|
Abstract:
|
This paper addresses problems in trend estimation when the main aim is the quick and accurate detection of turning points. The current economic uncertainty creates a demand for reliable estimates of turning points. Unfortunately, most statistical agencies do not produce trend estimates because of the uncertainty associated with them. We investigate a number of methods that are widely used, such as Census X-11, SEATS, Structural Time Series, as well as methods not commonly used, such as the Direct Filter Approach. Each trend estimation method is assessed against the following criteria: (i) accurate and quick identification of turning points; (ii) minimization of the number of false turning points; (iii) minimal revisions; (iv) unbiased revisions. The trend estimation methods are compared through simulation studies to assess their performance, as well as applications in real time series.
|
