IMPLEMENTING UNIT VALUE INDICES IN THE ANNUAL OECD INTERNATIONAL TRADE IN COMMODITY STATISTICS (ITCS) DATABASE Handling missing values, outliers in unit value variations, representativity of the indices, data conversion across classifications Working Party on Trade in Goods and Services 7-9 November 2011
31
Embed
IMPLEMENTING UNIT VALUE INDICES IN THE ANNUAL OECD INTERNATIONAL TRADE IN COMMODITY STATISTICS (ITCS) DATABASE Handling missing values, outliers in unit.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
IMPLEMENTING UNIT VALUE INDICES IN THE ANNUAL OECD INTERNATIONAL TRADE IN COMMODITY STATISTICS (ITCS) DATABASE
Handling missing values, outliers in unit value variations, representativity of the indices, data conversion across classifications
Working Party on Trade in Goods and Services
7-9 November 2011OECD Statistics Directorate
Outline of presentation
• Background• Data universe and formulae• Supplementary units or net weights?• Conversion issues• Estimation of missing values• Dealing with outliers• Specific products
Background 1/2
Advantages of UVIs and Quantities Indices: • proxies for X & M price indices and for volume measures• available at product detail level and in a timely manner• international comparability• analysis of :– terms of trade– price and non-price competitiveness in X &M– quantity effect from the transmission of inflation via foreign trade
• Context• follow-up of the WPTGS 2010• recommandation of the IMTS 2010• results of the short Survey WPTGS 2010
Background 2/2
• Stocktaking of the implementation:– on tracks with PWB– process with SAS– chained Laspeyres Paasche Fisher indices– on 34 countries– HS88 at a total level / partner World
• Issues that are being looking at : – missing values (of quantities and trade values) – outliers– conversion across HS classifications
Data Universe 1/3
ITCS database :– UNSD / OECD joint data processing
system– Contains Information on
– Trade values in US dollars– Quantities (liters, m²…) – Net Weight (kg)
• Estimation of missing quantity/weight by the UNSD method :
1) use of the available info (i.e. net weight proxied from quantity)
2) use of a Standard Unit Value
• Exclusion of 6 digits commodities without information on weights (whole chapter 99 )
Data Universe 3/3
•UV = Value in US dollars Quantity
•Unit Values Indices are sensitive to exchange rates fluctuations; not Quantity Indices …
(Pt)
Computation of Quantity indices
LASPEYRES
PAASCHE
FISHER
111
11
11
11/ *
*
*
*
*)(
t
t
tt
tt
tt
tttt Q
Q
QP
QP
QP
QPQL
t
t
tt
tt
tt
tttt Q
Q
QP
QP
QP
QPQP 1
11/ *
*
*/1
*
*)(
)(*)()( 1/1/1/ QLQPQF tttttt
Quantity ratio form
Weighing system
Computation of Unit Value indices
LASPEYRES
PAASCHE
FISHER
111
11
11
11/ *
*
*
*
*)(
t
t
tt
tt
tt
tttt P
P
QP
QP
QP
QPPL
t
t
tt
tt
tt
tttt P
P
QP
QP
QP
QPPP 1
11/ *
*
*/1
*
*)(
)(*)()( 1/1/1/ PLPPPF tttttt
Quantity ratio form
Weighing system
Defining the denominator of the UV
Supplementary Units or Net Weights ?
Should we use supplementary units?
• Supplementary units more accurate than Net weights for some commodities BUT
How to handle changes of quantity units if for instance the quantity unit is one year, ‘’number of items (5)’’ and
the next year ‘’ thousands of items (9)’’
On long series, net weights are more reported UNSD estimated that 75% of supplementary units is KG
=> OECD choice : net weight in kg
Representativity of the sample:Issue with historical series
• 2009 :– <50% Israel X (80% for Israel M)– >75% for All others OECD countries (M and X)
• 1999 : more problematic – <25% Canada(M and X) , Australia M and USA M – <60% + New Zealand (M), USA (X), Norway (X),
Australia (X)– <75% + Japan (X), Norway (M), Mexico (M), New
Zealand (X) and Netherlands (M & X)
• 1989 / 21 countries– <75% Canada (M & X), USA (M&X), New Zealand (M)
and Australia (M & X), Japan (X) , Norway (M & X)
=> Thresholds on UVI for coverage ?
Conversion issues
use of HS 1988conversion of
HS 1996, HS 2002 HS 2007
Conversion of HS
Conversion issues: looking at HS1988
Handling gaps in series
Estimating Missing Values and Quantities:
Unconvincing tests of 2 methods
Estimation of Trade Values cmd 010111
Estimation of Quantities based on Standard Unit Values (SUV)
• SUV– Compiled by UNSD– median unit value of – each 6-digit commodity/year/flow – after elimination of outliers– of a sample of Unit Value of– available data of the latest reporting year– that respect a certain number of criteria
(box 1)
Missing values estimation conclusions
• Thresholds for estimation for trade values: 1% of values estimated for the whole chapter
• Calculated indices present dubious movements :• Large fluctuations for chained type
Indexes at 6 digits level=> Issues for more disaggregated indices (2&4
digits)
Outliers detection
Outlier detection
• “an unrealistic price growth in the product specific distribution of unit value ratios”
CEPII indices :“ Unit values ratios are compared with the
product specific median change in unit values computed over the whole period”
• .
Outlier detection • “Value that lie far from the
middle of the distribution in either direction”
• Mexico and Italy :>100 obs : Asymmetric Fence
Method<100 obs : Mean Absolute Deviation• UNSD : Tukey Method
AFM and MAD Formulae
• Assymetric Fence Method
• Mean Absolute Deviation
UV =logarithm of the unit valueQ1, Q2; Q3 1st 2nd 3rd quartiles of log unit values of trade distributionc, k, A parameters
|)|,max(
|)|,max(
22338
21281
qcqqkquv
qcqqkuvq
AFMihs
AFMihs
||*||
*||
228
828
qAquv
MADkquv
ihs
hsMADihs )( 288 quvmedianMAD i
hshs
Specific products
chapters
% of outliers within a chapter using symetric yearly variation detection method
% of outliers within a chapter using AFM MAD detection method