RSI Strategy: First Threshold Cross vs. Cross-BackMicron Technology, Inc.BATS:MUTrendAdvantageRSI Strategy: First Threshold Cross vs. Cross-Back 📌 The question When RSI reaches oversold territory, many traders wait for it to recover above the threshold before buying. They do the reverse when RSI reaches overbought territory, waiting for it to fall back below the threshold before selling. This study tested whether waiting for that signal confirmation actually improves performance. ________________________________________ 🧪 What we tested Two simple rules, applied to the same stocks under the same conditions: • Rule A (act right away): buy when RSI drops into oversold, sell when it rises into overbought. • Rule B (wait for confirmation): buy only after RSI drops into oversold and then climbs back out, and sell only after it rises into overbought and then falls back. We ran both rules on 107 large US stocks from 2000 to 2025 (not every stock has data covering the entire period), a period that includes the dot-com crash, the 2008 crisis and the 2020 crash. All trades were long-only, and three common level settings were used: 30/70, 20/80 and 35/65. All levels were chosen in advance, with no optimisation. Trading costs were included (0.10% per side, covers slippage), and buy and hold served as the benchmark. ________________________________________ 🔬 Statistical validation The primary result was tested using paired stock-by-stock comparisons, with 95% confidence intervals, Wilcoxon signed-rank tests, and Benjamini–Hochberg correction for multiple comparisons. A synchronized calendar-time block bootstrap provided an additional robustness check; the CAGR disadvantage of waiting remained statistically significant at all three RSI settings. ________________________________________ 📊 Results: the typical stock (median across 107 stocks, 1 Jan 2000 – 31 Dec 2025) Each figure is the median of that metric across the 107 stocks, so the figures in a column come from different stocks. Values are shown as A (crossover) | B (cross-back) | Buy and hold. 🔹 RSI(14) 30/70 • Net CAGR: 5.76% | 4.85% | 11.36% • Cumulative return: 272.9% | 212.5% | 1218.5% • Maximum drawdown: 56.1% | 58.5% | 66.8% • Sharpe ratio: 0.37 | 0.33 | 0.53 • Martin ratio: 0.40 | 0.29 | 0.52 • Ulcer Index: 17.01 | 17.86 | 22.27 • Profit factor: 3.70 | 2.92 | – • Win rate: 80.8% | 77.8% | – • Average trade return: 8.34% | 7.30% | – • Average MAE: -14.42% | -14.39% | – • Trades per stock: 22 | 22 | 1 • Time in market: 38.3% | 38.5% | 100% 🔹 RSI(14) 20/80 • Net CAGR: 3.64% | 3.48% | 11.36% • Cumulative return: 133.4% | 112.1% | 1218.5% • Maximum drawdown: 43.8% | 45.2% | 66.8% • Sharpe ratio: 0.32 | 0.32 | 0.53 • Martin ratio: 0.41 | 0.42 | 0.52 • Ulcer Index: 8.22 | 8.52 | 22.27 • Profit factor: 4.62 | 4.57 | – • Win rate: 100% | 100% | – • Average trade return: 40.92% | 38.16% | – • Average MAE: -21.32% | -22.65% | – • Trades per stock: 3 | 3 | 1 • Time in market: 23.8% | 23.9% | 100% Note: at 20/80, most stocks traded only a few times in 26 years, and 7 never traded at all. The 100% win rate reflects those few long trades, not a reliable edge. 🔹 RSI(14) 35/65 • Net CAGR: 6.67% | 4.98% | 11.36% • Cumulative return: 370.3% | 219.6% | 1218.5% • Maximum drawdown: 57.0% | 58.8% | 66.8% • Sharpe ratio: 0.40 | 0.33 | 0.53 • Martin ratio: 0.35 | 0.26 | 0.52 • Ulcer Index: 17.76 | 18.10 | 22.27 • Profit factor: 2.93 | 2.44 | – • Win rate: 77.3% | 75.7% | – • Average trade return: 4.73% | 3.85% | – • Average MAE: -11.36% | -11.09% | – • Trades per stock: 43 | 43 | 1 • Time in market: 39.9% | 40.9% | 100% ________________________________________ 🧺 Results: if you held all 107 stocks (equal-weight portfolio) The typical stock is not what an investor actually holds. We also tested an equal-weight portfolio of all 107 stocks, as displayed in the chart above, to examine how diversification affects overall returns, risk, and the comparison between the two RSI rules. Values are shown as A (crossover) | B (cross-back) | Buy and hold. 🔹 RSI(14) 30/70 • Net CAGR: 7.57% | 6.54% | 15.82% • Maximum drawdown: 34.3% | 35.4% | 45.3% • Sharpe ratio: 0.62 | 0.56 | 0.86 🔹 RSI(14) 20/80 • Net CAGR: 4.99% | 4.82% | 15.82% • Maximum drawdown: 20.8% | 22.7% | 45.3% • Sharpe ratio: 0.73 | 0.71 | 0.86 🔹 RSI(14) 35/65 • Net CAGR: 8.16% | 6.84% | 15.82% • Maximum drawdown: 34.5% | 36.7% | 45.3% • Sharpe ratio: 0.64 | 0.57 | 0.86 Diversification cuts the drawdowns by about a third to a half and raises the Sharpe ratios. The ranking does not change: B trails A at every setting, and both trail buy and hold. ________________________________________ 📖 How to read the tables • Net CAGR: compound average yearly growth after costs. • Maximum drawdown: the largest drop from a peak. • Sharpe ratio: return per unit of volatility (higher is better). • Ulcer Index: how deep and long the drawdowns were (lower is better). • Martin ratio: return per unit of Ulcer Index (higher is better). • Profit factor: total profits divided by total losses. • Average MAE: the typical worst point reached during a trade before it closed. ________________________________________ 🔍 What we found 1️⃣ Waiting reduced returns at all three threshold settings. Across the 107 stocks, Rule A had a higher median CAGR than Rule B at 30/70, 20/80, and 35/65. The paired statistical tests support this result at all three settings, although not every secondary metric showed a statistically significant difference. 