We Backtested 11,683 RSI Combinations to Find Out What Really Work

Finance-specific summary (markets / investing / signal takeaways)

The video describes a systematic backtest of a triple RSI (Relative Strength Index) rule applied to SPY (S&P 500 ETF). The goal is to determine whether the original strategy’s performance reflects true edge or merely parameter luck.

Instruments / tickers mentioned

• SPY (SPDR S&P 500 ETF Trust)
• RSI indicators
  • RSI length varied in the tests
  • Mentions “RSI 1/2/3” style conditions as a future experiment
• 200-day moving average (200DMA) of SPY (used as a regime filter)

Backtest timeframes and key results

Original triple-RSI setup (baseline reproduced)

• Period: Feb 1993 to Sep 2026
• Trades: 90
• Win rate / profitability: 88.9% profitable
• Average transaction yield: 1.25%
• “Profitability ratio”: 5.53
• Average holding period: just under 5 trading days

Robustness test expansion

• Total configurations tested: 11,683 (almost all defining parameters changed)
• In a group of 1,125 similar configurations:
  • Median rate of return: 3.86
  • ~94.8% had a return ratio ≥ 2

Extreme / likely-overfit behavior (then addressed)

Some parameter combinations showed:

• Return ratio > 20
• One configuration > 30

To reduce fluke/overfitting, they added a constraint:

• Require at least 100 historical agreements (minimum number of signals/trades)

After stricter filtering:

• Extreme outcomes mostly disappeared
• A few strong configurations remained (e.g., return ratios above 6)

Performance across decades (original strategy breakdown)

• 1993–2002: return ratio 6.31
• 2003–2012: return ratio 5.57
• 2013–2022: return ratio 11.39
• 2023–Sep 2026: return ratio 2.07

Implied conclusion: the strategy is still historically profitable, but clearly weaker recently, consistent with changing market regimes/trends.

Step-by-step / methodology framework (as described)

Original “triple RSI” concept

Regime filter

They require:

• RSI 5 below 30
• RSI declining for three sessions
• Previous RSI reading below 60
• SPY above its 200-day moving average (200DMA)

Entry / exit rules

• Buy: on the close
• Exit: when RSI rises above 50

Intuition

Identify short-term weakness (oversold RSI conditions) within a longer-term uptrend (SPY above 200DMA).

Parameter search / robustness process

They kept the core idea but changed:

• RSI length
• Oversold threshold (e.g., “below 30”)
• Required number of consecutive RSI declines (e.g., “three sessions”)
• Prior RSI “ceiling” (e.g., “previous RSI below 60”)
• Exit threshold (e.g., “RSI above 50”)

Then they evaluated reliability by:

• Checking median returns
• Measuring how many configurations clear thresholds (e.g., return ratio ≥ 2)
• Using a heat map to see whether profits appear across a region or only at a narrow setting

Reducing overfitting risk

To avoid “extreme” results caused by chance:

• Require ≥ 100 historical agreements
• If a setting were genuinely special, they expected a single “bright cell” (needle-in-haystack)
• Instead, they observed a broader profitable region, interpreted as more consistent with a real effect than pure luck

Forward-looking next experiment (announced)

They plan to test RSI indicators one, two, and three separately using the same system to determine whether:

• each additional RSI condition adds value, or
• complexity is being increased without incremental edge

Explicit recommendations / cautions stated

• They emphasize skepticism toward extremely high optimized results (e.g., return ratio > 20 or > 30) after large parameter searches, as these may be fluke/overfitting.
• They would not automatically replace the original strategy with the best parameter set because:
  • the original existed before the search, and
  • new leaders appeared only through searching.
• They highlight weakened recent performance, especially 2023–Sep 2026, implying market regime change risk.

Disclosures / disclaimers

• “Historical results are hypothetical”
• “Do not take into account commissions and slippage”
• “Not a guarantee of future performance”
• (No explicit “not financial advice” line appears in the subtitles beyond the stated disclaimers.)

Presenters / sources

• No presenter name or external source is mentioned (only a reference to “join our community on School” / “Link in profile”).