By: |
Caroline Jardet (Centre de recherche de la Banque de France - Banque de France);
Baptiste Meunier (Centre de recherche de la Banque Centrale européenne - Banque Centrale Européenne, AMSE - Aix-Marseille Sciences Economiques - EHESS - École des hautes études en sciences sociales - AMU - Aix Marseille Université - ECM - École Centrale de Marseille - CNRS - Centre National de la Recherche Scientifique) |
Abstract: |
Although the Covid-19 crisis has shown how high-frequency data can help track
the economy in real time, we investigate whether it can improve the nowcasting
accuracy of world GDP growth. To this end, we build a large dataset of 718
monthly and 255 weekly series. Our approach builds on a Factor-Augmented MIxed
DAta Sampling (FA-MIDAS), which we extend with a preselection of variables. We
find that this preselection markedly enhances performances. This approach also
outperforms a LASSO-MIDAS—another technique for dimension reduction in a
mixed-frequency setting. Though we find that a FA-MIDAS with weekly data
outperform other models relying on monthly or quarterly data, we also point to
asymmetries. Models with weekly data have indeed performances similar to other
models during "normal" times but can strongly outperform them during "crisis"
episodes, above all the Covid-19 period. Finally, we build a nowcasting model
for world GDP annual growth incorporating weekly data that give timely (one
per week) and accurate forecasts (close to IMF and OECD projections but with
1- to 3-month lead). Policy-wise, this can provide an alternative benchmark
for world GDP growth during crisis episodes when sudden swings in the economy
make usual benchmark projections (IMF's or OECD's) quickly outdated. |
Keywords: |
big data,high frequency,large factor models,mixed frequency,nowcasting,variable selection |
Date: |
2022 |
URL: |
http://d.repec.org/n?u=RePEc:hal:journl:hal-03647097&r= |