The computation of the first estimates Mr. Mika Sirviö, Statistics Finland mika.sirvio@stat.fi
Compilation of turnover indices Releases of preliminary turnover indices with the delay less than 75 days VAT data not yet available The first releases are based on the sample only Year-on-year change is estimated from the sample
The sample of turnover indices The sample contains 2000 enterprises Accounts for the majority of economic activity in Finland Year-on-year changes calculated from the sample correlate with aggregate changes for all enterprises
Development of different size categories However, the development varies between different enterprise size categories Turnover has grown at a faster annual rate in small and medium sized manufacturing enterprises than in large enterprises, 5.0% vs. 2.7% In construction sector the difference was even greater, 10.7% vs. 4.5%
Development (trend) of turnover in enterprises of different sizes in manufacturing industry 1995-2006
Revisions of the first estimates Different development by size classes together with the slow accumulation of the data causes revisions in the indices of turnover Other reasons for revisions: changes in classification category (e.g. change of industry) changes in value or source data company reorganisations enterprise openings and closures
Revisions of the year-on-year change of the first estimates by industry
Problems at the turning points of the economy The problems become more crucial at the turning points of the economy Growth and recession usually affect companies of different size at different points in time The business cycle is reflected first in the large enterprises engaging in foreign trade Small enterprises are less likely to survive from economic fluctuation Or they tend to reach the growth path slower than large enterprises
Current imputation methods Five different imputation rules are used for enterprises with missing data (t = month to be estimated): Mean annual change Geometric mean of monthly changes Previous turnover Mean Turnvoer Turnover of comparison month
Current imputation methods The rules are tested with data concerning the five latest months The rule with the smallest largest prediction error is chosen as the best rule
Imputation and recession Recession inflicts also on imputation of missing values, which is based on historical turnover values of enterprises Imputation does not perform well for the enterprises with abnormal observations In the current situation imputations and outlier treatment are controlled more carefully than normally
Goals Improvement of the quality of data acquired from the direct data collection Aim to take into account the influence of the companies left outside the inquiry methodologically Use of regression models to estimate the rest of the population as a whole
Proposed methods for estimation Robust regression models that utilize the information on the wages of the current month