Capturing Market Volatility: A Comparative Analysis of IBM TinyTimeMixers, Time Series Transformer, and ARIMA in Nifty 50 Forecasting

Abstract

This study explores the application of advanced time series models, including IBM TinyTimeMixers (TTMs), Time Series Transformer (TST), and ARIMA, for anomaly detection and forecasting in financial data analysis of the Nifty 50 index. Utilizing a comprehensive dataset spanning November 9, 2015, to July 12, 2024, with daily market variables such as Open, Close, High, Low, Shares Traded, and Turnover (RS crores), we evaluate the efficacy of these models in identifying market anomalies and predicting trends. IBM TTS, a lightweight pretrained model designed for time series forecasting, leverages channel-independent and channel-mixing architectures to process temporal patterns efficiently. The results demonstrate that these models, with their pre-trained adaptability, outperform traditional methods in capturing volatility, trading patterns, and unexpected deviations in India’s benchmark stock index. This work offers valuable insights for financial analysts, researchers, and machine learning enthusiasts aiming to enhance decisionmaking in dynamic market environments.