Poverty PPPs - World Bankpubdocs.worldbank.org/...Doc...Poverty-PPPs-S3-PPT.pdfBackground 3 ICP...
Transcript of Poverty PPPs - World Bankpubdocs.worldbank.org/...Doc...Poverty-PPPs-S3-PPT.pdfBackground 3 ICP...
Poverty PPPsMay 26, 2016
MIT Sloan, Cambridge
Presentation Outline
Background
Project scope and overview
Input data: sources and validation
Methodology and Preliminary Results
Conclusions
Challenges and Way Forward
2
Background
3
ICP PPPs: World Bank (WB) global poverty measures rely on ICP
private consumption PPPs to:
i. [construct WB IPL]: convert selected national poverty lines into a
common currency, at PPP, to then derive the WB IPL
ii. [apply WB IPL]: convert the WB IPL back to LCUs of each country
Critique: concern about using ICP PPPs for global poverty analysis on
the grounds that the poor may face different [item] prices and (surely)
have different consumption patterns than what is reflected in the
national average (item) prices and aggregate (NA) expenditures used
in ICP
Project scope and overview
Explore the sensitivity of the (2011) PPPs to the choice of expenditure weights and products, with a focus on global poverty analysis
Poverty expenditures and item-level information sourced from standardized Household Expenditure Surveys (HES) and underlying (2011) ICP data
Initiate with Sub-Saharan Africa, other regions will follow
4
Standardized surveys were originally prepared for Deaton and Dupriez (2011). The number of surveys has expanded since and now includes additional countries (approx. 102) and more recent surveys
Surveys standardized by applying a common data dictionary to variables, annualizing consumption data, fixing outliers and mapping items to ICP basic headings
Our project uses 28 standardized surveys from different countries in Sub-Saharan Africa: All sub-regions are represented Oldest: 2005 (COD); Most recent: 2014 (LBR); Modal survey year: 2010
No. of items covered in the 28 surveys ranges from 30 (NAM) to 1296 (BFA); median 360. On average, 15 of the 108 HHC ICP basic headings that could be present in HHS are missing
Input data: sources
5
Input data: validation of expendituresBasic Headings (BH’s)
In several countries, aggregate expenditures from ICP National Accounts (NA) and Household Survey’s (HHS) show fairly similar patterns in terms of which BH’s have the highest exp. shares (out of total national HH expenditure)
Expenditures of poor households (1.90 WB IPL; 2011 PPPs) are concentrated around fewer basic headings than agg. expenditures that include all HH’s, regardless of income e.g. after averaging BH exp. shares across all surveys and ranking them by expenditure, largest to smallest, approx. 50% of the top expenditure of the poor can be found in 8 BH’s; versus 12 BH’s in national aggregates
Estimates of household consumption levels
differ between the NA (WB WDI) and HHS,
for the survey year. Unclear which source
provides a more accurate measure of
consumption
6
Ratio of survey means from HHS to HHC/capita from National Accounts(for survey year; 28 countries)
Input data: validation of expendituresItems
As with BHs, expenditures of poor households are concentrated among fewer items than in aggregates that incl. all households, regardless of income
In the surveys explored, many of the items with the highest exp. share (out of total national HH expenditure) by poor households are food staples (e.g. cereals, such as maize, millet, sorghum); exceptions: firewood and housing; also, mobile phone related fees
Still, uncertainty remains on what exactly (and where) the poor buy. This adds to the difficulty of compiling objective ICP lists, especially when poverty items are of concern
Yet, in many of these surveys, items with high exp. shares (out of total national HH expenditure) by poor households can be found in ICP lists. So, overall, ICP item lists seem to be moving in the right direction for poverty analysis. [Caveat: most HHS in SSA lack enough item-level detail to explore item varieties in depth]
A handful of items that in some surveys appear as having important exp. shares shares by poor households are either missing from ICP lists (milling services for cereals), or the given variety is not listed (cassava flour). These and other findings from HHS will be taken into account in upcoming ICP item list revisions
7
Ex: items consumed by poor HHs (TZA HHS, 2011)
50%
60%
8
Sum of
weights
consumed by
the poor
(below the
poverty l ine) ICP - Basic Heading name
Tanzania 2011 HHS - Item name
(in the order of largest consumption by the poor to the smallest)
Global
Core
(GCL)
Africa
Regional
(AFR) Note
Other cereals, flour and other products Maize Flour White
Rice Rice ICP has more varieties
Frozen, preserved or processed vegetables and vegetable-based products Cassava flour ICP has other varieties
Other fuels Firewood
Fresh or chilled vegetables other than potatoes Beans ICP has more varieties
Fresh or chilled potatoes Sweet Potatoes
Preserved or processed fish and seafood Dried sardines
Fresh or chilled fruits Cooking bananas, plantains
Telephone and telefax services Mobile telephone bill (including top-up cards) ICP has more varieties
Beef and veal Beef with bones ICP has more varieties
