The effects of persistence and non-response in short-term statistics

Abstract

Short-term business statistics are usually based on surveys with a rotating panel design in order to reduce the statistical burden for respondent. The partial overlapping of two periods allow for two different estimators for the year on year growth rate: (i) one that is based on the estimated population totals and (ii) one that is solely based on firms respondent in both periods. Indicators related to short-term statistics, such as the quarterly turnover for service sector, are a clear example on this way. Moreover these indicators are characterized by an high level of persistence which could have an effect on the performance of the two estimators. This paper aims to provide evidences on the two estimators in framework characterized by different degrees of persistence and non- response both modeled at a micro level. First the differences are shown by means of a simulation exercise. Secondly we provide an application based on the quarterly index of italian turnover for the Warehousing and Support Activities for Transportation. The results suggest that the estimator based on population totals outperforms the one based on firms respondent when the persistence and non-response are high as the one observed in the empirical case.