Build and test a strategy Lesson 8 of 14
Strategies · Mean reversion
RSI mean reversion: the two-day RSI rule, tested
The RSI(2) pullback rule on SPY and the 30 Dow stocks: exact rules, a two-year backtest with costs, why most trades win and where the losses come from.
On this page
Mean reversion is the idea that a price pushed far from its recent level tends to come part of the way back. The best-known rule built on it uses the Relative Strength Index with a period of two days instead of the usual fourteen. A two-day RSI is jumpy: two weak closes in a row send it close to zero. The rule buys at that moment and sells a few days later, when the price is back above its five-day average. Larry Connors and Cesar Alvarez popularized it in a 2008 book.
This lesson states the rule exactly and tests it on two years of daily bars: first on SPY, the oldest S&P 500 exchange-traded fund, then on the 30 stocks of the Dow Jones Industrial Average. Two years is all the daily history the test has, so it shows how the rule behaves, not whether it works. What it shows is still useful: most trades win a little, a few lose a lot, and on single stocks the rule trailed simply holding them.
The idea
A price series has two kinds of movement mixed together. Over months, stock indexes have tended to drift and trend. Over days, they tend to overshoot: a sharp two-day drop is often followed by a partial rebound as the selling that caused it runs out. A rule that buys weakness and sells the rebound is betting on the second kind of movement.
The RSI compares the size of recent up-closes with the size of recent down-closes and scales the answer from 0 to 100. With fourteen days it moves slowly and rarely leaves the 30 to 70 range. With two days it becomes a short-memory gauge: two down days in a row put it near zero, two up days near 100. "RSI(2) below 10" is a compact way of saying "the last two days were almost all down."
Notice what the rule does not know. It has no idea why the price fell. A two-day dip in a calm market and the first two days of a crash produce the same reading.
The rules
Checked once a day, at the close:
- Compute the two-day RSI from closing prices (Wilder's formula, the one on every chart).
- If no position is held and the RSI is below 10, buy at that close.
- If a position is held and the close is above the average of the last five closes (including today), sell at that close.
- Hold one position at most. A new signal while a position is open is ignored.
- Otherwise hold cash.
There is no stop-loss and no time limit. The only way out is a close back above the five-day average, however long that takes.
Connors' published version adds a filter: trade only when the price is above its 200-day average, so that dips are bought in uptrends only. The main test here leaves the filter out, because 200 days would use up 200 of the 500 bars available. A shorter 100-day filter is shown among the neighbouring settings below.
A worked example
On April 3, 2025 SPY fell 4.9% in a day to close at $536.70, and the two-day RSI was 8.6. The rule bought at that close.
The fall was not over. SPY closed at $505.28 the next day and at $496.48 on April 8, which put the trade 7.5% under water. It never closed above its five-day average on the way down, so the rule did nothing. On April 9 the market jumped, SPY closed at $548.62, far above the five-day average of $518.29, and the rule sold: a gain of 2.0% after costs.
The trade ended well. It is also the rule's deepest fall in the whole test, and it shows the risk plainly: the rule sat through a 7.5% loss in three trading days with no plan except waiting.
What the backtest shows
The test, stated in full so it can be checked:
- Data. SPY daily closes from November 18, 2024 to November 17, 2026: 501 bars, the two years of daily history the Finabase chart data provides.
- Trading. The signal and the trade both use the same day's close.
- Costs. 0.10% of the amount on every buy and every sell, to stand for the spread and any commission.
- Left out. Dividends, interest on cash and taxes. The rule is in cash 84% of the time, so interest would have added to its result; every gain is short-term, so in a US taxable account taxes would have taken more from it than from holding.
| Measure | RSI(2) rule | Buy and hold |
|---|---|---|
| Total return | 26.5% | 24.5% |
| Return per year (CAGR) | 12.6% | 11.7% |
| Deepest fall from a peak | −7.6% | −19.0% |
| Longest time below a peak | 26 trading days | 88 trading days |
| Time invested | 16% | 100% |
| Trades | 29 | 1 |
| Trades with a gain | 79% | n/a |
The rule ended slightly ahead while holding the fund on 80 of 500 days. Those 80 days were unusually good ones: the average day spent invested returned 0.38%, against 0.05% for the average day of the period. That is the whole effect, and it is a statement about these two years.
The 23 winning trades gained 1.2% on average and the six losing ones lost 0.8%. The last trade was still open at the end of the data and is counted at the final close.
Costs matter here. With no cost the rule returns 33.9%; at 0.10% per trade it returns 26.5%. Fifty-seven buys and sells at 0.10% took more than seven points, about a fifth of the gross gain. A rule whose average trade makes around 1% cannot afford much friction: at 0.25% per trade, half the average gain would be gone.
The two halves of the period
| Period | RSI(2) rule | Buy and hold |
|---|---|---|
| November 2024 – November 2025 | 10.4% | 12.2% |
| November 2025 – November 2026 | 14.6% | 11.0% |
The rule trailed in the first year and led in the second. The first year held the sharp fall and recovery of spring 2025; the second was choppier, with more short dips that reversed.
