Portfolio

Selected analysis, with the methods and the numbers behind it. The work below spans U.S. state, county and metropolitan economies as well as international development institutions. Three questions run through most of it: what artificial intelligence does to employment and to the tax base that employment supports, how to measure an economy in something close to real time, and whether a project’s promised economic impact survives a careful look.

AI, Labor, and the Macroeconomy

Exposed Twice: AI, Labor, and the Public Purse

AI displacement is usually framed as a worker problem. It is also a revenue problem. Applying a full labor tax schedule to 14.3 million formal wage employees shows that the occupations most exposed to automation account for 19 percent of private formal employment but 25 percent of labor tax revenue. Governments lose receipts precisely when demand for social spending rises. Where displaced workers land turns out to matter more for the fiscal result than how far wages fall: informalization drives roughly three-fifths of the revenue loss in the central scenario.

Fiscal exposure index — tax-revenue share divided by employment share. Automation-risk occupations sit at 1.29, meaning they carry considerably more of the tax base than of the workforce.

Household labor force microdata; occupational AI exposure framework; tax-schedule simulation. August 2026.

Nowcasting US GDP Growth with Machine Learning

A regularized regression model that turns monthly indicators into a current-quarter GDP estimate weeks ahead of the official release. Seven FRED series — payrolls, industrial production, retail sales, consumption, housing starts, unemployment and sentiment — bridged to quarterly output. It tracks the 2020 collapse and the recovery that followed without systematic bias, which is the test that matters: a nowcast that only works in calm quarters is not a nowcast.

Nowcast diagnostics. Predicted against actual growth, plus residuals over time and in distribution — the panels that show whether a model is calibrated or merely fitted.

Ridge regression with L2 regularization; FRED high-frequency indicators; bridge equations. September 2025.

Economic and Fiscal Impact Analysis

Six Questions for a Billion-Dollar Promise

What a dollar actually does in Montgomery County, Maryland. A county economic model built from public federal employment and tax data, tested against a hypothetical data center priced from a comparable nearby facility. It returns multipliers of 1.29 to 1.63 where the published figure is 1.7, and the gap is structural rather than a rounding difference: the county has no local manufacturing base to absorb construction spending. Between 37 and 52 percent of construction wages leave the county untaxed, carried home by commuters and invisible in a headline job count.

County economic model from federal employment and tax data; multiplier decomposition; leakage analysis. August 2026.

Quantifying the 2021 Texas Freeze with Satellite Econometrics

An independent loss estimate that does not depend on utility self-reporting. Harris County sits on the ERCOT grid and El Paso County sits on the Western Interconnection — same storm, different grid, which makes them a usable treatment and control pair. Nighttime radiance fell 20.37 percent in Houston during Winter Storm Uri, implying $7.6 to $15.8 billion in losses depending on the valuation applied.

Average nighttime radiance, Harris County against the El Paso control, with the February 2021 freeze marked. Difference-in-differences estimate of −6.83 nW/cm²/sr (p = 0.004).

Difference-in-differences; VIIRS nighttime lights; Google Earth Engine; Python. November 2022.

Pricing Carbon in Maryland: An Open-Source Toolkit

Four Python scripts, two public APIs and a laptop. A partial-equilibrium toolkit for state carbon pricing that projects coverage, revenue, emissions reduction and household rebates, with Prince George’s County as the distributional case study. Released open source, on the view that a policy model whose assumptions cannot be inspected is an assertion rather than an analysis.

Maryland CO₂ emissions by sector, 2021. Transportation alone accounts for 52.6 percent, which is what makes coverage design the decisive choice in any state carbon price.

Partial equilibrium modeling; FRED and EPA Greenhouse Gas Reporting Program APIs; open source on GitHub. February 2026.

