We use the pyfolio-reloaded package to compute standard metrics like:
- sharp ratio
- annual return
- alpha
- beta
For a more detailed explanation refer to their page.
They support a lot of metrics.
The custom metrics implemented in this repository are the following:
- PA = Price Advantage
- GLR = Gain Loss Ratio
- AD = Action Distance
- ADS = Action Distance from the Start
- ADE = Action Distance from the End
- cash_used_on_last_tick = The cash that was used at the end of an episode
- T = Tactics, which is just
T = ADS / cash_used_on_last_tick - num_actions = The number of the actions the agent has taken
NOTE: PA & GLR are used for the Order Execution Task. For a more detailed explanation
please refer to the following paper: Universal Trading for Order Execution with Oracle Policy Distillation.
I also wrote a summary of the paper on Medium.