Skip to main content

Blog

Market regime tagging: why a real edge can average out to nothing

Most setups only work in one kind of market. Blending trend, chop and news days into one win rate hides that. How to tag regime and split the numbers.

Published by TradeFlow Quantum. Filed under analytics, setups, process.

There is a specific way a good setup can look worthless in a journal. It works in one kind of market and fails in another, you trade it in both, and the summary row averages the two into something barely above breakeven. Nothing in the numbers is wrong. The numbers are answering a question — "how does this setup do overall?" — that has no useful answer, because there is no such thing as overall when the conditions change underneath it.

The fix is not a better indicator or a tighter entry. It is one more field on each trade: what kind of market you were in when you took it. That field is what lets you ask the question that does have an answer — how does this setup do in a trend, and how does it do in chop — and the two answers are usually nothing alike.

Quick answer: record the market regime before you enter, using a small fixed vocabulary you never expand mid-quarter. Then split every setup's numbers by that tag instead of reading the blended row. Expect the split to cut your sample per bucket by a third or more, which is the cost of the finding.

What the averaging actually costs

Take a setup you have traded 80 times, 40 of them on trending days and 40 on range-bound ones. In the trend it averages +0.4R, so those 40 trades are worth 40 × 0.4 = +16R. In the chop it averages −0.3R, so those are 40 × −0.3 = −12R. Add them: +4R over 80 trades, or +0.05R a trade.

Read as one number, +0.05R a trade is a setup not worth the screen time — after costs it is arguably negative. Read as two numbers, it is a setup with a real +0.4R edge attached to a filter you have not applied yet. Same trades, same journal, opposite conclusion. The only difference is whether the regime was recorded at the time.

This is also why "my strategy stopped working" is so often a misdiagnosis. A strategy that was fine for three months and then went flat has frequently not decayed at all — the market spent those three months in the regime it suits, and then stopped. Without the tag there is no way to tell that apart from genuine edge decay, and the two call for opposite responses: one says wait, the other says stop.

Pick three or four regimes and then stop inventing them

The vocabulary matters more than the taxonomy. Any set of labels works if you apply it consistently; no set works if it grows every week, because each new label splits the sample again and the buckets stop being comparable to the ones you recorded in March.

  • Trend — the instrument is making higher highs or lower lows on your reference timeframe and pullbacks are being bought or sold.
  • Range — price is rotating between levels that have held more than once, and moves out of the range keep failing.
  • News — a scheduled release, an earnings reaction, or an unscheduled headline is the dominant driver rather than structure.
  • Unclear — you could not tell before the entry. Recording this honestly is what keeps the other three clean.

The "unclear" bucket is the one people leave out and the one that does the most work. Without it, every ambiguous day gets forced into whichever label feels closest, which is a decision made after the fact and contaminates the very buckets you are trying to read. With it, you get a fourth number that is often the most informative one in the table — a lot of traders find their unclear days are where the money leaks out.

Tag it before the entry or don't bother

A regime tag written after the exit is not a regime tag. By then you know how the trade went, and the label bends toward the outcome: winners get filed under the regime you believe suits the setup, losers under the one that doesn't. The result looks like a strong finding and is entirely circular.

The tag has to be part of the pre-trade decision — written down at the same moment as the thesis, the invalidation level and the size. If it feels hard to name the regime before entering, that is information rather than an obstacle: it usually means the answer is "unclear", which is a legitimate reading and a legitimate reason to pass.

Split the numbers, then check the sample behind them

The reason to split is to compare like with like, and the reason to be careful is that splitting is expensive. A hundred trades across one setup is a thin sample already. The same hundred split three ways is three samples of thirty-odd, and thirty-odd trades will happily show you a twenty-point win-rate gap that is noise.

The scale of that noise is worth internalising once. The 95% confidence interval on a 60% win rate computed from 20 trades runs roughly 40% to 78% — the same 20 trades are consistent with a losing strategy and with an excellent one. A regime split that leaves 20 trades in a bucket has not measured that bucket; it has produced a number with an interval wide enough to contain the opposite conclusion.

Two buckets differing by ten percentage points on thirty trades each is not a finding. Treat an early split as a hypothesis you keep collecting against, and re-read it at 50 and 100 trades per bucket before you change how you trade.

How TradeFlow Quantum handles this

There is no dedicated market-regime field in the app, and that is worth saying before anything else — nothing here classifies the market for you or reads the tape to decide what kind of day it was. What exists is a free-text setup label on every trade (up to 100 characters) and a setup tag on the pre-trade plan (up to 60), and the practical way to carry regime is to encode it in that label as a compound tag: breakout:trend, breakout:range, fade:news. The tag is yours; the app only groups by it.

Once the tags exist, the analytics read them. The cohort view pivots your trades by setup — or by hour of day, day of week, side, psychology score, R-multiple bucket or symbol — and reports P&L, win rate, average R and trade count per cell (average R only over trades that carry one, and a broker sync or CSV import supplies none), with win rate denominated on closed trades rather than on everything in the cell. The compare view is the more direct tool for this particular question: it takes two filter sets (setup, side, date range, psychology range, minimum R) and runs them head to head, which is exactly the shape of "this setup in a trend versus this setup in chop". Until November 10, 2026 this paragraph did not say that the R figures are empty for imported trades.

The setup breakdown page adds one thing the raw pivot does not: it flags setups whose recent win rate has fallen 15 percentage points or more below their all-time figure. That flag is descriptive, not a verdict — a setup can show up there because it decayed, or because the market has been in the wrong regime for it lately, and the regime tag is what tells those two apart.

The honest limits, since free-text tags have sharp edges. trend and Trend are two different buckets, and a typo silently creates a third; nothing normalises them for you. Splitting by a tag you only started recording last month means the comparison is also a comparison of two time periods. And a compound tag is a convention rather than a schema — it works exactly as well as your consistency in typing it.

Honest disqualifier

If you take a handful of trades a month, this will not pay for itself for a long time — three regime buckets against 60 trades a year is 20 trades each, which the interval above says is not yet a measurement. If you trade one instrument in one direction on one timeframe, the regime split may just reproduce your day-of-week or hour-of-day numbers, and those are already available without any extra tagging. And if you have not been recording regime until now, this is not a question you can ask of your existing history — the tag has to be written before the entry, so the first honest answer is a few months out.

Related posts

Not financial advice. This post reflects the author’s opinion based on publicly-available information at the time of writing. Mention of third-party products is not an endorsement; product features and prices change over time. Past performance does not guarantee future results.