Prognostisering av statens nettolånebehov : En jämförelse mellan SARIMA-modeller och Riksgäldens prognoser

Sammanfattning

This thesis aims to examine forecast accuracy for models used to predict the Swedish government’s net borrowing requirement. The study compares the published forecasts of the Swedish National Debt Office (Riksgälden) with forecasts generated by classical time series models, with a particular focus on Seasonal Autoregressive Integrated Moving Average (SARIMA) models. Monthly data on the primary net borrowing requirement are analyzed, reflecting the pronounced seasonal patterns present in government spending and revenues. A SARIMA model as well as four different SARIMAX models are estimated and evaluated using a rolling-origin out-of-sample forecasting framework. Forecast accuracy is assessed using standard error measures, including root mean squared error (RMSE) and mean absolute error (MAE). Statistical differences in forecast performance are further evaluated using the Diebold-Mariano test. The inclusion of exogenous variables as well as principal component analysis (PCA) is explored to assess whether increased model complexity within the SARIMA/SARIMAX class of models improves predictive performance. The results suggest that a relatively simple SARIMA model can produce lower RMSE than Swedish National Debt Office’s published forecasts. The published forecasts from the Swedish National Debts office produce lower MAE. More complex model specifications, such as including additional regressors or PCA-based components, do not lead to lower forecast errors and in several cases perform worse out of sample. These findings imply that the dominant predictive structure of the primary net borrowing requirement is driven by a strong seasonal pattern, which can be effectively captured by a transparent statistical model. In conclusion, this thesis demonstrates that classical time series models can replicate and improve some forecast measures, raising important questions regarding the tradeoff between model complexity, transparency and practical applicability in public-sector forecasting.