The Remittance Blueprint: Data-driven Intelligence for Sri Lanka

This repository contains the dataset, analytical code, and research findings for "The Remittance Blueprint: Data-driven Intelligence for Sri Lanka", a comprehensive study analyzing 32 years (1994–2025) of Sri Lankan labor emigration and worker remittances. The study leverages a 384-month harmonized dataset, employing exploratory data analysis, stationarity-corrected econometric modeling (ADF, Johansen Cointegration, VECM), and supervised machine learning (Ridge Regression) to analyze and forecast remittance inflows.

---

📄 Abstract & Executive Summary

Worker remittances consistently account for 6–9% of Sri Lanka's GDP, serving as a critical buffer for foreign exchange reserves and debt sustainability. This study addresses key gaps in existing research by developing a unified 32-year, 384-month harmonized dataset from eight authoritative sources.

Our empirical findings reveal that remittance inflows are primarily driven by external macroeconomic variables (specifically exchange rate dynamics and global oil prices) rather than domestic economic indicators:

• Vector Error Correction Model (VECM): Confirms a stable long-run equilibrium. The system exhibits a speed-of-adjustment of $\alpha \approx -0.15$, correcting roughly 15% of disequilibrium monthly and returning to stability in ~6.7 months.
• Impulse Response Functions (IRFs): Reveal that LKR currency depreciation shocks trigger an immediate short-term (1–3 months) contraction in official inflows due to capital diversion into informal parallel networks (Hawala/Undiyal), before recovering by month 6.
• Predictive Modeling: Multivariate machine learning models significantly outperform univariate baselines. Ridge Regression achieves a 73.8% accuracy improvement over SARIMA, reducing testing RMSE from USD 1,889 Million to USD 494.8 Million.
• 2026 Forecast: Under stable conditions, the optimized framework projects a 2026 remittance inflow of USD 9,001 Million (± USD 970 Million at 95% confidence).

---

👥 Contributors

• Dhinanjaya Fernando
• Dinura Ginige
• Kalana Lakshan
• Chanupa Gurusinghe
• Lasana Pahanga
• Subavarshana Arumugam
• Sandeepa Weerasekara
• Sandareka Wickramanayake
• Nisansa de Silva

University of Moratuwa, Department of Computer Science & Engineering

---

📈 Key Empirical Insights

1. Demographic Shift and Skill Polarization

• Gender Transition: Sri Lankan migration has transitioned from a highly feminized flow in the mid-1990s (1994: 72.8% female, mostly low-skilled domestic workers) to a male-dominated flow stabilizing around a 60:40 male-to-female ratio post-2008.
• Contract Duration Length: Average contract duration nearly doubled from 1.8 years (1994) to 3.5 years (2025). This shows an extremely high correlation with the LKR/USD exchange rate ($r = 0.951$), suggesting that currency depreciation acts as a strong economic constraint pushing workers into longer overseas commitments.
• Geographic Concentration: Emigration pre-departure networks have increasingly concentrated in the Kurunegala district ($HHI$ rising by +9.00/year), bypassing other districts and highlighting spatial inequalities in migration access.

2. Macroeconomic Transmission Channels

• Oil Price Dependency: Emigration is strongly correlated with Brent crude oil prices ($r = 0.707, p < 0.001$), acting as a proxy for Gulf Cooperation Council (GCC) economic activity and hiring capacity. This creates a pro-cyclical vulnerability for Sri Lankan labor export.
• Remittances and Exchange Rate: An income and valuation effect operates in tandem: depreciating exchange rates encourage more remittances ($r = 0.665$), but sudden domestic devaluations trigger short-term official remittance contraction as funds divert to parallel markets.

3. Quantitative Summary Table

The table below displays the key Pearson correlation coefficients ($r$) from the 32-year historical analysis:

Variable Pair	Coefficient ($r$)	Significance	Direction & Interpretation
Emigration ↔ Oil Price	+0.707	***	Higher Gulf oil wealth drives demand for Sri Lankan labor.
Contract Duration ↔ Dollar Rate	+0.951	***	Rupee weakness strongly drives longer contract commitments.
Remittances ↔ Dollar Rate	+0.665	***	Long-term rupee depreciation is associated with higher remittances.
Remittances ↔ Oil Price	+0.626	***	Gulf economic health directly influences remittance inflows.
Average Age ↔ Remittances	+0.847	***	Older/experienced cohorts generate higher remittance yields.
Male Emigration % ↔ Dollar Rate	+0.819	***	Rupee weakness accelerates skilled male emigration.
Female Emigration % ↔ Dollar Rate	-0.687	***	Rupee weakness is associated with declining shares of female migration.

