Kronos: A Foundation Model for the Language of Financial Markets (arxiv.org)

A groundbreaking foundation model named Kronos has been introduced to enhance the analysis of financial markets, specifically through the modeling of financial candlestick (K-line) data. Unlike existing Time Series Foundation Models (TSFMs), which have struggled with the intricacies of financial data and critical tasks like volatility prediction, Kronos leverages a unique tokenizer that converts continuous market data into token sequences. This allows it to capture essential market dynamics and trade patterns. Pre-trained on a vast corpus of over 12 billion K-line records from 45 global exchanges, Kronos excels in a zero-shot setting across various financial tasks. The significance of Kronos lies in its outstanding performance, demonstrating a 93% improvement in price series forecasting accuracy over the leading TSFM and an 87% advantage over the best non-pre-trained baselines. Additionally, it shows a 9% reduction in mean absolute error for volatility forecasting and a 22% enhancement in generating synthetic K-line sequences. These advancements position Kronos as a versatile and powerful tool for comprehensive financial time series analysis, making it a vital resource for the AI/ML community focused on quantitative finance. The pre-trained model is publicly available, encouraging further exploration and development in this domain.