Hello everyone!
I'm exploring new things and trying to get the best out quantra. I recently saw that they have a video where they barely touch Reinforcement learning. It was an introduction course.
After this, a couple of questions came into my mind.
1. Is RL a way to make algos learn and then make it learn in a continuously way through getting new experiences? I think A2C (Actor-Critic) is actually needed for this, right?
2. Does RL with A2C make an algo actually learns after training and validating? I mean, does the algo learns while live trading? Is this a real thing? If RL+A2C can't do that, how it is done?
3. Does quantra has a course focusing on Reinforcement Learning and using A2C?
Hello Mario,
"Is RL a way to make algos learn and then make it learn in a continuously way through getting new experiences? I think A2C (Actor-Critic) is actually needed for this, right?" - Yes, RL is increasingly being used in quant finance problems. One example is live portfolio rebalancing based on the current state of multiple stocks. Another is optimally liquidating stock ( by market makers ) without changing the prices much. There are a variety of architectures that can be used for this. A2C, DQN, DDPG etc
"Does RL with A2C make an algo actually learns after training and validating? I mean, does the algo learns while live trading? Is this a real thing? If RL+A2C can't do that, how it is done?" - so RL is essentially this - selecting the best action for a given state. Here state will be the past and present prices and returns of a group of stocks. As the state changes so does the optimal action ( buying or selling a particular stock). This is done live. You need to try out various architectures before you settle for the one that works optimally.
"Does quantra has a course focusing on Reinforcement Learning and using A2C?" - not at the moment. But we're working towards creating such offerings.