Classifying Changes in the Consolidation Perimeter in the German Group Ac-Counts Statistics Using Statistical Learning

Raulf F

Published on: 2019-09-14

Abstract

The Deutsche Bundesbank publishes aggregated change rates of the revenue, the earnings before interest, taxes, depreciation and amortisation and the earnings before interest and taxes of the German non-financial groups listed in the Prime Standard of the Frankfurt Stock Exchange. Due to changes in the consolidation perimeter it is crucial to adjust these growth rates to obtain an unbiased impression of the economic development. However, an adjustment is just possible if it is known for which groups a significant change in the group structure has happened. To find evidence for changes in the perimeter, different statistical learning approaches are tested and combined.

This is the first paper which classifies changes in the consolidation perimeter for data quality maintenance. A committee of a random forest and a weighted nearest neighbors approach which are both oversampled provides the best performing results. The approach avoids over-looking changes in the consolidation perimeter in an efficient an affordable way.

Keywords

Statistics; Stock Exchange; Bundesbank

Introduction

The non-financial group statistics compiled by the Deutsche Bundesbank contain roughly 230 listed groups from 2005 up until the latest available data. All groups are listed in the Frankfurt Stock Exchange’s Prime Standard segment. The statistics include income statement, balance sheet and cash flow statement items, mostly on a quarterly basis. As groups are obliged to publish half-yearly, not quarterly, selected balance sheet and income statement positions are published every six months.

The balance sheet positions are published in aggregated values and in relation to total assets – from 2007 until the latest available data. These positions are not seasonally adjusted. By contrast, the aggregated positions of the income statement and their change rates are seasonally adjusted. Furthermore, the revenue, the earnings before interest, taxes, depreciation and amortization (EBITDA) and the earnings before interest and taxes (EBIT) are also adjusted for significant changes in the consolidation perimeter (changes in the group structure). For example, a merger or a disinvestment causes structural breaks within the time series of affected positions – based on each group.

Data on the biggest groups included in the statistics are gathered manually by Deutsche Bundesbank analysts, while data on the remaining groups are collected by a commercial da-ta provider. During manual data gathering, the analysts look for structural changes in the group reports. Sometimes it is difficult to decide whether or not there is a significant change in the perimeter, since the reports contain only limited information regarding groups’ activities. It is easy to overlook a change in the perimeter. Depending on its magnitude, an over-looked change in perimeter can cause a significantly biased aggregate change rate – based on all groups.

After some adjustments and quality checks (mostly plausibility checks), the manually gathered and the commercial data are merged to obtain the new complete dataset for all groups under consideration. The abovementioned statistics are created from this new dataset.

The statistics thus obtained are used for analysis and for publication. Mistakes in the information relating to changes in the perimeter are generally revealed during the process of assessing the data quality or during the plausibility checks. But sometimes (not uncommonly) a change in perimeter is overlooked throughout the process of data creation. In this case, the change is discovered in the final analysis, which requires a revision of the whole process of data creation.