Machine Learning for Trading
I wanted to implement some Machine Learning techniques for trading. As I am just beginning, which algorithms would be good to start?
I wanted to implement some Machine Learning techniques for trading. As I am just beginning, which algorithms would be good to start?
Career Resources
Hey Shagufta -Tahsildar, sorry about the delay, but are any of these useful:
Fingers crossed that one of those helps you.
It's not about starting off with an algo. Let's say I gave you algo, would you understand the perils of over fitting, the framework, forecasting, volatility targeting, position sizing? No... I recommend quantopian. Pretty much has all the resources to give you a good start. You need to learn programming, you need to grasp some understanding of math beyond just algebra. The site has resources to teach you.
Linear regression would be a great "algorithm" to start with, good luck!
scikit-learn (the machine learning library quantopian offers) has a variety of models available...so you can just plug in you chosen data, and it will spit out a model that you can use. Unfortunately, LSTM (the most powerful model for analyzing time series data that i am aware of) is not one of them. To use an LSTM model, you will need to use either TensorFlow, PyTorch, CNTK, or another......none of which are available in Quantopian (at this time).
I would also suggest that you explore more than just price as an input to your machine learning models (other things like volume, trades, and alternative data sources).
Unless you have gazillion of data points, I find using a form of linear + regularized models such as Ridge or Lasso work best ( many research papers supporting this ). The problem with prediction with machine learning in finance is the extremely low signal to noise ratio. Powerful and flexible models such as Neural Network ( i.e, anything "deep" in the deep learning ) will tend to over-fit the noise and hence require a lot of data. If you are doing intra-day with tick data NN based would be a possibility. For anything end-of-day, it is always better of with linear model.
It depends on the type of data, what you are predicting and what metric you are optimizing. From my experience xgboost and lgbm was always better than any NN architectures or linear models, but this is when the target is boolean value.
I love trees and forest. XGBM and LGBM are always my go to models for Kaggle competitions. But for trading alpha strategies, most of the times the predicted variables are either returns or some form of price time-series, hence the linear model.
Veniam aliquid quos omnis illo cum provident maiores. Maxime aut rem veniam et. Unde illo quae voluptatem sit fuga corrupti voluptatem. Accusantium eos cum enim dolor laboriosam exercitationem eos.
Quaerat aspernatur ut quia corporis. Laborum ipsam aut nemo quo eaque. Commodi ipsam et iste vel. Molestias enim dolore ut sunt eaque molestiae maiores. Culpa harum ea fugit et. Nihil et cupiditate eos rerum autem unde. Expedita aperiam inventore ex laborum rem.
Sint molestiae esse impedit iusto laborum qui. Explicabo a omnis ut aut. Quos itaque est ut minima et delectus distinctio non.
See All Comments - 100% Free
WSO depends on everyone being able to pitch in when they know something. Unlock with your email and get bonus: 6 financial modeling lessons free ($199 value)
or Unlock with your social account...