End-Point Bias in Trend-Cycle Decompositions : An Application to the Real Exchange Rates of Turkey

Estimating a robust and stable trend is an important challenge for policy analysis. In this paper, we compare alternative approaches by estimating the cyclical component of the real exchange rate series of Turkey. Comparison criterion is the sensitivity of the estimated cycle to additional data points. A formal test reveals that cycle values obtained with all methods change substantially upon new data arrivals. To rank the performance of alternative methods, additional measures underlining the comovement of real-time cycles with the cyclical values when additional data are available, and the magnitude of end-point bias are developed. These criteria show that an unobserved components approach, which assumes orthogonal trend and cycle innovations and a fixed share of trend shocks on the real depreciation fluctuations of 10 percent, dominates alternative filtering methods.