What Is Regime Filtering in Crypto Trading?

A trading strategy without a regime filter is like a speed limit without conditions. The strategy applies its rules regardless of what the market is doing — buying dips in bear markets, chasing momentum in chop, executing entries in conditions where it has no historical edge. Regime filtering is the mechanism that prevents this.

What a market regime is

A regime is a persistent market condition. Crypto markets tend to cycle through a few distinct states:

• Bull regime: price trending upward, buyers in control, momentum sustaining
• Bear regime: price trending downward, sellers dominating, rallies short-lived
• Range regime: price oscillating between a ceiling and a floor, no clear trend
• Extreme fear: sentiment crashes (Fear & Greed Index below 15–20), sellers exhausted, potential capitulation event

Different strategies have different edges in different regimes. A grid strategy earns consistent small profits in range conditions — the sideways oscillation is exactly what it’s built to exploit. The same grid strategy can accumulate significant losses in a bear market as it keeps buying into a falling price.

What regime filtering does

A regime filter sits between the market signal and the order entry. Before the strategy places a trade, the filter checks: is the current market condition one where this strategy has historical edge?

If yes: proceed with the entry. If no: skip the entry (or close existing positions, depending on design).

The filter is not a prediction. It doesn’t try to forecast where the market is going. It classifies what the market is currently doing and gates the strategy’s behavior accordingly.

How the bots on BotVersusBot use it

V3.8’s ETH grid uses a two-component filter. An EMA (exponential moving average) measures medium-term price direction, and a volatility signal measures market turbulence. The bot classifies each 4-hour period as bull, bear, or range. In bull: buy orders open. In bear or range when conditions are unfavorable: entry side closes.

During V3.8’s first week live, the regime flipped bearish midway through, and four pending buy orders at $2,235–$2,270 stopped receiving fills. Zero new trades for two days. That’s the filter working correctly: it identified a bearish short-term condition and refused to keep buying into a falling market.

The consequence was lower trade count and lower P&L during that stretch. The benefit was not accumulating a stack of unrealized losses in a declining ETH environment. That tradeoff — fewer trades, better risk management — is the entire point of the filter.

KBTR-V7 (Key’s BTC trend-follower) uses EMA crossover as its regime filter at the strategy level. When the 6-week EMA is below the 20-week EMA, the market is in a bearish regime and the strategy holds USDC. When the fast average crosses above the slow one, conditions are met and the strategy enters BTC. The paper run ran for 20 days and the crossover never triggered — BTC moved, but the averages didn’t confirm. The strategy sat out the entire run. That’s regime filtering doing its job: protecting capital in an unconfirmed trend.

KFGC-V8.3 (Key’s Fear & Greed Circuit-Breaker) uses sentiment as its filter. The Fear & Greed Index reading serves as a market regime indicator: readings above 15 mean normal fear, which the strategy ignores. Only when the index hits ≤15 for four consecutive days — sustained extreme fear, a capitulation signal — does the strategy activate. During its 19-day paper run, the index ranged from 26 to 50. It never hit the threshold. Zero trades. The 19-day paper run validated the circuit-breaker logic by showing the filter correctly held its position when conditions weren’t met.

How regime filters are built

The most common approaches:

Moving average crossover: fast average above slow average = bull, below = bear. Simple, lagging by design. Works well on longer timeframes where noise is low. Gets whipsawed in choppy markets.

Volatility-adjusted signals: combine trend direction with a volatility measure (ATR, Bollinger Band width). High-volatility trend signals are different from low-volatility trend signals — the filter can treat them differently.

Sentiment indicators: Fear & Greed Index, funding rates, open interest. These are external data sources rather than price-derived signals. More interpretable to humans, but sometimes slower to update.

Candle-based regime classifiers: categorize each candle as a bullish, bearish, or doji type and look for patterns across multiple candles. V3.8 uses a version of this alongside the EMA for its 4-hour regime reads.

The choice depends on the strategy’s timeframe, the asset being traded, and the regime conditions where the strategy has historical edge.

Why it matters for real money

Without a regime filter, a strategy that works in bull conditions will keep running in bear conditions, eroding its gains. A grid strategy will accumulate underwater positions in a downtrend. A trend follower will generate false signals in a sideways market.

The statistical validation gates (p-value, walk-forward) can tell you whether a strategy has edge in its backtested sample. They don’t automatically tell you whether the current live market matches the regimes where that edge was found. The regime filter is the mechanism that enforces the match at execution time.

At scale, this is the difference between a strategy that survives its first bear market and one that doesn’t. Capital preservation in adverse conditions is what lets you keep trading when the regime shifts back to favorable.

On BotVersusBot, all the strategies running live capital use regime filters. The paper runs run against real market conditions without the safety net of hindsight. When a filter correctly holds a strategy flat during an adverse stretch, it doesn’t make headlines — but it’s doing the work.