Mean reversion and the RSI2 strategy: buying the dip, tested

Mean Reversion is the assumption that a price that has stretched strongly in one direction in the short term tends to partially revert. The most famous strategy in this family is Larry Connors' RSI2: a stock in an uptrend (above 200 average), RSI two-day RSI dropping below 10, buy, and sell as soon as the close returns above the 5-day average. Simple rules, hundreds of trades per year, and a win rate that makes every beginner fall in love.

When we ran it on 184 stocks over the last three years, it was correct in 62% of trades and earned +14% Overall, versus +95% the index, with an even larger maximum drawdown than its own. The full story, including why this happens and when the approach does work, is below.

The rules we ran

• Trend filter: Close above 200-day average. Mean reversion against the trend is like catching a falling knife, and Connors doesn't suggest that either.
• Entry: RSI(2) below 10 at the close, buy at the next day's open. Up to 10 positions, 10% of capital per position, priority to the lowest RSI.
• Exit: Close above 5-day average, or after 10 bars. A disaster stop of 10% below entry.
• Regime: Entries only when S&P 500 is above 200 average. The measurement: 184 American stocks and ETFs, daily bars, May 2023 to September 2026 (840 trading days), execution at the open following the signal, cost of 0.05% per side, long only and without leverage. The universe is today's and not 2023's — stocks that collapsed and disappeared are not included, which biases every result upward. A rising market for most of the period.

The result

Why a high win rate earned so little

• The gains are small, the losses are not. The average trade earns less than a fifth of a percent. A single loss at a 10% stop wipes out fifty such trades. A profit factor of 1.05 means that all the wins barely covered the losses.
• Two thousand trades means two thousand commissions. At a cost of 0.05% per side, costs alone ate up about a third of the gross profit. With an average Israeli commission, the result is negative.
• When the market really falls, "the dip" doesn't come back. 2025 ended at-10.9%: buying corrections, and the corrections turn into a trend. A 10% stop is too far to protect and too far to trade.
• The historical evidence is aging. RSI2 was measured and published on indices in the 2000s, when short-term mean reversion was especially strong. A published edge erodes, and a new sample doesn't have to repeat the old one.

Where mean reversion does have its place

As an entry tactic within a trend strategy: instead of chasing a breakout, wait for a short pullback toward the average and enter there, with a stop below the pullback. That way the risk is small and the target is far. And as logic on broad indices, which almost always revert, not on individual stocks, which sometimes don't. The TechAnalyzer Pro engine does not generate mean-reversion setups; when we measured entries at RSI extremes on our signals, they hurt the track record, and we blocked them.