This paper uses a frequency domain approach to gain insight into the correlation between survey indicators and year-on-year GDP growth.
Using the Baxter-King filter, we split up each series into three components: a short-term, a business cycle (oscillations between 18 and 96 months) and a long-term component. We then calculate how much of the variation of the survey series and GDP growth can be ascribed to these different components. Finally, we use this information together with an analysis of the correlation between survey indicators and year-on-year GDP growth at the different frequencies to explain their overall correlation.
We show that survey indicators, similar to year-on-year GDP growth, do not perfectly reflect business cycle movements but contain cycles of other frequencies. Long-term cycles, in particular, are a nontrivial part of the series' variance. Furthermore, there exist some clear relations between the weight of these cycles in the survey indicators and their correlation with GDP growth. In general, the larger the business cycle component, the larger the correlation, while the opposite is true for the short-term component. The evidence for the long-term component is mixed: although a long-term component seems necessary as the correlation at this frequency is the highest, strong or weak long-term components are typically idiosyncratic, dragging down the overall correlation between the indicator and year-on-year GDP growth.
The paper applies this methodology to the euro area countries (EC survey indicators) and to Belgium separately (NBB business survey indicators). The results are highly comparable.