Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

A Comparative Evaluation of Forecasting Techniques for Public Sector Food Prices: From Statistical to Modern Methods

Wirfelt, Måns LU and Björklund, Vilhelm LU (2024) DABN01 20241
Department of Economics
Department of Statistics
Abstract
This paper examines the performance of traditional statistical models (SARIMA & SARIMAX) and modern machine learning models (XGBoost) for forecasting Swedish public sector food prices organized by a hierarchical index structure. The depth of the analysis is made possible by collaborating with Matilda Foodtech, who gave access to anonymized food procurement data from Swedish public organizations. A dataset spanning from 2015 to 2023, containing over a hundred food categories, has been enriched with macroeconomic indicators to serve as predictors. The results indicate that hierarchical models and modern advanced models can outperform statistical models, but that ensembles of the three yield the best forecasting performance. The study... (More)
This paper examines the performance of traditional statistical models (SARIMA & SARIMAX) and modern machine learning models (XGBoost) for forecasting Swedish public sector food prices organized by a hierarchical index structure. The depth of the analysis is made possible by collaborating with Matilda Foodtech, who gave access to anonymized food procurement data from Swedish public organizations. A dataset spanning from 2015 to 2023, containing over a hundred food categories, has been enriched with macroeconomic indicators to serve as predictors. The results indicate that hierarchical models and modern advanced models can outperform statistical models, but that ensembles of the three yield the best forecasting performance. The study highlights the importance of adapting methods and predictors to the time and place to enhance forecasting precision. These insights are crucial for public sector stakeholders to improve budget planning and optimize procurement strategies. They can also guide future researchers in understanding underlying dynamics of food price inflation based on categorical properties of foods that constitute food price indices. (Less)
Please use this url to cite or link to this publication:
author
Wirfelt, Måns LU and Björklund, Vilhelm LU
supervisor
organization
course
DABN01 20241
year
type
H1 - Master's Degree (One Year)
subject
keywords
Time series forecasting, Hierarchical forecasting, Gradient boosted trees, Food price inflation
language
English
id
9155689
date added to LUP
2024-09-24 08:37:10
date last changed
2024-09-24 08:37:10
@misc{9155689,
  abstract     = {{This paper examines the performance of traditional statistical models (SARIMA & SARIMAX) and modern machine learning models (XGBoost) for forecasting Swedish public sector food prices organized by a hierarchical index structure. The depth of the analysis is made possible by collaborating with Matilda Foodtech, who gave access to anonymized food procurement data from Swedish public organizations. A dataset spanning from 2015 to 2023, containing over a hundred food categories, has been enriched with macroeconomic indicators to serve as predictors. The results indicate that hierarchical models and modern advanced models can outperform statistical models, but that ensembles of the three yield the best forecasting performance. The study highlights the importance of adapting methods and predictors to the time and place to enhance forecasting precision. These insights are crucial for public sector stakeholders to improve budget planning and optimize procurement strategies. They can also guide future researchers in understanding underlying dynamics of food price inflation based on categorical properties of foods that constitute food price indices.}},
  author       = {{Wirfelt, Måns and Björklund, Vilhelm}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{A Comparative Evaluation of Forecasting Techniques for Public Sector Food Prices: From Statistical to Modern Methods}},
  year         = {{2024}},
}