Forecasting Financial Inclusion in Ethiopia

A forecasting system that tracks Ethiopia's digital financial transformation using time series methods.

📋 Project Overview

This project builds a forecasting system that predicts Ethiopia's progress on the two core dimensions of financial inclusion as defined by the World Bank's Global Findex:

• ACCESS — Account Ownership Rate
• USAGE — Digital Payment Adoption Rate

Business Context

Ethiopia is undergoing rapid digital financial transformation:

• Telebirr has grown to over 54 million users since launching in 2021
• M-Pesa entered the market in 2023 and now has over 10 million users
• Interoperable P2P digital transfers have surpassed ATM cash withdrawals
• Yet only 49% of Ethiopian adults have a financial account (2024 Global Findex)

🚀 Project Setup

Prerequisites

• Python 3.10+
• pip or conda

Installation

1. Clone the repository

git clone https://github.com/your-username/Forecasting-Financial-Inclusion.git
cd Forecasting-Financial-Inclusion

2. Create a virtual environment

python -m venv .venv
source .venv/bin/activate  # On Linux/Mac
# or
.venv\Scripts\activate  # On Windows

3. Install dependencies

pip install -r requirements.txt

Required Packages

pandas
numpy
matplotlib
seaborn

📁 Project Structure

Forecasting-Financial-Inclusion/
├── data/
│   ├── raw/                           # Original starter dataset
│   ├── processed/                     # Enriched analysis-ready data
│   │   ├── event_indicator_matrix_refined.csv  # Calibrated impact estimates
│   │   └── forecast_2025_2027.csv     # Forecast table with CIs
│   ├── ethiopia_fi_unified_data*.csv  # Main unified dataset
│   ├── reference_codes*.csv           # Valid codes reference
│   ├── Additional Data Points Guide*  # Enrichment guidance
│   └── data_enrichment_log.md         # Documentation of additions
├── notebooks/
│   ├── 01_data_exploration_enrichment.ipynb  # Task 1 notebook
│   ├── 02_exploratory_data_analysis.ipynb    # Task 2 notebook
│   ├── 03_event_impact_modeling.ipynb        # Task 3 notebook
│   └── 04_forecasting_access_usage.ipynb     # Task 4 notebook
├── src/
│   └── __init__.py
├── dashboard/
│   └── app.py
├── models/
├── reports/
│   ├── figures/                       # All visualizations
│   ├── interim_report.md              # Interim submission
│   ├── eda_summary_report.md          # EDA findings
│   ├── event_impact_methodology.md    # Impact modeling methodology
│   └── forecast_executive_summary.md  # Forecast executive summary
├── tests/
│   └── __init__.py
├── requirements.txt
└── README.md

📊 Task 1: Data Exploration and Enrichment

Objective

Understand the starter dataset and enrich it with additional data useful for forecasting ACCESS and USAGE indicators.

Key Findings

Dataset Structure

The unified schema uses record_type to categorize data:

Record Type	Count	Description
observation	30	Measured values from surveys, reports, operators
event	10	Policies, product launches, market entries, milestones
target	3	Official policy goals (NFIS-II targets)
impact_link	14	Modeled relationships between events and indicators

Account Ownership Trajectory (Core ACCESS Indicator)

Year	Rate	Change
2011	14%	—
2014	22%	+8pp
2017	35%	+13pp
2021	46%	+11pp
2024	49%	+3pp

Data Enrichment Summary

Addition Type	Count	Examples
New Observations	10	2011 baseline, bank branches, smartphone penetration
New Events	6	NPS Proclamation, CBE Birr launch, Agent Banking Directive
New Impact Links	6	Event-indicator relationships for ACCESS, USAGE, GENDER

Outputs

• 📓 notebooks/01_data_exploration_enrichment.ipynb — Full exploration and enrichment code
• 📄 data/data_enrichment_log.md — Detailed documentation of all additions
• 📊 data/processed/ethiopia_fi_unified_data_enriched.csv — Enriched dataset

� Task 2: Exploratory Data Analysis

Objective

Analyze patterns and factors influencing financial inclusion in Ethiopia.

Key Insights

1. The Account Ownership Paradox

Despite 65M+ mobile money registrations (Telebirr 54.8M + M-Pesa 10.8M), account ownership only grew +3pp (46% → 49%) from 2021-2024.

