license: mit
This dataset was presented in TLOB: A Novel Transformer Model with Dual Attention for Stock Price Trend Prediction with Limit Order Book Data
Code: https://github.com/LeonardoBerti00/TLOB
TRADES-LOB: a new synthetic LOB dataset: In market microstructure research, one major problem is the unavailability of a large LOB dataset. The only TRADES-LOB comprises simulated TRADES market data fmany and Intel for two days. Specifically, the dataset is structured into four CSV files, each containing 50 columns. The initial six columns delineate the order features, followed by 40 columns that represent a snapshot of the LOB across the top 10 levels. The concluding four columns provide key financial metrics: mid-price, spread, order volume imbalance, and Volume-Weighted Average Price (VWAP), which can be useful for downstream financial tasks, such as stock price prediction. In total, the dataset is composed of 265,986 rows and 13,299,300 cells, which is similar in size to the benchmark FI-2010 dataset.