Back to search

Πανεπιστήμιο Μακεδονίας

Machine learning applications in equity prediction and portfolio management

Abstract

dc:description

This thesis addresses two fundamental problems in quantitative equity investing: directionaltrading and portfolio construction. Despite advances in machine learning, existingapproaches often optimise predictive models using statistical accuracy metrics that are weaklyaligned with real-world investment performance, and rely on modelling frameworks that donot scale effectively across large asset universes. For directional trading, this thesis proposesa profit-driven model selection framework in which model configurations are optimised directlywith respect to realised trading performance. The approach is model-agnostic andincorporates a distribution-aware allocation mechanism that adjusts position sizing basedon the predicted return distribution. Empirical results across major U.S. equity indicesover a decade-long out-of-sample period show materially improved performance relative totraditional model selection methods and passive benchmarks. For portfolio construction, thethesis introduces the max-one framework, which resolves the trade-off between scalability andper-asset optimality by assigning to each asset its individually optimal model configurationwithin a single cross-sectional training procedure. Applied to a broad universe of U.S. equitiesover a twenty-year out-of-sample period, the framework delivers significantly higher returnsthan standard machine learning approaches and market benchmarks, after transaction costs.Overall, the results demonstrate that aligning predictive modelling with economic objectivesand enabling scalable model specialisation leads to substantial improvements inreal-world investment performance. The proposed methodologies are evaluated on long horizondatasets spanning multiple market regimes, large cross-sectional universes, andrealistic transaction cost assumptions, thereby addressing key shortcomings in the empiricalfinance literature.

Degree

thesis:*
Grantor dc:publisher
Πανεπιστήμιο Μακεδονίας
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Chalvatzis, Chariton
  • Χαλβατζής, Χαρίτων

Subjects

dc:subject × 15

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/63069

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Chalvatzis, Chariton; Χαλβατζής, Χαρίτων. Machine learning applications in equity prediction and portfolio management. Πανεπιστήμιο Μακεδονίας, 2026. http://hdl.handle.net/10442/hedi/63069