Neighbouring settings
| Setting (SPY, same two years) | Total return | Deepest fall | Trades | Time invested |
|---|---|---|---|---|
| RSI(2) below 5 | 22.4% | −2.5% | 12 | 5% |
| RSI(2) below 10 (the rule) | 26.5% | −7.6% | 29 | 16% |
| RSI(2) below 20 | 22.9% | −9.8% | 44 | 27% |
| RSI(3) below 10 | 18.8% | −2.0% | 9 | 4% |
| RSI(2) below 10, exit above 10-day SMA | 15.7% | −15.8% | 24 | 25% |
| RSI(2) below 10, only above 100-day | 13.0% | −2.1% | 16 | 8% |
| RSI(14) below 30 | 13.3% | −1.8% | 4 | 2% |
Every version made money on SPY in this period, which says more about the period than about the settings. The published setting happens to be the best of the seven, and that is exactly the result not to lean on: it is one history, and a slower exit cut the return by 11 points.
The same rule on 33 symbols
One fund over two years is 29 trades. Running the identical rule on more symbols gives a fairer picture.
| Symbols, daily, same two years | Rule, median return | Holding, median return | Rule gained on | Rule beat holding on |
|---|---|---|---|---|
| SPY, QQQ, IWM | 26.5% | 24.5% | 3 of 3 | 2 of 3 |
| 30 Dow stocks | 8.6% | 24.8% | 22 of 30 | 10 of 30 |
On the three index funds the rule did about as well as holding. On single stocks it did not: the median stock returned 8.6% under the rule and 24.8% held. Across the 30 stocks it made 765 trades; 65% gained, the average trade made 0.32% after costs, the best made 12.1% and the worst lost 30.6%.
The stock list is today's Dow membership. Companies that did badly enough to be removed are missing, which flatters both columns (survivorship bias).
On five years of weekly bars, with the same thresholds applied to weeks instead of days, the rule returned 21.7% on SPY against 60.9% for holding, and it beat holding on 10 of the 30 Dow stocks. In a market that mostly rose, being invested a fifth of the time is a large handicap.
Where it fails: the losing periods
- A fall that keeps falling. The rule buys the second down day and waits. UnitedHealth (UNH) lost a third of its value over these two years in a series of drops; the rule bought into them 24 times and lost 38.6%, worse than the 33.9% lost by holding. An index recovers from most dips. A single company in trouble may not.
- Slow declines. The worst SPY trade was bought on March 3, 2026 at $680.33 and sold ten trading days later at $670.79, a loss of 1.6%. The price drifted down slowly enough that the five-day average followed it, and the exit came on a small bounce at a lower level. The February 21, 2025 trade lost 1.2% the same way.
- Strong trends. Apple (AAPL) rose 54.3% in the period. The rule made 2.0% on it. It was in cash for most of the rise, because a rising stock rarely prints two weak days in a row.
- The shape of the results. Many small gains and a few large losses is the opposite of trend following, and it is uncomfortable in a specific way: the record looks excellent until the one trade that does the damage. A 79% win rate says nothing about the size of the next loss.
- Costs and taxes. The rule traded about once every three and a half weeks. Wider spreads than 0.10%, or short-term capital gains tax, cut directly into an average gain of about 1%.
Compared with its neighbours
| Approach | Buys when | Typical trade | Win rate in this test | Main risk |
|---|---|---|---|---|
| RSI(2) rule | Two weak days in a row | A few days, about 1% | 79% (SPY) | A fall that does not bounce |
| Bollinger Band rule | Close below the lower band | One to four weeks | Tested on its own page | The same, held longer |
| 10-month moving average (trend rule) | Price above its average | Months to years | 64% (SPY, 1993–2026) | Whipsaws in choppy markets |
| Buy and hold | Once | The whole period | n/a | The full size of every fall |
How Finabase does it
No Finabase tool runs this rule for you or tells you when to trade. Three tools show the pieces.
- Chart. The RSI study takes any period. The link above opens SPY with a two-day RSI, so the dips below 10 can be seen against the price.
- Backtest. The Backtest tool trades the RSI study's own signals: a bullish signal when the RSI drops below 30 and a bearish one when it rises above 70. The period can be edited to 2, but the levels stay at 30 and 70, and the exit is the opposite signal, not the five-day average. It also buys a fixed $10,000 on each signal and charges no trading cost. Its results will not match the table above.
- Strategy. The Strategy tool compares an indicator with a constant, so "SPY: RSI (2) is below 10" can be written as a condition. It decides what to hold from the current values each day and replays two years; it has no memory of an open trade, so the "sell above the five-day average" exit has to be approximated with a second condition.
Frequently asked questions
Why a two-day RSI and not the standard fourteen?
