Key Details
The IMF working paper Real-time Poverty Monitoring Using GDP Nowcasts: Evidence from India addresses a measurement gap: economic indicators are available frequently, but household surveys — the basis for measuring poverty — are less frequent.
2025 estimate: The authors’ model puts India’s extreme poverty rate at 2.36% under the international US$3-per-person-per-day line, measured in 2021 purchasing-power-parity terms. This is an estimate of a past year, not a survey result.
Starting point: The model uses household consumption data from 2022–23. The World Bank’s poverty estimate based on that survey was 5.25% at the same line.
How it works: Frequently available indicators are used to estimate GDP growth. The authors then use that growth to update the earlier household consumption distribution and calculate the share likely to fall below a poverty line.
Model test: When the researchers estimated the 2022–23 extreme-poverty rate using older survey data, their average result was 4.78%, against the 5.25% survey-based benchmark.
Additional data: Google search trends and satellite measures generally added little improvement to estimates based on conventional economic indicators.
The authors also examine poverty at the US$4.20 and US$8.30 daily lines. Their models indicate declines at both thresholds through 2025.
Monthly Information Can Update an Annual Poverty Estimate
Household consumption surveys provide the direct evidence needed to measure poverty, but they are conducted and processed less frequently than policymakers may need. The paper tests a way to fill the interval: it uses indicators such as industrial production, prices and fuel consumption to estimate GDP growth as new information arrives.
The estimated growth then adjusts the last survey-based distribution of household consumption. Although the calculation can be updated each month, each update is a revised estimate of an annual poverty rate—not a separate poverty measurement for that month.
The Model Points to Further Decline, with a Clear Assumption
Using the 2022–23 consumption survey as its starting point, the authors project extreme poverty at 2.36% in 2025. Their models also point downward at the higher international poverty thresholds.
The method assumes that a fixed share of GDP-per-person growth passes through to household consumption and that households across the distribution experience proportional growth. This makes frequent estimates possible, but national growth can miss differences between places and groups. The resulting figure is most useful as a timely indication of direction, pending survey-based verification.
More Complex Data Did Not Add Much
In the authors’ tests, machine-learning approaches — particularly XGBoost and Random Forest — performed well in estimating GDP growth and poverty. Adding Google search trends and satellite data produced only modest gains in some specifications and worsened accuracy in others. The paper therefore makes a practical case for testing the value of new data sources rather than assuming they improve a poverty estimate.
What Is Nowcasting?
Nowcasting uses information already available to estimate a measure whose direct data are delayed or unavailable. In this paper, it means estimating an annual poverty rate from economic indicators and an earlier household survey. Because 2025 has passed, the 2.36% result is a retrospective model estimate — not a forecast of future poverty.
Policy Relevance
The approach could give India a timelier national signal of changes in poverty between survey rounds, particularly when economic conditions shift quickly. It would be most useful alongside — not in place of — household surveys, which reveal how people’s consumption actually changed.
A validation step is immediately relevant: India’s 2023–24 Household Consumption Expenditure Survey was published after the 2022–23 baseline used in this paper. Testing the model against that newer survey would help establish whether its projected direction and magnitude held up before the method informs policy decisions. National nowcasts also cannot, by themselves, identify which districts or households need support.
Follow the Full Working Paper Here: Real-time Poverty Monitoring Using GDP Nowcasts: Evidence from India