Title: Comparative mean value with an application to personal financial data analysis
Authors: Andrej Svetlosak - University of Edinburgh (United Kingdom) [presenting]
Miguel de Carvalho - School of Mathematics, University of Edinburgh (Portugal)
Raffaella Calabrese - University of Edinburgh (United Kingdom)
Abstract: A statistical method is proposed for comparing a reference subject to a group of peers, where peers are selected based on similarity of values of a comparison variable. The proposed methodology is motivated by the need of comparing personal finances of a reference subject to subjects with similar characteristics. Comparative mean values are introduced so to identify if behaviour of an individual varies from the average of their peers. The similarity can be controlled by a parameter. We analyse the properties of the construct in respect to this parameter. First, we introduce our construct, the comparative mean value, in a time \textendash invariant setting. A time \textendash varying version is developed by taking advantage of non \textendash parametric regression methods. The construct is tested in a simulation study. Results support an overall good performance of the methods. We then apply the proposed method to data provided by a major financial service provider in the UK, Money Dashboard. The dataset consists of 10,689 customers of a financial services provider and contains information on customer accounts from 49 financial institutions in the UK. We showcase how the time \textendash varying comparative mean expenditure can be used to identify the spending behaviour of an individual in comparison to their peers.