This dataset provides a 384-month (32-year, Jan 1994 – Dec 2025) harmonized time series describing Sri Lanka's international labour emigration, worker remittances, and the domestic/global macroeconomic environment surrounding them. It was built by integrating eight authoritative sources — the Central Bank of Sri Lanka, World Bank Open Data, the Sri Lanka Bureau of Foreign Employment (SLBFE), the ILO, FRED (St. Louis Fed), the Department of Census and Statistics Sri Lanka, the U.S. EIA, and World Bank PPP indicators — into a single machine-learning-ready table.

The dataset underpins the peer-reviewed study "The Remittance Blueprint: Data-driven Intelligence for Sri Lanka" (MERCon 2026), which combines exploratory data analysis, stationarity-corrected time-series econometrics (ADF/KPSS, Johansen cointegration, VAR/VECM, impulse response analysis), and supervised machine learning to explain and forecast remittance inflows. A Ridge Regression model trained on this dataset improved forecasting accuracy by 73.8% over a SARIMA baseline (RMSE reduced from USD 1,889 Mn to USD 494.8 Mn) and projected 2026 remittances at USD 9,001 Mn (± USD 970 Mn, 95% CI).

Rows: 384 (monthly observations) · Columns: 35 · Period: January 1994 – December 2025

Motivation

Worker remittances account for 6–9% of Sri Lanka's GDP and have served as a critical foreign-exchange and crisis buffer, most notably during the 2022 sovereign debt default. Existing Sri Lankan remittance research has largely relied on descriptive statistics or annual, cross-sectional data that cannot capture short-term macroeconomic shocks or non-stationary dynamics. This dataset was built to close that gap by combining:

Demographic migration indicators (volume, gender, skill composition, destinations)
Domestic macroeconomic indicators (GDP, inflation, employment, wages, interest rates, poverty)
Global/external indicators (exchange rate, Brent crude oil price, destination-country growth, PPP)

into a single monthly panel suitable for causal inference, econometric modelling, and ML forecasting.

Dataset Structure

Each row represents one calendar month. Migration and socioeconomic statistics that are only available annually (e.g. slbfe_total_annual, poverty_rate_annual) are aligned to all twelve months of their corresponding year; genuinely monthly series (e.g. dollar_rate_monthly, brent_oil_monthly) vary within each year.

Date	Migration (annual)	Inflation (annual)	Exchange Rate (monthly)	Oil Price (monthly)
1994-01	...	...	...	...
1994-02	...	...	...	...

Features

Column	Description
`date`	Observation month (YYYY-MM)
`slbfe_total_annual`	Total annual overseas departures
`slbfe_male_annual`	Annual male migrant workers
`slbfe_female_annual`	Annual female migrant workers
`slbfe_skilled_annual`	Annual skilled migrants
`slbfe_lowskilled_annual`	Annual low-skilled migrants
`male_skilled_pct_annual`	% of male migrants classified as skilled
`male_lowskilled_pct_annual`	% of male migrants classified as low-skilled
`female_skilled_pct_annual`	% of female migrants classified as skilled
`female_lowskilled_pct_annual`	% of female migrants classified as low-skilled
`avg_age_annual`	Average migrant age
`avg_contract_years_annual`	Average employment contract duration
`top1`–`top5`	Top 1–5 destination countries by migrant share
`top1_perc`–`top5_perc`	Migrant share (%) to each top destination
`remittances_annual_usd_mn`	Annual foreign remittances (USD million), World Bank totals interpolated to monthly frequency
`gdp_usd_bn_annual`	Sri Lanka GDP (USD billion)
`inflation_rate_annual`	Annual inflation rate
`unemployment_rate_annual`	Annual unemployment rate
`employment_ratio_annual`	Employment-to-population ratio
`wage_all_workers_annual`	Average worker wage
`central_bank_interest_rate_annual`	Central Bank of Sri Lanka policy interest rate
`poverty_rate_annual`	National poverty rate
`CCPI_b2021`	Colombo Consumer Price Index (base 2021)
`PPP_annual`	Purchasing Power Parity (GDP per capita, log-transformed prior to normalisation due to right-skew)
`dollar_rate_monthly`	Monthly average USD/LKR exchange rate
`dest_gdp_growth_avg_annual`	Average GDP growth across major destination countries (Gulf-weighted)
`brent_oil_monthly`	Monthly Brent crude oil price (USD/barrel)

Target Variable

Primary target:

remittances_annual_usd_mn

(or its monthly-flow derivative, remittances_monthly_usd, used for the supervised forecasting task in the paper)

Additional plausible targets:

slbfe_total_annual, slbfe_male_annual, slbfe_female_annual, slbfe_skilled_annual

Data Collection

Data were sourced from eight institutions, with cross-validation against at least one independent source per indicator:

Central Bank of Sri Lanka — cbsl.lk/eResearch
World Bank Open Data — data.worldbank.org
Sri Lanka Bureau of Foreign Employment (SLBFE) — slbfe.lk/statistics
ILO — ilo.org
FRED / St. Louis Fed — fred.stlouisfed.org
Department of Census and Statistics, Sri Lanka — statistics.gov.lk
U.S. Energy Information Administration (EIA) — eia.gov
World Bank PPP indicators — data.worldbank.org/indicator/NY.GDP.PCAP.PP.CD

Collection combined automated web extraction, OCR-based digitisation of SLBFE annual reports (1994–2005), and structured CSV/XLSX imports.

