Optimal HP filtering for South Africa

Among the various methods used to identify the business cycle from aggregate data, the Hodrick-Prescott filter has become an industry standard – it ‘identifies’ the business cycle by removing low-frequency information, thereby smoothing the data. Since the filter’s inception in 1980, the value of the smoothing constant for quarterly data has been set at a ‘default’ of 1600, following the suggestion of Hodrick and Prescott (1980). This paper argues that this ‘default value’ is inappropriate due to its ad hoc nature and problematic underlying assumptions. Instead this paper uses the method of optimal filtering, developed by Pedersen (1998, 2001, and 2002), to determine the optimal value of the smoothing constant for South Africa. The optimal smoothing constant is that value which least distorts the frequency information of the time series. The result depends on both the censoring rule for the duration of the business cycles and the structure of the economy. The paper raises a number of important issues concerning the practical use of the HP filter, and provides an easily replicable method in the form of MATLAB code.

C22, E32

Hodrick-Prescott filter, Spectral analysis, Ideal filtering, Optimal filtering, Distortionary filtering, Business cycles, MATLAB