Option price prediction based on optimized decomposition-dynamic ensemble and text mining

This paper primarily focuses on the decomposition-ensemblehybrid model and proposes an optimized decomposition-dynamic ensemble algorithm. Additionally, it employs text mining techniques to construct the Investor Sentiment Index (ISI), which is ultimately applied to the prediction of option prices. Through empirical analysis of option data from the SSE 50 ETF, CSI 300 ETF, and CSI 500 ETF, we find that the dynamic ensmble method within the decomposition-ensemble hybrid model not only enhances predictive accuracy but also improves the stability of the model. Furthermore, the effectiveness of the dynamic hybrid model is validated through a rolling window decomposition-prediction framework. In terms of using ISI for option price prediction, the constructed ISI significantly enhances the predictive performance of both single and hybrid models. Notably, the ISI generated by FinBERT outperforms that constructed using traditional dictionary methods. Moreover, in the context of hybrid models, ISI effectively improves the low-frequency and high-frequency components across various hybrid models, while exerting relatively minor influence on the mid-frequency modal components. • Introduces an optimized decomposition-dynamic ensemble algorithm. • Confirms effectiveness via a rolling window decomposition-prediction framework. • ISI generated by FinBERT outperforms traditional dictionary methods in predictive performance. • ISI improves low-frequency and high-frequency components in hybrid models, with minor effects on mid-frequency components.