Application of Machine Learning to Fundamental Analysis of Securities

Panteleev, Aleksandr (author), Marinšek, Denis (mentor), Možina, Martin (comentor)

This inquiry tests whether a classic, fundamentals-based valuation philosophy can be systematically validated in today's data-rich environment. My approach centers on estimating a security's intrinsic worth using linear and non-liner models and maintaining a buffer for error, a contrast to methods that prioritize price momentum. I constructed and historically simulated statistical models on two decades of corporate data from 2003 to 2023 to derive these rational value estimates. The process incorporated a novel signal processing technique and was rigorously designed to avoid the analytical illusion of survivorship bias. The findings from the simulation were unambiguous: portfolios of the most favorably priced assets consistently generated returns superior to a broad market proxy. Crucially, simpler, interpretable models performed on par with their complex counterparts. The work thus validates these enduring principles for a modern context, while noting the growing challenge of finding a unique edge in automated markets.

• Language: English
• Keywords: Benjamin Graham, value investing, machine learning, fundamental analysis, fair value, survivorship bias, financial accounting, time series smoothing
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FE - Faculty of Electrical Engineering
• Year: 2025
• PID: 20.500.12556/RUL-174951
• COBISS.SI-ID: 253453315
• Publication date in RUL: 10.10.2025

Secondary language

• Language: Slovenian
• Title: Uporaba strojnega učenja pri analizi vrednostnih papirjev

Abstract

V tej raziskavi s pomočjo sodobnih tehnik strojnega učenja preverjam, ali je mogoče načela vrednostnega investiranja Benjamina Grahama uspešno uporabiti na današnjih trgih. V nasprotju s pogostimi pristopi, ki se osredotočajo na cenovne trende, sem razvil regresijske modele za izračun ocenjene poštene vrednosti (OPV) delnic. Modeli temeljijo na Grahamovih načelih, kot sta notranja vrednost in marža varnosti. Modele sem testiral na podatkih ameriških delnic med letoma 2003 in 2023, pri čemer sem uporabil inovativno tehniko glajenja fundamentalnih podatkov in dosledno upošteval pristranskost preživelih.

• Keywords: Benjamin Graham, vrednostno investiranje, strojno učenje, fundamentalna analiza, poštena vrednost, pristranskost preživelih, finančno računovodstvo, glajenje časovnih vrst