|
Abstract:
|
The amount of data available at high frequency has dramatically increased during the last years. At the beginning they were essentially coming from monetary and financial sectors and a minor part from the commodity and energy sectors. Later on, this set has increased covering information much more related to the real economic activity such as scanner price, credit card utilization, mobile phone data and also web related data. Often these data are unstructured and need to be subject to transformation to be put in the form of time-series. At this point, in order to be really useful for macro-economic nowcasting, they need to be filtered in order to remove all deterministic non-linear components as well as all the annual and sub annual movement according to the target variables they are supposed to be related with. Unfortunately, this exercise is not a trivial one and creates some conceptual and computational problems which impair the application of traditional seasonal adjustment methods. In this paper we are presenting some solutions to produce seasonal adjusted data at high frequency level. Then we are comparing their performance within a small-scale simulation exercise.
|
