Does RSI(2) Work? Oversold and Overbought Tested on S&P 500 Stocks
Does RSI(2) predict a bounce? We tested 568,460 signals across 483 S&P 500
stocks, 2006 to 2026.
STUDY 012
The short answer
When RSI(2) fell below 10 in a stock above its 200-day average, the stock rose a median 0.48% over the next 5
days, 0.19 percentage points more than on an ordinary day.
When RSI(2) rose above 90 in a stock below
its 200-day average, the stock rose a median 0.07%, 0.33 percentage points less than on an ordinary day.
Four of the eight comparisons we ran showed a clear gap; the other four did not.
“Clear” means the interval excluded zero under every test setting we ran. Overbought and oversold readings are tested on both sides of the 200-day average.
RSI(2) below 10 was followed by a slightly stronger 5-day returnRSI(2) below 10 means the stock has fallen hard over the last two days.In stocks above their 200-day average, the next 5 days were 0.19 percentage points better than an ordinary day.5 days later the stock was higher 56% of the time, against 54% on ordinary days.
RSI(2) above 90 was followed by a slightly weaker 5-day returnRSI(2) above 90 means the stock has risen hard over the last two days.In stocks below their 200-day average, the next 5 days were 0.33 percentage points worse than an ordinary day.5 days later the stock was higher 51% of the time, against 54% on ordinary days.
The 200-day average changed the oversold resultMany mean reversion rules, such as Connors’ RSI(2) rule, take oversold readings only in stocks above their 200-day average.After oversold, against an ordinary day: 0.19 percentage points more in stocks above the average,
0.05 percentage points more in stocks below it.The oversold result was clearer in stocks above their 200-day average.
An event study: it measures what followed each signal. No entry, exit, position size or trading cost is simulated; the mean reversion backtest of these signals is a separate study.
Every table and assumption behind the summary above
After RSI(2) fell below 10
Over the 10 trading days before the signal, stocks above their 200-day average had fallen a median
1.5% to 3.2%. Over the next 5 days they rose a median 0.48%, against +0.28% on an ordinary day: a gap of 0.19 percentage points.
Each panel is one trend and volatility state. Solid lines: medians; dashed: the middle 75% of days. Calm and volatile compare the stock’s average true range with its own median over the whole period, a label that uses hindsight.
After RSI(2) rose above 90
Over the 10 trading days before the signal, stocks below their 200-day average had risen a median
2.2% to 3.6%. Over the next 5 days they rose a median 0.07%, against +0.40% on an ordinary day: a gap of 0.33 percentage points.
Each panel is one trend and volatility state. Solid lines: medians; dashed: the middle 75% of days. Calm and volatile compare the stock’s average true range with its own median over the whole period, a label that uses hindsight.
Interpretation
Each line is one trend and volatility state. Above zero, signal days did better than ordinary days.
The gaps are small next to the usual spread of returns, the dashed lines in the charts above: the largest clear
gap was 0.33 percentage points over 5 days, measured across many thousands of signals.
A single trade can land far from it.
How we tested it
Question: after a stock’s RSI(2) crosses into or out of its zones, does
its median return over the next 5 trading days differ from ordinary days for the same stocks in the same
trend?
Part
Definition
Indicator
RSI(2) on daily closes, Wilder’s smoothing (the app’s RSI)
Entering a zone
The first close below 10 (or above 90) after a close outside it
Leaving a zone
The first close back above 10 (or back below 90)
Measured
From the signal day’s close, over the next 5 trading days
Compared with
Every eligible day of the same stocks on the same side of the 200-day average
Stocks
483 current S&P 500 members, 2006 to 2026, price returns
Leaving oversold happens on an up close, so the up day itself is outside the
measured window.
The design was set on 6 October 2026 in our internal study record, before any return was read. The record is internal, not a public preregistration. A logged preview on part of the data was run after the “leaving the zone” signals were added; no figure from it is used here.
The 568,460 signals are not 568,460 independent observations: signals repeat in the same
stock, the 5-day windows overlap, and stocks move together. The 95% intervals come from a calendar-block bootstrap
that resamples whole stretches of the calendar, rerun with several block lengths and designs. A gap is called
clear when every interval excluded zero on the same side. The eight comparisons are not adjusted for
being tested together. Signal days are part of the ordinary-day pool, as in the app’s event study, which makes each
gap slightly smaller.
