Most indicator strategies are folklore. RSI-2 is one of the few with a proper pedigree: it was popularised by Larry Connors, who published the rules and the stats instead of just the sales page. The pitch is simple, in a rising market, sharp two-or-three-day sell-offs are usually noise, and buying that panic has historically been paid.
The internet version of RSI-2 quotes win rates north of 80% and leaves out everything else. Let's define the actual rule, then run it live with costs on and let you see the whole scoreboard, win rate and what the wins were worth.
RSI in one paragraph
The Relative Strength Index measures how one-sided recent price action has been, on a scale of 0–100. Lots of up-days pushes it toward 100 (overbought); lots of down-days pushes it toward 0 (oversold). The standard setting is 14 days, which makes it a slow, smooth gauge, decent for describing a trend, terrible for timing a two-day dip. RSI-2 uses a 2-day lookback instead: hyper-reactive on purpose. It pins near 0 after a couple of hard red days and near 100 after a couple of green ones. That twitchiness is useless as a trend gauge and exactly right as a panic meter.
The rule
The classic Connors formulation:
- Only trade with the long-term trend. The market must be above its 200-day moving average. You're buying dips in uptrends, not catching knives in bear markets. Skip this filter and the strategy's history gets much uglier. (In our tool this is the Moving average stack rule under Chart setup, and we run it stricter than Connors' original: it wants the full 20 > 50 > 100 > 200 average stack at entry, i.e. an unambiguous uptrend. The link below opens with it already on.)
- Buy the panic. When the 2-day RSI closes below 10 (some traders use 5 for fewer, deeper signals), buy.
- Sell the bounce. Exit when the RSI closes back above 70, the panic has washed out and the snap-back you were paid to catch has happened. Typical hold: a few days.
Note what's not in the rule: no profit target in percent, no discretionary chart reading. It's mechanical, which means it's testable.
Why it has worked
Equity indexes have spent most of the last few decades mean-reverting on a multi-day horizon: sharp short dips inside rising markets got bought. RSI-2 is just a precise way of defining "sharp short dip". The strategy's character follows from that:
- High win rate, small wins. Most dips bounce, so most trades win, but each win is a modest few-day move. The quoted 80%+ win rates are real and misleading: a pile of small wins can be erased by a few large losses.
- The losses are the story. When a dip doesn't bounce, when the two red days are the start of a real break, RSI-2 is holding through it. The average loss is a multiple of the average win. This is the number the marketing leaves out, and it's why the trend filter matters so much.
- Costs bite hard. A strategy of many small trades pays fees on every one. A rule that's marginally profitable with free fills can be a steady loser at retail costs, so we test with fees on.
The live backtest
Here's RSI-2 on the S&P 500 ETF: 2-day RSI, buy below 10, sell above 70, the uptrend entry filter on, fees and slippage modelled. Run live by our engine on real prices. Click through and the tool opens with this exact configuration loaded:
RSI-2 reversion on SPYloading live result…Past result of the mechanical rule, fees on, not a prediction.Open this exact setup in the backtesterHow to read it honestly:
- Win rate versus expectancy. A high win percentage with a mediocre total return means the losers are doing the damage, exactly the RSI-2 signature. If you add a stop-loss in the tool, the expectancy-in-R figure appears and tells you the average result per unit risked, which is the number that actually decides whether a system makes money.
- Time in market. RSI-2 is in cash most of the time. Its return should be judged against that: being flat 80% of the time while the index compounds is a real cost, and it's why the benchmark line often wins the raw-return race.
- Try the last five years specifically. Mean reversion's golden era was 2000–2010. Whether the edge has decayed since is the live question, and a live backtest is the right way to keep asking it.
Tweaks worth testing
All of these are one field in the RSI-2 backtester:
- Buy threshold 5 instead of 10, fewer, deeper panics. Historically better per-trade, fewer trades.
- Exit above 90 instead of 70, hold the bounce longer, give more of it back.
- A stop-loss: Connors' original had none ("the exit is the RSI"), which is exactly what produces the rare large loss. Adding an ATR stop changes the character completely; test both and look at the drawdown line, not just the return.
- A single stock instead of the index, individual names mean-revert less reliably than the index that averages them. The same rule on your favourite ticker is usually a different, worse strategy. Prove it to yourself in one run.
The honest verdict
RSI-2 earned its reputation in an era when buying index dips was the most reliable trade in the world, and it remains the cleanest teaching example of what mean reversion is: many small wins, rare ugly losses, most of your money made in a few days a year. Whether it still clears costs today, on the ticker you'd trade it on, is not something an article from 2008, or this one, can settle. The run above is live. Change the thresholds, put the fees at your broker's, and let the engine answer for your case.