马达加斯加20122021年全国贫困率趋势估算（英）

Estimating a trend in the nationalpoverty rate 2012-2021 in Madagascar Nobuo Yoshida,Francis Mulangu,Ana Maria Oviedo,Haoyu Wu,Tomoyuki Sho,Danielle AronandOlive UmuhireNsababeraPublic Disclosure Authorized September2025 Keywords:Poverty measurement;International poverty line;Inflation andprice indices;Survey-to-survey (S2S) imputation. Public Disclosure Authorized Development Data GroupDevelopment Research GroupPoverty and Equity GlobalDepartment GLOBALPOVERTYMONITORINGTECHNICALNOTE48 Abstract The standard global poverty estimation method—relying on household survey data andCPI-based inflation—suggests a sharp decline in Madagascar’s international poverty rate(measured at $3 per day in PPP terms) between 2012 and 2021. This paper, however,applies a survey-to-survey imputation method—SWIFT (Survey of Wellbeing via Instantand Frequent Tracking)—to show that official CPI estimates significantly understateactual inflation. Once corrected, the data reveals poverty stagnation rather than decline,aligning more closely with other socio-economic indicators. The analysis also extendsthe poverty series back to 2001, uncovering a trajectory of volatility and stagnation, incontrast to the steady improvement implied by standard methods. While survey-to-survey imputation has been used to restore comparability of poverty trends, this studyis the first to employ the approach for correcting inflation estimates. 1.Introduction Two major challenges compromise the accuracy of estimating the national poverty trend inMadagascar between 2012 and 2021/22. Typically, such analyses presume that consumption datafrom successive surveys are comparable and that inflation rates, based on Consumer Price Indexes(CPIs), are accurately calculated. However, these assumptions may not hold true for Madagascar.Firstly, the comparability of consumption data across the last two survey rounds is questionable.The Enquête Périodique auprès des Ménages (EPM) 2021-22, adhering to LSMS guidelines,featured an extensive consumption module. In contrast, prior to 2021, the national statistics officeconducted a household budget survey with a more limited scope for measuring poverty. Beegle etal. (2012) demonstrate that broader item inclusion in surveys tends to inflate reported householdexpenditures, irrespective of actual consumption changes. Consequently, this shift in survey designcould potentially lead to an overestimation of poverty reduction.Secondly, CPI-based inflationratesmay have underestimated actual price increases between 2012 and 2021.Thisunderestimation could, in turn, lead to an overestimation of poverty reduction. In low-incomecountries(LICs),CPI measurements are particularly susceptible to external shocks andmethodological limitations, making inflation estimates more prone to bias (Ha et al. 2019). The standard poverty trend analysis for Madagascar from 2012 to 2021 indicates a significantreduction in poverty levels, potentially influenced by changes in survey modules and anunderestimation of inflation through CPI calculations. When the consumptiondata from EnquêteNationale sur leSuivi des OMD (ENSOMD) 2012 and EPM 2021/22 are examined with the 2021National Poverty Line (NPL), adjusted for CPI, the poverty rate, measured at the national povertyline (NPL), appears to plummet from 99% to 75%(Table 1). However, this dramatic decrease doesnot align with other socio-economic measures such as the Multidimensional Poverty Index (MPI),GDP per capita, and employment rates, which all indicate a stagnation in economic growth andpoverty reduction during thesame timeframe. This mismatch suggests that the observed substantialdrop in poverty may be due to changes in survey design and potential inaccuracies in inflationmeasurement. To address the survey design change and possible underestimation of the CPI-based inflation, asurvey-to-survey (S2S) imputation technique called the SWIFT (Survey of Wellbeing via Instantand Frequent Tracking) approach is taken. The S2S methodology trains a model for imputinghousehold expenditure in a survey and applies it to impute household expenditures in anothersurveywhere its household expenditures are not comparable to the other survey’s.ForMadagascar, SWIFT trains urban, capital city, and rural imputation models in EPM 2021/22 dataand imputes household expenditures in ENSOMD 2012/13. According to Yoshida et al. (2022),the imputed household expenditures in ENSOMD 2012/13 data are comparable to the rawconsumption expenditures in EPM 2021/22 data in the 2021 prices. In other words, both surveydesign effects and the bias in the CPI-based inflation are addressed if the imputed householdexpenditures for ENSOMD 2012/13 are used. The new estimates show that the poverty ratemeasured at NPL increased only marginally from 72.9 percent to 75.2 percent between 2012 and2021(Table 1), which is consistent with other socio-economic indicators. The reliability of theSWIFT-based poverty estimates is summarized in Box 1. A simil