
• A short measurement can pass even extra checks and still be a property of the window. Repeatability matters more than care.
• Slightly more than half the trades up with a small typical gain is break-even, not an advantage.
• If the average rests on one rare big case while the typical trade is negative, that is a lottery. In a bad month it loses.
• Before you pay for somebody else's signals, count them yourself. The author almost always shows their best window, without actually lying.