Estimation methods for computing a branch’s total value added from incomplete annual accounting data. National Bank of Belgium, Working Paper No. 371

Vansteelandt, Stijn and Coppens, François and Reynders, Dries and Vackier, Mario and Van Belle, Laurent (2019) Estimation methods for computing a branch’s total value added from incomplete annual accounting data. National Bank of Belgium, Working Paper No. 371. [Working Paper]

PDF - Published Version Download (1731Kb)

Abstract

Timely monitoring of the economic performance of a particular sector is generally hindered by the fact that not all companies have deposited their annual accounts by the time that an evaluation is made. In view of this, we develop several imputation strategies that each enable predicting a company’s value added based on available information from past and current years for those companies where the value added was not timely reported. For each proposed strategy we discuss the assumptions which must be fulfilled for unbiased estimation and calculate the estimation uncertainty. In particular, the proposed imputation procedures all rely on an assumption of missing at random, namely that the values added in companies that did not yet deposit their annual accounts are similar (in some way) to those in companies with the same characteristics (e.g. the same historical data) that did deposit their accounts by the evaluation date. We show how to retrospectively assess the validity of this assumption, and how to adjust the imputation procedure in case the assumption fails. The importance of the availability of the uncertainty margins should not be underestimated because they will result in faster and higher quality publications. Finally we retrospectively apply each strategy to data from the Belgian Port sector and compare their performance at several evaluation dates. All the proposed methods show good results on these data. The method using (ordinary least squares) regression is preferred because it is very flexible in the use of auxiliary variables, requires weaker assumptions, has smaller estimation uncertainty and is easily automatable.

Actions (login required)