Title: Seasonal and calendar adjustment of daily time series
Authors: Daniel Ollech - Deutsche Bundesbank (Germany) [presenting]
Abstract: The methods for seasonal adjustment of official data explicitly supported by Eurostat are X-11 and SEATS. Both methods do not allow for adjustment of data with a higher than monthly frequency, even though an increasing number of time series is available with a weekly or daily periodicity. Possible examples range from data on air pollution, web-search keywords, and traffic jam data to economic variables such as exchange rates, online prices, and the amount of Euro banknotes in circulation. The aim is the development of a procedure that makes it possible to estimate and adjust for regularly and periodically reoccurring systematic effects and the influence of moving holidays and trading days in time series with daily observations. To this end, an STL based seasonal adjustment routine is combined with a regression model with ARIMA errors for the estimation of calendar and outlier effects. The latter will also be used for forecasting of the original time series, so as to be able to compute forecasted seasonal factors. The proposed procedure successively estimates and adjusts intra-weekly, intra-monthly and intra-annual periodic movements. The prediction of the original series is based on the regARIMA model which uses trigonometric functions to incorporate monthly and annual seasonality. In addition, the intra-weekly seasonal factors are extrapolated using exponential smoothing. The procedure is evaluated empirically using the currency in circulation in Germany.