How good is good? Probabilistic benchmarks and nanofinance+

TL;DR

This paper introduces a two-step methodology combining noise filtering and machine learning to establish probabilistic benchmarks in noisy datasets, demonstrated through nano-finance+ data analysis.

Contribution

It proposes a novel two-step approach for calculating probabilistic benchmarks in noisy data, integrating hyperbolic undersampling and relevance vector machines.

Findings

Higher discrimination power with macro-economic variables.

Distinct benchmarks for rural versus urban groups.

Effective noise reduction in KPI data.

Abstract

Benchmarks are standards that allow to identify opportunities for improvement among comparable units. This study suggests a 2-step methodology for calculating probabilistic benchmarks in noisy data sets: (i) double-hyperbolic undersampling filters the noise of key performance indicators (KPIs), and (ii) a relevance vector machine estimates probabilistic benchmarks with denoised KPIs. The usefulness of the methods is illustrated with an application to a database of nano-finance+. The results indicate that-in the case of nano-finance groups-a higher discrimination power is obtained with variables that capture the macro-economic environment of the country where a group operates. Also, the estimates show that groups operating in rural regions have different probabilistic benchmarks, compared to groups in urban and peri-urban areas.