Introduction
This repository contains the datasets and codes described in the paper "Deep LOB Trading: Half a second please!". We propose a system model contains training, predicting and trading, and evaluate it on the simulation data and the empirical data.
Data
In our paper, we evaluate our system model on both simulation dataset and historical stock data of Chinese A-share market.
- The simulation dataset: We simulate three hypothetical market sentiments (uptrend, downtrend, and flat) and one dataset under a mixture of different market sentiments by a zero intelligence agent-based model with the codes. If you would like to reproduce our trading system, you could download the simulation datasets here and have a taste.
- The CS-20 dataset: Benchmark dataset is publicly available and downloaded.
- The proprietary dataset of CS-100: Provided by the Fintech company TradeMaster.
Code
The codes written by Python are provided in the fold for the simulation dataset under a mixture of different market sentiments.
- DCNN.py: Labelling, training and predicting.
- invest_strat.py: Trading strategy with and without optimization.