Market regimes — directional movement, step size and rolling windows

In a hypothetical illustration, Anna applies one mean-reversion rule to both ranges and trends. A regime filter changes which cases she trades, but any improvement in hit rate or annual result must be established on a predefined test with costs. The example does not report an actual trader’s EUR 2,000-to-EUR 8,000 performance improvement.

What a market regime is

A market regime is a label assigned to chosen properties of a specified data window. Directional movement, step size and execution conditions are different dimensions; trend and high volatility need not exclude one another. Define the measurements, thresholds and observation time before using the label in a strategy.

Separate directional movement from the size of steps

In the teaching classifier below, use five completed closes, giving four changes. Let E be the absolute first-to-last change divided by the sum of the four absolute changes. Record the net direction separately. Let A be that absolute-change sum divided by four, in pips. A is a mean close-to-close step, not ATR or return standard deviation; it cannot see intrabar High and Low.

Choose demonstration thresholds before inspecting the sequence: E ≥ 0.7 is high directionality, E ≤ 0.3 low, and the interval between them remains ambiguous. A ≥ 15 pips means larger steps in this example. If all four changes are zero, record E = 0 and “no movement”; missing observations are not zero prices. These thresholds are hypothetical settings, not recommendations for a pair or timeframe.

Hypothetical EUR/USD closes are 1.1000, 1.1040, 1.1030, 1.1020, 1.1010 and 1.1000; one pip = 0.0001. At t4, steps are +40, −10, −10, −10 pips: net +10, sum 70, E = 10/70 ≈ 0.1429 and A = 17.5 pips. At t5, the window loses the old jump: four −10-pip steps give net −40, sum 40, E = 1 and A = 10. The newest change is no larger than the preceding three falls. The label changed because the window changed, not because a lasting new regime has been established.

How to identify the regime — three indicators

ADX measures trend strength, with direction assessed separately. The familiar 20/25 thresholds are interpretation conventions, not guarantees. Multiple transformations of the same price history need not be independent evidence; a two-out-of-three rule needs its own definition and evaluation. Read the calculation context in ADX as a trend-strength measure.

• The slope of a selected EMA. Specify period, source, initialisation and the interval used to measure slope. A rising average is a property of that smoothed series, not proof that every current price change points upward.
• The Average Directional Index (ADX). Read its value as the selected measure of trend strength, not direction. An interpretation threshold does not establish future persistence or profitable strategy performance.
• Band width and ATR as a volatility measure. Specify band and ATR settings before comparing windows. ATR uses true range, including gaps, and does not determine direction. Neither a small range nor a narrow band guarantees a breakout.

Compare two synthetic windows: 1.1000 → 1.1015 → 1.1030 → 1.1045 → 1.1060 has E = 1 and A = 15 pips. Alternating 1.1000 → 1.1015 → 1.1000 → 1.1015 → 1.1000 has E = 0 and the same A = 15. Moving the A threshold from 15 to 16 changes both step-size labels without changing a price. Directionality and step size should remain separate measurements.

The trending market strategy — follow, never anticipate

A trend-following variant could specify a pullback to a chosen average or price level and an ATR-based stop. The chosen average, retracement and stop multiple are rule parameters; neither a “trend” label nor a 38.2% retracement establishes a future reward/risk of 5 or any win rate.

For a hypothetical exit variant, close one third at +2R, another third at +4R and the remaining third at −1R: the gross aggregate result is (2 + 4 − 1)/3 = 5/3 R, before costs. A trailing exit may give a different result; its final price is not known when the trade begins. Define fills and remaining volume separately from the classifier. More context: trend-following systems.

The range-bound strategy — mean reversion

A range variant might buy near predefined support and sell near predefined resistance, leaving the middle untraded. Its win rate and reward-to-risk distribution depend on precise entry, exit, costs and sample. No universal 60–70% hit rate or maximum 1:2 reward-to-risk ratio follows from the regime label.

A range variant needs a fixed level-selection rule, an entry and invalidation condition, and executable-side fills. An RSI threshold or reversal candle can be an added condition, but does not establish the strongest confluence. A 10–20-pip stop buffer is a chosen model parameter, not a universal range-market requirement or evidence of historical opportunities.

The high-volatility regime — smaller size, shorter horizon

In a high-volatility scenario, distinguish the planned risk budget from position size: a percentage of account equity is a risk budget, whereas position size is stated in lots or units. If the chosen stop widens, calculate a smaller volume from the monetary budget and account-currency pip value. Any stop multiple, holding horizon or decision to avoid overnight exposure is a tested strategy rule rather than a universal requirement.

Preserve the label known at the decision time

The t4 classification above becomes available only at the t4 close; it cannot justify a trade placed at that candle’s start. Keep t4’s result when the rolling window moves to t5. A provider revision or a changed rule may require a documented recalculation, but a future price alone must not overwrite the old decision record.

Regime labels are often uncertain while conditions change. Conflicting timeframe readings or a changing ADX can prompt review, but neither proves a particular regime transition. Predefine any reduction in planned risk and review the strategy against its tested drawdown limits; a 5–10 percent drawdown is not automatically normal or acceptable.

Practical takeaway: A regime label summarizes chosen conditions and can change. Test the complete rule rather than treat the label as a forecast.

The most common mistakes

• Inferring profitability from a label. Compare the complete rule with and without the filter, including omitted trades and costs. A category name does not prove that a strategy must lose under other conditions.
• Treating agreement as independent evidence. Several indicators derived from the same prices can overlap in information. Specify the classifier and what each measurement adds before interpreting agreement.
• Confusing risk budget and volume. With an unchanged monetary budget, a stop distance doubled under otherwise identical pip value and costs gives half the theoretical volume. A doubled ATR alone does not guarantee that the chosen stop doubled or that a fixed loss percentage follows.

What to do tomorrow — three concrete steps

1. Record inputs and thresholds. Calculate E, A and direction only after the chosen five-close window completes. Keep the closes with the label so another person can reproduce it.
2. Test a rule with memory if needed. For demonstration, enter high directionality at E ≥ 0.70 and leave only at E < 0.50, initially outside. E = [0.69, 0.71, 0.68, 0.72, 0.49, 0.71] gives states no, yes, yes, yes, no, yes: three changes. A single 0.70 threshold gives no, yes, no, yes, no, yes: five changes. Fewer switches do not establish better performance.
3. Separate classification from trade rules. Preserve the classifier’s initial state and both thresholds. Specify risk, entry, exits and costs separately; the demonstration does not prescribe half volume, tighter stops or smaller targets.

For a deeper view of how strategy choice depends on conditions, the trading strategies section on ForexMechanics walks through the same regime questions in long form; on the theory side, see the AQR and Bank for International Settlements (BIS) papers listed below.