QF4199 Honours Year Project: Construct Individualized Portfolios for Robo-advisors with Goal-based Portfolio Optimization and Conditional Quantile Estimation on Target Returns

Introduction

This repository contains the implementations and tests for my thesis.

Step-by-Step Guide

There are several ways to use this repository:

1. View results that have been previously generated
2. Run the algorithm again on a completely new dataset

View results that have been previously generated

1. Go to the bottom of evaluation.py and choose one of the following:

   • Uncomment lines 191 to 219 for evaluation_1()

     evaluation_1() was done with a specific value of tau = 0.05 on a slightly larger sample to observe how the conditional quantile estimates vary with the covariates.

   • Uncomment lines 224 to 236 for evaluation_2():

     evaluation_2() was done with various values of tau = 0.01, 0.05, 0.1, 0.25, 0.5 on a slightly smaller sample, to observe how portfolios vary with tau.

   • Uncomment lines 241 to 257 to get a portfolio for a new investor.

2. Re-run on a completely new dataset:

   1. Use the cross_validation method in nonparametric_conditional_quantile_estimator.py to optimize the bandwidth parameters.
   2. Use the estimate_quantiles method in evaluation.py to estimate the conditional quantiles.
   3. Use plot_estimated_quantiles method in evaluation.py if you wish to visualize how the conditional quantile estimates vary with the covariates.

   Refer to examples in evaluation_1() and evaluation_2() on how this can be done in a similar fashion.