1. Extrapolation of the ICP 2011 PPPs
Sergey Sergeev, Consultant
3rd Regional Coordinating Agencies meeting
October 28-30, 2015 Washington, DC

2. ICP 2005 – ICP 2011 – ICP 2017 (?) What in the meantime?
5/9/2018
Introduction
ICP 2005 – ICP 2011 – ICP 2017 (?)
What in the meantime?
Users are interested in the time–series
The extrapolation / retrapolation of the ICP benchmark PPPs is a possible way
WB decided to collect input data from the Regions and to undertake estimation of regional and global PPPs for non-benchmark years => for and (2016?)

3. Tasks The following specific tasks were addressed:
5/9/2018
Tasks
The following specific tasks were addressed:
Reviewing the input data and determining the most feasible estimation approach;
Conducting estimation of the non-benchmark year PPPs for the main ICP regions;
Conducting estimation of the non-benchmark year global PPPs;
Preparing a report on the estimation approach as well as key findings, to be reviewed by the WB and the ICP Regional Coordinating Agencies

4. No. of countries by the Regions*
5/9/2018
No. of countries by the Regions*
Region
R-code
Reg-Abbr
No.
Africa 1
AFR 50
Asia and the Pacific 2
ASI 23
CIS 3 9
Eurostat-OECD 4
EUO 47
Latin America 5 LA 17
Western Asia 6 WA 12
The Caribbean 7
CARIB 22
Singleton 8
SIN
Pacific Islands
Pac_Isl
Non-Benchmark 10
Non-participants 15
TOTAL
220
*EGY / EGZ (AFR / WAS), SDN / SDO (AFR/WAS), RUS / RUT (EUO/CIS) are double participants

5. Indicators used in the PPP extrapolation
5/9/2018
Indicators used in the PPP extrapolation
DATA CATEGORIES
Code
Name
CPI
Consumer price indices (2011 =100)
DEF
National account deflators (2011 = 100)
EXP
National accounts expenditures (LCU)
POP
Population
PPP
Purchasing power parities (Global ICP 2011, USD =1) XR
Exchange rate (USD = 1)

6. 5/9/2018
Classification

7. Aspects of the analysis / validation of input data
5/9/2018
Aspects of the analysis / validation of input data
Official ICP 2011 PPPs => without changes – only adjustment for the countries who changed national currency (LT, LV)
WB finished the collection of input data from the Regions (Extrapolation Database) at the beginning of Aug’2015
The following aspects of the analysis/validation:
Availability
Within-country consistency
Inter-country comparability

8. Availability of input data
5/9/2018
Availability of input data
The initial analysis => not all Regions could supply input data in the necessary details =>
To include in the further detailed analysis for the Global extrapolation the following Regions:
*EGY / EGZ (AFR / WAS), SDN / SDO (AFR/WAS), RUS / RUT (EUO/CIS) are double participants
Region
R-code
Reg-Abbr
No.
Africa 1
AFR 50
Asia and the Pacific 2
ASI 23
CIS 3 9
Eurostat-OECD 4
EU-O 47
Latin America 5 LA 17
Western Asia 6
WAS 12
Singleton (GEO, IRN) 8
SIN
TOTAL
157

9. CPI data Covering: 12 main product groups
5/9/2018
CPI data
Covering: 12 main product groups
General points (reservations):
Applicability as NA deflators?
Methodological differences (different price concepts – Health, Education) => Consistency between HH deflators and CPI-Total
Applicability as PPP extrapolators?
(significant) differences between PPP and CPI baskets
country’s peculiarities in the methodology, etc. =>
the comparability of the CPI figures between the countries?
Technical point:
CPI figures should be presented with the base 2011 = 100

10. (M2/ NPISH are combined with HFCE)
5/9/2018
NA deflators
Covering: 6 main GDP aggregates (M1, M3-M7)
(M2/ NPISH are combined with HFCE)
Availability
No big problems (C13 and M7 – yearly XRs)
Reliability
Validation should be done very carefully
(even for the figures from the official national websites)
Within-country consistency of NA deflators and other price indices should be checked
(see next slide with detected problematic points)

