I'm trying to apply reinforcement learning as a trading strategy. I have a problem with the environment. After some try and error, we realize that it's a multi-agent environment (very obvious now) and the single-agent approach (where only an action performed by my agent will change the environment) will not work (I'm not sure of this too) cause more agents are changing the price and status of a giving stock.

Which kind of reinforcement learning should I apply? It is necessary a multi-agent approach?

  • 1
    $\begingroup$ Whole books could be written about this. RL as a single agent is not such a bad match, since you can do nothing about the policies of other agents, you can treat them in some ways as "the environment". The caveat though is this environment is very complex, has lots of hidden state, and evolves, including reacting to perceived results from your agent (if it has high enough impact). That means RL joins a whole heap of other potential algorithms that can kind of be made to work in automated trading, but that require a lot of domain knowledge and are not without risk. $\endgroup$ – Neil Slater Aug 9 '18 at 8:05
  • $\begingroup$ @NeilSlater Wich other potencial algorithms are you referring, anything in particular? Thanks! $\endgroup$ – exsnake Aug 12 '18 at 14:52
  • $\begingroup$ Nothing in particular, but algorithmic trading has been a mainstream thing for a long time. There are lots of possible approaches. $\endgroup$ – Neil Slater Aug 12 '18 at 15:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.