VARUN KSHIRSAGAR, PhD
Economist & Data Scientist with two decades of post-PhD experience applying econometrics, machine learning, and geospatial analysis in data-constrained contexts. I design and implement analytical workflows, from problem formulation and data engineering, through modeling and validation, including diagnosing where models fail, to writing and presentation. I combine fragmented household (consumer), administrative (government records), operational, market, and satellite data with clients’ own data and contextual knowledge to produce interpretable measures of economic shocks and risk. My work uncovers economically consequential local differences in risks, opportunities, and living standards.
SKILLS
Model Validation, Interpretable ML, Cost-Benefit Analysis, Time Series Forecasting, Household Survey Analysis, Network Analysis, Spatial Statistics, Causal Inference, Geospatial Data Engineering, Remote Sensing, Tree-Based ML, Clustering, Google Earth Engine, R, Python, SQL, Stata, AWS, Docker, OSRM, Git
EXPERIENCE: Consultant (Aug 2005 – Present) Bracketed labels link to documents, the other blue links are sources and related materials.
WORLD BANK (Economist/Spatial Data Scientist; deliverable-based contracts) (Mar 2019 – Present)
• Developed and implemented a scalable multi-country framework to predict household living standards from limited survey data; benchmarked econometric and machine-learning methods, identified the information needed for reliable prediction, diagnosed model failures, and translated the workflow into requirements for an AI-assisted monitoring system (in progress).
• Designed and implemented a scalable satellite-based, interpretable method to identify deprived, job-isolated neighborhoods for cities without census data; tested out-of-sample across seven cities in four countries and applied to held-out Hanoi with a built-in reliability flag. [reproducibility package]
• Developed a spatial framework benchmarking five major South African cities against 20 global peers with a data-driven business-district delineation from open satellite building data, and the first five-city neighborhood-scale measurement of formal job exposure and volatility by wealth from geocoded tax records and census data; quantified the employment penalty of distance from jobs. [paper] [blog] [reproducibility package]
• Identified a policy gap and developed a data-driven proof-of-concept for climate-contingent health financing in South Africa, integrating local health, satellite rainfall, and deprivation data across 205 local municipalities to estimate climate-related health risks and design early-warning and budget triggers.
• Linked 557,000 households (15 survey rounds, Thailand) to gridded climate and satellite data to construct early- and whole-season weather shocks; rainfall shortfalls lower consumption and raise inequality, with social assistance partly cushioning the inequality and poverty effects. [paper] [reproducibility package]
• Linked satellite nighttime lights with provincial sectoral accounts in Türkiye; found that both the 2008/09 and 2018/19 recessions disproportionately affected less-remote provinces, while the post-2009 construction boom favored remote provinces, and that lights mainly track construction and manufacturing. [paper] [report]
• Developed an interpretable-machine-learning framework for drought-responsive social protection in data-scarce contexts that augments beneficiary identification from household characteristics with remotely sensed weather data across five African countries; showed that transparent decision trees and regressions matched black-box ML in out-of-sample accuracy while producing feasible operational rules. [paper] [slides]
• Co-developed an open-source R/Stata toolkit that automated integration of household surveys with crop-cycle-aligned satellite drought indicators, spatial cross-validation, and estimation of drought-welfare relationships. The codebase was subsequently adapted for a co-authored World Bank analysis of close to 100,000 households across eight African countries. [user guide] [paper] [reproducibility package]
• Developed rapid-response geospatial analyses during COVID-19 to quantify lockdown impacts: using Google Earth Engine, tracked near-real-time NO₂ in nine global cities and found declines of 40 to 70 percent that pass 2019 placebo tests; for India, combined NO₂ with nighttime lights, vegetation, and population for all 2,337 sub-districts, built a five-group typology from satellite indicators, and found that NO₂ fell most in large cities but at least 10 percent across three-quarters of the country. [blog: global cities] [blog: India]
PPI, IPA (Jan 2017 – Feb 2019)
• Developed an early interpretable-machine-learning approach to household poverty measurement, combining regularized regression and stability selection to produce a ten-question scorecard; the paper won Best Paper at a NeurIPS 2017 workshop; the method underlies scorecards in 20+ countries. [extended tutorial] [paper]
• Presented at a NeurIPS workshop and to officials from 15 countries in Southern Africa. [proceedings]
Kimetrica/USAID (FEWS NET) (July 2015 – Dec 2016)
• Developed an automated early-warning framework to detect food-market anomalies: learned each market’s seasonal and trend pattern from its own data, scored anomalies against short-, medium-, and long-term forecasts, and classified them as local or widespread, across 61 markets in eight African countries. [report]
BOOZ ALLEN HAMILTON/USAID SERA Tanzania (Feb 2014 – July 2016)
• Co-developed a framework challenging the emphasis on global markets by quantifying domestic drivers of maize prices in Tanzania, including satellite-derived weather shocks, harvest cycles, remoteness, transport costs, and trade policy; domestic factors accounted for two-thirds of explained price variation. [paper] [blog]
• Designed a market-network approach combining bilateral price-linkage models with network centrality to identify the origins and transmission paths of external food-market shocks, showing that border and secondary port markets, rather than Dar es Salaam (the largest city), were the main conduits. [paper] [brief]
• Led three training sessions on food market analysis to Tanzanian government officials. [training materials]
Biospheric Sciences Division, NASA (subcontract through Sigma Space Corp.) (Apr 2012 – Aug 2014)
• Built a state-space forecasting model to analyze food price dynamics across 554 markets in 51 countries, incorporating satellite-derived weather data; found that domestic weather disturbances affected about a fifth of local food markets in the short run, twice the share affected by international price shocks. [paper]
UNITED NATIONS, FOOD and AGRICULTURE ORGANIZATION (May 2013 – June 2013)
• Performed cost-benefit analysis for a 520 million USD multi-donor basket-fund loan to the Government of Tanzania. Articulated the rationale and modeled the mechanisms through which gains would be realized.
