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.

Why it sounded right

A spike is when a coin adds in a few minutes what it normally covers in a week. The candle is vertical, turnover is several times normal, the ticker shows up on every gainers list. A move like that is not an accident: somebody is doing it, and doing it fast.

Then comes the momentum logic. If big money went into the coin inside a minute, that money could not have gone in all at once — so the buying will continue. If the ticker landed on the gainers lists, thousands of people will see it, and they will bring a second wave. If somebody was not afraid to push price up aggressively, they have a reason you do not have. All three arguments are independent and all three point the same way: buy.

Watching it also backs this up. Moves like that almost always continue — by eye. And they really do continue: if you look not at the result but at the high, then after a spike half the coins manage to push noticeably further up. So the feeling "it kept going" is not imaginary. The mistake is elsewhere: seeing the high and taking it are different things.

How we tested it

We took the live spike-detection rule as it is, with no tweaking of thresholds, and ran it over several months of minute data across the whole universe of coins. We opened a paper buy at the close of the exact minute the signal fired, and held for a fixed time: fifteen minutes, half an hour, an hour, two hours. Repeat signals on the same coin within a few hours were collapsed into one case — otherwise a single wild coin would hand us three "independent" trades in four minutes and skew everything.

The control is a random minute in the same coins on the same days. So we compared not against zero but against "you just entered this same coin on this same day, with no signal". That matters: on days when a coin is churning, a random entry pays something on its own, and without that comparison it is easy to declare the rule a winner. We also split the sample into two halves by time and into four groups by spike strength, to see whether the conclusion holds everywhere or lives in one window.

How it ended

This is not "no advantage", this is a loss. The typical trade closes down on every hold length, and the share of trades that finished up is well below a random entry on the same days. A useful technique came out of this: the average showed nothing — the spread is enormous, and you cannot talk about it. But the share of trades that finished up showed a difference too large to be chance. On wide-spread samples, look at the share, not the average.

The conclusion holds up under every check. Both halves of the window have the same sign. All four spike-strength groups are negative — so "buy the most vertical ones" is not better, it is worse. Within half an hour a sizeable part of the trades go deep against the entry, and there are more of those than there are trades with a comparable gain.

The only thing arguing for buying: about half the coins do manage to push up further after the signal. But it does not last until the trade closes. The high exists, you cannot take it — it comes and goes inside those same minutes.

What this teaches

• "It usually keeps going" is a memory of the high, not of the result. The result is counted at the exit price.
• The check "what would a random entry in the same coin on the same day have paid" is what separates a rule from simply landing on a busy day.
• When the spread is enormous, the average says nothing and the share of trades that finished up says a lot. Look at both.
• Repeat signals on the same coin have to be collapsed. Otherwise one coin will act out a whole statistic for you.
• If a rule is worse than a random entry, that is useful knowledge: it tells you which direction to be looking in at all.