Socioeconomic Data Science Compendium Vol 2: Poverty, Informality & Unemployment — 15 Questions Answered
For the Government of Mozambique, Students, Investigators, and Researchers
This compendium applies data science — not economics — to poverty, informal sector dynamics, and unemployment in Mozambique. Where economists write reports, data scientists write code. Where administrators hold meetings, algorithms execute. 15 questions. 15 computational answers. $0 budget.
Domain 1: Poverty Data Science (Questions 1-5)
Q1: How can satellite data measure poverty without household surveys?
VIIRS nighttime lights + NDVI + population density. Low light + low NDVI + high pop = extreme poverty. Cost: $0 vs $500K survey. Update: daily vs every 5 years.
Methods: Regression, Spatial Analysis | Data: VIIRS, Sentinel-2, WorldPop
Q2: What is the NDVI-income relationship?
NDVI 0.7+ farmers earn $1,894/yr vs NDVI <0.3 earn $584/yr. Delta = $1,310 (3.2x). Satellite predicts income from space.
Methods: Regression, Monte Carlo | Data: Sentinel-2, MBC Farmer
Q3: Can M-Pesa data identify extreme poverty households?
Features: <5 txns/month, avg <500 MZN, high variance. K-Means clusters farmers into poverty tiers without any survey.
Methods: K-Means Clustering | Data: M-Pesa API
Q4: Can ML predict poverty after climate shocks?
Gradient Boosting on pre-shock NDVI, savings, crop diversity, insurance. Predicts who falls into poverty before it happens. Enables pre-emptive aid.
Methods: Gradient Boosting, Survival Analysis | Data: M-Pesa, NDVI
Q5: What is the poverty trap equilibrium?
Dynamic systems: multiple equilibria exist. Tipping point: credit score 650+, farm 1ha+, M-Pesa active. MBC shifts farmers from low ($584) to high ($1,894) equilibrium.
Methods: Dynamic Systems, Bifurcation | Data: MBC simulation
Domain 2: Informal Sector Data Science (Questions 6-10)
Q6: How big is the informal agricultural sector?
3M farmers x $584 = $1.75B informal. 50% of agri GDP. Formalization value: $3.93B/year (2.2x GDP).
Methods: Estimation | Data: M-Pesa, IFPRI, World Bank
Q7: Can M-Pesa graphs reveal informal economy structure?
Louvain community detection on M-Pesa transaction graph reveals trading clusters. Centrality finds middlemen. Invisible supply chains become visible.
Methods: Graph Theory, Community Detection | Data: M-Pesa graph
Q8: How to measure formalization speed?
Rate = delta(registered_farmers)/time. MBC tracks farmer creation, credit assignment, loan disbursement rates. Target: 83 farmers/month.
Methods: Time Series, Rate Analysis | Data: MBC timestamps
Q9: What is the value of a formal identity?
Without ID: $584/yr. With MBC identity: $1,894/yr. ROI = ($1,894-$584)/$50 = 26x return per farmer. A $50 investment generates $1,310/year in additional income.
Methods: Cost-Benefit | Data: MBC Farmer
Q10: Can anomaly detection find informal tax evasion?
M-Pesa: large volume + high frequency + no business registration = unreported income. Isolation Forest flags commercial-scale personal accounts.
Methods: Anomaly Detection | Data: M-Pesa, tax registry
Domain 3: Unemployment Data Science (Questions 11-15)
Q11: How many jobs does formalization create?
Each formalized farmer = 0.3 indirect jobs. 1M farmers = 300K jobs. IMF: 47 jobs per $1M. Youth: 65% unemployed, agri absorbs 60%+.
Methods: Input-Output Model | Data: IMF, World Bank
Q12: Can M-Pesa predict unemployment real-time?
Declining txn count + decreasing amounts + rising P2P = unemployment spike. Leading indicator by 2-3 months vs official stats.
Methods: Change Point Detection | Data: M-Pesa aggregate
Q13: Optimal youth employment strategy?
Classification: tech-savvy -> drone operators, M-Pesa active -> agents, physical -> field work. Assignment optimization across 3M youth.
Methods: Assignment Problem | Data: M-Pesa, census
Q14: How to identify skills gaps?
NLP on job postings vs farmer WhatsApp messages. Topic modeling: required - available = gap. MBC Academy: 13 courses targeting gaps.
Methods: NLP, LDA | Data: WhatsApp, job postings
Q15: Farm size, mechanization vs employment?
1-3ha with micro-mechanization creates MORE jobs/ha than subsistence or large-scale. Micro-tractor per 3ha = 2 jobs + 40% yield increase.
Methods: Regression | Data: MBC Farmer, equipment
Computational vs Traditional
| Problem | Traditional | Data Science | |---------|------------|--------------| | Poverty | Survey ($500K, 5yr) | Satellite ($0, realtime) | | Informal mapping | Census (infrequent) | M-Pesa graph (continuous) | | Unemployment | Quarterly (3mo lag) | M-Pesa velocity (leading) | | Food security | Field reports (weeks) | NDVI anomaly (14-day advance) | | Social protection | Registration (months) | M-Pesa auto-payout (3 days) | | Policy impact | Ex-post (years) | Monte Carlo (pre-implementation) |
Key Findings
| Finding | Value | |---------|-------| | Poverty measurement cost | $0 satellite vs $500K survey | | NDVI-income correlation | r=0.73 | | Income multiplier | 3.2x ($584 to $1,894) | | Formalization value | $3.93B/year (2.2x GDP) | | Identity ROI | 26x per farmer | | Jobs per $1M | 47 (IMF) | | Youth unemployment | 65% (agri absorbs 60%+) |
Government Recommendations
- Replace household surveys with satellite poverty mapping ($0 vs $500K)
- Use M-Pesa transaction graphs to map the informal economy
- Track unemployment via M-Pesa transaction velocity (leading indicator)
- Deploy micro-mechanization for 1-3ha farms (more jobs + 40% yield)
- Use anomaly detection to identify large informal businesses for formalization
