Say you’re a trader, add “AI” and “Machine Learning” into your conversation, and a large group of people befriend you on social media, ask you a large number of questions, and later say they want to do it but it is too costly and complicated.
Yes, understanding AI is hard. But for the dedicated it is an enjoyable exercise. It can make you think about data differently, and can help you come up with some interesting (and sometimes surprising) strategies.
One reason AI is seen as hard is because it is a huge subject with many branches. Fortunately, when you apply it to a particular trading strategy many of the branches do not need to be looked at. You’ve taken a big problem (learning ML) and broken it into a smaller one (learning how to apply ML to a particular problem).
For example: you want to develop a model based on how earnings reports from large companies in a sector affect smaller ones. If you had all the data electronically, what features of the earnings reports would you look at? This is an excellent use for machine learning. Dimensionality reduction can help you determine which items in an earnings report are often important, and which ones can be ignored.
Okay, you’ve got a handful of items to look at in earnings reports that seem to affect stock prices. But by how much? What normally happens to stock X when Y has a stellar quarter and Z’s results are abysmal? Look at “linear regression” and “random forest“.
Now that you have some theories under your belt, how do you actually do the work? That is where things get harder. The loop can look like the following:
Acquiring the data, massaging the data, shoving it through the right algorithm to get an answer, interpreting that answer, passing that answer to another algorithm to do further work
- Acquiring the data
- Massaging the data
- Shoving the data through the right algorithm to get an answer
- Interpreting that answer
- Passing that answer to another algorithm to do further work
- Lather, rinse, repeat
Trading is easy. Trading well is hard. But if you think about it, discretionary traders do the above loop all the time, with varying rates of success. Having a computer do it will give you answers quickly, also with varying rates of success.
So what is the next step? I suggest we go from theory to practice. We start with naive models and work our way up, just as you probably did when you began your trading journey. Future posts will take on the challenge.