Skip to content
BranchesTags

Repository files navigation

KES Exchange Rate Time Series Analysis (2015-2025)

A comprehensive statistical and machine learning analysis of the Kenyan Shilling exchange rate against major international currencies (USD, EUR, GBP) using advanced time series methods and deep learning forecasting.

Live Demo: Interactive Dashboard

Table of Contents

Overview

This project provides a complete end-to-end analysis of Kenyan Shilling (KES) exchange rates over a 10-year period (September 2015 - September 2025). The analysis combines traditional econometric methods with state-of-the-art machine learning techniques to deliver accurate forecasts and actionable insights for:

  • Corporate treasury departments
  • Import/export businesses
  • Financial institutions
  • Portfolio managers
  • Policy makers

Data Source

All exchange rate data sourced from the Central Bank of Kenya (CBK) official website: [www.centralbank.go.ke](https://www.centralbank.go.ke/)

Data Specifications:

  • Frequency: Daily indicative rates
  • Period: September 29, 2015 - September 29, 2025
  • Observations: ~2,610 per currency
  • Quality: Official CBK rates based on market transactions
  • Variables: Date, Price (Close), Open, High, Low, Volume, Change %

Key Achievement: LSTM neural networks achieved 95-99% accuracy (R²) in exchange rate prediction, dramatically outperforming traditional statistical methods.

image

Live Dashboard

Access the live interactive dashboard here: https://kes-usd-forex-timeseries-rbgqw6jkakwstauwsptzhu.streamlit.app/

The dashboard provides:

  • Real-time currency selection and analysis
  • Interactive visualizations with zoom and hover capabilities
  • LSTM-powered 6-month forecasts
  • Strategic recommendations by stakeholder type
  • Mobile-responsive design
image

Project Structure

kes-exchange-rate-analysis/
│
├── data/
│   ├── USD_KES Historical Data.csv      
│   ├── EUR_KES Historical Data.csv      
│   └── GBP_KES Historical Data.csv      
│
├── kesforextimeseries.ipynb             # Main Jupyter notebook with complete analysis
├── app.py                               # Streamlit web dashboard
├── requirements.txt                     
├── README.md

Features

Comprehensive Statistical Analysis

  • Stationarity Testing: ADF and KPSS tests with first-order differencing
  • Time Series Decomposition: Trend, seasonal, and noise component separation
  • Correlation Analysis: Price and returns correlation matrices
  • Volatility Modeling: Rolling 30-day standard deviation analysis
  • Seasonality Detection: Monthly pattern identification
  • Signal-to-Noise Ratio: Forecast reliability assessment
image image image

Advanced Machine Learning Forecasting

  • Random Forest Regressor: Ensemble learning with 200 trees
  • Gradient Boosting: Sequential error correction approach
  • LSTM Neural Networks: Deep learning for temporal sequences
  • Performance Metrics: MAE, RMSE, R², MAPE evaluation
  • 6-Month Forecasts: Extended predictions with confidence intervals
newplot (3)

Interactive Visualizations

  • Plotly-based interactive charts
  • Normalized comparison (base year 2015 = 100)
  • Yearly performance bar charts
  • Monthly seasonality patterns
  • Time series decomposition plots
  • Actual vs predicted comparisons
  • LSTM forecast visualizations with confidence bands

Installation

Prerequisites

  • Python 3.8 or higher
  • pip package manager
  • Jupyter Notebook or JupyterLab (for notebook)

Setup

  1. Clone the repository:
git clone https://github.com/yourusername/kes-exchange-rate-analysis.git
cd kes-exchange-rate-analysis
  1. Create virtual environment (recommended):
python -m venv venv

# On Windows
venv\Scripts\activate

# On macOS/Linux
source venv/bin/activate
  1. Install dependencies:
pip install -r requirements.txt

Dependencies

pandas>=1.5.0
numpy>=1.23.0
matplotlib>=3.6.0
seaborn>=0.12.0
plotly>=5.11.0
statsmodels>=0.14.0
scipy>=1.9.0
scikit-learn>=1.1.0
tensorflow>=2.10.0
streamlit>=1.25.0
jupyter>=1.0.0

Usage

Accessing the Live Dashboard

Simply visit: https://kes-usd-forex-timeseries-rbgqw6jkakwstauwsptzhu.streamlit.app/

No installation required! The dashboard is fully hosted and accessible from any device.

