Paper: arXiv 2501.07580
Authors: Konstantinos-Leonidas Bisdoulis
Abstract
Fluctuations in the stock market rapidly shape the economic world and consumer markets, impacting millions of individuals. Hence, accurately forecasting it is essential for mitigating risks, including those associated with inactivity. Although research shows that hybrid models of Deep Learning (DL) and Machine Learning (ML) yield promising results, their computational requirements often exceed the capabilities of average personal computers, rendering them inaccessible to many. In order to address this challenge in this paper we optimize LightGBM (an efficient implementation of gradient-boosted decision trees (GBDT)) for maximum performance, while maintaining low computational requirements. We introduce novel feature engineering techniques including indicator-price slope ratios and differences of close and open prices divided by the corresponding 14-period Exponential Moving Average (EMA), designed to capture market dynamics and enhance predictive accuracy. Additionally, we test seven different feature and target variable transformation methods, including returns, logarithmic returns, EMA ratios and their standardized counterparts as well as EMA difference ratios, so as to identify the most effective ones weighing in both efficiency and accuracy. The results demonstrate Log Returns, Returns and EMA Difference Ratio constitute the best target variable transformation methods, with EMA ratios having a lower percentage of correct directional forecasts, and standardized versions of target variable transformations requiring significantly more training time. Moreover, the introduced features demonstrate high feature importance in predictive performance across all target variable transformation methods. This study highlights an accessible, computationally efficient approach to stock market forecasting using LightGBM, making advanced forecasting techniques more widely attainable.
Complexity vs Empirical Score
- Math Complexity: 4.0/10
- Empirical Rigor: 7.5/10
- Quadrant: Street Traders — empirical and applied, lower math complexity
Why this score: The paper focuses on practical feature engineering and LightGBM optimization for stock forecasting, with detailed empirical setup, dataset description, and performance metrics, but lacks advanced mathematical derivations or heavy theoretical formalism.
Research Flowchart
flowchart TD Goal["Research Goal: Achieve accurate<br>stock forecasting with low<br>computational requirements"] --> Methodology["Methodology: Optimize LightGBM<br>with Feature Engineering &<br>Transformation Methods"] Methodology --> Inputs["Inputs: Stock Market Data<br>+ Engineered Features<br>(Slope Ratios, EMA Divisions)"] Inputs --> Process["Computational Process:<br>Test 7 Feature/Target Transformations<br>(Returns, Log Returns, EMA Ratios, etc.)"] Process --> Outcomes["Key Outcomes:<br>1. Log Returns/Returns/EMA Diff Ratio = Best Transformations<br>2. Engineered Features = High Importance<br>3. Standardized Versions = Slower Training<br>4. EMA Ratios = Lower Accuracy"] Outcomes --> Conclusion["Conclusion: Accessible &<br>Efficient LightGBM Approach<br>Validated for Stock Forecasting"]
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 optimizing LightGBM for stock forecasting by introducing novel feature engineering and evaluating multiple transformation methods.
Source sentence
In this study we seek to optimize the performance of LightGBM, shifting away from the reliance on powerful hardware typically required for Deep Learning models.
Assumes 5
- LightGBM requires stationary input data for effective splitting. model
Source sentence
As LightGBM uses a tree-based algorithm, the input data needs to be stationary.
- An 80%-20% train-test split is used for evaluation. data
Source sentence
We use an 80%-20% train-test split of the dataset.
- Rolling train-validation split with TimeSeriesSplit avoids data leakage. execution
Source sentence
With the scope of robustly generalizing the model, we opt for a rolling train-validation split, using TimeSeriesSplit from scikit-learn to avoid any data leakage.
- Outlier handling may not significantly affect LightGBM performance. model
Source sentence
The proposed outlier handling method doesn’t offer much benefit for LightGBM, although testing was limited in that regard.
Show 1 more
- Standardization of features is counterproductive for tree‑based models. model
Source sentence
The standardized variants of transformation methods slightly decreased accuracy but greatly increased training time, underscoring that standardization is counterproductive for tree‑based models like LightGBM.
Data
- AAPL daily prices market-prices unknown
Source sentence
testing on the AAPL daily prices.
Cboe
Costs
Not applicable
Finds 4
- Log Returns, Returns and EMA Difference Ratio achieve the best overall performance, outperforming the random walk by more than 10% in every metric. performance
Source sentence
Log Returns followed by Returns and then EMA Difference Ratio achieve the best performance overall, although not deviating notably from the other transformation methods.
- Standardized transformation variants decrease accuracy and increase training time. performance
Source sentence
The standardized variants of transformation methods slightly decreased accuracy but greatly increased training time, underscoring that standardization is counterproductive for tree‑based models like LightGBM.
- Novel features, especially the overnight price gap, consistently have high feature importance across models. predictive-accuracy
Source sentence
Features transformed with EMA Ratios, EMA Differenced Ratios and Returns scored particularly high in feature importance throughout the models. Of crucial importance proved the novel features, especially the overnight price gap, consistently outperforming conventional features in terms of importance.
- All models reduce RMSE by more than 30% compared to the benchmark, mitigating large unexpected movements. performance
Source sentence
In all models but the benchmark, RMSE decreased by more than 30% highlighting the models’ ability to mitigate the effects of large, unexpected movements such as crisis.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 31 related papers). Coverage is partial, so this is neither replication nor refutation.
1 paper in the archive : Alpha Discovery via Grammar-Guided Learning and Search
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.