• Mobile money-only users are rare (~0.5% of adults)
• Most MM users already have bank accounts (complementary, not substitute)
• Activity rate is only 66% (many dormant registrations)

2. The Digital Crossover Milestone

P2P transactions surpassed ATM withdrawals in FY2024/25 — a historic first for Ethiopia:

• P2P: 128.3M transactions (+158% YoY)
• ATM: 119.3M transactions (+26% YoY)
• Crossover ratio: 1.08

3. Persistent Gender Gap

• Account ownership gap: 18-20pp (56% male vs 36% female)
• Women hold only 14% of mobile money accounts
• Phone ownership gap: 24% (86% male vs 65% female)

4. Infrastructure as Leading Indicator

• 4G coverage doubled: 37.5% → 70.8%
• Bottlenecks: Smartphone (24%), Mobile internet (26.9%)
• Traditional banking very sparse: 0.49 branches, 0.65 ATMs per 100k

5. NFIS-II Target Gap

• Current: 49% | Target: 70% by 2025 | Gap: 21pp
• At current trajectory (+1pp/year), would reach 70% by 2046

Data Quality Assessment

Metric	Value
High confidence data	~75%
Medium confidence data	~25%
Temporal coverage	2011-2025
Core indicators sparse	Findex every 3 years

Outputs

• 📓 notebooks/02_exploratory_data_analysis.ipynb — Full EDA with visualizations
• 📄 reports/eda_summary_report.md — Key findings summary
• 📊 reports/figures/ — All visualizations

🎯 Task 3: Event Impact Modeling

Objective

Model how events (policies, product launches, infrastructure investments) affect financial inclusion indicators.

Methodology

Functional Forms for Event Effects

Form	Use Case	Example
Step	Permanent regulatory changes	NPS Proclamation
Ramp	Infrastructure buildout	4G rollout
Impulse-Decay	Price shocks	FX reform
S-Curve	Technology adoption	Telebirr launch

Comparable Country Evidence

Impact estimates derived from:

• Kenya: M-Pesa (+22% ownership), M-Shwari (first mobile credit)
• Tanzania: Vodacom M-Pesa (+15% ownership)
• India: Jan Dhan Yojana (+20% ownership), UPI (+25% digital payments)
• Rwanda: Agent banking (+12% ownership)
• Bangladesh: bKash (+18% ownership)

Key Findings

Event-Indicator Association Matrix

Created comprehensive matrix showing estimated impact of 14 events on 9 core indicators across ACCESS, USAGE, AFFORDABILITY, and GENDER pillars.

Validation Results

Indicator	Observed Δ	Predicted Δ	Error
Mobile Money Accounts	+4.75pp	+19.25pp	Over-predicted 4x
Account Ownership	+3pp	+10.25pp	Over-predicted 3x
4G Coverage	+33.3pp	+34.5pp	Accurate (1.2pp error)

Key Insight: Mobile money registrations ≠ survey-measured account ownership. Mobile money complements existing bank accounts rather than substituting for them.

Refined Impact Estimates

Applied adjustment factors based on validation:

• ACCESS indicators: Reduced by 50-70% (complementarity effect)
• USAGE indicators: Kept as-is (transaction data validates estimates)
• 4G Coverage: Accurate, no adjustment needed

Confidence Assessment

Level	Count	Description
High	4	Validated, <30% error
Medium	11	Comparable evidence
Low	5	Theoretical only

Outputs

• 📓 notebooks/03_event_impact_modeling.ipynb — Full analysis notebook
• 📄 reports/event_impact_methodology.md — Detailed methodology documentation
• 📊 data/processed/event_indicator_matrix_refined.csv — Calibrated impact estimates
• 📈 reports/figures/ — Impact visualizations (4 new figures)

� Task 4: Forecasting Access and Usage

Objective

Forecast Account Ownership (ACCESS) and Digital Payment Usage for 2025-2027.

Methodology

Given sparse data (5 Findex data points over 13 years), we use:

1. Trend Regression - Linear model on historical Findex data (R² = 0.97)
2. Event-Augmented Model - Trend + expected event effects from Task 3
3. Scenario Analysis - Pessimistic, Base, and Optimistic scenarios

Key Forecast Results

Account Ownership (ACCESS)

Year	Trend Only	Base Scenario	Range (Pess - Opt)	95% CI
2025	54.8%	61.8%	57.8% - 64.4%	[42.9%, 80.8%]
2026	57.7%	73.7%	64.7% - 79.5%	[53.9%, 93.4%]
2027	60.5%	82.5%	70.0% - 90.6%	[61.9%, 103.1%]

