Enhancing Saudi Investment Decisions Using Artificial Intelligence (AI) and Market Cycles

Abstract

The study investigates the integration between the advanced Arabic transformer-based sentiment models, namely: AraBERTv2, CAMeL- BERTDA, MARBERTv2 and market cycle detection (FFT, DPO, Hurst exponent). To enhance the predictive analytics of the Tadawul All Share Index (TASI) in Saudi Arabia, an important dataset of Arabic financial tweets and daily TASI price data are adopted to assess the effectiveness of ensemble sentiment models and cycle-based approaches for market timing and volatility forecasting. The results indicate that, while BERT-based sentiment models show robust in- sample classification accuracy, the predictive power of TASI returns is limited and unstable over time. In particular, the findings are in line with previous studies associated with the sentiment of the Saudi market. In contrast, cycle detection and Hurst exponent analysis consistently indicate actionable trading windows and confirm persistent, nonrandom market dynamics (H > 0.6), challenging the efficient market hypo study. In general, the findings highlight the superiority of hybrid cycle-based analytics over sentiment-only models for tactical investment strategies in the Saudi market. Furthermore, the results provide important insights for investors and policy makers and offer a novel and interesting framework for AI-driven financial analytics in emerging markets.