KangOxford / Volume-Forecasting

Time Series Prediction of Volume in LOB

Time-Series-Forecasting

Trading Target(Volume/Volatility) Time Series Prediction in LOB

Forecasted Trading Volume is crucial in the VWAP replicating strategies and could be used in the Optimal Execution, such as Optimal Scheduling with Predicted Trading Volume. We have the environment: AlphaTrade built in Jax which can automatically paralleling the env and train/testing our algorithm on thousands of stocks at the same time, with a speed up over 100 times.

Timeline:

Week 2 Feature Engineering

• Target: percentage change of volume
• Imbalance:
  $\frac{x_{bid} - x_{ask}}{x_{bid}+x_{ask}}$
• Lookback window
  • disjoint lookback window
  • overlap lookback window

$$ \left{ \begin{array}{lr} (t-1), & 20 features, \\ (t-2) \sim (t-5),& 20 features\\ (t-6) \sim (t-20),& 20 features, & \end{array} \right. $$

• Indicators
  • Implemented Indicators
• Literature
  • Order Behavior In High Frequency Markets
  • Order Exposure in High Frequency Markets