WORLD BANK (Consultant; agriculture, research, and analytical units) (Aug 2005 – June 2015)
• Examined cereal productivity growth in India at the district level and found that anomalous rainfall patterns, rather than policy neglect, explained the slowdown in productivity growth. [report chapter] [note]
• Found that Afghanistan’s opium-dependent regions were not those where poverty reduction or agricultural support was most urgent, so those programs could not double as counter-narcotics policy. [note]
• Synthesized the literature and proposed a Sri Lanka sector strategy aimed at structural constraints (land rights, water, trade and subsidy policy), rather than household programs. [strategy note]
• Examined local food price volatility in 50 developing countries and concluded, against the prevailing belief, that local food price volatility had not systematically increased after the 2007/08 spike and crash. [note]
• Performed ex-ante economic analyses for a 120 million USD loan to finance Uganda’s Agricultural Technology and Agribusiness Advisory Services Project and a proposed 115 million USD loan for Kenya’s Arid Lands Resource Management Project, and ex-post economic analyses for two Uganda project completion reports, using matching and difference-in-differences to correct for selection bias.
• In Kenya, used remotely sensed NDVI and rainfall data to assess drought conditions (ex-post) for the Arid Lands operation. In Uganda, highlighted for the operational team: i) complementary investments in production and marketing, ii) the interaction of transport costs and local effects, iii) evaluating the intervention process, not just the outcome. The report anticipated three priorities of later operations. [field report]
• Conducted applied economic and data analysis on the poverty impacts of food-price shocks, global commodity markets, African small enterprises, and microfinance institutions and regulation.
EDUCATION
07/98 – 12/04 Ph.D. in Economics, NORTH CAROLINA STATE UNIVERSITY, Raleigh, North Carolina
Fields: International Economics, Econometrics, and Industrial Organization
07/94 – 06/97 Bachelor of Arts (First Class, Economics) BOMBAY UNIVERSITY (St. Xavier’s College), Mumbai, India
SELECTED PUBLICATIONS AND WORKING PAPERS [Google Scholar]
South Africa’s Fragmented Cities: The Unequal Burden of Labor Market Frictions. Co-authored with Javier E. Baez (World Bank). World Bank Policy Research Working Paper 11285
The Distributional Impacts of Climatic Variability on Welfare in Thailand. Co-authored with Sailesh Tiwari (World Bank) and Emmanuel Skoufias (National U. of Singapore). World Bank Policy Research Working Paper 11253
A Spatial Perspective on Booms and Busts: Evidence from Türkiye. Co-authored with Javier E. Baez (World Bank) and Cigdem Celik (World Bank). World Bank Policy Research Working Paper 10553
Drought-Sensitive Targeting and Child Growth Faltering in Southern Africa. Co-authored with Javier E. Baez (World Bank) and Emmanuel Skoufias (National U. of Singapore). World Development, Volume 182, October 2024
Household poverty classification in data-scarce environments: a machine learning approach. Co-authored with Jerzy Wieczorek (CMU), Sharada Ramanathan and Rachel Wells. Won the best paper award at the NeurIPS 2017 Workshop on Machine Learning for the Developing World. December 2017.
Shocks to Food Market Systems: A Network Approach. Co-authored with John Baffes (World Bank), Agricultural Economics, January 2020 (Volume 51, Issue 1)
Weather and International Price Shocks on Food Prices in the Developing World. Co-authored with Molly Brown (NASA). Global Environmental Change 35, 31-40 (2015).
What Drives Local Food Prices? Evidence from the Tanzanian Maize Market. Co-authored with John Baffes (World Bank) and Donald Mitchell (USAID). World Bank Economic Review, February 2019 (Volume 33, Issue 1)
Sources of Volatility during Four Oil Price Crashes. Co-authored with John Baffes. Applied Economics Letters, 2016
WORK AUTHORIZATION: U.S. Citizen