Online Program

Return to main conference page

All Times ET

Program is Subject to Change

Tuesday, June 15
Tue, Jun 15, 9:30 AM - 11:00 AM
Modelling of Time Series Data - Challenges and Applications to Economic Statistics

Seasonal adjustment of daily and weekly data published in the UK during the COVID-19 pandemic (310030)

*Duncan Elliott, Office for National Statistics 

Keywords: seasonal adjustment, COVID-19, high-frequency

The COVID-19 pandemic increased demand for time series data at a higher frequency than the more usual monthly and quarterly indicators regularly published by the Office for National Statistics (ONS) in the UK. Experimental high frequency estimates, such as weekly labour market data, daily road traffic count data and other indicators published in a faster economic indicators publication have sought to meet this demand. As with monthly and quarterly time series, daily and weekly time series often exhibit regular periodic fluctuations which can obscure movements of interest. The European Statistical System Guidelines on Seasonal Adjustment recommend to use either signal extraction methods based on ARIMA models, for example as implemented in TRAMO-SEATS or the semi-parametric approach using the X-11 algorithm in software such as X-13ARIMA-SEATS. While these approaches are the most commonly used methods for seasonal adjustment of official statistics using software such as X-13ARIMA-SEATS or JDemetra+, they are not designed to handle daily or weekly data and there is no consensus on methods for seasonal adjustment of higher frequency time series. The sudden demand for higher frequency time series and interest in providing seasonally adjusted and trend estimates to accompany these series lead to experimental publications of higher frequency data using a modified version of TRAMO-SEATS implemented in the experimental rjdhighfreq package. Two case studies are used to identify and discuss some of the challenges of publishing high-frequency seasonally adjusted estimates in official statistics.