Ensemble Investment Strategies Based on Reinforcement Learning

Table 1

Pseudo-code for the A2C model.

---

Input: environment of the stock market

Output: estimated optimal strategy

Initial setup of actor and critic networks

Repeat

 For episodes = 0, 1, 2, …, N do:

  Get state and calculate to get action

  IF the episode does not end there:

   Get with reward

   Using critic networks to obtain return values to estimate Q

   Calculating the gradient using Q values and updating the actor network

   Updating the critic network to reduce the difference

   Update status

  End

 End

To convergence

---