How is machine learning used in quantitative finance?

Finance Tips

How is machine learning used in quantitative finance?

I am a professional trader and have traded billions of dollars worth of stocks through electronic trading systems.  Let me tell you about three ways that I have used machine learning.

   To improve what I'm already doing.  When I first applied serious data analysis to my trading, I identified 36 different trading strategies that I use and calculated on four aspects of applied trades: what is risk, when to take profit,  Profit to loss ratio, slippage.  Through this analysis, I placed more emphasis on some trades than others, increased or decreased risk tolerance, and so on.  The total profit increased by 50%.

   To automate what I'm already doing.  Computers don't have the breadth of decisions I can make, but they can learn to copy my best stuff.  We did what's called supervised learning, which means we showed him a bunch of our trades and asked him to create a simulated algorithm.  He finds more opportunities than me and can manage more positions, although he does not trade them either.  The total profit increased by 30%.

   To expand on what I'm already doing.  We started with the things I already trade and added some useful things that I was skeptical about.  We fed it through the neural network, which means it combines these inputs and assigns different weights to each of them, and then rinses and iterates several times to create a new strategy.  The better you do it, the faster it becomes efficient, so the first rollout brings a little profit, the next few, and then a lot at the end.


   While doing all this, I actually learned two important things.  We tried to fill the ocean with pipettes until we focused our thoughts on how to proceed.  We needed a tool more suitable than a pipette, and a target that was more focused than the sea.

   The key skill of a trader is to separate signals from noise.  use it.  Most people think that they will load a bunch of raw data into the computer and it will come with great trades.  The possibilities for this are immense.  The trader's skill must be applied to reduce problems by an order of magnitude.  What predictions will we use?  What dynamic parts of the strategy do we want to change?  How are we going to group the strategies?

   Determining what a good strategy is is harder than you might think.  "Maximizing profits" or "minimizing losses" sounds great, but the life of any trading strategy has its ups and downs.  It matters whether his losses are huge or frequent, or if his significant victories are very rare.  So, instead of relying on our own choosing the best strategy, we started comparing the profit/loss distributions of the strategies applied to the sample data with our own eyes.

   Best wishes!

Post a Comment

Post a Comment (0)