magistrsko delo
Tilen Kavčič (Author), Tomaž Hovelja (Mentor)

Abstract

Zaradi velike zaskrbljenosti glede inflacije v razvitih gospodarstvih so natančne napovedi stopenj inflacije bistvenega pomena za dobro informirane odločitve o fiskalni in monetarni politiki. To magistrsko delo analizira uporabo različnih modelov za boljše napovedovanje kratko, srednje in dolgoročnih sprememb inflacije. Opredeljene so tradicionalne metode, kot je model SARIMAX, ter sodobni modeli, ki temeljijo na umetni inteligenci in ekonometriji. Predlagan je model, ki temelji na nevronskih mrežah z dolgim kratkoročnim spominom. Določen je tudi osnovni model AR(1). Vzpostavljen je zanesljiv okvir za ocenjevanje definiranih modelov. Ta vključuje testiranje modelov z uporabo navzkrižnega preverjanja časovnih vrst na podatkih treh različnih gospodarstev. Analiza v vseh modelih prikaže pomanjkanja. Najbolje se osnovnemu modelu približa SARIMAX, medtem ko LSTM izraža potencial pri napovedovanju z večjo količino podatkov. Univerzalno napovedovanje inflacije ostaja nerešeno vprašanje, dokler raziskovalcem ne bodo na voljo obsežni makroekonomski in mikroekonomski podatki z modeli, učinkovitimi v tem okolju.

Keywords

napovedovanje inflacije;napovedni modeli;ekonometrija;računalništvo in informatika;magisteriji;

Data

Language: Slovenian
Year of publishing:
Typology: 2.09 - Master's Thesis
Organization: UL FRI - Faculty of Computer and Information Science
Publisher: [T. Kavčič]
UDC: 004.8:338.27:336.748.12(043.2)
COBISS: 181837059 Link will open in a new window
Views: 79
Downloads: 18
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Inflation forecasting with long short-term memory neural networks
Secondary abstract: Given the high inflation concerns in advanced economies, accurate forecasts of inflation rates are essential for well-informed fiscal and monetary policy decisions. This master's thesis analyses the use of different inflation forecasting models to better predict short, medium and long term changes. Traditional methods such as the SARIMAX model are identified, as well as modern models based on artificial intelligence and econometrics. A model based on neural networks with long short-term memory is proposed. AR(1) model is defined as a baseline. A reliable framework for the evaluation of defined models is established. It includes model testing using time series cross-validation on data from three different economies. The analysis shows deficiencies in all models. SARIMAX comes closest to the baseline model, while LSTM shows potential in forecasting with a larger amount of data. Universal inflation forecasting remains an open question until researchers have access to comprehensive macroeconomic and microeconomic data with models that work in this environment.
Secondary keywords: inflation forecating;inflation;artificial intelligence;neural networks;predictive models;LSTM;computer science;computer and information science;master's degree;Nevronske mreže (računalništvo);Umetna inteligenca;Inflacija;Računalništvo;Univerzitetna in visokošolska dela;
Type (COBISS): Master's thesis/paper
Study programme: 1000471
Embargo end date (OpenAIRE): 1970-01-01
Thesis comment: Univ. v Ljubljani, Fak. za računalništvo in informatiko
Pages: 68 str.
ID: 22331029