Title: Aggregation of bank transaction data into a weekly Index of economic activity
Authors: Feliciaan De Palmenaer - Universiteit Gent and Vrije Universiteit Brussel (Belgium) [presenting]
Milan van den Heuvel - Ghent University (Belgium)
Kris Boudt - Vrije Universiteit Brussel and VU Amsterdam (Belgium)
Koen Schoors - Ghent University (Belgium)
Abstract: Bank transaction data gives a high-frequency overview of monetary transfers between firms, aggregating certain transactions, the output of a firm can be constructed for a week. When aggregating all firms over a week, an index can be constructed to show the week-over-week growth or changes in economic activity of the Belgian economy. The anonymized dataset contains all bank transactions of firms that are clients of a Belgian bank. Going from raw transactions data to the weekly index is a multi-step process with some challenges. Aggregating is done in three steps, first as much as possible, filtering, seasonal adjustments and cleaning is made on a per firm basis. The next step is to group all firms in the same sector, using NACE-codes, an EU wide classification nomenclature, and take the mean of all firms' outputs. The final step is to aggregate these sectoral indices to one weekly economic index. The first correction we had to on our data was to add sectoral seasonality adjustments to firms' outputs to prevent cyclical deviations, using the geometric mean of the weekly outputs of firms in this sector for 10 years. Instead of weeks, pseudo-weeks are used where each month consists out of 4 weeks. To correct for the longer weeks, all weekly data is normalized. And to prevent differences between the weight of a sector in the overall economy and the weight in the bank index, correct weights were used from the Belgian statistical office when constructing the index.