Evaluating The Predictive Performance Of LSTM Models On Chilean Pension Funds: A Comparative Study Using The S&P 500 Index

Gajardo Montecino, Cristian (2025) Evaluating The Predictive Performance Of LSTM Models On Chilean Pension Funds: A Comparative Study Using The S&P 500 Index. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (331kB) | Preview

Abstract

The expanding need for sustainable pension systems requires data-driven forecasting tools which help both financial planning and institutional decision-making. This research examines how three Long Short-Term Memory (LSTM) architectures—Multivariate LSTM, Attention-Based LSTM, and Multi-Input LSTM, perform in forecasting the daily quota values of “Fondo A” which is managed by AFP Habitat in Chile. The S&P 500 index served as an exogenous input to address traditional model limitations in volatile financial environments. The preprocessed historical time series data underwent standardized normalization followed by sliding window techniques before Python models with TensorFlow/Keras were trained for evaluation through RMSE, MAE and R² metrics. Model architecture proved to be a decisive factor in accuracy because the Attention-Based LSTM enhanced temporal feature extraction while the Multi-Input LSTM delivered the highest performance (RMSE: 909.62, R²: 0.9820) by processing local and global signals independently. The research demonstrates that specific architectural improvements produce superior results than conventional methods which provide a flexible framework to enhance pension fund predictions in emerging market environments. The study faces limitations because it uses data from a single AFP and only one external indicator; future research should include more macroeconomic variables and expand datasets while investigating real-time deployment to maximize operational value.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Khan, Sallar UNSPECIFIED
Subjects:	H Social Sciences > HG Finance Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning H Social Sciences > HD Industries. Land use. Labor > Issues of Labour and Work > Pensions H Social Sciences > Economics > Wealth > Income > Retirement income > Pensions
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	01 Jul 2026 10:26
Last Modified:	01 Jul 2026 10:26
URI:	https://norma.ncirl.ie/id/eprint/9426

Actions (login required)

View Item