Title: Robust methods for ranking using composite indicators
Authors: Kris Boudt - Vrije Universiteit Brussel and VU Amsterdam (Belgium)
Valentin Todorov - UNIDO (Austria)
Wenjing Wang - Vrije Universiteit Brussel (Belgium) [presenting]
Abstract: Composite indicators (CIs) are widely used for evaluating and ranking multidimensional individual performance, such as countries industrial competitiveness or well-being. Their composite nature necessitates standardizing the individual indicators prior to the aggregating. Popular choices include the use of so-called z-scores and min-max standardization. The former is popular in combination with arithmetic aggregation, while the latter is often used in conjunction with geometric aggregation. Both the z-score and min-max standardization approaches have in common that, because of the standardization, the presence of an extreme observation in the time series leads to an explosion of the scale statistics (standard deviation, range) and thus an implosion to zero of most observations in the standardized time series of that indicator. From the viewpoint of ranking using individual indicators, this has of course no effect on the ranking, but, as we show, it can have large effects on the ranking obtained using the composite indicator. We document in detail the robustness issues of standard approaches to ranking using composite indicators, and propose alternatives that are less sensitive to extreme observations in the data. The main contribution is that we propose a framework to use distribution-based winsorization methods to reduce the impact of outliers on ranking using composite indicators.