On Adjusting the Hodrick-Prescott Filter for the Frequency of Observations

Abstract

Abstract-This paper studies how the Hodrick-Prescott filter should be adjusted when changing the frequency of observations. It complements the results of Baxter and King (1999) with an analytical analysis, demonstrating that the filter parameter should be adjusted by multiplying it with the fourth power of the observation frequency ratios. This yields an HP parameter value of 6.25 for annual data given a value of 1600 for quarterly data. The relevance of the suggestion is illustrated empirically.

Testing the Relationship between Government Expenditure and Private Investment: The Case of Small Open Economies

“…where  -the penalty parameter is set to 6.25 (according to the recommendations of [17] [19] (GDP, millions of EUR; GDP, EUR per inhabitant; Investment, EUR per inhabitant; Private investment, % of GDP), Annual Macroeconomic Database of the European Commission [20] (General government total expenditure, gross fixed capital formation by private sector, Price deflator of gross fixed capital formation, Price deflator of real total expenditure of general government), World development indicators [21] (Trade, % of GDP) databases. The study covers the period from 1996 till 2012, using annual data.…”

Weight of the Crisis: Evidence from Newborns in Argentina

“…Since we are using monthly data, we choose a smoothing parameter of 129,600 (Ravn and Uhlig, 2002). Our …ndings are not sensitive to the method used, as our use of other …ltering methods attests.…”

Day After Tomorrow: Designing an Optimal Fiscal Strategy for Libya

“…If there is no noise, the series is fully informative and the weight -λ -should be equal to zero. While a λ of 100 is typically the choice for annual data in the empirical literature, Baxter and King (1999) argue that a value of around 10 is more reasonable, and Ravn and Uhlig (2002) recommend a λ of 6.5 for estimations using annual data. After experimenting with a range of smoothing parameters, we find marginal computational differences in the empirical analysis and adopt a λ of 100.…”