Fresh or chilled vegetables other than potatoes Other leafy vegetables ICP has more varieties
Other edible oil and fats Other cooking oil ICP has more varieties
50.3 Other fuels Kerosene
Fresh or chilled vegetables other than potatoes Tomatoes, Round
Other cereals, flour and other products Sorghum, flour ICP has other varieties
Fresh, chilled or frozen fish and seafood Fresh, chilled or frozen fish ICP has more varieties
Actual and imputed rentals for housing Paid or estimated monthly rent (Owner or joint owner of dwelling) ICP has more varieties
Fresh milk Fresh cow milk ICP has more varieties
Water supply Water ICP has more varieties
60.0 Passenger transport by road Transport by road (bus and taxis) ICP has more varieties
Pharmaceuticals products Pharmaceutical products (medicines, serum, vaccines) ICP has more varieties
Sugar Brown sugar ICP has more varieties
Catering services Cafes, bars and the like (consumed outside home) ICP has more varieties
Fresh or chilled potatoes Potatoes
Fresh or chilled vegetables other than potatoes Spinach / Lettuce
Preserved or processed fish and seafood Dried or salted fish/ shellfish ICP has more varieties
Other cereals, flour and other products White Maize grains
Catering services Konyagi (consumed outside home) National specific
Fresh or chilled potatoes Cassava, Fresh
Telephone and telefax equipment Mobile phone
Other cereals, flour and other products Green maize cob ICP has other varieties
Other fuels Charcoal
70.48 Garments Kitenge National specific
ICP List
HHS Item Name(in order of largest to smallest share out of total national HHC)ICP Basic Heading
ICP 2011 List
Methodology
9
• 108 ICP BHs were included in calculations to ensure consistency with the ICP results
• Iterative scheme was employed to deal with circularity in order to identify poor households across the surveys, as PPPs use expenditure weights of the poor who in turn are identified as poor on the basis of PPP-converted poverty lines (see, e.g. Deaton & Dupriez, 2011)
Methodology: identification of poor households
[Steps (1) through (4) are repeated until the poverty (weighted) PPPs, for each country, converge]
10
BH shares of the poor
Unweighted BH-level PPPs
Select initial poverty line: (WB IPL: 1.90)
Convert IPL to LCU using PPPs
(Bring IPL to year of HHS by
using General CPI)
Identify poor HHs in the HES (below IPL or within an
interval of the IPL)
Estimate average basic heading expenditure
shares of theidentified poor= Poverty (weighted) PPPs
1 2
34
+
• Dealing with circularity to consistently identify poor households across the surveys
Main principles
• HHS mapped to ICP 108 basic headings (harmonized)
• Standard ICP aggregation methods are used (CPD and GEKS-Fisher)
• Baseline– Published regional results (NA weights)– subsample of SSA – 28 countries (NA weights)– subsample of SSA – 28 countries (HHS weights)
• Scenarios- below IPL (using HHS weights, plutocratic)- around IPL (using HHS weights, democratic), using two kernels – quartic and
uniform
11
Main principles
• Country-product-dummy (CPD), with country and product dummy variables
• GEKS-Fisher
12
Two kernels used
• Quartic (biweight) kernel
• Uniform kernel
13
Results: Below PL PPPs vs. around PL PPPs
• As bandwidth increases the kernel estimated PPPs are getting closer to the below PL PPPs; however, the changes are relatively small compared to the overall distance between below PL and around PL PPPs
14
0.97
0.98
0.99
1
1.01
1.02
1.03
Around PL (quartic kernel) vs. below PL
bw=1.0 bw=0.50 bw=0.25
s.d.bw= 1.0 0.72%
bw= 0.50 0.72%bw= 0.25 0.77%
Results: PPP sensitivity to kernel form
• Ratio of PPPs estimated with two kernels, with different bandwidths
Effect is smaller as bandwidth increases:
15
0.96
0.97
0.98
0.99
1
1.01
1.02
1.03
Effect of kernel (quartic vs. uniform)
bw=0.5 bw=0.25
s.d.bw= 0.5 0.13%bw= 0.25 0.33%
Results: PPP sensitivity to bandwidth
• PPP changes with bandwidth using quartic kernel
Effect is smaller as bandwidth increases:
16
0.97
0.98
0.99
1
1.01
1.02
1.03
Effect of bandwidth (quartic kernel)
1.0 to 0.5 0.5 to 0.25
s.d.1.0 to 0.5 0.11%
0.5 to 0.25 0.20%
Results: Poverty PPPs vs ICP PPPs
• Differences among various Poverty PPP methods are insignificant when compared against ICP PPPs (baseline on the graph)
17
0.85
0.9
0.95
1
1.05
1.1
1.15
Poverty PPPs vs. ICP PPPs
quartic kernel uniform kernel below PL
s.d.quartic kernel 2.69%
uniform kernel 2.69%below PL 2.99%
Results: BH weights according to different kernels
• Effect of kernels on BH weights is relatively small (BH shares for South Africa is shown below)
18
0.00%
2.00%
4.00%
6.00%
8.00%
10.00%
12.00%
14.00%
16.00%
1 3 5 7 91
11
31
51
71
92
12
32
52
72
93
13
33
53
73
94
14
34
54
74
95
15
35
55
75
96
16
36
56
76
97
17
37
57
77
98
18
38
58
78
99
19
39
59
79
91
01
10
31
05
10
7
Bandwidth .25
uniform exp.weighed biweight
Challenges and way forward
• More exploration at the item-level using HHS data
• Further inspection on the quality of national accounts
• Investigate behavior of different kernel forms, under different bandwidths
• Further harmonize HHSs, reduce effect of bad survey data
• Expand analysis to more regions
19
THANK YOU
20