2️⃣ The cost of waiting depended on the thresholds. • 30/70: Rule B had approximately 0.9 percentage points less annualized return, with Rule A outperforming on 94 of 107 stocks. • 35/65: The gap was approximately 1.2 percentage points, with Rule A outperforming on 85 of 107 stocks. • 20/80: The gap was smaller, approximately 0.2 percentage points. Trading was infrequent, with a median of only three trades per stock and seven stocks recording no trades. These are historical results, not forecasts of future performance. 3️⃣ Cross-back confirmation changed timing, not the number of trade episodes. Both rules produced the same number of one-to-one paired trade episodes. However, Rule B generally entered at higher prices. Waiting for the cross-back did not improve overall CAGR in this test. 4️⃣ Waiting did not improve drawdown outcomes. Median maximum drawdowns were slightly deeper with Rule B at all three threshold settings. The difference was not statistically significant at 20/80, so the evidence does not support a claim that waiting consistently worsened drawdown risk at every setting. 5️⃣ The threshold settings mattered. The choice of RSI levels substantially changed trade frequency and time in the market. The 20/80 setting traded much less frequently and spent about 76% of the time out of the market, while 35/65 generated roughly twice as many trades as 30/70. 6️⃣ Buy and hold earned more, but also experienced deeper historical drawdowns. At 30/70, the median stock's annualized return was 5.8% with Rule A, compared with 11.4% for buy and hold. Median maximum drawdown was 56.1% versus 66.8%, respectively. This is a return-versus-drawdown trade-off, not evidence that the RSI strategy is superior overall. Because the test uses prices without dividends, the comparison with buy and hold should be interpreted cautiously. ________________________________________ 🧩 Where does the cost of waiting come from? Additional testing was performed to explore which half of the cross-back rule causes the loss. We tested two mixed versions: one that uses the cross-back only on the buy, and one that uses it only on the sell. Median Net CAGR, shown as A (first cross) | Cross-back on sell only | Cross-back on buy only | B (cross-back on both): 🔹 RSI(14) 30/70: 5.76% | 5.36% | 4.88% | 4.85% 🔹 RSI(14) 20/80: 3.64% | 3.56% | 3.56% | 3.48% 🔹 RSI(14) 35/65: 6.67% | 6.14% | 5.53% | 4.98% • The cross-back on the buy caused most of the loss: about 0.8 percentage points a year at 30/70, 0.15 at 20/80 and 1.1 at 35/65. • The cross-back on the sell alone cost much less: about 0.2 points a year at 30/70, with no reliable difference at 20/80 or 35/65. • The cross-back usually came quickly, a median of 2 days after the first cross. Yet it bought cheaper than Rule A on only about 31% of paired trades, roughly 3 times in 10. • When waiting did save money, the median saving was about 1.4%. When it cost money, the median extra paid was about 1.8% at 30/70 and 35/65 (2.1% and 2.0% at 20/80). In short, the cross-back protected against a falling price only about 3 times in 10, and the protection didn't make up for the higher prices paid the rest of the time. If you prefer to wait for a cross-back, these results suggest it costs far less on the sell than on the buy. ________________________________________ 💡 What this means for the trader • Confirmation has a cost. In this test, waiting for RSI to cross back out of the zone generally meant entering at a worse price, without compensating improvement in overall returns. • Thresholds matter. The choice between 20/80, 30/70, and 35/65 changed trading frequency and time in the market substantially. • Risk reduction is not the same as market outperformance. The tested RSI rules had lower median returns than buy and hold, although they also had lower median maximum drawdowns. • Test the rule on your own instruments. Different markets, periods, entry and exit rules, and risk controls may produce different outcomes. ________________________________________ 🔭 Where might cross-back confirmation work better? Although waiting for cross-back confirmation underperformed in this test, it may be worth investigating whether the results change under different market conditions. For example, combining cross-back signals with a trend filter, or testing them during high-volatility periods and markets prone to false reversals, might improve entry timing. These are hypotheses for future testing, not established advantages. Any improvement would need to be validated out of sample and after trading costs. ________________________________________ 🏁 Bottom line In this 107-stock historical test, waiting for RSI to cross back out of the oversold or overbought zone did not improve performance. Rule B generated lower median CAGR than Rule A at all three tested threshold settings, and the primary differences were statistically significant. At 30/70, the estimated annualized-return disadvantage of waiting was approximately 0.9 percentage points. That is evidence against the added confirmation step in this particular setup—not proof that cross-back confirmation cannot work under different rules or market conditions. RSI Cross-Back Study — Educational Reference | TrendAdvantage