Data Preprocessing

Missing data were handled hierarchically: columns with >40% missingness were dropped; isolated gaps (<5%) received median/mean imputation; consecutive gaps were linearly interpolated; temporal edge gaps used forward/backward fill.
Outliers (|z| > 3) were visually confirmed and cross-referenced against historical events. 2022 crisis-year observations were deliberately retained (not clipped) as analytically significant structural breaks, while genuine data-entry errors were corrected via interpolation.
Logical constraints were enforced (e.g., male % + female % = 100%), categorical labels standardised, and duplicates removed.
Annual-to-monthly harmonisation: remittances and other annual totals were linearly interpolated to monthly frequency via y(t) = y_n + (y_{n+1} - y_n) * (t - t_n) / (t_{n+1} - t_n). Exchange rates were harmonised via within-month averaging, SLBFE departures via uniform monthly allocation, and CCPI via chain-linking.
Normalisation: all variables were Z-score normalised; right-skewed variables (e.g. GDP, PPP) were log-transformed prior to normalisation.
Stationarity: primary series (remittances, USD/LKR, CCPI, oil prices) were confirmed integrated of order one, I(1), via ADF/KPSS testing; first-differencing is required before correlation, Granger causality, or VAR/VECM analysis.
Collinearity screening: Pearson screening (|r| > 0.9) found no fully redundant variable pairs; three variables with VIF > 10 (CCPI, GDP, wages) were retained for their distinct analytical roles rather than dropped.

Key Empirical Findings (from the accompanying paper)

These are provided as context for users deciding how to model the data — they are not properties of the raw file itself, but results obtained from it.

GDP–remittance correlation: r = 0.980 (R² = 0.961); GDP–emigration correlation is much weaker (r = 0.454), suggesting remittance growth reflects rising migrant earning capacity and formal-channel use more than headcount.
Emigration–oil price correlation: r = 0.707 (p < 0.001), reflecting Gulf labour-market dependence.
Remittances–exchange rate correlation: r = 0.665 (p < 0.001).
Johansen cointegration confirms a stable long-run equilibrium among remittances, USD/LKR, oil prices, and CCPI; a VECM speed-of-adjustment coefficient of α ≈ -0.15 implies the system corrects 15% of disequilibrium per month (6.7 months to re-equilibrate).
Impulse response analysis: a depreciation shock to USD/LKR suppresses official remittance inflows for ~3 months (consistent with diversion to informal Hawala/Undiyal channels) before recovering; a positive oil-price shock produces a sustained increase in remittances over 12 months.
Forecasting: Ridge Regression (multivariate, with lagged/differenced macro features) reduced test RMSE from USD 1,889 Mn (SARIMA baseline) to USD 494.8 Mn (a 73.8% improvement), projecting 2026 remittances of USD 9,001 Mn (± USD 970 Mn, 95% CI) under stable conditions.
A long-term gender shift (toward male-dominated flows) and skill polarisation (declining overall skilled share, −0.69/yr by Theil–Sen estimate) were also identified as structural risks to per-capita remittance yields.

Full statistical tests, model specifications, and figures are in the paper (see Citation below).

Intended Uses

Time-series forecasting (SARIMA, VAR/VECM, Ridge/Lasso, LSTM/GRU, Temporal Fusion Transformer, Prophet, etc.)
Regression and feature-importance analysis
Causal/structural analysis (Granger causality, Johansen cointegration, impulse response functions)
Explainable AI (SHAP/LIME) on migration and remittance drivers
Migration-trend and policy research
Teaching/benchmarking time-series econometrics on a real, non-stationary macro dataset

Limitations

Annual variables repeated monthly: many socioeconomic and demographic indicators are annual statistics replicated across all 12 months of a year — they carry no genuine within-year variation and should not be treated as independently informative at monthly resolution.
Interpolated remittances: remittances_annual_usd_mn originates as a World Bank annual total, linearly interpolated to monthly frequency. This assumes uniform intra-year distribution, injects synthetic autocorrelation, attenuates true intra-year volatility, and partially explains why seasonal decomposition shows negligible seasonal strength (Sf < 0.1) — a property that is partly tautological rather than a genuine absence of seasonality.
Registered migration only: SLBFE statistics capture only documented departures and do not account for undocumented/irregular migration, which likely biases skill-composition ratios toward the registered, higher-skilled segment and underestimates true labour outflow.
Aggregated destination indicators: destination-country economic variables are aggregated averages across top destinations, not country-specific series.
Linear model boundary conditions: downstream linear models (e.g. Ridge Regression) cannot capture non-linear regime transitions; sudden geopolitical shocks (e.g. changes to the Gulf Kafala sponsorship system) could invalidate forecasts derived from this data.
Scope: the dataset focuses on labour migration and remittances and does not capture all dimensions of Sri Lanka's external sector.

Ethical Considerations

This dataset contains only aggregated, national-level statistics and contains no personally identifiable information (PII), presenting minimal privacy risk. As with any macroeconomic/policy-relevant dataset, model outputs (e.g. forecasts) should not be used as the sole basis for migration or fiscal policy decisions — they should complement, not replace, domain expertise and qualitative judgement, particularly given the structural-break and informal-channel dynamics documented in the accompanying study.

Citation

If you use this dataset, please cite the accompanying paper:

License

Released under CC BY 4.0 unless otherwise specified. Attribution to the original eight data sources (CBSL, World Bank, SLBFE, ILO, FRED, DCS Sri Lanka, U.S. EIA) is appreciated alongside citation of the paper above.

Contact

For questions or suggestions regarding the dataset, please contact the maintainers through the Hugging Face repository.

Downloads last month: 81

Total file size:

111 kB

Paper for DS-research/Sri-Lanka-Emigrant-Remittance

The Remittance Blueprint: Data-driven Intelligence for Sri Lanka

Paper • 2606.28190 • Published 14 days ago