An Exploration of Trend-Cycle Decomposition Methodologies in Simulated Data
Abstract
This paper uses simulations to explore the properties of the HP filter of Hodrick and Prescott (1997), the BK filter of Baxter and King (1999), and the H filter of Hamilton (2018) that are designed to decompose a univariate time series into trend and cyclical components. Each simulated time series approximates the natural logarithms of U.S. Real GDP, and they are a random walk, an ARIMA model, two unobserved components models, and models with slowly changing nonstationary stochastic trends and definitive cyclical components. In basic time series, the H filter dominates the HP and BK filters in more closely characterizing the underlying framework, but in more complex models, the reverse is true.
Makes the point that the HP filter tends to dominate the Hamilton filter when dealing with data that doesn’t follow relatively simple time series models like random walk or autoregressive processes.
References
Why You Should Never Use the Hodrick-Prescott Filter
The Disappointing Recovery of Output after 2009