Running the Jupyter Notebook

  1. Start Jupyter:
jupyter notebook kesforextimeseries.ipynb

  1. Run all cells sequentially to perform complete analysis

  2. Customize analysis by modifying parameters in configuration cells

Running the Dashboard Locally

  1. Launch Streamlit app:
streamlit run app.py

  1. Access in browser: Opens automatically at http://localhost:8501

  2. Navigate pages: Use sidebar menu to explore different analysis sections

Quick Start Example

import pandas as pd
import numpy as np
from statsmodels.tsa.statespace.sarimax import SARIMAX

# Load data
usd_data = pd.read_csv('USD_KES Historical Data.csv')
usd_data['Date'] = pd.to_datetime(usd_data['Date'])
usd_data = usd_data.sort_values('Date')

# Fit SARIMA model
model = SARIMAX(usd_data['Price'], order=(1,1,1), seasonal_order=(1,1,1,12))
results = model.fit()

# Generate forecast
forecast = results.get_forecast(steps=180)
print(forecast.predicted_mean)

Analysis Components

1. Data Preprocessing

  • Date parsing and sorting
  • Missing value handling (forward-fill, backward-fill)
  • Returns calculation (daily percentage changes)
  • Feature engineering for ML models

2. Stationarity Analysis

Price Level Tests:

  • ADF Test: All p-values >0.60 (non-stationary)
  • KPSS Test: All p-values <0.01 (non-stationary)

After First-Order Differencing:

  • ADF Test: All p-values <0.001 (stationary)
  • KPSS Test: All p-values >0.05 (stationary)
  • Validates SARIMA(p,1,q) differencing parameter

3. Decomposition

Components Extracted:

  • Trend: Long-term directional movement (2-2.5% annual depreciation)
  • Seasonal: 12-month recurring patterns
  • Residual: Random noise after removing predictable patterns

Signal-to-Noise Ratios:

  • USD/KES: 4.56 (Strong signal, most predictable)
  • GBP/KES: 2.91 (Moderate signal)
  • EUR/KES: 2.52 (Weak signal, least predictable)

4. Correlation Analysis

Price Correlations: Very high (>0.90) - limited diversification Returns Correlations: Moderate (0.30-0.68) - some tactical opportunities

image

5. Trend Analysis

  • Linear trend: Quantifies long-term depreciation
  • Polynomial trend: Captures acceleration/deceleration
  • Trend strength: 52.8-68.9% (USD strongest)

6. Seasonality

Depreciation Months (High FX Demand):

  • January: +1-4% (year-end demand)
  • April-May: +2-5% (tax payments, school fees)
  • December: +2-6% (holiday imports)

Appreciation Months (High FX Supply):

  • February-March: -3 to -9% (diaspora remittances)
  • September: -2 to -3% (export earnings)

Machine Learning Models

Model Architectures

Random Forest:

  • 200 decision trees
  • Features: 30 lags + rolling statistics (7d, 30d) + EMAs
  • Max depth: 20
  • Min samples split: 5

Gradient Boosting:

  • 200 estimators
  • Learning rate: 0.1
  • Max depth: 5
  • Subsample ratio: 0.8

LSTM Neural Network:

  • Architecture: Input(60) → LSTM(50) → Dropout(0.2) → LSTM(50) → Dropout(0.2) → Dense(25) → Dense(1)
  • Optimizer: Adam
  • Loss: MSE
  • Early stopping: Patience 10

Model Performance

Currency Model MAE RMSE R² MAPE (%)
USD/KES Random Forest 4.18 5.02 0.73 3.04
USD/KES Gradient Boosting 4.70 5.31 0.70 3.45
USD/KES LSTM 1.37 2.17 0.95 1.01
EUR/KES Random Forest 4.44 5.99 0.71 2.92
EUR/KES Gradient Boosting 4.90 6.45 0.66 3.24
EUR/KES LSTM 1.00 1.49 0.98 0.68
GBP/KES Random Forest 4.97 6.87 0.64 2.78
GBP/KES Gradient Boosting 5.15 7.09 0.62 2.89
GBP/KES LSTM 0.83 1.20 0.99 0.47

Key Achievement: LSTM achieved 60-80% improvement over traditional SARIMA methods.

image image

Results

Historical Performance (2015-2025)

