We went looking for a bottom signal. We found something else.

Thirteen markets, four hundred thousand candles, nine years of history and rules written down before the first run. None of the factors we started with predicts a trend reversal. One simple principle, however, halves the drawdown — and most of that effect does not come from the indicator.

In October ETH bounced from around 2,300 dollars. Fear was extreme — about 5 out of 100 — and three levels most of us keep on the chart sat right under the low: the monthly Pivot Boss L3, the daily RMA 55, and on bitcoin the MA 55 exactly at the bottom. A beautiful setup. I looked at it and thought what everyone thinks: this could have been seen coming.

So we checked. Not on one bounce, but on thirteen markets, roughly four hundred thousand candles and nine years of history. With rules written down before the run, with a control sample that was off limits, and with one test I recommend to anyone who studies anything: the method must fail to find an effect in random data. Our first version found one. We fixed the random number generator and went back to zero.

What does not work

Extreme fear does not raise the odds of a bounce. After days with Fear & Greed below 10, the market went up exactly as often as on any other day after a decline. The whole apparent effect came from 2024–2025; in 2020 the same rule lost 38%.

Levels do not work. We tested Pivot Boss (L3, L4, L5 — monthly and weekly), ten moving averages I use myself, the ATR lines from our script and the "current support" rule the script draws. A touch of a level, a confluence of three levels, a confirmed support: all within the base rate. Reversals after a "new support" — 36%. Base rate — 36%.

Candle patterns work a little on the S&P 500 and not at all in crypto. 15-minute trend lines precede 92% of bottoms — and 95% of false bottoms, because they precede everything. BTC and ETH are coupled so tightly that 36 of 46 bottoms fall on the same day, yet neither one leads the other.

The only thing that carries information is the state of the market: how far below the high, how far from the average, how volatile. In the best fifth of days the chance of a reversal rises from 33% to 49%. That is not enough for a signal. Nobody should buy because "it's 49%".

What works

Since reversals cannot be predicted, we tested the simplest thing: be in the market while the trend lasts, and get out when it breaks. Our script draws an ATR Trailing Stop for exactly this. Daily close above the stop — you are in. Below — you are out.

It works, but not the way people imagine. In bull years such a rule gives back part of the profit. Its point shows in a bear market: in 2022 bitcoin lost 64%, the rule 0%. Solana −94%, the rule −53%. Which indicator does not matter — Donchian, 50/200 averages, EMA 21/55 all give the same picture. The principle matters.

And here is the catch this text is about. In the "all in or all out" version, that principle still produces 55–80% drawdowns and years under water — ETH: five and a half years without a new equity high. Nobody survives that. Diversification does not help, because these markets move together.

Something else helps: position size scaled to the volatility of the whole portfolio. When the market is turbulent, you hold little. When it is calm — more, but never beyond what the risk budget allows. On a basket of the five major coins in 2024–2026 this construction produced a 22–24% drawdown at 17–22% a year. Holding: a 62% drawdown.

The honest part

Most of that improvement does not come from the signal. It comes from holding a small position. Someone who simply keeps 35% of capital in these coins and 65% in cash gets a 35% drawdown instead of 74% and 25% a year instead of 61% — with no indicator at all. The trend filter adds 5–10 points a year on top and shortens the time under water. We tested it on six coins that played no part in the design: it works on four, not on two. On a basket of altcoins alone the risk control does not hold — 2025 is −34%. On stock indices and gold the same rule is worse than simply holding a small position — this is a crypto phenomenon, not a universal law. What it does not depend on is luck with the start date: of 24 quarterly starts since 2019, all 24 came out better than a constant position, including a start at the very top of 2021.

The parameters are frozen and since 16 October we have been tracking this live. In a quarter I will write up what came out, whatever comes out.

What follows

If you remember one thing from this text: in crypto, scale your position to volatility, not to conviction. That is the only thing we have proven beyond doubt. There are no bottom signals. There are market states in which a bounce is a little more likely, and there is a trend you can follow, paying for it with part of the bull-market profit.

Educational material, not investment advice. Past results do not determine future ones. Method: definitions written before the run, self-test on random data, 2024+ control sample, test on markets unused in the design, 24 start dates. Data: Binance 1D/1H/15m since 2017, Yahoo Finance, alternative.me. Polish original: /edukacja/badanie-odwrocen.