Do Microfinance Institutions Prioritise Need? Evidence from Loan Allocation and Repeat Borrowing Patterns

Replication files & code for an ECO225 course project.

---

📄 Project Description

This repository contains the paper PDF, conference presentation PDF, code, and supporting files, for a regression and machine learning analysis of microloans from Kiva.org. The project aims to use loan descriptions classified into low-need and high-need loans to investigate whether microfinance institutions allocate loan capital in proportion to borrower financial need, and whether repeated borrowing grows or shrinks.

---

💻 Code & Analysis

Main File:

• ECO225_Code.ipynb – A Jupyter Notebook containing:
  • Data cleaning and preprocessing
  • Visualisations of key trends
  • Feature engineering
  • Machine learning pipeline for loan classification
  • Regression analysis of loans and disbursement amounts

To Replicate:

1. Download the required datasets (see below)
2. Place them in the appropriate folder
3. Open the notebook and run all cells sequentially

---

🗂️ Datasets

External (must be downloaded):
📥 Kiva Kaggle Dataset

Required files:

• loans.csv – Primary loan-level data
• loan_coords.csv – Geographical coordinates for each loan

Included in this repository:

• gdp_data.csv – Country-level GDP
• mpi_data.csv – Multidimensional Poverty Index
• llm_subsample.csv – Subsample of 2,000 LLM-labeled loans used in classification tasks
• ECO225_Paper.pdf – Paper PDF for the project
• Tech-Econference_Presentation.pdf – Presentation PDF for the project at the 5th Annual Econ-Tech Conference (2025), University of Toronto

---

⚙️ Dependencies

• Python 3.8+
• Jupyter Notebook
• Core packages used: pandas, numpy, sklearn, matplotlib, seaborn

---

📌 Notes

Due to GitHub’s file size limits, only essential supporting files are included. For full replication, refer to the Kaggle dataset linked above.