Inflation Forecasting with Machine Learning - A study on the US with big data - Master Thesis

Aug 3, 2026

This thesis evaluates whether machine learning improves inflation forecasting accuracy. Two datasets are used: a monthly dataset widely adopted in prior studies and a weekly dataset constructed from scratch to incorporate the most recent information. Forecast performance is examined across different forecasting horizons to determine whether forecasts improve or deteriorate. The study also tests whether higher data granularity increases forecast consistency. Findings show machine learning does not outperform a strong ARIMA baseline, with Random Forest the main competitive model.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages59

Reading Time2 h 29 min

AccessFree to download

Introduction

Literature review

Data

Financial and commodity variables in inflation forecasting

Methodologies

What is Machine Learning?

Lasso and Ridge Regression

Random Forest

An overview of Deep Learning and Neural Networks

ARIMA and ARIMAX models

Results

Monthly dataset

LSTM models performance

Machine Learning and Univariate models performance

Weekly dataset

LSTM models performance

Machine Learning and Univariate models performance

Discussion

Conclusions

References

Master of Science in Business Analytics and Data Science Department of Business Administration Master Thesis Inflation Forecasting with Machine Learning: A study on the US with big data Author: Konstantinos Manetas Violetas Thesis Supervisor: Konstantinos Tarampanis Acknowledgements To my family, whose constant support and encouragement are key determinants for every step I make. To Professor Konstantinos Tarampanis and Assistant Professor Evangelos Kalampokis, whose guidance and comments have been crucial throughout every step of the research process. To my friends and relatives, who have always believed in me. To my nephews, who are growing fast and already make their impact to the world. Master Thesis Inflation forecasting with Machine Learning: A study on the US with big data. Konstantinos Manetas Violetas 􀀍 Under the supervision of Prof. K. Tarampanis† July 2025 Abstract The aim of this thesis is to investigate the extent to which Machine Learning techniques can enhance the accuracy of inflation forecasting. By using two different datasets, one with monthly data widely used in the literature and one with weekly data that I construct from scratch, I try to update the findings of the literature with the newest available data, examine how inflation forecasts for the US improve or deteriorate when using different forecasting horizonsand assess how higher data granularity can affect the consistency of the forecasts. I find that Machine Learning algorithms do not produce better forecasts than a plain ARIMA with a strong competitor being only the Random Forest model, while higher data granularity does not seem to play any role for the production of more accurate inflation forecasts. 􀀍 UoM, [bad24004@uom.edu.gr](bad24004@uom.edu.gr)[ ](bad24004@uom.edu.gr)† UoM, [kat@uom.edu.gr](kat@uom.edu.gr) Table of Contents 1. Introduction..................................................................................................... 1 2. Literature review ............................................................................................ 3 3. Data .................................................................................................................. 7 3.1. Financial and commodity variables in inflation forecasting ............... 8 4. Methodologies............................................................................................... 11 4.1. What is Machine Learning?.................................................................. 12 4.2. Lasso and Ridge Regression................................................................. 13 4.3. Random Forest ...................................................................................... 14 4.4. An overview of Deep Learning and Neural Networks ..................... 15 4.5. ARIMA and ARIMAX models ............................................................. 19 5. Results............................................................................................................ 20 5.1. Monthly dataset .................................................................................... 21 5.1.1. LSTM models performance .......................................................... 21 5.1.2. Machine Learning and Univariate models performance ........... 22 5.2. Weekly dataset ...................................................................................... 32 5.2.1. LSTM models performance .......................................................... 32 5.2.2. Machine Learning and Univariate models performance ........... 37 6. Discussion...................................................................................................... 43 7. Conclusions ................................................................................................... 44 8. References...................................................................................................... 45 List of

FAQs

What is the primary goal of the thesis?

To assess how far machine learning techniques can enhance the accuracy of inflation forecasting.

How many datasets are used, and how do they differ?

Two datasets are used: a monthly dataset from prior literature and a weekly dataset constructed from scratch using higher-frequency data.

Do machine learning models outperform ARIMA for US inflation forecasting?

No. The results indicate machine learning does not produce better forecasts than a plain ARIMA baseline, with Random Forest being the strongest competitor.

Aug 3, 2026

Upload by Theodora