Momentum Oscillators

A momentum oscillator summarizes how a recent move is behaving

Momentum is the tendency of a price move to continue in its current direction. An oscillator does not observe that tendency directly. It transforms a chosen slice of price history into a compact reading about change, speed, gain-loss balance, or location within a range.

It is called an oscillator because the transformed value rises and falls around a reference scale rather than tracing price itself. Some oscillators move around zero, some are bounded between fixed limits such as 0 and 100, and some are unbounded but interpreted relative to a center. The scale determines what a level means, so the number cannot be read without knowing the calculation.

The result is always conditional on the calculation. A ten-bar Rate of Change answers a different horizon from a fifty-bar version. A stochastic oscillator can sit near the top of its range even when the percentage return is small. Two oscillators can therefore disagree without either calculation being wrong.

Oscillators are also different from trend filters. A trend filter usually asks where price sits relative to a smoothed reference or which direction that reference points. An oscillator asks how forceful, extended, recent, or range-bound the move appears. Price can remain above a rising moving average while a momentum oscillator falls because the advance is continuing at a slower pace.

Most oscillator mistakes come from turning a description into a prediction. “RSI is 75” describes the balance of gains and losses under one lookback and smoothing convention. It does not say that price must fall, that it will keep rising, or that 75 represents a probability.

Six different measurements sit inside the family

Each type should be read on its own terms:

These are not merely alternative scales for the same fact. Rate of Change uses two price endpoints. Relative Strength Index compares average gains with average losses. Stochastic and Williams %R depend on the high-low range. Commodity Channel Index uses typical price and its mean deviation. Moving Average Convergence Divergence compares smoothed horizons. Each calculation can retain or discard different information from the same path.

One advance can produce several different readings

Illustrative sequence. Imagine price advances from 100 to 108 over several bars, then continues to close near 108 without making much additional progress.

The readings differ because the questions differ. None of them, alone, tells the reader whether the pause will become continuation or reversal.

Read the level, transition, duration, and direction separately

An oscillator produces a value, but a strategy often acts on a derived event. Those layers should not be collapsed:

• Level: the current value, such as RSI at 75 or Rate of Change at +4%.
• Change: whether the value is rising, falling, or accelerating.
• Transition: a one-observation event such as crossing above a threshold.
• Duration: how long the oscillator has remained inside a declared state.
• Divergence: disagreement between a defined price extreme and a defined oscillator extreme.

For MACD, positive MACD, a cross above its signal line, and a growing positive histogram are three different observations. For a bounded oscillator, entering an extreme zone, remaining there, and leaving it are likewise different events.

“Overbought” does not mean “must fall”

An oscillator near the top of its range is often labeled overbought. In a strong trend, it can stay there because repeated gains are exactly what the calculation measures. Shorting every high reading can amount to repeatedly fighting persistence.

The same observation can support opposing hypotheses:

• Continuation: strong momentum confirms that the move is persistent.
• Reversion: an extreme reading identifies a temporarily stretched move.
• State filter: another setup is allowed only when momentum is neither too weak nor too extended.

The strategy must declare which hypothesis it tests and what additional event turns the reading into an action.

The shaded interval is a state; its two edges are events. A rule that enters on the first crossing, a rule that remains eligible throughout the state, and a rule that waits for the exit crossing will produce different decisions from the same oscillator. This is why “use RSI above 70” is not a complete specification.

A simple RSI interpretation

Arizmic's shipped SMA RSI uses simple rolling averages of gains and losses rather than Wilder's recursive smoothing.

Illustrative example. For a completed 14-bar window:

• The level says average gains were three times average losses under this exact window and smoothing convention.
• The transition through 70 identifies one completed-event crossing.
• The duration above 70 describes persistence in that state.
• A later drop below 70 can occur while price remains above its earlier level.

Buying, selling, or exiting on those events are separate strategy choices. The indicator does not select one.

Bounded and unbounded outputs need different comparisons

Cross-asset research should normalize outputs appropriately. A five-dollar change is not comparable across price scales; even percentage momentum can have different volatility across markets.

Divergence is an observation, not hidden intent

A divergence occurs when price makes a new extreme and the oscillator does not. This can indicate slowing momentum. It can also arise from window mechanics, smoothing, or one older observation leaving the calculation.

To test divergence:

• define the price and oscillator pivots without future leakage;
• specify the maximum distance between them;
• state when confirmation becomes knowable;
• include all attempted divergences, not only clean chart examples;
• compare with a simpler price-only reversal rule.

How to investigate an oscillator

1. Plot raw price change and the oscillator to understand the transformation.
2. Separate level, crossing, slope, and duration events.
3. Test continuation and reversion hypotheses independently.
4. Compare the threshold rule with a continuous or ranked use.
5. Inspect turnover and repeated crossings near the boundary.
6. Evaluate nearby lookbacks and thresholds on later data.
7. Check whether the output adds information beyond a simpler return measure.

False readings to avoid

• Treating an overbought label as a guaranteed reversal.
• Calling multiple price-derived oscillators independent confirmation.
• Using a pivot before later bars confirm it.
• Comparing raw momentum units across instruments.
• Ignoring warm-up and smoothing initialization.
• Optimizing thresholds without accounting for the search.
• Assuming a bounded scale has the same historical distribution in every market and regime.

Shipped signals in this family

Try it in Arizmic

Further reading

• Wilder, New Concepts in Technical Trading Systems (1978) — Gives the original Relative Strength Index construction and its smoothing method. It is the clearest primary reference for understanding what RSI measures before treating its thresholds as strategy rules.
• Neely, Rapach, Tu, and Zhou, “Forecasting the Equity Risk Premium: The Role of Technical Indicators” (2014) — Evaluates technical indicators alongside macroeconomic predictors in an out-of-sample forecasting setting. It provides modern empirical context without turning any one oscillator reading into a universal trading instruction.
• Sullivan, Timmermann, and White, “Data-Snooping, Technical Trading Rule Performance, and the Bootstrap” (1999) — Reassesses large families of technical rules while accounting for selection across them. It is an important warning that choosing the most attractive oscillator, threshold, and horizon after inspection changes the evidence.