Cliquez pour modifier le style des sous-titres du masque
Cliquez pour modifier le style des sous-titres du masque
Summary
2
1. C ompos ite indic a tors and ranking
4. S ome examples of implementing C FAR -m
2. Traditiona l methods for c ons truc ting c ompos ite indic a tors and their w eaknes s
3. The C FAR -m a lg orithm
Cliquez pour modifier le style des sous-titres du masque
3
1. Composite indicators and ranking
A C ompos ite Indic a tor is an ag reg ate index that s ummarizes a larg e amount of information g iven by s ing le indic ators .
Competitivity (Global Competitivity Index - FEM) Country risk (ICRG-PRS group) Well-being (Health System Achievement Index-WHO) Environment (Environmental Sustainability Index- WEF) Governance (The Corruption Perceptions Index - Transparency International) Innovation (Technology Achievement Index- UN)
C ompos ite Indic ators are inc reas ing ly being us ed to meas ure multidimens iona l performanc e and to rank c ountries , firms , c lients , ins titutions , etc ., in many fields , s uc h as :
What is it ?…
2 reas ons , bas ic a lly :
1- C omplexity of modern ec onomy : jus t one, or a s et of s ing le indic a tors is not enoug h any more.
2- Development of IC Ts : it means that a hug e mas s of information has to be proc es s ed
4
A real interest …
1. Composite indicators and ranking
Demand for, and produc tion of C ompos ite Indic ators are rapidly g row ing .
Google search results for "composite indicators"
Cliquez pour modifier le style des sous-titres du masque
5
2. Traditional methods for constructing composite indicators and their weakness A great number …
Most used weighting schema in aggregation methods:
E qua l w eig hts
Weig hts bas ed on s ta tis tic a l models
Da ta E nvelopment Ana lys is (DE A) P rinc ipa l C omponent Ana lys is (PC A) U nobs erved C omponents M odels (U C M )
B udg et a lloc ation
Weig hts bas ed on the s ta tis tic a l qua lity of da ta
S tandard devia tion
Weig hts bas ed on experts ’ opinions
Cliquez pour modifier le style des sous-titres du masque
6
Drawbacks of traditional methods :
They are exogenous They are linear
They lose information
They offer a no positive capability to assist decision-making processes
Many problems…
2. Traditional methods for constructing composite indicators and their weakness
An orig ina l method bas ed on artific ia l intellig enc e for the c ons truc tion of c ompos ite indic ators tha t a llow s to perform relevant ranking .
The w eig hting s c hema of s ing le indic ators is g enerated throug h a learning proc es s , from informationa l c ontent of the variables thems elves and their interna l dynamic s .
7
3. The CFAR-m algorithm
Innovation
Our solution …
C -FAR m w orks in three s tag es that a re s truc tura lly c ombined :
8
S tag e 1 : Firstly, it carries out a c la s s ific a tion (self-organization) of objects (records, points, cases, samples, entities, or instances) through a lea rning proc es s that takes into account interactions between the attributes (variables, fields, characteristics, or features) in homog eneous c lus ters .
3. The CFAR-m algorithmOur solution …
Preliminary stage : Preparing the data base
Stage 1 : Classification
9
S tag e 2 : S ec ondly, an appropria te w eig hts vec tor is g enerated for eac h objec t.
3. The CFAR-m algorithmOur solution …
Stage 1 : Classification
Stage 2 : Generating weights: one vector is defined for each object
10
S tag e 3 : Thirdly, w eig hts vec tors a re applied to the orig ina l data to c ompute C FAR -m c ompos ite indic ators and fina lly to c a rry out the overa ll ranking of objec ts .
3. The CFAR-m algorithmOur solution …
Stage 2 : Generating weights: one vector is defined for each object
Stage 3 : Computing the composite indicators and rankingthe objects
11
Stage 3 : Computing the composite indicators and rankingthe objects
Stage 2 : Generating weights: one vector is defined for each object
Stage 1 : Classification
Preliminary stage : Preparing the data base
3. The CFAR-m algorithmOur solution …
3. The CFAR-m algorithm
Our s olution is bas ed on an orig ina l tec hnique tha t us es neura l netw orks and, unlike exis ting methods , pres ents the follow ing c harac teris tic s :
Objectivity There is no manipula tion of w eig hts . The Weights used to aggregate single indicators are generated automatically from the database through a learning process. Our model provides a fundamental solution to the main aggregation problem.
12
S pecificity E ac h objec t has a s pec ific equation to compute its composite indicator.
Decision support I t a llow s perform ing of s imula tions and therefore, can help to decide on appropriate actions and corrections.
4. Some examples of implementing C-FARm
13
Case study 1 : Computing a CFAR-m Human Development Index(comparison with the UNDP aggregation methodology based on equal weights)
Case study 2 : Computing a CFAR-m indicator Governance Index
(comparison with the MINEFI-France aggregation methodology using weights based on
statistical quality of data )
Case study 3 : Computing a CFAR-m Country Risk Index
(comparison with the PRS Group aggregation methodology based on expert opinion)
Cliquez pour modifier le style des sous-titres du masque
Case study 1 : Computing a CFAR-m Human Development Index
(comparison with the UNDP aggregation methodology based on equal weights)
14
4. Some examples of implementing C-FARm
Cliquez pour modifier le style des sous-titres du masque
H ea lth
Educ ation
S tandard of L iving
HDI
Dimens ions
In its firs t Human Development Report (1990), the U nited N ations Development P rog ram (U N DP) introduc ed a new index : H uman Development Index (HD I).
