Does the RSI 2 Strategy Work? Our S&P 500 Test Since 1999

Does the RSI 2 strategy work? In our lab test it does, at the level of single trades. We ran Larry Connors' RSI(2) dip-buying rule on every stock that was in the S&P 500 at the time, from 1999 to 2026, delisted companies included. Across 60,107 trades it earned an average of +0.218% per trade after costs. Taking the same trade at random times earned +0.024%.

There's a catch, though. An edge on each trade isn't the same thing as a better portfolio. When we turned the signals into a simple 10-slot portfolio, it did its best work in bad decades, mostly by staying out of crashes. Our verdict: RSI-2 is one of the very few retail rules with a real per-trade edge in our tests, but as a stand-alone portfolio it has not beaten simply holding the S&P 500 since 2009.

What the RSI 2 strategy actually is

RSI stands for Relative Strength Index, a momentum gauge that runs from 0 to 100. Most charting software sets it to 14 days by default. Connors shortened it to 2 days, so it hits extreme readings after just a couple of sharp down closes.

StockCharts' ChartSchool lays out the version most people learn:

• Only look for buys when the stock is above its 200-day moving average. That's the trend filter.
• Buy when RSI(2) falls to 5 or lower. Some versions use 10, but ChartSchool notes that readings below 5 produced higher returns.
• Exit when price moves back above its 5-day moving average.
• Don't use a stop loss. ChartSchool says Connors found stops hurt performance on stocks and stock indices. It also warns that trading without stops can lead to large losses and deep drawdowns.

The rule spread widely through Connors' 2008 book Short Term Trading Strategies That Work. CXO Advisory's review of the book sums it up as a set of mean reversion tests on daily data from 1995 to 2007: buy pullbacks rather than breakouts, buy above the 200-day average, and buy fear rather than greed. TradingMarkets has published articles built on the same triggers, such as a 2010 piece by Richard Miller that reports win rates above 80% on a screened list of stocks. That article doesn't mention trading costs or drawdowns.

The idea has academic roots as well. A paper by Narasimhan Jegadeesh in the Journal of Finance (1990) found strong short-term reversal in individual stock returns: a stock's return one month tended to be followed by an opposite move the next month. That's a monthly effect, not a two-day one, so it doesn't prove RSI-2 works. It does show that the idea of buying losers to catch a bounce has been studied for decades.

CXO's review also pointed out weak spots in the book's testing. It found no correction for data mining, no out-of-sample test, and little attention to drawdowns or costs. Those are exactly the gaps our test is designed to cover.

How we tested it

Synergy Trading runs a private research lab that puts every strategy through the same checks: realistic costs, a random-entry control, a look-ahead check, and a correction for testing many variants. About 290 ideas have gone through it so far, and only 2 are clean survivors.

For RSI-2 we used point-in-time S&P 500 membership from 1999 to 2026, so the test includes companies that later dropped out or went bust. A signal only counted while the stock was in the index. Trades filled at the next day's open. Costs were 2 basis points per side, where a basis point is 0.01%. If a stock was delisted while a trade was open, we closed that trade at the stock's last price.

The random-entry control matters most. It takes the same trade at random times on the same eligible days, using the same trend filter, and repeats that 20 times. The result tells you how much luck alone would have made.

The per-trade result: a real edge

Over 60,107 trades the average net return was +0.218% per trade. The median trade, the one right in the middle, made +0.65%, and 63% of trades were winners. When the median sits well above the average, most trades win small and a smaller number of losers are big enough to drag the average down. That's the usual pattern for dip buying.

The weekly t-statistic was 5.19. A t-statistic measures how confident we can be that the average return isn't really zero. Researchers usually want to see about 2 or higher, so 5.19 clears the bar comfortably. The random-entry control made just +0.024% per trade. The real rule's result is about 34 standard deviations above the control, a z-score of about 34. Luck doesn't explain a gap that size.

The edge also held up over time. By era it averaged +0.175% per trade in 1999 to 2008, +0.216% in 2009 to 2018 and +0.26% in 2019 to 2026. Splitting the data at 2013 gives +0.27% and +0.18%, so both halves were positive.

