Machine Learning

I completed Andrew Ng's Deep Learning Specialization consisting of five courses by deeplearning.ai on Coursera.

1. Empirical Asset Pricing

Most of the work is done during the 10-week internship in Summer 2023 at PineBridge Investments, where I served as a Quantitative Research Consultant. I will present below an autoregressive machine learning model to predict and rank stock prices, using publicly available data. The method was introduced in the paper "Empirical Asset Pricing via Machine Learning." A sample code used to generate these results are available in github.

Due to limited time and computational resources, I am not able to implement the full model presented in the paper. Nonetheless, even as a "first" try using a three-layer neural network (NN3), my ML model achieves an overall SR (Sharpe Ratio) of 1.30 for equal weighted portfolios (competitive with the paper's SR of 1.20 for value weighted portfolios and 1.57 for equal weighted portfolios excluding stocks below 20th percentile on NYSE cap weights).

Conclusions of comparing three-layer Neural Network (NN3) and Ordinary Least Squares using 3 factors (OLS3)

• NN3 shows 30% SR increase over all stocks

• NN3 does better on large caps and worse on small caps; however, in both cases, NN does better on the top decile, i.e., NN finds stocks with high returns better than OLS3 does

• NN achieves an SR of 0.6 for large cap stocks, tripling the SR of 0.2 in the case of OLS3

Here are some detailed results. 

Pred: predicted monthly returns; Avg, Std: average, standard deviations of realized monthly return. All numbers are annualized; Equal weighted portfolios; Red Number: H-L SR

2. Stable Diffusion (to be updated)

This project (Slides) is supervised by Professor Tyler Maunu. A more detailed writeup will be updated in the future.