Wavelet-Based Dual-Domain Phase Alignment for Predicting Stock Index Trends from Investor Sentiment Cycles

The dynamics in the financial markets are complicated and non-stationary, with a significant influence of the wave of investor sentiment. The recent changes in sentiment-based stock prediction have shown promising results, but the current research is to a large extent based on individual domain analysis, constant correlation, or the traditional machine learning framework, which constrains its capability to elucidate multi-scale temporal dynamics and phase-based lead-lag relationships. In order to overcome these weaknesses, a new Cross-wavelet Sentiment-driven Dual-domain Phase Alignment, abbreviated as CS-D²PA, is introduced for stock index trend prediction. The suggested structure has 91.8, 90.6, 92.1, 91.3, and 93.5 accuracy, precision, recall, F1-score, and trends consistency rate, respectively, proving to have a better predictive stability and a better classification performance in sentiment-driven stock trend forecasting. The non-stationary and multi-scale behavior of financial markets is dictated by non-periodic changes in investor sentiment cycles. This work proposes a Cross-wavelet Sentiment-based Dual-Domain Phase Alignment model (CS-D2PA) of predictive modeling of stock index trend. The framework combines the feature extraction by the continuous wavelet technique with the cross-wavelet phase difference estimation, as well as the structured alignment in time and frequency domains. The explicit modeling of the sentiment-price phase synchronization of the approach makes it possible to identify lead-lag interaction early and increases the predictability of the forecasts in volatile market conditions, which increases their interpretability.