Dealing with a Break in Series - Job Vacancies
Seasonal adjustment is designed to remove predictable calendar (or seasonal) patterns from time series data; where exactly the same thing is measured at regular time intervals. The time series of the ABS Job Vacancies Survey (JVS) was broken in the five quarters from August 2008 to August 2009. During this period, ABS temporarily ceased the collection of the Job Vacancies data and, in November 2009, reinstated the survey. The break in series coincided with the Global Financial Crisis which made it difficult to predict the missing data using historical data.
Availability of long and unbroken series is key to undertaking time series analyses, including seasonal adjustments. In order to continue the production of seasonally adjusted and trend estimates for the Job Vacancies series, ABS decided to fill up the data gap using an econometric model. Conceptually, job advertisement was considered closely related to the job vacancies because they measure a similar concept. Initially, job advertisement time series from non-ABS sources were evaluated in the modelling exercise. However, there is no co-integration relationship between these series and the ABS Job Vacancies series, meaning that the job advertisement time series do not contain sufficient information about the Job Vacancies trend direction.
The Time Series Analysis (TSA) section found that there was a six-month lag between the business cycles of the Job Vacancy rate and the unemployment rate. The scale of changes in the Job Vacancy rate was consistent with the scale of changes in the unemployment rate. So TSA also evaluated autoregressive (two-quarter lag) models relating the Job Vacancy estimates to Employment estimates, the Full Time Equivalent estimates, and the Hours Worked estimates. After consultation with Treasury -- the main client who uses the job vacancy series in the TRYM for macroeconometric modelling -- TSA chose the last option.
TSA reintroduced seasonality and used these modelled estimates in the gap between measured JVS estimates. Thus series continuity is mostly restored.
The modelled data may be obtained from the TRYM Modeller's database, on the ABS website under cat no. 1364.0.15.003.
For more information, please contact Rachel Barker on (02) 6252 6183 or firstname.lastname@example.org.