RSI and Momentum Indicators: What They Measure and What They Miss

In physics, momentum is mass times velocity. In markets, price momentum is the rate of change of price over a window of time. If EURUSD moves from 1.0800 to 1.0900 in five days, that is +100 pips of change, or about +0.93%. The first derivative of price – how fast it is moving – is what every momentum indicator attempts to capture and visualise.

Momentum indicators share a family resemblance: they take recent closing prices, compare recent gains to recent losses, and produce a bounded line that oscillates between a floor and a ceiling. Because the line is bounded (most cap at 0 and 100), it appears "reversible" – price cannot go to infinity, but an oscillator near 100 must come back down eventually. That intuition is the source of both the appeal and the danger of these tools.

The four indicators covered below – RSI, Stochastic, CCI, and Williams %R – all measure the same underlying quantity (recent momentum normalised to a range) using slightly different denominators. In practice their signals correlate 0.75–0.95, so stacking four of them on a chart adds visual confidence but almost no new information.

The Relative Strength Index was published by J. Welles Wilder Jr. in New Concepts in Technical Trading Systems (1978). The formula:

RSI = 100 − (100 ÷ (1 + RS)), where RS = Average Gain ÷ Average Loss over the last N periods.

Default N is 14 periods. Average gain is the mean of up-close magnitudes; average loss is the mean of down-close magnitudes. Both use Wilder's smoothing (similar to an exponential moving average with α = 1/N), so RSI reacts smoothly to new data rather than jumping on every bar.

Because RS is a ratio of two positive numbers, RSI is mathematically bounded between 0 (all losses) and 100 (all gains). When average gain equals average loss, RS = 1 and RSI = 50. When gains are three times losses, RS = 3 and RSI = 75. The boundedness is a feature – it lets traders compare momentum across instruments without re-scaling – but it is also the reason traders wrongly expect "reversion to the middle".

The conventional reading – RSI >70 is overbought, RSI <30 is oversold – is the single most misused idea in technical analysis. Wilder himself wrote that these levels were meant as alert thresholds, not trade triggers. In a strong uptrend, RSI can sit above 70 for weeks: any trader who shorted Bitcoin every time its daily RSI crossed 70 during the 2020–2021 bull run would have bled to zero.

Empirical studies confirm this. Cheol-Ho Park and Scott H. Irwin's 2007 meta-study "What Do We Know About the Profitability of Technical Analysis?" reviewed 95 modern studies of technical rules and found that simple overbought/oversold RSI rules produced no statistically significant excess returns after transaction costs on major equity indices post-2000. The rule works in range-bound markets and fails in trending ones – a regime-dependence that oscillator-only strategies cannot solve.

A more robust reading: extreme RSI values indicate conditions, not directions. RSI >80 means "recent gains have dominated" and is a valid warning to tighten stops, but it is not a short signal without additional context – a trend-exhaustion pattern, a failed breakout, a divergence.

Divergence occurs when price makes a new high (or low) but the oscillator does not. The market is making a fresh extreme on the surface, but the internal momentum has weakened. Four types exist:

Regular bearish divergence – price prints a higher high, RSI prints a lower high. Suggests the up-leg is running on fumes and a reversal is possible.

Regular bullish divergence – price prints a lower low, RSI prints a higher low. Selling pressure is fading even as price explores new lows.

Hidden bearish divergence – price prints a lower high, RSI prints a higher high. Typically a trend-continuation signal within a downtrend (the bounce lacks momentum).

Hidden bullish divergence – price prints a higher low, RSI prints a lower low. Trend-continuation within an uptrend – pullback losing steam.

Academic backtests of divergence signals (Lento 2007; Marshall, Young & Rose 2006) find weak but non-zero edge on equity indices and FX majors – typically Sharpe 0.3–0.5 when used as a standalone signal, improving to 0.8–1.2 when combined with a trend filter. The divergence is useful because it describes deceleration, which is causally linked to reversals in a way that a raw level like "RSI 70" is not.

The four classic momentum indicators differ in their internals but share >0.8 correlation on daily FX data. A quick field guide:

Notice that Stochastic and Williams %R are mathematically related – they both normalise the close within the recent high-low range. CCI is the outlier in that it is unbounded and measures deviation from a moving average rather than up/down balance. For traders working through a simulated evaluation, using two or three of these simultaneously is redundant; pick one and learn its quirks deeply.