A fourteen-day RSI reacts too slowly for a trade that lasts a few days. On SPY, buying below 30 produced only four trades in the two years tested. The two-day version fell below 10 on 29 occasions that led to a trade, which is enough to act on and to measure.
Is a high win rate a sign of a good strategy?
No. Win rate ignores size. This rule won 79% of its SPY trades by taking small gains quickly and holding losers until they recovered. On the Dow stocks the same habit produced one loss of 30.6%, which wipes out a long run of 1% gains.
Does the rule work on individual stocks?
Less well than on index funds in this test. On the 30 Dow stocks it made money on 22 and beat holding on 10. An index is an average of many companies and tends to recover from a dip; a single stock can fall for a reason that does not go away.
Why not add a stop-loss?
A stop is a reasonable thing to test, and it changes the strategy: it turns some trades that would have recovered into realized losses, and limits the ones that would not. Connors' published tests reported that stops reduced the results of this kind of rule. With 29 trades on SPY, this data cannot settle the question either way.
Summary
The RSI(2) rule buys after two weak days and sells on the first close above the five-day average. On SPY over the two years to November 2026 it returned 26.5% after costs against 24.5% for holding, while invested 16% of the time, and its deepest fall was 7.6% against 19.0%. Costs took a fifth of its gross gain. On the 30 Dow stocks it trailed holding on 20, and its worst trade lost 30.6%, because the rule has no exit for a fall that does not reverse. Two years is too short to say the rule works. It is long enough to show its shape: frequent small gains, rare large losses, and a result that depends on prices bouncing.
Check your understanding
5 questions on this lesson. Each answer comes with the reason, and nothing is scored.
SPY has closed lower two days in a row and its two-day RSI is 6. No position is held. What does the rule do at today’s close?
- Nothing, because the price is below its five-day average
- Buys, because the two-day RSI is below 10
- Waits for the RSI to rise back above 10 first
Show the answer
B. Buys, because the two-day RSI is below 10 The entry looks only at the two-day RSI. The five-day average is the exit: the position is sold at the first close above it.
The rule won 79% of its SPY trades. Why is that not enough to call it a good strategy?
- Because a win rate ignores how large the gains and the losses are
- Because 79% is below the level most strategies reach
- Because the win rate was measured after costs
Show the answer
A. Because a win rate ignores how large the gains and the losses are The rule takes small gains quickly and holds losing trades until they recover. On the Dow stocks one trade lost 30.6%, far more than a typical gain of about 1%. A high win rate with rare large losses can still lose money.
On SPY the rule returned 33.9% before costs and 26.5% after a 0.10% cost per trade. What does that say about this kind of rule?
- Costs are too small to matter for short-term rules
- A rule that trades often for small gains gives up a large share of them to costs
- The backtest double-counted the costs
Show the answer
B. A rule that trades often for small gains gives up a large share of them to costs The rule made 29 round trips in two years with an average gain of about 1%. A 0.20% round-trip cost takes about a fifth of that, and a wider spread would take more.
Which situation does the most damage to the rule?
- A dip of two days that reverses on the third
- A long steady rise with no dips
- A stock that keeps falling for weeks after the buy
Show the answer
C. A stock that keeps falling for weeks after the buy There is no stop. The only exit is a close above the five-day average, so a fall that continues is held all the way down. A steady rise only keeps the rule in cash.
The rule beat holding on SPY but on only 10 of the 30 Dow stocks, over the same two years. What is the sound reading?
- The rule should be used on index funds only, and it will keep working there
- Two years and one fund is a small sample; the wider test shows the result is not dependable
- The Dow stocks were tested incorrectly
Show the answer
B. Two years and one fund is a small sample; the wider test shows the result is not dependable A single two-year result on one symbol is 29 trades. Running the same rule on more symbols showed a much weaker picture, which is why the page reports the spread instead of the best case.
Exercise
One task to do in Finabase. Nothing is graded or saved.
Open the one-year SPY chart with the two-day RSI. Find five separate days when the RSI dropped below 10. For each, look at the price over the next week or two: did it bounce at once, drift lower first, or keep falling? Could you have told, from the RSI reading on the day itself, which of the three it would be?
Open the SPY chart with the two-day RSIWhat you should find
On an index fund most such dips have been followed by a bounce within days, which is where the high win rate comes from. But the reading looks the same whether it is a two-day wobble or the start of a longer fall, and the rule has no stop: it waits for a close above the five-day average however far the price goes first. The few dips that kept falling are the trades that decide the result.
Sources and further reading
- Larry Connors and Cesar Alvarez, Short Term Trading Strategies That Work (TradingMarkets, 2008) — the two-period RSI pullback rule
- J. Welles Wilder Jr., New Concepts in Technical Trading Systems (Trend Research, 1978) — the RSI
- State Street — SPDR S&P 500 ETF Trust (SPY), fund page: inception January 22, 1993
- Finabase chart data — SPY, QQQ, IWM and the 30 Dow stocks, daily bars November 2024 to November 2026 and weekly bars from November 2021; rule and results computed by Finabase