11. NA HFCE deflator vs. CPI-Total: inconsistencies
5/9/2018
NA HFCE deflator vs. CPI-Total: inconsistencies

12. GDP Expenditure data Covering: 6 main GDP aggregates (M1, M3-M7) +
5/9/2018
GDP Expenditure data
Covering: 6 main GDP aggregates (M1, M3-M7) +
13 main product groups for HFCE (C1-C13)
Availability
At least GDP data was mostly available
Missing data – the use of the structure of a benchmark year
Reliability
Numerous current problematic points => Validation of should be done very carefully
Efficient way: the comparison of the structures for similar countries
General point:
SNA’93 / ESA’95 vs. SNA’08 / ESA’10?

13. Several validation rounds of input data (Aug – Sept’15)
5/9/2018
First actions
Several validation rounds of input data
(Aug – Sept’15)
Extrapolation of Global and Regional PPPs
(GDP, M1 + M3-M6; C1 – C12)
Main aim of the 1st set of the experiments
To evaluate - What the aggregated level (GDP deflator, C1-C13+M1-M7, M1-M7) is the most appropriate / practicable for the extrapolation?

14. Extrapolation of Global and Regional PPPs (GDP, M1 + M3-M6; C1 – C12)
5/9/2018
Extrapolation of Global and Regional PPPs (GDP, M1 + M3-M6; C1 – C12)
Extrapolated Global PPP to USD: 
Global PPP “Country X / USA” for year (2011+t) =
Global PPP “Country X / USA” for year 2011 * (Def X2011+t/2011 / Def USA2011+t/2011)
Regional PPPs to Base currency: 
Regional PPP “Country X / Reg.B” for year (2011+t) =
Regional PPP “Country X/Reg.B” year 2011*(Def X2011+t/2011/Def Reg.B2011+t/2011)
Extrapolation of Global and Regional is done separately for GDP, C1-C12, M1-M6
NA deflators - for GDP and M1-M7 aggregates
CPI data – for Consumer Headings (C1-C12)
Yearly XRs – for M7 („Net exports“) and C13 („Net touristic consumption“)

15. EKS aggregation of extrapolated PPPs: possible approaches
5/9/2018
EKS aggregation of extrapolated PPPs: possible approaches
Three approaches are more or less feasible: 
1) No aggregation = Global extrapolation => G (GDP) = G version
2) Intermediate version => The level of main aggregates M1 + M3-M => M version (6 categories)
3) Detailed version => The combination of C1-C13 & M3-M7 =>
=> C+M version (13+5 = 18 categories)
Further treatment for M and C+M versions:
The Global and Regional EKS aggregations for M version and/or for C+M version
Aggregated results for HFCE, AIC, GDP
Regional fixity (for the Regions with new Regional results) =>
CAR-Volume or CAR-PPP approach

16. EKS aggregation of extrapolated PPPs: technical aspects
5/9/2018
EKS aggregation of extrapolated PPPs: technical aspects
Global EKS (157 countries) includes the countries with very different price and expenditure structures
Bilateral F-PPPs different degree of reliability (or even senseless => negative L- or P- PPPs - due to high negative “Net export”) 
This problem occurred already in the ICP 2011 but this problem is more complicated in the Extrapolation =>
It is very difficult to adjust / correct input data from official sources: CPI and NA deflators
In effect, numerous paits of the countries have very high (> than 1.5) or very low (less than 0.95) Laspeyres-Paasche Spread (LPS)

17. EKS aggregation of extrapolated PPPs: technical aspects
5/9/2018
EKS aggregation of extrapolated PPPs: technical aspects
No. of bilateral PPPs with high / low LPS
Negative L- or P- PPPs are senseless; these can’t be used in the further EKS calculations => respective F-PPPs were excluded from the EKS aggregation
However – What to do with the pairs of the countries with very high / very LPS in a general case?

