Market Regimes: Trend Following Lost to Buy and Hold Every Up Year

Written by chwoo · Crypto trader, 7 years. Running an automated futures bot since spring 2026. · 2026-07-20 · updated 2026-09-07

Every trader eventually lives through the same mystery: a strategy that printed money for months suddenly bleeds for months, with no change in the rules. The usual explanation is not that the strategy "broke." It is that the market changed shape underneath it. Those shapes are called regimes, and understanding them explains more backtest results than any indicator setting.

This guide tests that claim with the site's engine. The default period windows (7 years, 5 years, 3 years and so on) are overlapping trailing spans, which makes them poor for comparing regimes, so we split by calendar year instead and computed each year as if it started fresh with $10,000.

The three shapes of a crypto market

• Trend: price grinds in one direction with only shallow pullbacks. Breakouts follow through, momentum entries are rewarded, and mean-reversion shorts get run over.
• Range: price oscillates between support and resistance. Fading the extremes works, and breakout entries buy the ceiling and sell the floor.
• Chop: violent, directionless noise, common in crypto after crashes or during thin liquidity. Almost nothing works, and the honest response is usually smaller size or no trading.

Split by year, and actually measured

We took one trend follower and one mean-reversion system and ran them year by year from 2019. The trend follower is Supertrend momentum trading both directions on BTC 4-hour candles; the mean-reversion system fades the lower Bollinger band on BTC 1-hour candles. Position size 90%, 1x leverage, 0.06% fee and 0.02% slippage per side. The left column is BTC's own return, so the regime of each year is an observation rather than an assertion.

Open the strategy this guide measures in the backtester →

The regime claim broadly holds. Trend following is profitable in all five years BTC rose, and loses in two of the three years it fell. Mean reversion is profitable in only two of eight years. The advice to know which bet you are making survives contact with the data.

But something unexpected turned up

Read the BTC column beside the trend-following column. Trend following underperformed simply holding in all five up years. In 2020 BTC returned +302% and the strategy returned +37.7%. In 2024 it was +121.3% against +40.7%. A strategy built to ride trends fell furthest behind in the largest trends.

The reason is structural. This system reverses on every flip signal, so it exits and re-enters several times even inside a large advance. Stops get hit in the pullbacks between, and re-entry happens at a higher price. Forty-five to fifty trades a year is the cost of that round trip.

So where is this strategy's value? 2022. BTC returned -64.2% and the strategy returned -27.0%. And in 2025 BTC was -6.3% while the strategy was +1.7%, opposite signs. The conclusion this table supports is that trend following exists not to earn more in a rising market but to lose less in a falling one. That is real value, but it is not the value usually advertised.

Cross-checked with a second trend follower

One sample is not a conclusion, so we repeated it with a different trend system: the EMA 50/200 golden cross on BTC 4-hour candles.

The pattern repeats. Profitable in all five up years, behind buy-and-hold in all five, and far more durable in the down years: -5.6% in 2022 and +17.8% in 2025.

This table has to be read together with its trade count. Three to seven trades a year. The +17.8% in 2025 rests on seven trades, so one or two different outcomes would flip its sign. Compare that with 134 to 478 trades a year for mean reversion in the previous table: the same single row carries a completely different amount of weight.

Counting the signs

Something honest has to be said about mean reversion here. The textbook says it works in ranges, but in this table it loses in most up years and in every down year. The likely cause is frequency: 134 to 478 trades a year. Multiply a 0.16% round trip by that many trades and the strategy struggles to clear its costs even when the regime cooperates. Regime alone does not explain everything.

Reading regimes out of a backtest

You do not need a regime-detection model. Run a strategy over seven years and look at the shape of the equity curve rather than its final value. Trend-following systems usually look like a staircase: flat grinding stretches where chop nibbles away small losses, punctuated by steep climbs where a trend pays for everything. Mean-reversion systems show the mirror image, steady gains interrupted by a cliff when a real trend refuses to revert.

Then cross-check against the period table. As in the tables above, the trending years and the falling years will disagree loudly, and that disagreement tells you which bet you are holding.

What to do about it

Three honest options. Filter: add a higher-timeframe trend or ADX condition so the strategy only trades in its preferred regime. Expect fewer trades and a smoother curve. Diversify: run a trend system and a reversion system side by side, each sized so neither can sink the account; when one earns, the other waits. Accept: keep the strategy as it is and size it knowing the bad regime will certainly return.

What does not work is re-optimising parameters every time the market shifts. That is overfitting in slow motion, and it is always one regime behind.

And do not dodge the question these tables raise. If a strategy cannot beat simply holding in an up year, why trade it? If the answer is "to lose less in a downturn", that is legitimate. If the answer is "to earn more", these tables say the answer is wrong.