RSI Strategies

The Relative Strength Index (RSI) is a bounded momentum oscillator introduced by J. Welles Wilder Jr. in 1978. It compares the size of recent gains to the size of recent losses over a lookback window — 14 periods in the original formulation — and normalizes the result onto a 0–100 scale. An RSI strategy is any rule set that converts that number into a decision: an entry when the reading falls below a threshold, an exit when it crosses back above one, a requirement that the oscillator agree or disagree with price, or a permission filter layered on top of a separate trigger.

The versions catalogued on this page span beginner teaching material, packaged indicator suites, discretionary chart methods and systematically backtested rule sets. They differ in lookback, thresholds, instrument, timeframe and confirmation logic, but all of them reduce to one question: how should normalized momentum be read?

## How the calculation behaves

RSI = 100 − 100 / (1 + RS), where RS is the average gain divided by the average loss across the lookback. Wilder's smoothing is the standard; some platforms substitute a simple or exponential average, which yields slightly different readings on identical data.

Two properties drive most design choices. First, the oscillator is bounded, so it compresses information at the extremes — a strong trend can hold RSI near 80 for weeks without producing the reversal an "overbought" reading is supposed to imply. Second, the value is relative to its own lookback, not to any absolute price level: a reading of 30 on a 2-period RSI describes a two-bar pullback, while 30 on a 14-period RSI describes a multi-week decline. The same number means different things.

## Main variants

**Mean reversion.** The largest family: buy when RSI drops under a lower threshold, exit when it recovers. Short lookbacks (2 to 5 periods) with aggressive thresholds belong here, as do the classic 14-period oversold/overbought readings.

**Momentum and trend continuation.** The inverse reading — treating a high RSI as strength rather than exhaustion, or using the 50 line as a regime boundary. Crossings of 50 or 60 serve as directional confirmation rather than fade signals.

**Divergence.** Comparing the slope of price highs and lows against the slope of RSI highs and lows to flag weakening momentum. Implementations vary widely in how a swing point is defined and how long a divergence remains valid.

**Adaptive or derived bands.** Instead of fixed 70/30 lines, thresholds are computed from the oscillator's own recent distribution, from volatility, or from bands drawn around the RSI series itself.

**RSI as a filter.** Here the oscillator does not generate the trade. A candlestick pattern, a breakout, a moving-average cross or a volatility trigger does, and RSI only gates which of those signals are taken.

**Non-standard inputs.** The oscillator applied to a smoothed price series, to an intermarket spread between two instruments, or paired with related oscillators (money flow, vigor, choppiness) that share the same normalization idea.

## What differentiates implementations

The entry threshold gets most of the attention, but the exit usually does more work. The same oversold entry paired with a fixed-bar exit, a threshold-crossing exit, a moving-average exit or a trailing stop produces four strategies with materially different behaviour. Beyond that: lookback length, whether both sides are traded or only longs, the instrument's natural character (index ETFs, single stocks, FX and crypto do not mean-revert alike), whether signals are evaluated intrabar or on the close, cost assumptions, and whether losing positions are added to rather than cut.

## Common mistakes

Treating 70/30 as universal constants rather than one arbitrary pair among many. Fading a strong trend because the oscillator is extreme, which is precisely the condition a trending market produces. Leaving the exit undefined, so results depend on an unstated assumption. Evaluating signals intrabar in a way that could not have been executed live. Stacking several momentum indicators — RSI, stochastics, money flow, rate of change — and mistaking their agreement for independent confirmation when they measure nearly the same thing. Reading divergences in hindsight, where the swing points are obvious only after the fact. And tuning lookback and thresholds on a single instrument over a single sample until the numbers look agreeable.

## How to evaluate a version

State the rules completely before testing: input series, lookback, smoothing method, thresholds, entry timing, exit condition, stop, position size. Anything left implicit will be filled in silently by the backtest engine, and the result will describe that engine's assumption rather than the idea.

Then test for robustness rather than a single number. Scan the parameter surface and look for a broad plateau instead of an isolated peak — a rule that works at RSI(2) below 7 but not below 6 or 8 is fitted to the sample. Reserve out-of-sample data or run walk-forward. Re-test with realistic commissions and slippage, which weigh disproportionately on short-lookback, trade-heavy mean reversion. Compare against baselines: buy-and-hold, and the same exit rule triggered by a random or unconditional entry — if the RSI condition adds nothing over that, the exit is doing the work. Finally, segment results by regime and by year, and inspect the distribution of individual trade outcomes rather than the average alone.

Strategies in this concept (177)

Frequently asked questions

Is RSI a leading or a lagging indicator?

It is derived entirely from past closes, so it lags in the strict sense — it cannot know anything price has not already done. It is often called "leading" because extreme readings sometimes precede a turn, but that is an interpretation of the reading, not a property of the calculation. Treat it as a description of recent momentum, not a forecast.

What RSI period and thresholds should I use?

There is no setting that is correct across instruments and timeframes. The lookback determines what kind of event the oscillator measures — short periods react to a few bars, longer periods to multi-week moves — and thresholds only have meaning relative to that lookback. Pick a period that matches the holding horizon you intend to trade, then verify that nearby parameter values behave similarly rather than choosing the single best-scoring pair.

Does RSI work better for mean reversion or for trend following?

Both readings appear throughout the catalog and they are not compatible: one buys weakness, the other buys strength. Which fits depends on the instrument's behaviour over the period tested — broad index products have historically shown more short-term mean reversion than trending single names or commodities. The honest answer is that the market being traded decides, so it needs to be tested per instrument rather than assumed.

Is RSI divergence reliable enough to trade on its own?

Divergence is hard to specify precisely, which is the core difficulty. Before it can be tested at all you must define what counts as a swing point, how far apart the two points may be, and how long the divergence stays active. Different definitions produce very different signal sets from the same chart, so a divergence rule should be written out as code and backtested rather than judged visually.

Can RSI be used alone, or does it need confirmation?

A complete strategy needs an entry, an exit, and a risk rule; RSI can supply the entry trigger by itself, but not the rest. When adding confirmation, prefer inputs that measure something different — trend direction, volatility, volume, or session/time context — since another momentum oscillator will largely repeat the information RSI already carries.

Why do two strategies with the same RSI entry produce different results?

Usually the exit, the costs, or the execution timing. A fixed-bar exit, a threshold-crossing exit and a trailing stop applied to identical entries are three separate strategies. Signal timing also matters: acting on an intrabar reading versus the confirmed close changes which trades exist at all, and short-lookback systems trade often enough that commission and slippage assumptions can dominate the comparison.

All strategy concepts · Explore all strategies