fagan2888 / portfolio-optimization-using-ai

PORTFOLIO

OPTIMIZATION USING

NLP

KARTIK SETHI
3rd Year Undergraduate
Indian Institute of Technology, Guwahati

WORKFLOW

Text Classifier

● Programmatically generated synthetics dataset of investor input ● Labelled the dataset as different constraints such as “Has Tobacco”,”Has military” etc. ● Used nltk library to generate different word synonyms ● Created a corpus of ~ 30,000 investor inputs ● Used a pre-trained spaCy text-classification model to predict different constraints ● Text classifier predicts the final constraints to be applied to the investor’s tradable portfolio

Portfolio Optimization using time series data

● Extracted time series data of the given stocks from Yahoo Finance
● Extracted exponentially weighted mean returns of the stocks from the time
series data of stock prices
● Covariance matrix of the stocks constructed using time series analysis
● Generated Markowitz Portfolio maximising Sharpe Ratio

Different portfolio optimization models

● Efficient frontier optimisation via quadratic programming ● Hierarchical Risk Parity, using clustering algorithms to choose uncorrelated assets ○ From a universe of assets, form a distance matrix based on the correlation of the assets. ○ Using this distance matrix, cluster the assets into a tree via hierarchical clustering ○ Within each branch of the tree, form the minimum variance portfolio (normally between just two assets. ○ Iterates over each level, optimally combining the mini-portfolios at each node. ● Markowitz Critical Line Algorithm ○ Robust alternative to the quadratic solver used to find mean-variance optimal portfolios ○ Unlike generic quadratic optimisers, the CLA is specially designed for portfolio optimisation ○ Guaranteed to converge after a certain number of iterations, and can efficiently derive the entire efficient frontier

● Black-Litterman Allocation ○ Takes a Bayesian approach to asset allocation ○ It combines a prior estimate of returns (canonically, the market-implied returns) with views on certain assets, to produce a posterior estimate of expected returns. ○ Can provide views on only a subset of assets and BL will meaningfully propagate it, taking into account the covariance with other assets. ○ Using Black-Litterman posterior returns results in much more stable portfolios than using mean-historical return