Survivorship bias. This is not the historical index. It is today’s S&P 500 members projected backwards, so companies that failed or left the index after a deep fall are missing. That can materially favour a signal that buys falls.
Limitations
Today’s S&P 500 members projected backwards (survivorship bias); one period; price returns, dividends excluded.
An event study: no trading rule, cost or short sale is simulated, so a weaker return after an overbought reading tests no short-selling rule.
The calm and volatile label uses each stock’s whole-period median, which is known only in hindsight.
No adjustment for the market’s own move or for other risk factors.
RSI(14), ConnorsRSI and Stochastic RSI are tested the same way in their own studies.
How to reproduce this
In the Events panel, add RSI as a custom signal with period 2, build the condition “crosses below 10”, set the horizon to 5 days and the trend precondition to above (or below) the
200-day average, and run it on your own stocks on daily bars.
Appendix
All results
Signal
Stocks
Signal days
Median 5-day return
Ordinary days
Difference
95% interval
Result
Enters oversold (below 10)
Above 200-day
67,627
+0.48%
+0.28%
+0.19 points
+0.11 to +0.29
clear
Enters oversold (below 10)
Below 200-day
59,830
+0.45%
+0.40%
+0.05 points
−0.10 to +0.21
not clear
Leaves oversold (back above 10)
Above 200-day
67,829
+0.37%
+0.28%
+0.09 points
−0.01 to +0.19
not clear
Leaves oversold (back above 10)
Below 200-day
59,544
+0.40%
+0.40%
0.00 points
−0.15 to +0.15
not clear
Enters overbought (above 90)
Above 200-day
115,912
+0.16%
+0.28%
−0.12 points
−0.17 to −0.07
clear
Enters overbought (above 90)
Below 200-day
40,901
+0.07%
+0.40%
−0.33 points
−0.48 to −0.19
clear
Leaves overbought (back below 90)
Above 200-day
115,299
+0.23%
+0.28%
−0.06 points
−0.11 to 0.00
clear
Leaves overbought (back below 90)
Below 200-day
41,518
+0.22%
+0.40%
−0.18 points
−0.35 to 0.00
some settings
Checks set in advance
Checks set in advance, after 1 day
Above the 200-day average
Below it
Enters oversold (below 10)
+0.08 points (+0.05 to +0.11)
+0.03 points (−0.03 to +0.09)
Leaves oversold (back above 10)
0.00 points (−0.05 to +0.04)
−0.07 points (−0.15 to +0.01)
Enters overbought (above 90)
−0.05 points (−0.07 to −0.03)
−0.07 points (−0.13 to −0.03)
Leaves overbought (back below 90)
0.00 points (−0.02 to +0.02)
−0.03 points (−0.09 to +0.03)
In the backtest
In the mean reversion
backtest, which buys the dip above the 200-day average and sells after the first close above the 5-day average,
RSI(2) came first of four entries, with a median total return per stock of
+45% over 2006 to 2026.
Connors’ stricter line
Connors’ stricter line, 5, gave +0.22 points (+0.09 to +0.35) in stocks
above their 200-day average from 32,699 signals, against +0.19 points at 10. The stricter 5 line produced a similar result, but the test did not show that it was better than 10.
Using this research. The analysis, tables and figures on this page are
GU Analyser Ltd’s own work, published under a Creative Commons BY-NC 4.0 licence. You are welcome to quote it, cite it and build on it,
with credit and a link. Suggested credit: GU Analyser, The Test Bench, Study 012, linking to
guanalyser.com/test-bench/rsi-2-oversold-overbought/. For commercial use, ask us at
hello@guanalyser.com.
Rebuild the test
Build the RSI(2) condition in the Events panel, set the trend precondition and the horizon, and measure what followed it on your own stocks.
These are historical analyses. They are neither advice nor forecasts. They describe
what happened on past data under stated, simplified assumptions. Past results do not prove
what will happen next, and nothing in the Test Bench is a recommendation to buy or sell an
investment. GU Analyser is an analytical and educational tool — no money is ever
traded here.