Smooth and timely business cycle indicators for noisy Swedish data

Noise in statistical time series is often overlooked when selecting the best forecasting model by minimizing forecast errors. An ‘error’ implies that one knows the true (noise-free) outcome. Instead of merely trying to forecast a noisy outcome, the authors construct entirely new indicators, based on business tendency survey data and statistical time series. False turning point signals are avoided by exponential smoothing. A special trigger is found in the joint behavior of model generated smoothed and unsmoothed forecasts, by which smoothing can be switched off in sharp turns, and this avoids late turning point signals that would occur with smoothed data.