Anticipatory Survey Data in Leading Indicators: Gains from Using Data at Sectoral Level and Responses in Non-Balanced Form

This paper suggests improvements on leading indicators based on survey data. Three main point are discussed. a) The use of survey data at sectoral level results in a longer lead of the indicator series. Exploiting the information contained in "no change" answers by applying the concept of "canonical coherence" and by running regressions on certain wave-lengths to higher cross-spectral coherencies between the indicator and the reference time series. b) Lead-lag relationships calculated from the frequency domain are compared with results of the time domain, thereby highlighting potential superpositions of seasonal and business cycles. c) "Out of sample" forecasts, finally, reveal an inferior performance of the traditional balance concept. It is dominated by a weighted average of "worse" and "equal" responses, and also by the application of the simple "worse" share. Surprisingly, the latter is even the overall dominating alternative.