Momentum is often described as a market anomaly, a behavioral inefficiency, or a source of return. These explanations are appealing because they imply persistence—something structural that can be harvested. Yet momentum survives even when its explanations fail, reappearing across assets, time periods, and implementations. This persistence suggests a different interpretation: momentum is not a return source at all, but a method of isolating certain types of price movement.

A useful way to think about momentum is not as alpha, but as a filter.

Momentum strategies are commonly evaluated as if the signal itself generates returns. In this framing, performance is treated as evidence of insight, while drawdowns are treated as evidence of decay. The underlying assumption is that momentum forecasts direction in an economically meaningful sense.

That assumption is rarely tested.

In practice, momentum often succeeds without identifying mispricing, behavioral bias, or fundamental information. It also fails in ways inconsistent with the narrative of alpha erosion. The mistake is subtle but consequential: mistaking signal extraction for economic insight.

From a signal processing perspective, prices are noisy, non‑stationary time series composed of overlapping components operating at different time scales. Momentum does not explain these components. It selectively emphasizes some of them.

Viewed this way, momentum behaves like a temporal filter:

This reframing removes any requirement that momentum be “correct” in a causal sense. The filter does not need to understand why prices move. It only needs to isolate periods when movement is coherent.

Because price aggregates multiple processes, momentum inevitably filters a mixture—an issue returned to later.

In this context, sustained coherence does not mean uninterrupted trends or accurate forecasts. It refers to periods in which price changes reinforce one another at a chosen time scale, such that aggregation reveals structure rather than cancellation.

Coherence is therefore relative: it depends on the horizon of the filter and the dynamics of the underlying process. A series can be noisy in absolute terms and still exhibit coherence when viewed through an appropriate temporal lens.

Momentum filters are designed to respond to this property. They do not require prices to move monotonically, nor do they require insight into why movement occurs. They only require that directional persistence persists long enough, and at the right scale, for the filter to respond.

The persistence of momentum across markets and decades is sometimes described as a time‑invariant anomaly. This is misleading. What remains invariant is not performance, but structure: markets repeatedly generate regimes in which price coherence emerges, disappears, and reappears.

The sources of coherence—flows, liquidity conditions, institutional behavior, macro constraints—change over time, but the existence of coherence as a feature of market dynamics does not. Momentum survives changes in participants, instruments, and narratives because the filter remains relevant as long as the type of structure it isolates continues to recur.

A generic momentum signal can be expressed as:

where r₍ₜ₋ₖ₎ denotes past returns, wₖ defines the filter shape, and mₜ is the resulting filtered signal.

This equation contains no information about economic value, fundamentals, or future payoffs. It is a transformation of the past, not a forecast of the future. Whether the signal generates returns depends on what it is paired with: execution, volatility conditions, leverage, and risk controls.

Momentum identifies when price movement is coherent—not why it exists.

This interpretation explains several empirical regularities:

These properties are difficult to reconcile with a narrow alpha interpretation, but natural under a filtering view. Momentum strategies tend to load on environments characterized by sustained movement and volatility clustering. In effect, many momentum strategies act as conditional exposure mechanisms, not predictors.

The persistence of momentum across asset classes and decades is often cited as evidence of behavioral bias or delayed information diffusion. Yet these explanations emerged only after the phenomenon had already been documented, and they vary widely by market and period. What remains consistent is not the narrative, but the structure: momentum appears wherever prices exhibit sustained coherence over time. When performance breaks down, it tends to coincide not with the disappearance of trends, but with shifts in the underlying regime—changes in volatility, liquidity, or market organization that alter how coherence translates into payoff. This is consistent with a filtering interpretation: the mechanism survives as long as the structure persists, and fails when the environment it operates in changes.

A useful parallel exists in institutional work on volatility forecasting. Volatility is widely regarded as more forecastable than returns because it exhibits persistent statistical structure—clustering, mean reversion, and regime dependence—while returns themselves remain weakly serial. The lesson is not that markets are predictable, but that certain properties of financial time series recur. Momentum operates under a similar logic. It does not predict returns; it isolates persistence in price movement at a given time scale. Like volatility filters, momentum strategies remain viable not because outcomes are stable, but because the type of structure they respond to recurs. When that structure shifts, performance breaks—not because the filter has decayed, but because the environment it was designed for has changed.

Filters operate within a design envelope. When the structure of the underlying signal changes, the filter misfires.

Momentum filters fail when:

These failures are not anomalies. They are the expected behavior of a filter applied to a process whose structure has shifted.

If momentum is a filter rather than alpha, then debates about whether it is “dead” miss the point. Filters do not decay because beliefs change; they fail when signal structure changes. The unsettling implication is that momentum can succeed without understanding markets—and fail precisely when understanding matters most.

Disclaimer

This material is provided for educational and informational purposes only. It does not constitute investment advice, a recommendation, or an offer to buy or sell any security or financial instrument. The views expressed are based on stated assumptions and simplified models and may not reflect real-world market conditions. Monoco does not manage assets or provide personalized financial advice.

© 2025 Monoco Research