In a recent project, I used Python to work with historical stock data. I implemented momentum-trading strategies and tested if they have the potential to be profitable. Let me go over the idea as a big picture. I’ll continue in another page with more details and examples:
I created two portfolios utilizing smart beta methodology and optimization, and evaluate the performance of the portfolios by calculating tracking errors. After these two stages and some research, I generated multiple alpha factors and applied various techniques to evaluate performance to pick the best factors for the stock portfolio.
Then I concentrated on AI Algorithms for Trading. In this phase, I worked with another set of data and use machine learning to generate trading signals. I ran backtests to evaluate my trading signals and used advanced techniques like following to combine my top performers.
Sentiment Analysis using NLP
I applied natural language processing on corporate filings, such as 10Q and 10K statements, covering everything from cleaning data and text processing to feature extraction and modeling.
Deep Neural Network with News Data
I built deep neural networks to process and interpret news data. Then constructed and trained LSTM networks for sentiment classification, ran backtests, and applied financial models to news data for signal generation.
Combining Trading Signals for Enhanced Alpha
I created a prediction model for the S&P 500 and its constituent stocks by performing a model selection for an extensive data set, which includes market, fundamental, and alternative data.
In the next post, I’ll write more details on each step and the actual results and progress in each step.
Kayvan Momeni, Jan 2019