Using Machine Learning to Investigate the Role of Real Estate in a Mixed-Asset Portfolio - thesis

Aug 4, 2026

Upload by Maeve

Investing in real estate delivers diversification and potential long-term appreciation, yet direct ownership often demands substantial capital that many individual investors cannot access. This work evaluates the added value of real estate within diversified portfolios through machine-learning-driven asset allocation. Predictive models including Linear Regression, Support Vector Regression, k-Nearest Neighbours, Extreme Gradient Boosting, and LSTM are used to forecast asset prices, enhanced by Technical Analysis Indicators. Portfolio weights are optimized with a Genetic Algorithm, comparing ML-based strategies against historical-data benchmarks using risk-adjusted performance and diversification effects.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages128

Reading Time5 h 23 min

AccessFree to download

Introduction

Motivation and Objectives

Novelty of Research

Thesis Structure

Publications

Background Information

Financial Markets

Modern Portfolio Theory

Machine Learning

ML for Optimisation

ML for Regression

Literature Review

Portfolio Optimisation

Financial Forecasting

Optimising Mixed-Asset Portfolios Including REITs

Problem Statement

Methodology

Experimental Setup

Results

Summary

ML for Real Estate Time Series Prediction

Methodology

Experimental Setup

Results

Using Machine Learning to Investigate the Role of Real Estate in a Mixed-Asset Portfolio Fatim Zahra Habbab A thesis submitted for the degree of Doctor of Philosophy in Computational Finance Centre for Computational Finance and Economic Agents School of Computer Science and Electronic Engineering University of Essex September, 2024 This thesis is dedicated to my mother, who offered me all the support possible, and to my father, whose memory continues to inspire and motivate me every day. Abstract Investing in real estate offers significant benefits, such as diversification and potential longterm appreciation, making it an attractive option compared to stocks and bonds. However, direct investments in real estate often require substantial capital, which is a barrier for many individual investors. To overcome this, investors often use Real Estate Investment Trusts (REITs), which allow for indirect investment in real estate through shares in companies that own income-generating properties. This study examines the added value of including real estate in a diversified investment portfolio, utilising innovative methods to optimise asset allocation. Instead of relying on historical data, it employs machine learning algorithms (such as Linear Regression, Support Vector Regression, k-Nearest Neighbours, Extreme Gradient Boosting, and LSTM Neural Networks) to predict future asset prices. The study also incorporates Technical Analysis Indicators (TAIs) to further improve predictive accuracy. Furthermore, a Genetic Algorithm (GA) is used to determine optimal portfolio weightings, considering the expected returns and risks of each asset class. The study compares the performance of portfolios constructed using price predictions with those based on historical data, assessing diversification benefits and risk-adjusted returns. Overall, by integrating machine learning techniques, technical analysis, and optimisation algorithms, the study aims to demonstrate the potential advantages of including real estate investments in a diversified portfolio, enabling investors to make more informed decisions and improve their investment outcomes. Acknowledgements First of all, I would like to thank my supervisor, Dr. Michael Kampouridis, who helped me acquire crucial skills for my future as a researcher. He generously gave me his time and worked with dedication and patience from the beginning. I couldn’t have asked for more! I would also like to thank my colleagues at the University of Essex, who have shared their knowledge and experience, making my research journey enriching and interesting. I am also grateful to my mother, who supported me in every possible way during my PhD journey and believed in me even when I could not. Additionally, I would like to thank my friends, both those I met before and during my PhD studies, who provided emotional support and made my journey unforgettable—they are like family to me. Last, but not least, I thank Allah for helping me through all the difficulties. I will continue to trust in you for my future. Contents Abstract iii Acknowledgements iv Figures and Tables viii 1 Introduction 1 1.1 Motivation and Objectives ............................ 1 1.2 Novelty of Research ............................... 2 1.3 Thesis Structure ................................. 3 1.4 Publications ................................... 4 2 Background Information 5 2.1 Introduction ................................... 5 2.2 Financial Markets ................................ 5 2.2.1 The Real Estate Market ........................ 6 2.2.2 The Stock Market ............................ 9 2.2.3 The Bond Market ............................ 11 2.3 Modern Portfolio Theory ............................ 12 2.4 Machine Learning ................................ 14 2.4.1 ML for Optimisation ........................... 15 2.4.2 ML for Regression ........................... 16 3 Literature Review

FAQs

Why is real estate considered in mixed-asset portfolios despite barriers to direct investment?

Real estate can improve diversification and offer potential long-term appreciation, but direct investment usually requires large capital. This motivates using REITs for indirect exposure through shares.

Which machine learning methods are used for asset price prediction in the study?

The study employs Linear Regression, Support Vector Regression, k-Nearest Neighbours, Extreme Gradient Boosting, and LSTM Neural Networks. It further incorporates Technical Analysis Indicators to improve predictive accuracy.

How are portfolio weights determined and how is performance evaluated?

A Genetic Algorithm optimises portfolio weightings using expected returns and risks for each asset class. The resulting portfolios are compared against historical-data-based benchmarks using diversification benefits and risk-adjusted returns.

Aug 4, 2026

Upload by Maeve