The bar chart below shows the empirical distribution of 14-day RSI readings on two simulated regimes across 2,000 daily closes – a strong uptrend (annualised return +25%, vol 15%) and a mean-reverting range (no drift, vol 15%). Notice how the uptrend shifts the whole distribution rightward: RSI >70 occurs on more than 25% of days in the trend regime, vs ~5% in the range regime. Using the same 70/30 thresholds across both regimes is a category error.

Practical implication: use moving averages or an ADX reading to classify the regime first. In a ranging regime (ADX <20), treat RSI 70/30 as reversal alerts. In a trending regime (ADX >25), ignore overbought/oversold and trade only divergences or trend-continuation pullbacks.

The Stochastic oscillator, popularised by George Lane in the 1950s, answers a slightly different question: where is the close sitting within the recent N-period high-low range? If price made a high of 1.2000 and a low of 1.1900 in the last 14 bars and closed at 1.1990, Stochastic %K = (1.1990 − 1.1900) ÷ (1.2000 − 1.1900) = 90, meaning close is in the top 10% of the range.

The raw %K is noisy, so most platforms show %D – a 3-period SMA of %K – as a signal line. Crosses of %K and %D (analogous to MACD crosses) are the most common signal. The full "slow stochastic" applies an additional 3-period SMA to %K before computing %D, producing a smoother line at the cost of lag.

Stochastic shines when markets are range-bound because it inherits the range directly in its denominator. In a strong trend it pegs at 100 (or 0) for long stretches and becomes a useless line on a chart. That is not a bug; it is the indicator telling you "no range to mean-revert to".

The Commodity Channel Index, developed by Donald Lambert in 1980, takes a different approach: it measures how far the current price sits from its moving average, scaled by the average absolute deviation. The 0.015 constant is chosen so that 70–80% of CCI values fall between −100 and +100 in most markets – a heuristic, not a mathematical truth.

CCI is unbounded, which means in a powerful trend it can print values of ±300 or more. Two common uses: (1) break of ±100 as trend-entry – entering long when CCI crosses above +100 from below, exiting when it crosses back through +100 from above, and (2) zero-line crosses as a coarser trend filter. Pure overbought/oversold readings (±200) fail as standalone signals in trending markets for the same reason RSI does.

Colby & Meyers' Encyclopedia of Technical Market Indicators (2002) tested CCI rules across 15 US equities and found break-of-100 rules produced modestly positive but not statistically significant returns after commissions. Like all oscillators, CCI is a component in a system, not a system.

1. Pick one, not three. Stacking RSI, Stochastic and CCI adds visual noise and almost no information. Most pros choose RSI for divergence or Stochastic for range-timing and stick with it.

2. Always classify regime first. Is this a trend or a range? ADX, higher-timeframe moving-average slope, or simple "is price above or below the 200-SMA" all work. Only switch into oscillator-triggered mean-reversion trades in ranges.

3. Use multi-timeframe alignment. A 1-hour RSI bullish divergence is meaningless against a daily downtrend. Require divergence to appear on the timeframe you trade AND absence of a stronger opposing signal one timeframe higher.

4. Treat extreme readings as warnings, not triggers. RSI 85 tells you recent buying was relentless; it does not tell you when buying stops. Enter on a pattern (failed test of the high, bearish engulfing bar, supply-zone reaction), not on the oscillator alone.

5. Log the edge honestly. Backtest your divergence or threshold rules over 500+ trades using walk-forward analysis. If the edge does not survive out-of-sample, it was curve-fit to your eyeball.

6. Combine with structure, not with more oscillators. Support and resistance, prior swing levels, volume profile POCs – these are causally linked to price behavior. Another oscillator is just a laggy echo of the first.

Mistake 1 – Fading strong trends. Shorting every RSI >70 in a bull market is the statistical definition of suicide-by-drift. In 2020–2021 this trade blew up thousands of retail accounts in BTC and ES futures.

Mistake 2 – Cherry-picked divergence. The human eye finds divergences everywhere in hindsight. A rigorous backtest with pre-defined pivot detection and minimum separation gives drastically lower divergence counts than visual inspection suggests.

Mistake 3 – Wrong period. Shortening RSI from 14 to 5 makes it whip around every bar. Lengthening to 21 makes it sluggish. 14 is a Wilder default, not a law; but at minimum backtest any change.

Mistake 4 – Treating the 50-line as support. RSI 50 is noise in a sideways market and a weak trend-continuation zone in a trend – but it is not a tradable level on its own, only in combination with price structure.

Mistake 5 – Using RSI on heikin-ashi or Renko. Smoothed price types feed already-smoothed data into Wilder's smoothing, producing a flat, useless RSI line. Use on standard candles.