Digital Payment Usage

Year	Trend Only	Base Scenario	Range (Pess - Opt)	95% CI
2025	48.6%	59.6%	53.3% - 63.6%	[35.8%, 83.4%]
2026	52.9%	82.9%	66.4% - 93.4%	[59.1%, 106.8%]
2027	57.3%	105.3%	79.1% - 122.0%	[81.5%, 129.1%]

Events with Largest Impact

Event	ACCESS Impact	USAGE Impact	Confidence
Interoperability Full Launch (2026)	+4pp	+16pp	Low
EthioPay Instant Payment (2025)	+3pp	+15pp	Low
Telebirr continued growth	+6pp	+9pp	Medium
Fayda Digital ID rollout	+6pp	+2pp	Low
M-Pesa market penetration	+3pp	+6pp	Medium

NFIS-II Target Assessment

• Target: 70% account ownership by 2025
• Current (2024): 49%
• 2025 Forecast (Base): 61.8%
• Gap: ~8pp
• Conclusion: ⚠️ Target is very unlikely to be met by 2025

Key Uncertainties

1. Data sparsity: Only 5 Findex data points; CI width of ±21pp
2. Event execution: Interoperability & EthioPay timing uncertain
3. Macro headwinds: FX volatility, inflation may slow adoption
4. Survey vs. Admin gap: Mobile money registrations ≠ Findex ownership (4x gap)
5. Gender gap: Women's adoption trajectory could drag overall rates

Outputs

• 📓 notebooks/04_forecasting_access_usage.ipynb — Forecasting notebook
• 📊 data/processed/forecast_2025_2027.csv — Forecast table with CIs
• 📄 reports/forecast_executive_summary.md — Executive summary
• 📈 reports/figures/forecast_scenarios.png — Scenario visualization
• 📈 reports/figures/forecast_decomposition.png — Trend vs event effects
• 📈 reports/figures/nfis_target_gap.png — Gap to NFIS-II target

� Task 5: Interactive Dashboard

Objective

Create an interactive dashboard enabling stakeholders to explore data, understand event impacts, and view forecasts.

Dashboard Features

The Streamlit dashboard (dashboard/app.py) includes four main sections:

📊 Overview Page

• Key metrics summary cards (Account Ownership, Digital Payment, P2P/ATM Ratio, etc.)
• Growth highlights with interactive charts
• Data summary and coverage statistics

📈 Trends Page

• Interactive time series plots with date range selector
• Channel comparison (Mobile Money providers)
• Infrastructure growth visualization (4G, Smartphone penetration)
• Gender gap analysis
• Data download functionality

🔮 Forecasts Page

• Forecast visualizations with 95% confidence intervals
• Model selection (Event-Augmented vs Linear Trend)
• Forecast summary table
• Key milestones and projected achievements

🎯 Inclusion Projections Page

• Progress toward 70% NFIS-II target (gauge chart)
• Scenario comparison (Pessimistic, Base, Optimistic)
• Answers to consortium's key questions
• Report download functionality

Running the Dashboard Locally

1. Ensure dependencies are installed

pip install -r requirements.txt

2. Ensure processed data exists The dashboard requires these files in data/processed/:

• ethiopia_fi_unified_data_enriched.csv
• forecast_2025_2027.csv
• event_indicator_matrix_refined.csv

If missing, run the notebooks in order (01-04) to generate them.

3. Start the dashboard

cd /path/to/Forecasting-Financial-Inclusion
streamlit run dashboard/app.py

4. Access the dashboard Open your browser to http://localhost:8501

Dashboard Screenshot Sections

Section	Description
Overview	8 KPI cards, P2P vs ATM chart, Account Ownership trajectory
Trends	Pillar-filtered time series, Gender gap, Infrastructure growth
Forecasts	Interactive forecasts with CI, Model toggle, Milestone timeline
Projections	Scenario gauge, Comparison chart, Key Q&A

Outputs

• 📱 dashboard/app.py — Complete Streamlit application
• 📄 Requirements updated with streamlit>=1.31.0 and plotly>=5.18.0

🔜 Upcoming Tasks

• Task 6: Final presentation and report

👥 Team

Tutors: Kerod, Mahbubah, Filimon

📅 Key Dates

• Challenge Introduction: January 28, 2026
• Interim Submission: February 1, 2026
• Final Submission: February 3, 2026

📚 Data Sources

• World Bank Global Findex
• IMF Financial Access Survey
• GSMA Intelligence
• National Bank of Ethiopia
• Ethio Telecom Reports
• EthSwitch Annual Reports

---

Selam Analytics — Financial Technology Consulting for Emerging Markets