Filtering techniques for big data and big data based uncertainty indexes - Products Statistical working papers

This work is concerned with the analysis of outliers detection, signal extraction and decomposition techniques related to big data. In the first part, also with the use of a numerical example, we investigate how the presence of outliers in the big unstructured data might affect the aggregated time series. Any outliers must be removed prior to the aggregation and the resulting time series should be checked further for outliers in the lower frequency. In the second part, we explore the issue of seasonality, also continuing the numerical example. Seasonal patterns are not easily identified in the high frequency series but are evident in the aggregated time series. Finally, we construct uncertainty indexes based on Google Trends and compare them to the corresponding Reuters-based indexes, also checking for outliers and seasonal components.