A trend-following strategy in a sideways market is a loss machine. A mean-reversion strategy in a strong trend gives back every gain and then some. This is not a fluke or a backtesting failure — it is a structural feature of how markets work. The signal is not wrong; the market regime is different.
Understanding this is the difference between a strategy that survives three years of live trading and one that dies in the first six months.
A market regime is the broad behavioural state of the market at a given point in time. At the coarsest level, markets are in one of two states:
More granular frameworks add sub-regimes: low volatility vs high volatility, risk-on vs risk-off, trending-up vs trending-down. The key insight is the same: the same signal in the same asset produces radically different outcomes depending on which regime is active.
Most strategy backtests run across a long historical window and report a single aggregate profit factor and win rate. That number is the average across all regimes. It hides something important: most strategies have a regime where they work well and a regime where they bleed.
A trend-following strategy with a three-year backtest PF of 1.31 might look like this on closer inspection:
The aggregate looks acceptable. The regime-split reveals that the edge is entirely concentrated in one type of market, and deploying the same strategy through the wrong regime is actively losing money.
This is one of the core reasons backtests flatter — the historical period happened to contain more of the regime the strategy likes, or the researcher (unconsciously) optimised the parameters on a regime-heavy period.
A regime filter is a rule that gates a strategy's entries to the market conditions where its edge has historically been strongest. Common forms:
None of these filters improve a strategy with no underlying edge. They improve a strategy that has edge in one regime by keeping it out of the regime where it doesn't.
Regime filters can also fail in the wrong direction — and this is a harder lesson. A filter that appears theoretically sound can, in live data, actively hurt the strategy it was designed to protect.
We encountered this directly. One of our trend strategies was running a BTC macro regime gate: it would only trade when BTC's primary trend read as positive. The logic was sound on paper — why take long-biased altcoin trades when the macro trend is down?
In live data, the gate was producing the opposite of its intended effect. The trades it allowed through — during positive-BTC periods — were mostly losing. The trades it blocked — during ambiguous or negative-BTC periods — would have been profitable. The filter was not protecting the strategy; it was inverting its edge.
We ran an A/B analysis: trades with the BTC gate active produced a profit factor near zero. Trades in the same period without the gate — filtered only by a coin-level trend filter — produced a profit factor of 1.45. We removed the BTC macro gate and kept the coin-level filter. The live results improved within the next 14 days.
The lesson: regime filters must be validated on live data, not just designed on backtest logic. A filter that sounds right and tests well historically can still fail in production. Validate, log, and be willing to kill the filter on evidence — the same way you'd kill a strategy.
When evaluating any strategy you might copy or follow, the regime question is one of the most important you can ask. Here's what to look for:
At a desk level, regime awareness is not just about individual strategy filters — it shapes which strategies are active at all. A desk running five strategies simultaneously should ideally have strategies with different regime profiles: some that do well in trending markets, some in ranging markets, some that are deliberately regime-agnostic (macro plays based on structural setups rather than momentum).
This diversification does not eliminate drawdown — strategies can correlate during stress periods — but it reduces the exposure to any single regime turning unfavourable.
The current macro context (BTC below its 50-week EMA, all major assets inside defined watch zones with no breakout in either direction) is a ranging regime at the weekly scale. Strategy selection and position sizing reflect that — the trend strategies run their rules unchanged, but we track the regime context alongside every live trade because it shapes the interpretation of results.
No strategy works in all regimes. The ones that survive long enough to build a real track record either have genuine regime filters validated on live data, or they happen to trade a signal that is robust across regime types — and even then, those periods of underperformance are visible in the data if you look for them.
Before you copy any strategy, ask: what regime does this work in, and what regime are we in now? If the answer is "I don't know," that is useful information — and the honest answer more often than not.
We publish the regime context alongside every strategy in our book — which regimes were live when the trades closed, which filters are active, and the results of filters we removed because the data said to.
The full track record, live and backtest, with methodology audit trail: glasshousedesk.com/lab