Adaptive outlier treatment technique for rapid recovery in time series
64th ISI World Statistics Congress - Ottawa, Canada
Format: CPS Abstract
Keywords: covid-19, outlier, seasonal_adjustment, time_series, transitory_change
Session: CPS 01 - Statistical methodology II
Monday 17 July 8:30 a.m. - 9:40 a.m. (Canada/Eastern)
In official statistics seasonal adjustment is often used to ensure the comparability of inter-annual data over time and across space. During this process the effects of the outliers need to be managed, and unobserved components of time series (like trend cycle, seasonal adjusted, etc.) need to be calculated.
Nowadays several situations (i.e. COVID-19, energy crisis) cause atypical time periods in our time series. Although these crises could handle by outliers, the standard types may fail to result the corresponding components of the time series. Thus, in HCSO a generalization was made in the practice of one type of outlier, which increase the quality of our seasonal adjustment process.
The outlier that begins with a great difference then it gradually decreases and at the end the difference disappearing called transitory change (TC). This outlier has a parameter (TC rate) which theoretically can take any value between 0 and 1. The parameter defines the rate of the gradually decreases. In practice, this parameter is fixed constant 0.7, but the smaller the parameter, the faster the time series returns to the initial level. In this way the length of an outlier affected time period can be modified.
During our analysis it was found that for example the effect of COVID-19 is often similar to what can be treated by TC but the rate of the gradually decreases did not fit in every case. Therefore, the adaptive TC which was mentioned before was used. We discuss the method that we use to find the suitable TC rate parameter and some interesting example is also described.