Fixed-Width Fractional Differencing for Stationary Price Features
Summary
The document explains fixed-width fractional differencing (FFD) as a way to make price series more suitable for statistical and machine-learning models while retaining some long-term dependence. It contrasts this with ordinary integer differencing, which can remove useful price memory, and describes constructing fractional weights by recurrence, truncating them at a threshold, and applying the reversed weights to a log-price window. The differencing order and cutoff jointly determine the lookback length and the balance between stationarity and memory retention.
It outlines an MQL5 engine and indicator, including initialization, bar-by-bar computation, buffer updates, and an example of supplying FFD as a model feature. The document reports a close numerical match with a Python implementation on EURUSD hourly data and gives a per-bar performance claim. These are implementation checks, not evidence that FFD improves trading results. It does not supply an ADF-based parameter-search procedure or trading evaluation; the optimal order depends on the series and stationarity criterion, and long windows require enough history.
Key ideas
- Fractional differencing uses a non-integer order to seek stationarity while retaining more price memory than integer differencing.
- A weight cutoff determines the finite lookback and therefore affects both memory retention and required history.
- The described implementation can compute log-price FFD values as an MQL5 indicator or as a model feature.
- A Python comparison checks numerical agreement, but the document does not show predictive or trading performance.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.