Connors RSI-2 oversold/overbought with a 200-SMA trend filter. Enter on extreme RSI readings, exit on reversion.
“Short-lookback RSI catches true exhaustion. Pair it with a long-term trend filter so you only mean-revert in the direction of the bigger tide.”
Every variant we walk-forward-tested collapsed out-of-sample (OOS SQN < 1.0 or negative). The in-sample numbers shown below are the RAW backtest — they did NOT transfer to unseen data. Reject for live trading.
This strategy: Mode A walk-forward on legacy_5m: all 3 rsi2 variants failed OOS (best was r2_10/90 with OOS SQN -0.85). Rolling Mode B: rsi2 variants never appeared in the top-5 train in any of the 4 windows. The featured legacy_15m result is in-sample only — the edge did not transfer to unseen data.
Catastrophic on TV multi-year 5m. -104k MYR, PF 0.24. The 30-tick SL against a 2-period RSI on low-volatility palm oil 5-min bars means stops get hit ~75% of the time before RSI can revert. The 1-month 62% WR was sampling luck.
Range-bound / low-volatility regimes. Avoid strong trending sessions.
{
"dataset": "legacy_15m",
"tf": "15m",
"params": {
"rsiPeriod": 2,
"longThreshold": 20,
"shortThreshold": 80,
"smaPeriod": 200
},
"trades": 47,
"winRate": 61.7,
"netProfitMYR": 3865,
"profitFactor": 2.03,
"sqn": 1.68,
"maxDrawdownMYR": 2015
}