Back to search

Institutional Repository of Vilnius University

Mašininio mokymosi modelių palyginimas ir interpretavimas prognozuojant Lietuvos bendrąjį vidaus produktą (BVP) realiuoju laiku /

Abstract

dc:description

Accurate and quick evaluation of the economic situation is important in both public and private sectors. Gross domestic product (GDP) is published every quarter and the publishing is delayed by several months, so decisions are often made without the latest data. Nowcasting tackles this problem by utilizing high-frequency indicators to obtain the GDP prognosis in advance. Although this approach is well analyzed in the literature, applications for small open economies such as Lithuania remain limited, and the interpretability of machine learning models in an econometric context is an important and often overlooked topic. This work aims to compare and interpret machine learning models for nowcasting Lithuania's GDP growth. The analysis utilizes a dataset of 101 macroeconomic and financial variables obtained from Eurostat and the European Central Bank, which covers the period from 2000 to 2023. A benchmark ARMA(2,2) model and two machine learning models - Elastic Net and XGBoost were built for the nowcasting task. The models were evaluated using RMSE and MAE metrics across five nowcasting vintages. For interpretive analysis at both global and local levels, SHAP values were applied with 13 economically meaningful variable groups. Results indicate that both machine learning models successfully outperformed the ARMA benchmark model across all information availability periods, while Elastic Net achieved the best overall accuracy. Interpretive analysis revealed that the models rely on economically meaningful, but different signals - Elastic Net relies on money market and credit indicators, while XGBoost trusts customer confidence the most. Local analysis of the COVID-19 period revealed that XGBoost produced a forecast closer to the actual economic downturn compared to Elastic Net. The results indicate that machine learning models are suitable for nowcasting Lithuanian GDP, and that SHAP analysis provides a way to ensure economic interpretability.

Degree

thesis:*
Grantor dc:publisher
Institutional Repository of Vilnius University
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Vaičiūtė, Miglė,

Rights

dc:rights
Statement dc:rights
  • info:eu-repo/semantics/openAccess
Language dc:language
lit

Identifiers

dc:identifier.*

Chain of custody

source
Harvested from
Vilnius University
Base URL
epublications.vu.lt/oai
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Vaičiūtė, Miglė,. Mašininio mokymosi modelių palyginimas ir interpretavimas prognozuojant Lietuvos bendrąjį vidaus produktą (BVP) realiuoju laiku /. Institutional Repository of Vilnius University, 2026. https://repository.vu.lt/VU:ELABAETD308335111&prefLang=en_US