Direct answer
Frama (Fractal Adaptive Moving Average) behaves differently when market price action changes in ways that affect the “roughness” of recent prices. In practice, that means its smoothness and responsiveness tend to shift across regimes such as persistent trending conditions versus choppy or highly variable conditions. Frama’s behavior can also change when you alter its input assumptions (for example, the lookback length or smoothing settings) or when your data conditions differ (for example, gaps, inconsistent sampling, or execution-time frictions). The important point is conditional behavior: Frama adapts based on recent variability, not in order to forecast reliably.
Mechanism and definition
A moving average converts a price series into a smoother series by using past values. Frama adds an adaptive layer: it attempts to estimate how “fractal” or rough the recent price path is, then uses that estimate to adjust how quickly the average can move.
So, the “market condition” that matters most is whether the recent price path is comparatively regular (smoother movement) or irregular (more jagged movement). If the last segment of prices is relatively smoother, Frama typically behaves like a more slowly responding average. If the last segment is more irregular, it tends to behave more like a faster or more reactive average—because the underlying variability estimate changes.
Two stable ideas help separate mechanics from context:
- The mechanics are deterministic given the input series and parameters.
- The market context changes the input series (how rough or smooth recent prices are), which changes the mechanics’ outcome.
Evidence or example (assumptions included)
Consider a simplified, non-live example to illustrate conditional behavior without claiming predictive power.
Assume you have the same asset sampled at a fixed interval (for example, 1-hour closes) and you run Frama with unchanged parameters. Create two backtest windows:
- Trend-like window (assumption): price moves in a relatively steady direction with smaller short-term reversals.
- Choppy window (assumption): price alternates direction more frequently, producing more visible jaggedness.
In the trend-like window, the estimated variability over recent periods is typically lower, so the adaptive smoothing often results in fewer abrupt bends. In the choppy window, estimated variability is typically higher, so the adaptive smoothing often results in more frequent turns or tighter following.
A second comparison is parameter-driven rather than market-driven. Keep the market data the same, but use different lookback lengths:
- A shorter lookback can increase sensitivity to short-term roughness changes.
- A longer lookback can reduce sensitivity and make behavior appear smoother.
This illustrates that “different behavior” can come from market regime shifts, parameter choices, or both.
Limitations and risks
A material limitation is that Frama’s conditional adaptation depends on recent data characteristics, which means it can change its behavior when market conditions shift again. That creates at least three common failure modes:
- Regime sensitivity can mislead interpretation. If you label a move based on how Frama looked during one regime, it may not match how it behaves after the next regime change.
- Overfitting to historical windows. If you tune parameters by trial-and-error on prior data, the apparent differences across regimes may not generalize.
- Data and execution differences. Frama computed from closes differs from Frama computed from other sampling rules; real trading adds costs and timing effects that are not captured by a simple indicator calculation.
Additionally, historical relationships do not establish future results. Even if Frama appears “better” in a past trending regime, you cannot assume the same relationship will hold after volatility patterns or liquidity conditions change.
Verification and next question
To independently verify the relevant facts, compare Frama’s outputs across multiple market regimes using the same sampling method and fixed parameters, then repeat with clearly documented parameter variations. Track how the Frama line’s smoothness and turning frequency change when the input series becomes more jagged or more regular.
A useful next question is: “Which parts of your input and parameter setup (sampling interval, lookback length, smoothing settings, and data cleaning choices) are most responsible for the behavior differences you observe?” That framing keeps the explanation conditional and testable, without treating Frama as a standalone forecast.