Full Name / Objective:
Kronos is a foundation model specifically designed for financial market candlestick charts (time series data such as OHLCV). It treats financial market candlestick sequences as a "language," modeling and predicting them through quantization, tokenization, and Transformer architecture.
Its purpose is to uniformly support multiple financial tasks, such as price prediction, volatility forecasting, and synthetic candlestick data generation.
Method / Core Design:
1. Tokenization / Quantization
Discretizes continuous candlestick data (open, high, low, close prices + trading volume / transaction amount, etc.) into hierarchically structured tokens. This transforms continuous numerical values into processable token sequences.
2. Autoregressive Transformer Model
Performs autoregressive training on these tokens (i.e., predicting the next token given historical tokens), enabling the model to learn the dynamic patterns of market time series.
3. Multi-task / Multi-purpose
Capable of not only predicting future price trends but also performing tasks such as volatility forecasting and data generation (simulating candlestick sequences).
4. Extensive Training Corpus + Large Scale
Reportedly utilizes candlestick data from over 45 global exchanges across multiple time granularities, with a total scale exceeding hundreds of millions of records.










Comments
No comments yet
Be the first to comment