Although high frequency data, i.e. data observed at infra-monthly intervals, could provide valuable information to official statistics, they are rarely modelled due to numerous problems with their estimation. One of the most crucial ones is a proper identification of the various periodicities. High frequency series often include multiple types of seasonality and many other effects that make a distinction between the various periodic components but also between the trend-cycle frequencies and the annual frequencies troublesome. Another challenge is a high volatility of such data, which influence on an identification and modelling of the outliers, breaks and calendar effects. As the availability of high-frequency data is rapidly growing and no officially recommended method for seasonal and calendar adjustment of high frequency time series exists, there is a growing pressure on developing efficient procedures to estimate all seasonal patterns with different periodicities. This paper focuses on a modelling of the daily time series of the currency in circulation in Poland. This series is an important factor that influences the level of banking sector liquidity and is vital for conducting monetary policy. Therefore, there is a need for its proper modelling and forecasting. The aim of this paper is to improve the models currently used by the National Bank of Poland for the currency in circulation. For this purpose the experimental R routines developed by Jean Palate (the National Bank of Belgium) were used. These algorithms allow for an estimation of a model that contain any number of periodicities, as well as an automatic outlier detection and generation of regression variables corresponding to holidays. The decomposition of the series is performed in an iterative way using a canonical decomposition (SEATS), STL or X11.