Seasonal adjustment of the Bank of Russia Payment System financial flows data

Sergey Seleznev
Natalia Turdyeva
Ramis Khabibullin
Anna Tsvetkova

We present a seasonal adjustment algorithm used by the Bank of Russia to clean the high-frequency payment data. Main features of the data, such as daily frequency, as well as lack of transactions on weekends and holidays made it difficult to use well-known packages like Facebook Prophet (Taylor and Letham, 2017). Following Taylor and Letham (2017), and taking into account the specific characteristics of financial flows data, limitations on the calculation time, we have developed a simple and fast procedure based on a set of trigonometric functions and dummy variables. The procedure shows good results in terms of various quality metrics and can be easily modified to work with more flexible model specifications.

Our smoothing algorithm for daily data can be used to solve a large class of applied problems, including, as was done at the Bank of Russia, the development of real time indicators reflecting changes in economic activity. This task is especially important for informed policy decisions during crisis.

Seasonal adjustment of the Bank of Russia Payment System financial flows data