The Hamilton regression in comparison: Evidence from German business cycles since 1950

Business-cycle adjustment is mostly determined via filter methods, especially the HP filter, or, e.g. within the EU fiscal rules, by a production function approach. James Hamilton put big doubt on the quality of the HP filter estimates, and proposed an alternative regression approach to decompose trend and cycle of time-series. We investigate how the new Hamilton filter compares to the common methods. We find that the Hamilton regression produces partly significantly different results. The average estimated output gap, and its variance, is significantly higher. As a consequence, the average identified cycle length is the shortest in comparison. By construction, the Hamilton regression produces odd results for periods of massive crisis, while it performs better in context of structural breaks. The highest correlation of the Hamilton gaps we find for the EU production-function approach. The identified business cycles, in contrast, do not differ in most cases since 1950. In an ex post evaluation, the HP filter with smoothing factor 20 performs well for Germany, and precisely fulfils the assumption of symmetry.