MIDAS: Forecasting quarterly GDP using higher-frequency data
Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE creditsStudent thesis
We forecast US GDP sampled quarterly over horizons ranging from one quarter to three years. Using AR-MIDAS models we study three lag polynomials: the Almon lag, the exponential Almon lag and the beta lag, and nine macroeconomic variables, sampled weekly or monthly. Our benchmark model is an AR(1) and we compare forecast errors using RMSE. In all instances the AR-MIDAS achieves lower forecast errors compared to the benchmark model. The predictor sampled weekly generally performs better compared to other predictors, which are sampled monthly.
Place, publisher, year, edition, pages
2015. , 18 p.
MIDAS, GDP, forecasting, mixed-frequency data
Probability Theory and Statistics
IdentifiersURN: urn:nbn:se:uu:diva-242554OAI: oai:DiVA.org:uu-242554DiVA: diva2:783891
Subject / course