Operational and Advisory Engagements

Alongside analytical assignments, we work the operational side: the economic analysis that has to justify a project or a plan before it is approved, the support that keeps it moving through implementation, and the measurement that establishes whether it worked. Recent engagements span social protection and youth employment in West Africa and regional development strategy in Southeast Asia.

Adaptive Social Protection in Sierra Leone

Designing social protection for a country carrying climate exposure and institutional fragility at the same time. The engagement ran across appraisal-stage economic analysis, implementation support, and monitoring and evaluation: costing a shock-responsive cash transfer system with pre-financed contingency mechanisms, and a digital public works component that put young workers to digitizing historical disaster records and mapping vulnerability zones. Climate impacts are projected to reduce national GDP by 9 to 10 percent by 2050, with exposure concentrated in flood-prone informal settlements. The analysis returned a per-beneficiary cost of roughly $450 against a net present value spanning $129 to $4,541 — a range wide enough to be the finding in itself, since it locates exactly which assumptions the case actually rests on.

Appraisal economic analysis; implementation support; monitoring and evaluation. Social protection and youth employment, Sierra Leone.

Social Safety Nets and Jobs in Ghana

Lead technical author of the Implementation Completion and Results Report for the Ghana Productive Safety Net Project (P164603), a $139 million operation combining productive inclusion, labor-intensive public works and cash transfers. The project was rated Satisfactory, with four of five outcome indicators exceeding target — beneficiary enterprise initiation at 96 percent against a 70 percent target, public works earnings at 180 percent of target. The single shortfall, payment timeliness at 35 percent against a 75 percent target, traced to the 2022 macroeconomic crisis constraining counterpart funding rather than to delivery failure, and the digitized payment systems the project built were what allowed emergency COVID-19 relief to reach a further 107,574 beneficiaries.

Earlier targeting work linked climate risk maps to program beneficiary rolls to test whether the households most exposed to shocks were the ones actually enrolled — a question programs are rarely built to answer about themselves. Supporting diagnostics covered labor market transitions in national panel data, where workers aged 15 to 35 proved roughly 20 percentage points less likely than older workers to move from unemployment into employment, and the structural transformation of employment out of agriculture.

Completion and results reporting; program targeting analysis; labor market diagnostics. Social protection and jobs, Ghana.

Regional Development Strategy in Malaysia

Economic analysis supporting regional development strategy for Malaysia’s East Coast Economic Region — a four-state area covering roughly half of Peninsular Malaysia’s land mass and a population of about 4.8 million. The work addressed sector prioritization, the economic case for public investment, and the growth and employment implications of competing development pathways for a region whose income levels have historically trailed the west coast.

Regional economic strategy; sector prioritization; investment case analysis. Delivered through a prime contractor. East Coast Economic Region, Malaysia.

Published Research

Published research with the IMF and the World Bank. Much of the labor market analysis on this page rests on household survey microdata, which makes whether those surveys still reach the population a live methodological concern rather than a footnote — the subject of the first paper below.

Fotopoulou, E., Kejji, L., Klyuev, V. and Nxumalo, M. (2026). Eroding Participation in Labor Force Surveys: Evidence, Drivers and Solutions. IMF Working Paper No. 2026/045.
Nxumalo, M. (2023). Chapter 6, “Social Protection for Online Gig Workers,” in Working without Borders: The Promise and Peril of Online Gig Work. Washington, DC: World Bank. Core team member.
Nxumalo, M. and Raju, D. (2022). Shocks and Social Safety Net Program Participation in Ghana: Descriptive Evidence from Linking Climate Risk Maps to Program Beneficiary Rolls. World Bank.
Nxumalo, M. and Raju, D. (2020). Structural Transformation and Labor Market Performance in Ghana. World Bank Jobs Working Paper No. 55.

Work with us

Economic and fiscal impact assessment, labor market analysis, and macroeconomic modeling for site selection teams, economic development organizations, and public agencies. The full archive of working notes sits at lelonxumalo.blogspot.com.

Discuss an engagement