Paper: arXiv 2507.15437
Authors: Matthieu Garcin, Karl Sawaya, Thomas Valade
Abstract
The linear fractional stable motion (LFSM) extends the fractional Brownian motion (fBm) by considering \(α\)-stable increments. We propose a method to forecast future increments of the LFSM from past discrete-time observations, using the conditional expectation when \(α>1\) or a semimetric projection otherwise. It relies on the codifference, which describes the serial dependence of the process, instead of the covariance. Indeed, covariance is commonly used for predicting an fBm but it is infinite when \(α<2\). Some theoretical properties of the method and of its accuracy are studied and both a simulation study and an application to real data confirm the relevance of the approach. The LFSM-based method outperforms the fBm, when forecasting high-frequency FX rates. It also shows a promising performance in the forecast of time series of volatilities, decomposing properly, in the fractal dynamic of rough volatilities, the contribution of the kurtosis of the increments and the contribution of their serial dependence. Moreover, the analysis of hit ratios suggests that, beside independence, persistence, and antipersistence, a fourth regime of serial dependence exists for fractional processes, characterized by a selective memory controlled by a few large increments.
Complexity vs Empirical Score
- Math Complexity: 8.5/10
- Empirical Rigor: 6.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced stochastic calculus and stable distribution theory (α-stable processes, spectral measure, codifference), indicating high mathematical complexity. It includes both simulation studies and real-data applications (FX rates, volatility) to validate the forecasting method, demonstrating solid empirical rigor, though the practical implementation details (e.g., coding, data specifics) are not fully elaborated.
Research Flowchart
flowchart TD
A["Research Goal: Forecast LFSM increments<br>from discrete observations"] --> B{"Input: High-Frequency FX &<br>Volatility Time Series"};
B --> C["Key Methodology:<br>Compute Codifference<br>(Serial Dependence)"];
C --> D{"Conditional Logic"};
D -- α > 1 --> E["Process: Conditional Expectation"];
D -- α ≤ 1 --> F["Process: Semimetric Projection"];
E & F --> G["Computational Output:<br>Forecasted Increments"];
G --> H["Key Findings:<br>Outperforms fBm in FX;<br>Isolates Kurtosis & Persistence;<br>Identifies 'Selective Memory' Regime"];
classDef default fill:#f9f,stroke:#333,stroke-width:2px;
class A,H default;
class B,C,D,E,F,G default;
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a forecasting method for linear fractional stable motions that uses the codifference to predict future increments, applying conditional expectation when α>1 and a semimetric projection otherwise.
Source sentence
We propose a method to forecast future increments of the LFSM from past discrete-time observations, using the conditional expectation when α>1 or a semimetric projection otherwise.
Assumes 6
- The process to be forecast follows a linear fractional stable motion (LFSM) with α‑stable increments. model
Source sentence
The linear fractional stable motion (LFSM) extends the fractional Brownian motion (fBm) by considering α‑stable increments.
- α is in the interval [1,2] for the empirical financial data. statistical
Source sentence
We note that we focus on α ∈[1, 2] because this interval contains all the estimated values of α in our financial dataset
- When α>1 the conditional expectation exists and can be used for forecasting. model
Source sentence
using the conditional expectation when α>1 or a semimetric projection otherwise
- Microstructure noise may bias estimation of the self‑similarity parameter, so the smallest clean time step is selected. data
Source sentence
It can for example be the smallest time step in the dataset, so that one can count on numerous observations, unless the dataset is disrupted by a microstructure noise
Show 2 more
- The codifference adequately captures the serial dependence needed for forecasting, even though it does not describe the full dependence structure. statistical
Source sentence
the codifference can be used as a measure of serial dependence, even though it does not capture the entire dependence structure
- Realized volatility increments are non‑Gaussian with heavy tails, justifying the LFSM model. data
Source sentence
series of the increments of the logarithm of realized volatilities have a significant non‑Gaussian behaviour, with a leptokurtic distribution
Data
- Realized volatility series of major equity indices and VIX market-prices unknown
Source sentence
The series starts on January 2000, except N225, which starts in February 2000, OSEAX in September 2001, and SMSI in July 2005. The end date of our sample is on the 12th April 2021.
Costs
Not applicable
Finds 5
- The LFSM‑based forecast outperforms the fBm‑based forecast in mean absolute error and hit‑ratio for most series, though the HAR model remains the best overall. predictive-accuracy
Source sentence
the LFSM outperforms the fBm, but these differences are statistically significant for only 4 out the 9 series.
- Four regimes of serial dependence are identified for fractional processes, adding a regime of selective memory when H<1/α and α is small. theoretical
Source sentence
we have been able to identify four regimes, instead of three: persistence
- Simulation results show that increasing the number of past observations (d) improves forecast performance, especially when H<1/α. performance
Source sentence
Last, when one increases d and thus enlarges the information set, Figure 5 shows an improved performance, especially when H < 1/α, that is for a negative dependence of increments.
- Empirical estimates of α and H for the volatility series lie within the range α∈[1,2] and display varying H, confirming the suitability of the LFSM model. statistical
Source sentence
We focus on α ∈[1, 2] because this interval contains all the estimated values of α in our financial dataset
- Hit‑ratio analysis indicates that LFSM forecasts are significantly better than fBm for six series, though differences are not always statistically significant. predictive-accuracy
Source sentence
When it comes to hit ratios, LFSM performs better than fBm for 6 series, but the difference is never significant, according to a McNemar test.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 57 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.