The RSI(2) Pullback Strategy, Backtested
The RSI(2) strategy buys a stock when its two-day relative strength index falls below 10 while the close is above the 200-day average, then sells on the first close above the five-day average. Backtested from 2015 to 2026, it held up on seven big winners and mostly failed on five that lagged the S&P 500.
The RSI(2) strategy is a short-term dip-buying rule. It uses a two-day version of the relative strength index, which swings to extremes after just two or three down closes, and it only buys when the longer trend is still up.
How it works
Three conditions, all on daily bars. The version tested here, run in TradingView’s Strategy Tester in August 2026, uses exactly these rules and nothing else:
- Trend filter: the close is above its 200-day simple average.
- Signal: the two-period relative strength index closes below 10.
- Exit: the first close above the five-day simple average.
Orders fill at the next day’s open, because the signal is only known once the day has closed. There is no stop loss at all, one position at a time, and each trade uses the whole account. Costs are 0.05% commission each way plus one tick of slippage.
Why two days, not fourteen
Why two days instead of fourteen? A 14-day index moves slowly and rarely reaches an extreme while a stock is above its 200-day average. The same test ran RSI(14) under 30 with the same trend filter, and it fired between 2 and 6 times per stock in eleven years. The two-day version fires often enough to measure, which is the whole reason it exists. It is the same short-lookback idea behind Connors RSI, without the extra components.
The bet underneath is mean reversion: a sharp two-day drop inside an uptrend tends to be partly taken back within a week. The average hold across the seven winners was 4 to 6 bars.
A worked example
Take Apple’s row, straight from the results file. From January 2015 to August 2026 the rule took 70 trades on AAPL. 45 of them closed higher, which is 64.29%. The profit factor was 1.592, meaning the winning trades together made $1.59 for every $1.00 the losing trades lost.
The average trade returned 0.52% and lasted about 5 bars. Now look at the costs. At 0.05% commission on the way in and 0.05% on the way out, every round trip pays about 0.10% before slippage. That is 0.10 / 0.52, or roughly a fifth of the average trade’s result, gone to costs.
That ratio is why this rule lives or dies on small edges. A stock where the average trade makes 2.28%, as NVDA’s did, can absorb the costs easily. A stock where it makes 0.13%, as Intel’s did, cannot: the costs are most of the result. Intel’s run finished +2.17% over the same eleven years.
A hypothetical to show the timing. Say a stock above its 200-day average closes lower on Monday and again on Tuesday, and Tuesday’s two-day index reads 7. The order fills at Wednesday’s open. It then holds until the first close above the five-day average, say Friday, and sells at the following Monday’s open. Every one of those nights is exposure with no stop behind it.
The original data
Seven stocks first: AAPL, MSFT, GOOGL, AMZN, NVDA, META and TSLA, January 2015 to August 2026. Across all seven, 305 of 443 trades closed higher, which is 68.85%. The profit factor ran from 1.405 on TSLA to 3.312 on MSFT. The test set three bars before calling a result good: at least 60% of trades closing higher, a profit factor of 1.3 or more, and at least 30 trades. RSI(2) cleared all three on seven of seven.
The survivorship test: five stocks that lagged
Then the same code ran on five large caps that lagged the S&P 500 over the same window: INTC, PYPL, DIS, NKE and WBA. Walgreens’ runs stop at 27 Aug 2025, its last trading day before it was taken private. Across those five, 112 of 194 trades closed higher, which is 57.73%. Leave out Walgreens, which traded only 10 times, and it is 108 of 184, or 58.70%. Outside Walgreens, the profit factors fell to between 1.032 and 1.380. It cleared all three bars on none of the five.
Said plainly: the seven-for-seven result was measuring the stocks, not the rule. Those seven are the biggest winners of the decade. A dip-buying rule on a stock that climbs for eleven years looks good almost regardless of its details. The second list was picked because it trailed the market, so it is just as biased in the opposite direction, and the truth for an ordinary stock sits somewhere between the two results. The full tables are published: the per-ticker results and the five-stock survivorship test.
The index test runs longer. On SPY from January 1993 to August 2026, the same rule took 227 trades, 65.64% closed higher, and the profit factor was 1.897. It finished +141.80%. Holding SPY over the same window made roughly +1,480%. The rule’s worst drawdown was 16.55%. It bought a smoother ride and gave up most of the return to get it. Those figures are in the SPY table.
Then the window was cut in half: 2015 to 2020 to build, 2021 to 2026 to check. Three stocks, one rule. Microsoft looked like the best result in the whole study in the first half, with 72.22% of 36 trades closing higher at a profit factor of 5.141. In the second half it slipped to 59.26% of 27 trades at 2.161. Apple went the other way, from 1.323 to 1.370. Only NVDA held its shape, 3.646 then 3.433.
These are past results from a small, hand-picked set of tickers, not a forecast. Nothing was optimized, which is good, but everything is still in-sample except that one split, which is published as the out-of-sample table. No figure here says what the rule will do next.
When it fails
It fails when the stock stops going up. That is the survivorship result in one line. The filter lets the rule buy dips while the close is above the 200-day average, but a stock can sit above that average for months while going nowhere. On Disney the rule made +7.25% in eleven years. On Intel it made +2.17%. On both, the costs of 53 and 47 round trips took a large share of a thin edge.
It fails on a gap. There is no stop. A stock bought after a two-day drop can open 10% lower the next morning on news, and the rule simply holds until a close above the five-day average. TSLA’s worst drawdown on this rule was 39.51%, far above the 10.93% on MSFT, from the same code.
It fails when a sample is small. Walgreens produced 10 trades and a profit factor of 10.217. That number is noise: four winners and six losers cannot support a ratio to three decimals.
It fails through selection after the fact. Picking the stock where the table looks best, then trading it, repeats the mistake the survivorship test exposed. Microsoft’s first half would have sold the rule to anyone, and the second half was much weaker.
And it fails against simply holding. On SPY, the rule’s 227 trades returned about a tenth of what buy-and-hold returned. A high share of winning trades is not the same as making more money, which the win rate page covers in full.
Related
The RSI page explains the indicator itself, and why its usual 70 and 30 lines mean little in a trend. Mean reversion is the idea this rule bets on, and it explains why the timeframe decides the outcome. And backtesting lists the errors that make a test like this one look better than it is, several of which this page had to correct for.
Run any short-term oversold rule on the stock you actually trade, over a stretch that includes a bad year for it, before trusting a table built on someone else’s tickers. The seven-stock result and the five-stock result come from the same code.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.