RSI Explained: RSI(14) and RSI(2)
RSI explained: the calculation, 30 and 70, and the 2-day RSI read with 10 and 90, with our tests of both settings on S&P 500 stocks.
The Test Bench › Mean reversion
What mean reversion is, why prices revert and when they do not, how a rule picks its entry and exit, and the risks, with what our tests found.
Mean reversion trading buys a stock after a sharp fall, expecting part of the move to reverse, and sells once the price has recovered some of it. Trend following takes the opposite view. Three choices shape a mean reversion rule: how the fall is measured, when the trade is closed, and which stocks are allowed in.
This page covers:
Mean reversion is the idea that a price which has moved far from its recent average may move part of the way back. A mean reversion strategy expects an unusually strong move to partly reverse. It may buy after a fall or sell short after a rise, then exit with a rule designed to capture some of the reversal.
The trades are usually short-lived, often a few days. Some mean reversion rules produce many small gains and fewer, larger losses: in our backtest, the four rules won 65% to 67% of their trades. The losses come from the falls that keep going. Because the rule waits for an extreme, it also spends most of its time in cash, so its return should be compared with how long its money was actually invested.
The idea is applied in other forms too: to the price gap between two related stocks (pairs trading), to whole indices, and on intraday bars. This page covers single stocks on daily bars, the setting of our tests.
One explanation is that short-term moves can overshoot. A burst of selling, a reaction to news or a fund cutting a position can push a price further than the news justifies, and part of the move comes back once the pressure passes. A mean reversion rule tries to buy at that point.
It fails when the fall has a lasting cause: a collapse in earnings, a change in the business, or a market-wide sell-off that keeps going. Then the price does not come back, and the rule holds a falling stock.
Trend following works the other way: it buys a stock that is rising, such as a close at a new 20-day high, and holds while the move continues. It often wins fewer trades and relies on a few large moves. Our Donchian breakout study tests a trend-following rule.
| Mean reversion | Trend following | |
|---|---|---|
| Buys | After a fall | After a rise |
| Typical signal | RSI(2) below 10 | A close at a new 20-day high |
| Typical hold | Days | Weeks to months |
| Wins | Can win often, with smaller gains | Often relies on fewer, larger gains |
| Loses on | Falls that keep going | Breakouts that reverse |
| Expects | The move to fade | The move to continue |
The two can sit in the same rule: many mean reversion rules buy short-term weakness only inside a longer uptrend.
The entry needs a way to say a stock has fallen far enough. The common choices are a short oscillator reaching an extreme, a close well below a moving average, a close below the lower Bollinger Band, or a run of down days. Our tests use four oscillators from the RSI family:
| Signal | Oversold line | What it measures | Trades in our backtest (typical stock, 20 years) |
|---|---|---|---|
| RSI(2) | Below 10 | Up and down closes over about two days | 132 |
| RSI(14) | Below 30 | Up and down closes over about three weeks | 6 |
| ConnorsRSI | Below 10 | A 3-day RSI, the down streak and how unusual today’s fall is | 35 |
| Stochastic RSI | Below 20 | Where RSI(14) sits in its own 14-day range | 209 |
A faster signal reaches its oversold line more often; a slower one reaches it only after a longer or larger decline. Each signal’s page explains its calculation with a worked example: RSI and RSI(2), ConnorsRSI and Stochastic RSI. A rule also chooses when to act: on the day the signal enters its zone, or on the day it leaves it, after the first up close.
Mean reversion exits are usually short, because the trade was opened for a rebound. The common ones:
Many rules take a signal only when the close is above its 200-day moving average, a sign of a long-term uptrend. The filter aims to keep the rule out of stocks in a long decline, where falls have more room to continue, at the cost of fewer trades.
A short-term rule trades often, so each trade’s cost is paid many times. To judge the result, compare it with buying and holding the same stock over the same period: a rule that is in the market for a small part of the year can win most of its trades and still end far behind a stock held throughout.
Each signal’s page explains the calculation with a worked example, compares it with its closest relative, and lists our tests of it.
RSI explained: the calculation, 30 and 70, and the 2-day RSI read with 10 and 90, with our tests of both settings on S&P 500 stocks.