There's a warning for anyone running their own backtest. When we tested only the stocks that are in the index today (survivors) over 2019 to 2026, the result jumped to +0.38% per trade. Survivor-only data overstated this edge by about 0.1% per trade, and many free backtests use exactly that kind of list.

Turning RSI 2 signals into a portfolio

Next we built a simple portfolio from the signals: 10 equal-weight slots, 1999 to 2026, with idle cash earning nothing. Below it's compared with SPY, the S&P 500 ETF, including dividends. CAGR is the average yearly growth rate. Sharpe ratio is return per unit of risk, so higher means a smoother ride for the same gain. Max drawdown is the worst fall from a peak to a low.

The book was invested 62% of the time. A random-pick control book reached a Sharpe of 0.23, and its 95th percentile was 0.29, so the real book's 0.74 passes clearly.

The catch: it has trailed SPY since 2009

Over the full period the two grew at almost the same rate, and the book took far less pain getting there. Split by era, though, the picture changes. The book's Sharpe was 0.97 in 1999 to 2008, 0.51 in 2009 to 2018 and 0.72 in 2019 to 2026. SPY's was 0.04, 0.81 and 0.91.

So the book's whole advantage comes from the bear-market decade. After 2009 it lagged plain SPY. In practice it worked mostly as a crash-avoider, sitting in cash while the market fell and missing part of the long bull run after that.

We also tested a 50/50 blend of SPY and the book, rebalanced monthly. It had a CAGR of 8.94%, a Sharpe of 0.72, a max drawdown of -37.4% and a 2008 return of -22.6%. Blending cut SPY's worst losses but didn't remove them.

A sister rule that failed

We tested a related idea with the same checks: buying pullbacks to the 200-day moving average. It made 2,969 trades with an average of +0.096% per trade. The median trade lost 1.08%, the win rate was 43% and the weekly t-statistic was 0.06, which is basically zero. In 1999 to 2008 it averaged -0.21% per trade. Verdict: dead. Rules that sound alike can test very differently.

What this test can't tell you

This test covers S&P 500 stocks from 1999 to 2026 and nothing else. We didn't test other markets, small caps or other time frames. Costs were set at 2 basis points per side with fills at the next open. Your broker, your order size and any slippage could make results worse. Idle cash earned 0% in the portfolio test. We started a paper forward test on 2026-11-03, and it doesn't have results worth quoting yet. Past results don't guarantee future ones.

Frequently asked questions

What are the basic RSI 2 rules?

The common version buys a stock trading above its 200-day moving average when its 2-day RSI falls to 5 or lower, then sells when price moves back above the 5-day moving average. Some versions use 10 instead of 5. Connors' original approach uses no stop loss.

Is RSI 2 better than buying and holding the S&P 500?

Not since 2009 in our test. Over 1999 to 2026 the 10-slot book had a higher Sharpe ratio (0.74 against 0.53) and a much smaller drawdown than SPY, but almost all of that advantage came from 1999 to 2008. After 2009, plain SPY did better.

Why do RSI 2 backtests online often look better than this?

Many backtests only use stocks that are in the index today. In our test, using survivors alone over 2019 to 2026 raised the result to +0.38% per trade, an overstatement of about 0.1% per trade. Leaving out costs, or skipping a random-entry control, can flatter results too.

Should you use a stop loss with RSI 2?

Connors reported that stops hurt performance on stocks and stock indices, so his rules leave them out. StockCharts warns that trading without stops can lead to large losses and deep drawdowns. It's a real tradeoff, and every trader has to weigh it against the risk they can take.

Sources

• StockCharts ChartSchool: RSI(2)
• CXO Advisory: A Few Notes on Short Term Trading Strategies That Work
• TradingMarkets: Buying Weakness and Selling Strength (Richard Miller, 2010)
• Jegadeesh, Evidence of Predictable Behavior of Security Returns, Journal of Finance, 1990 (EconPapers)

Want us to test a strategy you've seen online? Email info@synergytrading.net with the subject Verdict request. It's a waitlist with no price and no promised date, and nothing is for sale yet. Synergy Trading isn't a registered broker-dealer or investment adviser, and nothing here is a recommendation to buy or sell.

Educational research only, not investment advice. Past results do not guarantee future results.