RSI Indicator Strategy Guide

Q: What is the best RSI setting?

It depends entirely on the use. RSI(2) or RSI(3) for short-term mean reversion, RSI(14) as a general context measure, and longer settings for smoother momentum readings. The standard 14 is a reasonable default precisely because it has not been optimised to your data.

Q: Does RSI overbought mean I should sell?

No. In an uptrend, RSI spends long periods above 70 because recent gains genuinely exceed recent losses. Selling on that basis means fading strength, which is the opposite of what trend evidence supports. Overbought readings are useful as exhaustion signals only within a mean-reversion framework and only with a trend filter.

Q: Does the RSI(2) strategy still work?

The underlying effect, short-term reversal in index products within an uptrend, remains measurable, but returns have compressed since the approach was widely published. It also carries the mean-reversion tail risk: rare trades that do not revert can erase many small winners. Test it with the trend filter, a time stop, and a disaster stop included.

Q: Is RSI divergence reliable?

It is difficult to evaluate because it is rarely defined precisely enough to test, and informal versions appear frequently during trends without reversals. Treat it as a supporting observation within a tested framework rather than as a signal, and be sceptical of any presentation that only shows the instances that worked.

Q: Can I use RSI on any market or timeframe?

The calculation applies anywhere, but the thresholds must be recalibrated. A reading of 25 may be a meaningful extreme for a stable index and entirely routine for a volatile crypto asset. Examine the historical distribution of RSI values for your specific instrument and timeframe before choosing thresholds.

At a glance

What it measures: Average gains versus average losses over a lookback
Range: 0 to 100, bounded
Standard setting: 14 periods, with 70 and 30 as conventional thresholds
Best documented use: Short lookbacks (2 to 4) for mean reversion with a trend filter

Key takeaways

RSI is a bounded oscillator, so it reaches extremes routinely in trends and can stay there for months.
Selling because RSI reads 70 in a strong uptrend is one of the most reliable ways to lose money with an indicator.
Short-lookback RSI, typically RSI(2), combined with a 200-day trend filter, is the best-documented systematic use.
Divergence is popular, visually compelling, and difficult to define objectively, which makes it hard to test and easy to see in hindsight.
RSI is a timing tool within a context established by something else; it cannot establish the context itself.

How RSI is calculated

Over the lookback period N:
   Average gain = mean of up-day changes
   Average loss = mean of down-day changes (as a positive number)

   RS  = Average gain / Average loss
   RSI = 100 - (100 / (1 + RS))

Implications:
   All up days   -> average loss = 0  -> RSI = 100
   All down days -> average gain = 0  -> RSI = 0
   Equal gains and losses            -> RSI = 50

Shorter N makes RSI far more sensitive:
   RSI(14) rarely goes below 20
   RSI(2)  reaches below 5 regularly

The formula, which explains the behaviour.

The bounded nature is the source of most misuse. Because RSI cannot exceed 100, a sustained uptrend pushes it into the 70s and 80s and holds it there. The indicator is correctly reporting that recent gains dominate recent losses, which is what an uptrend is. It is not reporting that a reversal is due.

Why 70 and 30 are poor signals alone

The conventional thresholds were proposed for a 14-period setting in markets of a different era. Used as standalone reversal signals they perform poorly, for a structural reason: the strongest trends produce the most extreme readings, so the signal is most frequent exactly when fading it is most dangerous.

Use of RSI extremes	Typical outcome	Better approach
Short when RSI(14) > 70	Shorting strong uptrends	Use extremes only with a trend filter
Buy when RSI(14) < 30	Buying into sustained declines	Require price above a long moving average
Exit longs at RSI 70	Cutting winners in strong trends	Use a trailing exit instead
RSI > 70 as a strength confirmation	Reasonable in momentum systems	Treat extremes as strength, not exhaustion
RSI(2) < 10 above the 200-day MA	Documented mean reversion setup	This is the tested application

The RSI(2) mean reversion system

The best-documented systematic use of RSI employs a very short lookback to identify short-term extremes within an established uptrend.

Universe: Liquid index ETFs, or large-cap stocks with high average dollar volume. Avoid illiquid names where the bounce is consumed by the spread.
Trend filter: Close above the 200-day simple moving average. This is not optional; removing it produces a fundamentally different and much worse strategy.
Entry: Buy at the close when RSI(2) is below 10. Some versions use below 5 for fewer, stronger signals.
Exit: Sell at the close when RSI(2) rises above 70, or when price closes above the 5-day moving average.
Time stop: Exit unconditionally after 8 to 10 trading days.
Disaster stop: 3 x ATR(10) below entry. Rarely triggered; it is what makes the tail survivable.
Position size: Risk 0.5 percent of equity; maximum 3 to 4 concurrent positions, because signals cluster on the same days.

Typical characteristics: win rate of 65 to 75 percent, average win smaller than average loss, modest expectancy, and vulnerability to sustained declines. It is a mean reversion strategy and carries that family’s tail risk. Note also that it is widely published, so returns have compressed relative to earlier tests.

Divergence, honestly assessed

Divergence occurs when price makes a new high while RSI makes a lower high, which is interpreted as weakening momentum. It is visually compelling and appears frequently in hindsight at major turning points.

Definition problems. How much lower must the RSI high be? Over what window? Must the highs be consecutive? Different answers produce completely different signal sets, which is why published tests disagree.
Frequency problems. Divergence occurs many times during a strong trend without any reversal. The memorable instances are the ones that preceded turns.
Hindsight bias. After a top, the divergence is obvious. Before it, several divergences failed and one worked, and they looked identical.
Where it may help. As one input within a mean-reversion framework that already has a trend filter and a defined stop, rather than as a standalone reversal signal.
How to evaluate it. Define it numerically, code it, and measure the base rate. If you cannot define it precisely enough to code, you cannot know whether it works.

Practical guidelines

Choose the lookback for the job: 2 to 4 for mean reversion, 14 for general context, 20 or more for smoother momentum measurement.
Always pair RSI with a trend filter that comes from a different measurement, usually a long moving average.
Do not use RSI for exits in trend strategies; it will cut winners early by design.
Adjust thresholds by instrument. A level that is extreme for an index ETF is routine for a volatile stock.
Beware of stacking RSI with Stochastic and MACD: they measure overlapping properties and agreement between them is not independent confirmation.

Frequently asked questions

What is the best RSI setting?

It depends entirely on the use. RSI(2) or RSI(3) for short-term mean reversion, RSI(14) as a general context measure, and longer settings for smoother momentum readings. The standard 14 is a reasonable default precisely because it has not been optimised to your data.

Does RSI overbought mean I should sell?

No. In an uptrend, RSI spends long periods above 70 because recent gains genuinely exceed recent losses. Selling on that basis means fading strength, which is the opposite of what trend evidence supports. Overbought readings are useful as exhaustion signals only within a mean-reversion framework and only with a trend filter.

Does the RSI(2) strategy still work?

The underlying effect, short-term reversal in index products within an uptrend, remains measurable, but returns have compressed since the approach was widely published. It also carries the mean-reversion tail risk: rare trades that do not revert can erase many small winners. Test it with the trend filter, a time stop, and a disaster stop included.

Is RSI divergence reliable?

It is difficult to evaluate because it is rarely defined precisely enough to test, and informal versions appear frequently during trends without reversals. Treat it as a supporting observation within a tested framework rather than as a signal, and be sceptical of any presentation that only shows the instances that worked.

Can I use RSI on any market or timeframe?