Adaptive Weighted Genetic Algorithm-Optimized SVR for Robust Long-Term Forecasting of Global Stock Indices for investment decisions

Battle Count: 73

The paper is highly relevant for long-term investment decisions and portfolio management. The IGA-SVR model provides daily price predictions up to one year ahead with ~8.91% average MAPE across five global indices, significantly outperforming LSTM (11.12%) and OGA-SVR (17.83%). The computational efficiency (14s vs 272s for LSTM) makes it practical for regular retraining. However, it focuses on price level prediction rather than trading signals, direction prediction, or risk-adjusted returns. The model could be integrated into quantitative trading systems for position sizing, entry/exit timing, and portfolio rebalancing decisions.

The model is relatively straightforward to implement using scikit-learn's SVR and a standard GA library. The main components are: (1) data preprocessing with min-max scaling, (2) GA optimization of three SVR hyperparameters with defined bounds, (3) weighted MAPE calculation combining full dataset and recent 5-year performance, (4) SVR fitting and prediction. The pseudocode is provided. No custom neural network architectures or complex feature engineering required. Python implementation on Google Colab is mentioned.

Data is publicly available from Yahoo Finance. Python implementation on Google Colab using scikit-learn library. Pseudocode provided. However, no GitHub repository is mentioned, and some hyperparameter restrictions were determined by trial-and-error. GA parameters (population size, crossover/mutation rates) are not fully specified.