Significance level: *** $p < 0.001$

---

📂 Repository Structure

The codebase is organized into modular paths covering data processing, descriptive analysis, and econometric/machine learning modeling:

├── Dataset_Management/
│   ├── SriLanka_Migration_final.csv       # Harmonized 32-year monthly dataset (384 rows)
│   └── dataframe_Extract.py               # Data prep, interpolation, and percentage calculations
│
├── Analysis/
│   ├── Dhinanjaya/
│   │   ├── migration_remittance_analysis.ipynb  # Primary econometric & ML forecasting pipeline
│   │   ├── SriLanka_Migration_final.csv         # Local dataset copy
│   │   └── fig*.png                             # Statistical and model evaluation plots
│   │
│   ├── Lasana/
│   │   ├── descriptive_analysis.ipynb          # Descriptive metrics, distributions, and outliers
│   │   ├── eda_analysis.ipynb                  # Pairwise correlations, heatmaps, lag analysis
│   │   ├── descriptive_findings.txt            # Documented qualitative findings (descriptive)
│   │   ├── eda_findings.txt                    # Documented correlation metrics
│   │   └── desc_*.png, eda_*.png               # EDA visual outputs
│   │
│   ├── Chanupa/                                # Subfolders containing:
│   │   └── EDA_3, EDA_4, Hypothesis_2, 4       # Specific EDA routines & Hypothesis test notebooks
│   │
│   ├── Dinura/                                 # Subfolders containing:
│   │   └── EDA1, EDA2, Hypothesis1, 3          # Baseline EDA & Hypothesis test notebooks
│   │
│   └── Kalana/
│       ├── SriLanka_Migration_final.csv
│       ├── charts/                             # Visualizations repository
│       └── final_column_analysis/              # Domain-specific feature analysis
│
└── README.md                                  # Project overview and reproduction guide

---

🛠️ Installation & Setup

1. Clone the Repository

   git clone https://github.com/Dinurang/SriLanka-Remittance-DS-Project.git
   cd SriLanka-Remittance-DS-Project

2. Set Up a Virtual Environment (Recommended)

   python -m venv venv
   # On Windows:
   venv\Scripts\activate
   # On macOS/Linux:
   source venv/bin/activate

3. Install Dependencies Ensure you have the required analytical libraries installed:

   pip install pandas numpy matplotlib seaborn statsmodels scikit-learn scipy jupyter

---

Running the Code

1. Data Cleaning & Extraction

The main harmonized file is located in Dataset_Management/SriLanka_Migration_final.csv. If you wish to re-run the extraction script that prepares annual subsets and calculate migration ratios:

cd Dataset_Management
python dataframe_Extract.py

2. Exploratory Data Analysis (EDA)

Open the Jupyter Notebooks under Analysis/Lasana/ or target directories in Dinura/ and Chanupa/ to reproduce the descriptive analysis, correlation matrices, and gender distribution plots:

jupyter notebook Analysis/Lasana/eda_analysis.ipynb

3. Advanced Econometrics & Ridge Forecasting

To run the full modeling pipeline, including VECM, Cointegration testing, IRF shocks, K-Means clustering, and the predictive Ridge Regression model:

jupyter notebook Analysis/Dhinanjaya/migration_remittance_analysis.ipynb

---

📊 Analytical Pipeline & Methods

graph TD
    A[Data Sourcing from 8 Institutions] --> B[Harmonization & Linear Interpolation]
    B --> C[384-Month Unified Dataset]
    
    C --> D[Demographic Path]
    C --> E[Macroeconomic Path]
    C --> F[Time Series & ML Path]

    D --> D1[Theil-Sen Skill Trends]
    D --> D2[HHI District Concentration]
    D --> D3[Contract Duration Analysis]
    
    E --> E1[Pearson Correlation & Lags]
    E --> E2[Poverty-Migration Links]
    
    F --> F1[Augmented Dickey-Fuller / KPSS]
    F --> F2[Johansen Cointegration]
    F --> F3[VECM & Speed of Adjustment]
    F --> F4[Structural IRFs & Variance Decomposition]
    F --> F5[K-Means Regime Clustering]
    F --> F6[Ridge Regression vs. SARIMA Baseline]

Loading

Econometric Specifications

• Stationarity: All primary series are non-stationary in levels but stationary upon first-differencing, verifying they are integrated of order one $I(1)$.
• Vector Error Correction Model (VECM): $$\Delta Y_t = \Pi Y_{t-1} + \sum_{i=1}^{k-1} \Gamma_i \Delta Y_{t-i} + \epsilon_t$$ Where $\Pi$ represents long-run cointegrated relationships, and $\Gamma_i$ represents short-run dynamics.
• Ridge Regression: Fitted using L2 regularization to penalize collinear variables and prevent overfitting, capturing temporal autoregressive momentum via engineered historical lags (Lag-1 and Lag-12).

Citations