Comparative Analysis of Deep Leaarning Models for Predictiong Market Tail Risk

Financial data as volatility data is difficult to capture patterns and trends. Considering the unstructured pattern and volatility inherent in financial data, it is necessary to perform multi-faceted analysis by complementing data with unique characteristics to achieve a reliable level of prediction. Accordingly, we apply the SKEW index and VIX index, which measure market tail risk and volatility, to a deep learning model to analyze asymmetric volatility more comprehensively. Deep learning, which can process large-scale data, is effective in learning the complex patterns and nonlinearity characteristics of financial time series data. LSTM suitable for time series data analysis and Transformer, which enables efficient parallel processing, are applied, and TCN and N-HiTS, which have been proposed relatively recently, are applied together to compensate for the problem of gradient loss and inefficiency in long-term sequences. Some models have been confirmed to have excellent actual overall predictive ability because they can learn the dependence of very long sequences and have strengths in capturing various temporal patterns of time series data.