1. Consistency and Extrapolation of ICP Benchmarks: The Case of Asia
Yuri Dikhanov
3rd Regional Coordinating Agencies meeting
October 28-30, 2015 Washington, DC

2. Background
Asian comparisons as a part of the International Comparison Program (ICP)
Why extrapolation between benchmark is necessary?
Inconsistencies between PPPs: ICP benchmarks, national accounts deflators and CPIs
Available information for extrapolation and ICP-SNA (CPI) consistency studies

3. ICP-CPI inconsistencies
Biggest difference: PPPs are spatial and CPIs are temporal indices
Comparing CPIs to changes in PPPs over time involves many factors other than price movements.
CPIs are mostly [but not always] estimated using Laspeyres indices at the aggregate level, based on different years.
ICP uses the EKS (Fisher) index.
Addressing inconsistencies:
Bridge the two benchmarks (2005 and 2011) using the 12 COICOP category CPI components
Predict PPPs by applying CPI components to the corresponding PPP categories

4. Linking 2005 and 2011 ICP: Asia
Comparable in terms of:
Structured product descriptions for individual items of HH
Set of countries (22 out of 23)
Sampling framework
Classification systems used
Methodology
Consistent comparison across time and space is possible

5. Scope of ICP comparison
2005 comparison, 22 countries, 108 basic headings, EKS (Fisher) aggregation
Implicit ICP inflation
2011 comparison, 22 countries, 108 basic headings, EKS (Fisher) aggregation
Joint comparison

6. Scope of extrapolation
2005 comparison, 8 countries,
12 COICOP categories EKS (Fisher or Tornqvist) aggregation
CPI, 12 COICOP categories 8 countries
GDP deflators, 12 COICOP categories 5 countries
2011 comparison, 8 countries,
12 COICOP categories EKS (Fisher or Tornqvist) aggregation

7. Elementary aggregation in ICP
Country-Product Dummy method:

8. Aggregate index in ICP: EKS (Fisher)

9. Index number problem in extrapolation
ICP uses multilateral index: EKS (Fisher) and CPD at the elementary level
In Asia most CPI uses Laspeyres with varying base years (aggregate)
At the elementary level: some uses geometric, harmonic or arithmetic average
CPI uses national expenditure weights
ICP uses national accounts weights (in the context of international comparison)

10. Table 1. Consistency of PPPs Estimated at Different Levels of Aggregation
BAN
HKG
IND
MAL
PHI
SIN
SRI
THA SD
PPPs: single-year estimation (HKG = 1)
2005
ICP, 22 countries 108 BHs, EKS (Fisher)
3.387
1.000
2.072
0.276
3.277
0.188
5.100
2.404
8 countries, COICOP12, EKS (Törnqvist)
3.481
2.143
0.278
3.298
5.236
2.427
COICOP12-8 vs. ICP (geomean=1)
0.986
1.014
0.980
1.007
0.987
1.004
1.25%
8 countries, COICOP12, EKS (Fisher)
3.500
2.141
0.279
3.312
5.304
2.439
0.985
1.018
1.015
0.979
1.40%
2011
4.281
2.586
0.270
3.188
0.191
7.099
2.181
4.339
2.649
0.272
3.220
7.278
2.203
0.998
1.012
1.002
1.001
0.88%
4.345
2.654
3.219
0.190
7.317
2.202
0.997
1.016
0.982
1.10%

11. ICP 2005 and 2011: Consistency of Joint vs. Single-year aggregation
BAN
HKG
IND
MAL
PHI
SIN
SRI
THA SD
ICP2005, EKS, 108 BH
3.387
1.000
2.072
0.276
3.277
0.188
5.100
2.404
ICP2005, EKS, 108 BH Joint
3.287
2.025
0.273
3.201
0.187
5.019
2.387
ICP2011, EKS, 108 BH
4.281
2.586
0.270
3.188
0.191
7.099
2.181
ICP2011, EKS, 108 BH Joint
4.405
2.637
3.255
0.192
7.217
2.207
2005, Joint vs. Individual
0.985
1.015
0.992
1.003
0.991
1.009
0.999
1.007
0.96%
2011, Joint vs. Individual
0.986
1.006
0.997
0.990
1.002
0.998
0.87%

12. CPI vs. Implicit ICP Deflators (Joint)
BAN
HKG
IND
MAL
PHI
SIN
SRI
THA SD
CPI, official
1.678
1.176
1.655
1.177
1.329
1.197
1.774
1.199
22 countries 108 BHs, EKS (Fisher)
1.681
1.255
1.634
1.253
1.276
1.288
1.804
1.160
8 countries, COICOP12, EKS (Törnqvist)
1.658
1.620
1.257
1.815
1.161
8 countries, COICOP12, EKS (Fisher)
1.648
1.626
1.272
1.807
1.158
difference from official CPI
1.002
1.067
0.987
1.065
0.960
1.076
1.017
0.968
4.33%
0.988
0.979
1.068
1.023
4.52%
0.982
0.983
0.957
1.019
0.966
4.40%

13. CPI vs. Implicit ICP Deflators

14. Table 4. ICP-CPI Consistency: Extrapolation vs. Actual Benchmark
extrapolation, 2011 to 2005, with CPI components
BAN
HKG
IND
MAL
PHI
SIN
SRI
THA SD
ICP, 22 countries 108 BHs, EKS (Fisher)
3.387
1.000
2.072
0.276
3.277
0.188
5.100
2.404
8 countries, COICOP12, EKS (Törnqvist)
3.278
2.093
0.277
2.980
0.187
5.035
2.280
COICOP12-8 vs. ICP (geomean=1)
1.009
0.977
0.967
0.974
1.075
0.982
0.990
1.030
3.40%
8 countries, COICOP12, EKS (Fisher)
3.292
2.101
2.971
5.076
2.282
1.007
0.978
0.965
1.079
0.985
0.983
1.031
3.55%
extrapolation, 2005 to 2011, with CPI components
4.281
2.586
0.270
3.188
0.191
7.099
2.181
4.606
2.739
0.274
3.574
7.537
2.348
0.976
1.051
0.992
1.034
0.937
1.050
3.78%
4.646
2.751
0.275
3.595
7.626
2.364
0.973
1.056
0.993
1.035
4.07%
Detailed CPI extrapolation: Eliminate both differences in index numbers between ICP and CPIs and index number differences among CPIs. > Apply detailed CPI by 12 COICOP categories to both benchmarks > Compare the results to the actual benchmarks

15. Precision: CPI vs. GDP (HHCE) Deflator Components

16. Comparative Measures of Inflation: CPI, GDP (HHCE) deflators, ICP implicit inflation

17. Summary
The current study shows that in extrapolating ICP benchmarks with CPIs and GDP deflators, we observe in Asia:
Accuracy of extrapolation
3.4 to 4.1% (CPI components)
5.1% to 5.9% (with GDP (HHCE) component deflators)
These discrepancies are irreducible further
Two distinct clusters observed:
Higher income showing price levels higher than predicted with the CPI;
Lower income having price levels close to or lower than their predicted values.

18. ANNEX: Case of Africa Time period: 2005-11 18 countries
COICOP 12 (most countries)
No joint comparison (so no direct estimate of ICP inflation)
No GDP details (HHCE) available

20. CPI-ICP inconsistencies, by component

21. Conclusions
The largest inconsistencies observed are for Education (CV – 59.1%), Communication – 86%;
Predicted vs. Benchmark ratio for Household Consumption ranged from 84.4% to 127.3%;
No distinct pattern or clustering observed