Scarb Glossary
live
← All sections
Tested — does not work

Shorting the spike, and the hunt for a better exit

The idea: if buying a vertical spike loses money, selling into it should make money, and all that is left is finding the right exit. We swept the whole space of exits and hit a ceiling that is a zero even in the best case imaginable.

Why it sounded right

The opposite of a losing idea usually looks like a winning one, and here there were real grounds for it, not just flipping a minus into a plus. A vertical spike on a small coin is almost always a crowd that arrived late. They buy at any price, with no plan, on emotion. A position like that is unstable by construction: nobody meant to hold, everybody meant to make money fast. All it takes is for the move to stop and the exodus starts on its own.

Add to that a direct confirmation you can see on the minute scale at moments like these: longs really are paying to hold, and paying dearly. So the lopsidedness is not an assumption, it is a fact you can read before entering. Selling into a place where the crowd is long and paying for the privilege is the clearest mechanism on the whole market.

And the first measurements supported it. Selling at the close of the signal minute really does pay a typical plus, really does beat a random entry, and does it not in one window but across two opposite market regimes. So the direction is not imaginary. It felt like one step was left — find a proper exit — because a raw hold by the clock is obviously suboptimal: some coins run further up first and fall later.

How we tested it

First on minute history across many months, with a random minute in the same coins and days as the control. Then — and this matters more — the rule was run live on paper: every alert it sent opened a trade, and the trade was managed to an exit. The point is that a run over history answers "what would the rule have done", while a paper run answers "what did the rule actually send". Those are different questions, and they had already diverged before.

For each trade we recorded not only entry and exit but the full minute-by-minute path of price and the times the extremes happened. That let us sweep exits honestly: any exit rule — by time, by target, by stop, by trailing the stop — was tested against the real path of the trade instead of being picked from memory. Separately we ran the rule over a quarter with the opposite market regime, because the original measurement was made in a rising market and we needed to know whether this was a property of the rule or a property of the moment.

How it ended

The live run came out a zero. The typical trade is up, more than half the trades are up — and the average is still indistinguishable from zero, because one candle on one coin eats dozens of winners. And the clustering is heavy: the trades pile up on a small number of coins, with a single coin supplying both a dozen entries and both of the main losses.

The exit sweep left no free knobs. A target breaks the rule: by clipping the winners you cut off exactly what everything rests on — and the result goes negative even though almost every trade closes up. Holding by the clock gives nothing better. A stop looked like gold: the tighter it is the better the result, in a clean staircase, across a wide plateau of settings, and identically on both halves of the sample. That is exactly what a real finding looks like.

And that is exactly where it fell apart. Count the exit not at the stop level's price but at the price of the minute when the break of the stop became visible, and the sign flips across the entire grid of settings. A coin that has just gone vertical goes straight through the stop: the slip past the level is about as large as the whole idea was worth. The truth lies between those two boundaries, and minute data cannot narrow it.

The last nail is the ceiling. Even if you assume an unreachable fill at the signal's own price, with no cost of getting in, the result is still indistinguishable from zero. The best case imaginable is already empty, so there is no point sharpening it with better data. The direction is real, the size is zero. On top of that, the test in a falling market showed that the ratio of how much you get to take against how much you have to sit through compresses to nothing once trading costs are paid, and the share of trades that go deep against you doubles.

What this teaches

More in this section

A long shelf, then a break upward
The idea: price sits in a well-worn range for months, and leaving it means a run to the next such range. When we tested it, breaking out of the shelf paid exactly as much as a random day on the same coin.
Big orders in the book as a reference point
The idea: a big order in the book is a level price will bounce off, or a level it will aim for. When we tested it, price reached such a wall no more often than it reached a randomly picked level at the same distance.
Buying into a vertical spike
The idea: the coin has gone vertical, so a move has started and you should get on board while it runs. When we tested it, buying the spike was not a zero but a loss: far fewer trades finished up than with a random entry.