18. Provisional considerations
5/9/2018
Provisional considerations
From the theoretical point of view, the C+M version would be preferable
However many countries have numerous spaces and the comparability of detailed figures between the countries is questionable
So, probably, version M is the best practicable version
[The Eurostat-OECD experience shows that if there is no new price data then the Global extrapolation at the GDP level brings very close results as the aggregation of extrapolated PPP]
It seems that the M version would be very similar to the next version of the Groningen PWT

19. Provisional experiments (1)
5/9/2018
Provisional experiments (1)
First of all, we compared GDP extrapolated "free" PLIs by three approaches:
Global extrapolation by the GDP deflators
Aggregation by the M version
Aggregation by the C+M version
Similar analysis was done also for AIC
For HFCE one can compare two versions:
by NA deflators vs. C1-C13 aggregation

20. Provisional experiments (2)
5/9/2018
Provisional experiments (2)
PLI by different approaches for each year were compared and the differences were calculated:
The PLIs of Global extrapolation was used as the basis
The original ICP 2011 PLIs were used for year 2011
All PLIs were presented the base World = 100 (unweighted GM)
Such presentation is more neutral and more appropriate for the evaluation than the PPPs with the base USD =1.
Presentation below contains Regional absolute average differences
(the country results are presented in Annex 2)
The Regional PLI averages (unweighted GM, World =100) were also calculated and compared
These average Regional PLIs can be used as quasi-Linking Factors = CAR-PPP approach (as it is used by the Eurostat-OECD, to keep sub-regional fixity)

21. Extrapolation of ICP 2011 PPP at different aggregated levels 2005-2006: Av Reg PLIs
5/9/2018

22. Extrapolation of ICP 2011 PPP at different aggregated levels 2005-2006: Av Reg PLIs
5/9/2018

23. Extrapolation of ICP 2011 PPP at different aggregated levels 2005-2006: Av Reg PLIs
5/9/2018

24. 5/9/2018
Extrapolation of ICP 2011 PPP: Av Reg PLIs as LFs PLI“Country/World”=PLI“Country/Region”*PLI “Region/ World”

25. Regional PLIs (World157 = 100) % Differences
Extrapolation of ICP 2011 PPP: Av Reg PLIs as LFs PLI“Country/World”=PLI“Country/Region”*PLI“Region/ World”
5/9/2018
Regional PLIs (World157 = 100) % Differences

26. Time-series obtained by the extrapolation=> Expectable problems
5/9/2018
Time-series obtained by the extrapolation=> Expectable problems
The reasons between original and extrapolated ICP results were described very well in the Ch.8 of the ICP Book
So, the potential significant differences between the Global extrapolations and original Regional results (if they were produced) should be not surprise
Even Eurostat PPP time-series obtained by the Rolling benchmark approach contain big inconsistencies in time – see next slide => VIpc for “Health” as an example. [“Health” comprises a very significant share of expenditure in GDP. Therefore the inconsistencies for “Health” have significant impact on AIC and even GDP results]
Eurostat plans also the comprehensive recalculation of the present PPP results due to the introduction of ESA 2010, more detailed COICOP and the changes in some other classifications. It is desirable to coordinate the ICP 2011 extrapolation works with the Eurostat revisions.

27. 5/9/2018
Time-series => VIpc (EU28=100) for “Health” (an extract from the Eurostat PPP database

28. Conclusions and recommendations
5/9/2018
Conclusions and recommendations
Present set of input data contains still problematic points
They should be reviewed further together with the ICP Regional coordinators
Problematic points should be solved and the extrapolated results should be recalculated
Provisionally, the version with the extrapolation and further aggregation of the NA main aggregates (M version) seems to be the most practicable [The final conclusion on this point should be made after the recalculated results are available]
Regional updates (PPPs, etc.) should be collected and the extrapolated results with regional fixity should be calculated after this

29. 5/9/2018
Thanks for listening