Market Regime

The prevailing character of a market over a given period: trending up, trending down, choppy (sideways), or volatile.

In plain English

A "regime" is just a label for what the market is doing right now. The same strategy can be a winner in one regime and a loser in another. Most retail traders fail because they apply a strategy in the wrong regime.

The four common regimes:

• Bull trend — sustained upward direction. Great for long trend following.
• Bear trend — sustained downward direction. Great for short trend-following.
• Chop / range — price oscillates around a level. Kills trend-followers (constant whipsaw), helps mean-reverters.
• High-vol expansion — wide swings in both directions. Dangerous for everything; favors breakout systems with wide stops.

Why it matters for this fleet

The backtest window (2020-09-14 → 2026-06-16, ~5.75 years) was one exceptional bull regime — a sustained rise across all three symbols. Buy-and-hold (just buying and holding the asset for the whole window) returned BTC +509%, ETH +392%, and SOL +2405% over that span.

That single fact explains the most lopsided number in the whole dossier. At 2× leverage (a position twice the size of the cash backing it), long strategies were 43% profitable while shorts were only 3% (a long profits when price rises; a short profits when price falls). When the market only goes up, the side betting on "up" wins and the side betting on "down" loses.

This is window-dependent, not a forward-looking claim. It does not mean longs are better than shorts in general. It means this particular window was the best possible regime for long trend-following — which is exactly what most of the fleet is. In a chop or bear regime the same fleet would likely look very different, probably worse.

This is the single biggest caveat on every metric in the dossier: the regime favored the strategy class.

Examples from the live fleet

• Long vs short at 2× (the regime fingerprint). Across the fleet at 2× leverage, 43% of long variants were profitable versus just 3% of shorts. That 14-to-1 gap is not skill — it is the bull regime showing up in the numbers. Read it as "this window rewarded buying," not "buying is a strategy."
• id628 — EMA 9/21 · BTC · 1m · 2× · short is the clearest casualty. It is a fast scalp (frequent in-and-out trades) on the noisy 1-minute candle, betting on price falling, run inside a market that mostly rose. Over N=10,574 trades its win rate (the share of trades that closed in profit) was just 10.6%, its worst losing streak hit 145 trades, and its max drawdown (the deepest peak-to-trough equity drop) reached −98%. A short scalp in a bull regime is fighting the regime on every candle.

How to defend against regime risk

• Out-of-sample testing. Run the same strategy on a different period (e.g. 2021–2022 included both a bull and a bear). If edge holds, regime-robustness is plausible.
• Diversify by direction. Combine long-only and short-only strategies so one of them is always aligned with the prevailing trend.
• Regime filters. Add a high-timeframe trend filter (e.g. only take long signals if the 1-week trend is up). Several "Robust" variants in the fleet attempt this with ADX.
• Watch out for regime change in live deployment. A strategy's metrics from backtest assume the regime continues; the moment it changes, performance can invert.

Refined 2026-06-22 — regime sets the benchmark bar, not just the strategy's odds

Regime doesn't only decide whether a strategy wins — it decides how high the buy-and-hold bar is. This surfaced when the fleet was switched to the long Binance window (~2021 → 2026) to maximize sample size, and suddenly ~97% of variants "lost to buy-and-hold."

The trap: "more data" and "fair benchmark" are two separate goals. Extending the window backward to raise sample size (more trades, better significance — a real win) had a side effect: it loaded the benchmark with the single largest beta (whole-market) run in these assets' history. The hold bar is regime-specific:

• Window 2021 → 2026: buy-and-hold returns BTC +509% / ETH +392% / SOL +2405%. Almost nothing beats it.
• Window Nov-2021 top → Nov-2022 bottom: buy-and-hold returns roughly −77%. Almost every strategy beats it.

Same strategies, same engine — the verdict flips because the regime flipped. So "97% lose to hold" is true about this window, not a timeless law (the in-sample / per-window caveat in docs/analytics.md:219). The fix is regime segmentation: run the fleet across labeled sub-windows (bull / bear / chop) and report the beat-rate per regime. A strategy that trails hold in the bull but beats it in the bear is doing exactly its job — a single-window read erases that signal. Captured as a simulation-feature seed (regime-segmented backtest windows).

See buy and hold, alpha, risk adjusted return.

Related

• ema cross
• trend following
• mean reversion
• whipsaw
• out of sample testing
• buy and hold
• alpha
• risk adjusted return

Sources

• wiki/qa-sessions/2026-05-17-session.md#q1 (first asked here)
• wiki/qa-sessions/2026-06-22-session.md#q1 (benchmark-regime refinement)
• wiki/2026-05-17-ema-cross-symbol-breakdown.md