ConnorsRSI explained: its three parts, how each is calculated, and the 10 and 90 zones, with our tests on S&P 500 stocks.
Stochastic RSI explained: how it places RSI in its own range, %K and %D, and the 20 and 80 zones, with our tests on S&P 500 stocks.
We tested four RSI signals as the entry of one dip-buying rule on 483 S&P 500 stocks. RSI(2) had the highest median total return per stock, +45% over 2006 to 2026; buying and holding returned +361% and was invested every day.
RSI(2) crossing 10 and 90 on 483 S&P 500 stocks, 2006 to 2026: 4 of 8 signal tests differed clearly from ordinary days over the next 5 days.
RSI(14) crossing 30 and 70 on 483 S&P 500 stocks, 2006 to 2026: 3 of 8 signal tests differed clearly from ordinary days over the next 20 days.
ConnorsRSI crossing 10 and 90 on 483 S&P 500 stocks, 2006 to 2026: 4 of 8 signal tests differed clearly from ordinary days over the next 5 days.
Four RSI entries under one dip-buying rule (above the 200-day average, out on a close above the 5-day average) on 483 S&P 500 stocks, 2006 to 2026, against buying and holding.
Stochastic RSI crossing 20 and 80 on 483 S&P 500 stocks, 2006 to 2026: no clear difference from ordinary days over the next 10 days in any of 8 signal tests.
A strategy that buys after a fall, or sells after a rise, on the expectation that the price will move back towards its recent average. It trades against the latest move and usually holds for days.
Mean reversion buys weakness and expects part of the move to reverse; trend following buys strength and expects it to continue. Mean reversion rules often win more of their trades with smaller gains; trend-following rules often win fewer and rely on a few large moves.
Indicators that mark a short-term extreme: oscillators such as RSI(2) and ConnorsRSI below 10, RSI(14) below 30 or Stochastic RSI below 20, a close far below a moving average, a close below the lower Bollinger Band, or several down days in a row. Most rules add a long-term trend filter such as the 200-day moving average.
The common exits are a close back above a short moving average (the 5-day in Connors’ RSI(2) rule), the oscillator moving back past a level, or a fixed number of days. A short exit closes the trade once part of the fall has recovered.
The falls that keep going. A stock bought after a drop can keep dropping when the cause lasts, and a rule without a stop holds it all the way. Frequent trading also pays costs often, and a rule that is mostly in cash can trail a stock held throughout.
Falls are often larger and longer in a bear market, so more dips keep going. Many rules use a long-term trend filter for this reason and stand aside while a stock trades below its 200-day moving average.
A condition that takes a signal only when the stock’s close is above the average of its last 200 closes, a common sign of a long-term uptrend. Mean reversion rules use it to buy short dips only in stocks above the average and to stay out of stocks in a long decline.
In our backtest of four RSI-based entries on 483 S&P 500 stocks, every rule finished with a positive return on most stocks and the rules won 65% to 67% of their trades, but none beat buying and holding on more than 61 of the stocks. The rules were invested only 0.5% to 14% of the time, against 100% for buying and holding. Against a position of the same average size, a comparison we added after the main results, every rule came out ahead on most stocks. The rules produced positive median returns under the stated assumptions; the result does not show that mean reversion is a better general approach than holding stocks.
No single answer holds for every rule. RSI(2) ranked first in this particular four-entry backtest on 483 S&P 500 stocks, with the highest median total return per stock over 2006 to 2026 as an entry (+45%), followed by Stochastic RSI (+44%), ConnorsRSI (+13%) and RSI(14) (+5%). The ranking applies to that rule — the same trend filter, exit and cost for every entry — these stocks and this period.
Often days, though no holding period is universal. Connors’ RSI(2) rule sells on the first close above the 5-day moving average, which often comes within a week; rules built on slower signals wait for slower moves.
Yes. In GU Analyser you choose the oversold signal, the trend filter and the exit on a visual canvas and run a backtest or an event study on daily bars, with no code.
Want to test your own mean reversion rule?
Choose the oversold signal, the trend filter and the exit, and measure what followed the signal across your stocks, or backtest the rule with a cost on each trade.
Create a free account and build your own strategies → or read how to backtest a strategy without code →The methods: how a backtest is built · how an event study works