Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Basic model for ranking microfinance institutions

Dmitry Dudukalov, Evgeny Prokopenko

TL;DR

This study tackles ranking microfinance institutions on an aggregator platform using post-click conversion data. It introduces a baseline ranking via a pairwise comparison of MFIs across five features—average rating, loan approval rate, fairness, service period, and earnings per click—implemented through a Markov-chain model that yields a stationary distribution $\pi$ for ranking. The approach is supported by analysis of three real Russian datasets (conversion, MFI descriptions, and clicks) and employs Bayesian normalization to handle sparse observations and cold-start issues. The work emphasizes interpretability and practicality, with public data releases on GitHub and potential applicability to microinsurance aggregators lacking personal data.

Abstract

This paper discusses the challenges encountered in building a ranking model for aggregator site products, using the example of ranking microfinance institutions (MFIs) based on post-click conversion. We suggest which features of MFIs should be considered, and using an algorithm based on Markov chains, we demonstrate the ``usefulness'' of these features on real data. The ideas developed in this work can be applied to aggregator websites in microinsurance, especially when personal data is unavailable. Since we did not find similar datasets in the public domain, we are publishing our dataset with a detailed description of its attributes.

Basic model for ranking microfinance institutions

TL;DR

This study tackles ranking microfinance institutions on an aggregator platform using post-click conversion data. It introduces a baseline ranking via a pairwise comparison of MFIs across five features—average rating, loan approval rate, fairness, service period, and earnings per click—implemented through a Markov-chain model that yields a stationary distribution for ranking. The approach is supported by analysis of three real Russian datasets (conversion, MFI descriptions, and clicks) and employs Bayesian normalization to handle sparse observations and cold-start issues. The work emphasizes interpretability and practicality, with public data releases on GitHub and potential applicability to microinsurance aggregators lacking personal data.

Abstract

This paper discusses the challenges encountered in building a ranking model for aggregator site products, using the example of ranking microfinance institutions (MFIs) based on post-click conversion. We suggest which features of MFIs should be considered, and using an algorithm based on Markov chains, we demonstrate the ``usefulness'' of these features on real data. The ideas developed in this work can be applied to aggregator websites in microinsurance, especially when personal data is unavailable. Since we did not find similar datasets in the public domain, we are publishing our dataset with a detailed description of its attributes.
Paper Structure (10 sections, 5 equations, 13 figures, 2 tables)

This paper contains 10 sections, 5 equations, 13 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Microloans in Russia
  3. Problem description
  4. Microloans Data
  5. Conversion dataset
  6. MFIs description dataset
  7. Click dataset
  8. Ranking algorithm
  9. Feature description
  10. Description of the ranking algorithm

Figures (13)

  • Figure 1: Status and loan type distributions.
  • Figure 2: Dependence of the number of applications on MFI page rank. Left: cumulative number of applications increases with time (for eight highest ranks); right: the number of applications at the last available timestamp.
  • Figure 3: Percentage of applications by MFI id and MFI page rank. At different times, and on different web pages, the rank of a fixed MFI may vary, which is the reason why there is no one-to-one correspondence between MFI id and MFI page rank.
  • Figure 4: Visualization of clients by their number of applications (left). MFI id chosen by clients who submitted more than $1$ application (right). Clients are ordered by decreasing number of applications. Color corresponds to the number of different MFIs for a client. It is easy to see that clients apply to various MFIs.
  • Figure 5: Income, number of sales and EPC for $4$ indicative MFIs.
  • ...and 8 more figures