I keep a strategy in a drawer that made a genuinely stupid amount of money over one stretch, then gave a chunk of it back over the next one, without a single line of code changing. Same entries, same exits, same position sizing. The only thing that moved was the market underneath it. That gap used to bother me, because it looks like the strategy broke. It did not break. It was doing exactly what it was built to do. The environment stopped rewarding that behavior and started punishing it.
Once you see strategies this way, a lot of the mystery around blowups goes quiet. A breakout system is not good or bad in the abstract. It is a bet that price, once it clears a level, keeps going. That bet pays when the market trends and it loses when the market chops. Mean reversion is the opposite bet, that a move away from the average snaps back, which pays in a range and gets steamrolled in a trend. Neither one is a permanent edge. Each is an edge on loan from a particular regime.
Why each strategy is chained to its regime
The reason is structural, not psychological, and it comes down to how each system distributes its wins and losses.
Trend and breakout systems have a specific shape to their returns. They lose small and often, then win big and rarely. You get stopped out on a dozen fakeouts, each a small paper cut, and then one clean move pays for all of them and then some. The whole model depends on those occasional large winners existing. In a trend they exist, because moves that start tend to continue, so the tail is fat and the math works. In a range that tail gets chopped off. Price clears the level, you enter, and then it reverses back through your stop because there was never any continuation to catch. Now you have all the small losses and none of the big wins. The exact mechanism that makes the system great, letting winners run while cutting losers fast, becomes a machine for donating money one small loss at a time.
Mean reversion has the mirror-image shape. It wins small and often and loses big and rarely. You fade the stretch away from the average, price snaps back, you book a modest gain, and you do this over and over so the equity curve looks almost boringly smooth. The catch lives in that rare large loss. When a real trend shows up, the thing you are fading does not snap back. It keeps going, and going, and the position you kept adding to because it looked cheap becomes the position that takes the quarter. A smooth curve with a hidden left tail is the signature of a mean-reversion book meeting a trend it refused to respect.
So the same underlying property, whether moves persist or revert, is what makes one system print and the other bleed. When that property flips, both systems flip with it. There is no version of the rules that survives both regimes, because the rules encode a directional bet about persistence, and persistence is exactly the thing the regime decides.
Telling the regimes apart before you deploy
The practical question is never which strategy is better in theory. It is which phase you are standing in right now, before you put size on. You do not need anything exotic to answer that. A few crude measures agree with each other often enough to be useful.
- Efficiency of the move. Take how far price traveled net over a window and divide it by the total distance it actually covered getting there. A trend moves in a fairly straight line, so net distance is a big fraction of total distance. A range covers a lot of ground and ends up nowhere, so that ratio collapses. High ratio, trend. Low ratio, chop.
- Where price sits relative to a longer average. In a trend, price stays stubbornly on one side of a slow moving average and the average itself has a slope. In a range, price crosses back and forth through a flat average constantly. If you cannot tell which side of the average you are on, that is your answer.
- Behavior of new highs and lows. Trends make a sequence of higher highs and higher lows, or the reverse. Ranges make highs and lows that keep landing in roughly the same zone. If the last several swing points are stacked in a band, you are in a range whatever the indicators say.
- Volatility expansion versus contraction. Trends tend to run with volatility expanding in the direction of the move. Persistent contraction, where each swing is smaller than the last, usually means the market is coiling sideways, which is hostile to breakout entries and friendly to fading the edges.
None of these is a magic number and you should not treat any single one as a signal. The point is to have two or three of them roughly agree before you commit. If efficiency is high, price is pinned on one side of a sloping average, and highs keep stepping up, deploy the trend system and put the mean-reversion book in the drawer. If efficiency is low, the average is flat, and swings keep landing in the same band, do the opposite. When they disagree, that is a transition zone, and transitions are where both strategies get hurt worst, so the honest move is smaller size or no size until the picture resolves.
Building the regime check into the process
The failure mode I see most is running one strategy permanently and blaming the market when it stops working. The fix is to make the regime read a required step, not an afterthought. A workable version looks like this. Before any deployment, classify the current phase with two independent measures. If they agree, run the strategy that matches and size it normally. If they conflict, cut size hard or stand down. Re-run the check on a fixed cadence rather than only when something hurts, because the regime rarely announces its turn and you want to be reducing exposure into the shift, not after it has already cost you.
You can also test the whole idea directly instead of trusting your gut. Tag your historical periods by regime using whatever measure you settled on, then look at how each strategy performed inside each bucket. If a breakout system shows a strong trend bucket and a bleeding range bucket, you have confirmed it is regime-bound, which is fine, that is most systems, and now you know exactly when to turn it off. We built a lot of the scorecard and backtesting tooling inside Blockcircle around this kind of question, because a single blended performance number hides the one thing that actually predicts whether tomorrow is friendly, which is the regime you are walking into.
The strategy in my drawer is still there, and I still like it. I just stopped asking it to be good all the time. It has a job, the job has conditions, and most of the edge turned out to be in noticing when those conditions are present and having the discipline to sit on my hands when they are not.