Stock Market Forecasting Using Restricted Gene Expression Programming

Bin Yang, Wei Zhang, Haifeng WangComputational 2019 · DOI

TL;DR

This paper introduces a novel stock market forecasting method using an S-system model optimized by Restricted Gene Expression Programming (RGEP) and a hybrid Brain Storm Optimization (BSO) and Particle Swarm Optimization (PSO) algorithm. The approach demonstrates superior prediction accuracy compared to several state-of-the-art methods across multiple stock indices for both short-term and medium-term forecasts.

What Problem Does This Paper Address?

Predicting stock market indices accurately is a notoriously difficult challenge due to the complex interplay of numerous economic and uncontrollable factors. While various machine learning and statistical models have been applied, many "black box" approaches struggle to provide interpretable insights into the underlying market mechanisms. Existing methods often use all available input variables, increasing the risk of overfitting. This paper aims to develop a more accurate and potentially more interpretable model for stock index forecasting by leveraging the strengths of differential equation models and advanced optimization techniques.

Methods

The core of the proposed method lies in the S-system model, a nonlinear differential equation framework known for its ability to capture dynamic system behavior. The challenge is to effectively define and optimize this model.

  • Restricted Gene Expression Programming (RGEP): This technique is used to automatically discover and optimize the structure of the S-system model. RGEP encodes potential S-system structures as chromosomes, where genes represent coefficients and exponents. It then uses evolutionary operators to search for the optimal structure that best fits the data. A key aspect is that RGEP can automatically select relevant input variables, mitigating the overfitting issue.
  • Hybrid Brain Storm Optimization (BSO) and Particle Swarm Optimization (PSO): Once the S-system structure is determined, the parameters (coefficients and exponents) need to be optimized. The paper proposes a hybrid BSO-PSO algorithm for this task. BSO is good at exploring diverse solutions, while PSO excels at fine-tuning. By combining them, the authors aim to achieve faster convergence and avoid getting stuck in local optima, which can be a problem for individual BSO or PSO.
  • S-system Model Solving: The S-system's differential equations are solved using the fourth-order Runge-Kutta method to obtain predictions for training and testing.

Datasets

The proposed method was evaluated using five well-known stock market indices:

  • Dow Jones Index (DJI)
  • Hang Seng Index (HSI)
  • NASDAQ Index (NASI)
  • Shanghai Stock Exchange Composite Index (SSEI)
  • SZSE Component Index (SZSEI)

For each dataset, 70% of the data was used for training and 30% for testing. The paper does not introduce new benchmark datasets but uses these established indices to compare against existing methods.

Key Insights and Ideas

  • Interpretable Modeling with Differential Equations: By using an S-system model, the approach moves beyond purely black-box models. The resulting S-system equations, while complex, offer a more structured and potentially interpretable representation of the underlying market dynamics compared to traditional neural networks.
  • Automated Feature Selection: RGEP's ability to automatically select a subset of input variables for the S-system model is crucial. This not only simplifies the model but also significantly reduces the risk of overfitting, a common pitfall in financial time series analysis.
  • Synergistic Optimization: The combination of BSO and PSO for parameter optimization is a key innovation. The paper suggests this hybrid approach offers a better balance between exploration and exploitation, leading to more robust and efficient convergence than using either algorithm alone.

Key Results

The proposed method demonstrated strong performance across all tested stock indices for both 1-week-ahead and 1-month-ahead predictions.

  • Superior Accuracy: The method consistently outperformed several state-of-the-art models, including Deep Recurrent Neural Network (DRNN), Flexible Neural Tree (FNT), Radial Basis Function Network (RBFNN), Backpropagation Neural Network (BPNN), and ARIMA.
  • Quantitative Improvements: For 1-week-ahead predictions, the proposed method showed significant improvements in metrics like RMSE (Root Mean Square Error), MAP (Mean Absolute Percentage), ARV (Average Relative Variance), MAPE (Mean Absolute Percentage Error), R² (Coefficient of Multiple Determinations), and VAF (Variance Accounted For). For example, it was 34.8% lower in RMSE than DRNN for the DJI dataset and achieved VAF values closer to 100% across the board.
  • Faster Convergence: The hybrid BSO-PSO algorithm exhibited faster convergence characteristics compared to standalone BSO and PSO, as visualized in error convergence plots.

Limitations and Future Work

  • Computational Cost: While the hybrid algorithm is faster than individual BSO/PSO, complex optimization tasks like this can still be computationally intensive.
  • Data Availability: The paper mentions that the stock index data can be downloaded from Yahoo Finance.
  • Future Work: The paper doesn't explicitly detail future work, but the success of the S-system model with RGEP and hybrid optimization suggests potential for application to other complex time-series forecasting problems or for exploring more sophisticated S-system structures. The interpretability of the S-system model could also be further investigated.

Summary updated

Abstract

Stock index prediction is considered as a difficult task in the past decade. In order to predict stock index accurately, this paper proposes a novel prediction method based on S-system model. Restricted gene expression programming (RGEP) is proposed to encode and optimize the structure of the S-system. A hybrid intelligent algorithm based on brain storm optimization (BSO) and particle swarm optimization (PSO) is proposed to optimize the parameters of the S-system model. Five real stock market prices such as Dow Jones Index, Hang Seng Index, NASDAQ Index, Shanghai Stock Exchange Composite Index, and SZSE Component Index are collected to validate the performance of our proposed method. Experiment results reveal that our method could perform better than deep recurrent neural network (DRNN), flexible neural tree (FNT), radial basis function (RBF), backpropagation (BP) neural network, and ARIMA for 1-week-ahead and 1-month-ahead stock prediction problems. And our proposed hybrid intelligent algorithm has faster convergence than PSO and BSO.
BibTeX
@inproceedings{yang2019stock,
  author = {Bin Yang and Wei Zhang and Haifeng Wang},
  title = {{Stock Market Forecasting Using Restricted Gene Expression Programming}},
  booktitle = {Computational},
  year = {2019},
  doi = {10.1155/2019/7198962},
}
Figure