Currency Pair Total Depreciation Annualized Rate Volatility Trend Strength
USD/KES +22.0% +2.06%/year 0.21% 68.9% (Strong)
EUR/KES +28.0% +2.49%/year 0.51% 64.0% (Moderate)
GBP/KES +10.0% +0.84%/year 0.61% 52.8% (Weak)
image *10-year historical performance showing three distinct phases*

Three Historical Phases

  1. Stability (2015-2019): Gradual depreciation, low volatility
  2. Crisis (2020-2023): COVID-19 impact, 2023 peak (27-35% depreciation)
  3. Recovery (2024-2025): Sharp correction (18-23% appreciation), stabilization

LSTM 6-Month Forecasts

Currency Current Rate 6M Forecast Expected Change
USD/KES 129.20 149.61 +15.80%
EUR/KES 151.51 168.96 +11.52%
GBP/KES 173.48 197.52 +13.86%

Interpretation: All forecasts project continued depreciation, with USD showing highest expected movement.

Strategic Recommendations

For Corporate Treasurers:

  • Hedge 60-80% of 6-month exposure
  • Prioritize USD hedging (highest forecast depreciation)
  • Use LSTM forecasts for decision-making

For Importers:

  • Execute forward contracts immediately
  • Negotiate USD-denominated contracts
  • Front-load imports in Oct-Nov (before seasonal peak)

For Exporters:

  • Delay EUR/GBP conversions until favorable periods
  • Accelerate USD collections
  • Diversify currency mix: 40% USD, 30% EUR, 20% GBP

For Investors:

  • Hedge 70-80% of KES-denominated assets
  • Favor USD-denominated securities
  • Reduce KES exposure when volatility >0.8%

Key Findings

  1. Long-term Trend: KES depreciates 2-2.5% annually against major currencies
  2. 2023 Crisis: Exceptional year with 27-35% depreciation (not the new normal)
  3. Current Valuation: Rates fairly valued near long-term trend lines
  4. Volatility Status: Normalized to historical levels after crisis period
  5. USD Advantage: Most stable (0.21% volatility), strongest trend (68.9%)
  6. Seasonal Patterns: Exploitable for 2-4% annual savings through timing
  7. ML Superiority: LSTM achieved 95-99% accuracy vs 40-65% for SARIMA
  8. Forecast Outlook: 11-16% depreciation expected over next 6 months

Contributing

Contributions are welcome! Here's how you can help:

Areas for Contribution

  1. Additional Features:

    • Incorporate macroeconomic indicators (interest rates, inflation)
    • Add sentiment analysis from news sources
    • Implement real-time data feeds
    • Develop high-frequency (intraday) analysis

  2. Model Enhancements:

    • Attention mechanisms for LSTM
    • Transformer architectures
    • Ensemble methods combining multiple models
    • Bayesian approaches for uncertainty quantification

  3. Visualization Improvements:

    • Additional interactive dashboards
    • Real-time monitoring alerts
    • Custom report generation
    • Mobile app development

  4. Documentation:

    • Tutorial videos
    • Academic paper write-up
    • API documentation
    • Use case examples

How to Contribute

  1. Fork the repository
  2. Create feature branch (git checkout -b feature/AmazingFeature)
  3. Commit changes (git commit -m 'Add AmazingFeature')
  4. Push to branch (git push origin feature/AmazingFeature)
  5. Open Pull Request

Guidelines

  • Follow PEP 8 style guide for Python code
  • Add tests for new features
  • Update documentation
  • Ensure reproducibility with random seeds

Contact

Email: aimemuganga07@gmail.com

Project Link: https://github.com/DevAime/kes-usd-forex-timeseries.git

Past performance does not guarantee future results. Users should:

  • Conduct independent due diligence
  • Consult qualified financial advisors
  • Consider their specific circumstances
  • Understand the risks of foreign exchange trading

The authors assume no liability for decisions made based on this analysis.

⭐ Star this repository if you found it helpful!

About

A comprehensive statistical and machine learning analysis of the Kenyan Shilling exchange rate against major international currencies (USD, EUR, GBP) using advanced time series methods and deep learning forecasting.

kes-usd-forex-timeseries-rbgqw6jkakwstauwsptzhu.streamlit.app/

Topics

data-sciencedeep-learningfinanceforexjupyter-notebooklstmmachine-learningpythonquantitative-financescikit-learnstreamlittensorflow

Resources

Readme
Activity

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages