Mean Reversion

An RSI-based mean reversion strategy: buy when RSI dips below oversold, sell when it rises above overbought.

Complete Example

from replaybt import BacktestEngine, CSVProvider, Strategy, MarketOrder, Side


class RSIScalper(Strategy):
    def configure(self, config):
        self._prev_rsi = None

    def on_bar(self, bar, indicators, positions):
        rsi = indicators.get("rsi")
        if rsi is None:
            return None

        prev = self._prev_rsi
        self._prev_rsi = rsi

        if prev is None:
            return None

        if not positions:
            # Long when RSI crosses below 25 (oversold)
            if rsi < 25 and prev >= 25:
                return MarketOrder(
                    side=Side.LONG,
                    take_profit_pct=0.04,
                    stop_loss_pct=0.05,
                )
            # Short when RSI crosses above 75 (overbought)
            if rsi > 75 and prev <= 75:
                return MarketOrder(
                    side=Side.SHORT,
                    take_profit_pct=0.04,
                    stop_loss_pct=0.05,
                )
        return None


engine = BacktestEngine(
    strategy=RSIScalper(),
    data=CSVProvider("ETH_1m.csv", symbol_name="ETH"),
    config={
        "initial_equity": 10_000,
        "skip_signal_on_close": False,  # allow re-entry after exit
        "indicators": {
            "rsi": {"type": "rsi", "period": 7, "method": "wilder"},
        },
    },
)
results = engine.run()
print(results.summary())

Key Differences from Trend Following

skip_signal_on_close = False

Mean reversion strategies often want to re-enter immediately after an exit. Setting skip_signal_on_close=False allows on_bar() to run even on bars where a position just closed.

config = {
    "skip_signal_on_close": False,  # re-entry on same bar as exit
}

RSI Method: Wilder's vs Simple

# Wilder's exponential RSI (recommended for volatile assets)
"rsi": {"type": "rsi", "period": 7, "method": "wilder"}

# Simple rolling RSI
"rsi": {"type": "rsi", "period": 7, "method": "simple"}

Wilder's RSI reacts faster to recent price changes due to exponential weighting. On volatile assets like meme coins, it generates different entry times that often catch more profitable mean reversion opportunities.

Wider SL Than TP

Mean reversion strategies typically use a wider stop loss than take profit. The idea is that most entries are near extremes and will revert, but you need room for the occasional continuation move:

MarketOrder(
    side=Side.LONG,
    take_profit_pct=0.04,   # 4% TP (tight — capture the reversion)
    stop_loss_pct=0.05,     # 5% SL (wider — tolerate noise)
)

Adaptive TP/SL with Volatility Regime

Adjust TP/SL based on recent volatility:

class AdaptiveScalper(Strategy):
    def configure(self, config):
        self._prev_rsi = None

    def on_bar(self, bar, indicators, positions):
        rsi = indicators.get("rsi")
        atr = indicators.get("atr")

        if rsi is None or atr is None:
            self._prev_rsi = rsi
            return None

        prev = self._prev_rsi
        self._prev_rsi = rsi

        if prev is None or positions:
            return None

        # High vol → wide TP/SL, low vol → tight TP/SL
        vol_ratio = atr / bar.close
        if vol_ratio > 0.0012:  # high volatility
            tp, sl = 0.04, 0.05
        else:
            tp, sl = 0.025, 0.03

        if rsi < 25 and prev >= 25:
            return MarketOrder(side=Side.LONG, take_profit_pct=tp, stop_loss_pct=sl)
        if rsi > 75 and prev <= 75:
            return MarketOrder(side=Side.SHORT, take_profit_pct=tp, stop_loss_pct=sl)
        return None