Paper: arXiv 2505.19243
Authors: Dominik Stempień, Janusz Gajda
Abstract
We compare traditional approach of computing logarithmic returns with the fractional differencing method and its tempered extension as methods of data preparation before their usage in advanced machine learning models. Differencing parameters are estimated using multiple techniques. The empirical investigation is conducted on data from four major stock indices covering the most recent 10-year period. The set of explanatory variables is additionally extended with technical indicators. The effectiveness of the differencing methods is evaluated using both forecast error metrics and risk-adjusted return trading performance metrics. The findings suggest that fractional differentiation methods provide a suitable data transformation technique, improving the predictive model forecasting performance. Furthermore, the generated predictions appeared to be effective in constructing profitable trading strategies for both individual assets and a portfolio of stock indices. These results underline the importance of appropriate data transformation techniques in financial time series forecasting, supporting the application of memory-preserving techniques.
Complexity vs Empirical Score
- Math Complexity: 6.5/10
- Empirical Rigor: 8.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced mathematical concepts like fractional differencing operators (Granger-Joyeux/Hosking) and ARFIMA/ARTFIMA model estimations, indicating high mathematical complexity. It also demonstrates strong empirical rigor through a rigorous backtesting framework covering 10 years of data, multiple assets, hyperparameter tuning, and evaluation via both forecast error and risk-adjusted trading metrics.
Research Flowchart
flowchart TD
Start(["Research Goal:<br>Compare Financial Data<br>Differentiation Techniques"]) --> Inputs
Inputs["Data Inputs:<br>4 Major Stock Indices<br>10-Year Period<br>+ Technical Indicators"]
Inputs --> Methods
subgraph Methods ["Methodology: Data Preparation"]
LR["Traditional Approach:<br>Logarithmic Returns"]
FD["Fractional Differencing<br>with Parameter Estimation"]
TE["Tempered Extension<br>of Fractional Differencing"]
end
Methods --> Processing
subgraph Processing ["Computational Process"]
ML["Advanced Machine Learning<br>(LSTM Neural Network)"]
end
Processing --> Evaluation
subgraph Evaluation ["Evaluation Metrics"]
FM["Forecast Error Metrics"]
TR["Risk-Adjusted Return<br>Trading Performance Metrics"]
end
Evaluation --> Findings(["Key Findings/Outcomes:<br>Fractional methods improve predictive<br>performance & generate profitable<br>trading strategies across assets"])
style Start fill:#e1f5fe
style Findings fill:#f3e5f5
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 using fractional differentiation methods as data transformations to improve LSTM forecasting performance and construct profitable trading strategies.
Source sentence
The findings suggest that fractional differentiation methods provide a suitable data transformation technique, improving the predictive model forecasting performance.
Assumes 4
- Many econometric time-series models and supervised machine learning algorithms require data stationarity. statistical
Source sentence
Many econometric time-series models and supervised machine learning algorithms require data stationarity in order to correctly forecast unobserved values.
- Trading signals are only acted upon when potential gains exceed transaction costs. execution
Source sentence
opening a new position or modifying an existing one takes place only if potential gains associated with a trade exceed transaction costs.
- The ARFIMA model can extract linear patterns from price series before feeding residuals to the LSTM. model
Source sentence
the ARFIMA model can extract linear patterns from data.
- Fractional differentiation parameters are estimated using multiple techniques and the smallest value achieving stationarity may not yield the best predictions. model
Source sentence
the smallest possible value of the differencing operator that achieves stationarity does not always lead to the most reliable predictions.
Data
- Four major stock indices market-prices public
Source sentence
Data was downloaded from two sources: Yahoo Finance with the use of Python’s yfinance library (S&P 500, DAX, and Nikkei 225) and the Stooq portal (WIG20).
Costs
Modeled
Transaction costs of 0.005% are included in the backtesting results.
Source sentence
Transaction costs of 0.005% are included (Kryńska and Ślepaczuk, 2022).
Finds 4
- Fractional differentiation techniques outperform other data preparation methods and the Buy & Hold benchmark for both Long‑Short and Long‑Only strategies. performance
Source sentence
The presented results strongly indicate that fractional differentiation techniques outperform not only the rest of the methods, but also the benchmark 𝐵𝑢𝑦 & 𝐻𝑜𝑙𝑑 approach.
- LSTM models trained on fractionally differentiated data achieve higher forecasting accuracy than those using logarithmic returns. predictive-accuracy
Source sentence
Based on the forecasting accuracy metrics, the LSTM models trained on fractionally differentiated data provided more accurate predictions compared to those using logarithmic returns.
- Trading performance metrics confirm the superiority of memory‑preserving (fractionally differentiated) inputs. performance
Source sentence
Trading performance metrics further confirmed the superiority of fractionally differentiated data, that is, memory-preserving techniques, as inputs to the predictive model.
- Fractional differencing with tempering removes excessive memory, leading to poorer forecasting and trading results. negative
Source sentence
Fractional differencing with tempering appeared to constitute an approach that tends to remove an excessive amount of memory from the time‑series, which negatively affected both forecasting accuracy and trading model performance.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 19 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.