NON-PARAMETRIC AND PARAMETRIC APPROACHES TO THE CZECH BUSINESS CYCLE DATING

When composite indicators are used for business cycle analysis their construction usually consists of five steps: 1. pre-selection phase, which is passed only by long time series of indicators that have justified economic relationship with the reference series, broad coverage of economic activity and high frequency of observations, 2. filtering phase, when the time series are seasonally adjusted and de-trended, 3. evaluation phase, when only the best individual indicators with the strongest relationship with the reference series are selected to be included in the composite indicators, 4. aggregation phase, when the composite indicators are created, and 5. presentation of the results. This paper focuses mainly on the evaluation phase of the composite indicators construction. In the evaluation phase cyclical components of all the evaluated individual indicators are compared to the reference series. Their relationship can be described by several methods: the average lead (lag) time between the turning points, cross correlations and number of extra and missing cycles. The analyses therefore depends largely on the correct turning points tracking. This paper describes non-parametric (Bry-Boschan algorithm) and parametric (Markov switching model) approach to tracking the turning points. It applies these methods on several indicators of Czech business cycle and compares their performance. It shows that BryBoschan algorithm is more suitable for analyzing the short time series like those analyzed in this paper.