Constructing leading indicators from non-balanced sectoral business survey series

This paper considers the construction of leading indicators based on monthly survey data from the Ifo Institute, Munich. The three main points covered in the paper are: (a) The use of survey data at the sectoral level results in a longer leading indicator. By taking a non-balanced form of the survey answers and exploiting the information contained in 'no change' responses through the use of canonical coherence, regressions on certain wave lengths lead to higher cross-spectral coherencies between the survey data and the actual business cycle, (b) Comparisons of frequency domain and time domain results for lead-lag relationships highlight the roles of seasonal and business cycles, (c) Out of sample forecasts reveal that the traditional balance concept is dominated by a weighted average of 'worse' and 'equal' responses. Surprisingly, the best results come from using the 'worse' share.