H D I is intended to s ummarize in one meas ure three dimens ions of the development proc es s : long evity, educ ationa l a tta inment, and s tandard of living .
15
V ariables (bas ic indic a tors )
Life expec tanc y a t birth
A dult literac y ra te
Primary, s ec ondary and tertiary s c hooling enrolment ra tios
G DP per c apita
Case study 1 : Computing a CFAR-m Human Development Index
Cliquez pour modifier le style des sous-titres du masque
16
To c ompute the HD I, U N DP c ons ider the s imple averag e (equa lly w eig hted s um) of the tree dimens ions .
The three dimens ions have the s ame w eig ht
Life expec tanc y index
E duc ation index
G DP
index
Case study 1 : Computing a CFAR-m Human Development Index
Cliquez pour modifier le style des sous-titres du masque
… … , thus , c omparis ons among different c ountries /reg ions a re c a rried out.
17
Case study 1 : Computing a CFAR-m Human Development Index
The main arguments against HDI:
Important dimens ions a re not c ons idered (freedom, human rig hts ,
g overnanc e, etc .)
H D I is hig hly c orrela ted to the G DP (0,89 ac c ording to M ac G illivray,
1991).
The three dimens ions a ls o a re hig hly c orrela ted to the G DP
Weig hting of the three dimens ions is too s ubjec tive
18
Case study 1 : Computing a CFAR-m Human Development Index
“The best known macro-indicator in the world is probably the Human Development Index (HDI) developed by the United Nations Development Program. It has been severely criticized for combining together indicators of income, health and education to create a composite index, both on the grounds that the weights are arbitrary and unjustified and on the grounds that the three components of the index are highly correlated and hence give redundant results”
Literature Review of Frameworks for Macro-indicatorsAndrew Sharpe (2004)
19
Case study 1 : Computing a CFAR-m Human Development Index
The main critics made to the HDI :
20
S tag e 1 : C ountry c la s s ific a tion
Case study 1 : Computing a CFAR-m Human Development Index
21
Weig hting s c heme differs from one c ountry to another : C FAR -m is non-linear
Life expec tanc y index
Educ a tion index
G DP
index
Case study 1 : Computing a CFAR-m Human Development Index
S tag e 2 : G enera ting w eig hts : one vec tor is defined for eac h c ountry
22
Each country has a specific equation to compute its development index :
S pec ific ity
CFAR-m allows the identification, for each country, of the dimension that most influenced the calculation of its index, and therefore its ranking :
Intens ity and S ig n
Weights are generated automatically through a learning process from the database :
Objec tivity
S tag e 2 : G enera ting s pec ific w eig hting s w ith C FAR -m
The ranking of C FAR -m w ill be both objec tive and relevant.
Case study 1 : Computing a CFAR-m Human Development Index
S tag e 3 : C omputing a C FAR -m Human Development Index and ranking c ountries
23
HDI dimensions - Year 2005 CFAR-m's results for year 2005
Countries topping the list
Life expectancy
indexEducation
indexGDP
index Country
CFAR-m rankt
CFAR-m rank minus
UNDP rank
Ic eland 0.941 0.978 0.985 IS L 1 0
N orway 0.913 0.991 1.000 N OR 2 0
Aus tra lia 0.931 0.993 0.962 AU S 3 0
C anada 0.921 0.991 0.970 C AN 4 0
I reland 0.890 0.993 0.994 IR L 5 0
S w eden 0.925 0.978 0.965 S WE 6 0
United S ta tes 0.881 0.971 1.000 U S A 7 5
S w itzerland 0.938 0.946 0.981 C HE 8 1
Japan 0.954 0.946 0.959 JPN 9 1
Case study 1 : Computing a CFAR-m Human Development Index
S tag e 3 : C omputing a C FAR -m Human Development Index and ranking c ountries
24
HDI dimensions - Year 2005 CFAR-m's results for year 2005
Countries closing the list
Life expectancy
index
Education index
GDP index
Country CFAR-m rank
CFAR-mrank
minus UNDP rank
Burundi 0.391 0.522 0.325 BDI 169 2
Central Afr. Rep. 0.311 0.423 0.418 CAF 170 1
Mozambique 0.296 0.435 0.421 MOZ 171 1
Guinea-Bissau 0.347 0.421 0.353 GNB 172 3
Chad 0.423 0.296 0.444 TCD 173 3
Mali 0.469 0.282 0.390 MLI 174 1
Sierra Leone 0.280 0.381 0.348 SLE 175 2
Burkina Faso 0.440 0.255 0.417 BFA 176 0
Niger 0.513 0.267 0.343 NER 177 3
Case study 1 : Computing a CFAR-m Human Development Index
25
C FAR -m as a dec is ion s upport s olution
As w eig ht s c hemas a re s pec ific , it a llow s to perform s imula tions
Case study 1 : Computing a CFAR-m Human Development Index
Number of ranks gained in overall ranking
Life expectancy index
Impact on the rank of an improvement 0.1 in one dimension (here, for North African countries)
26
Case study 1 : Computing a CFAR-m Human Development Index
Number of ranks gained in overall ranking
Life expectancy index
Education index
As w eig ht s c hemas a re s pec ific , it a llow s to perform s imula tions
Impact on the rank of an improvement 0.1 in one dimension (here, for North African countries)
As w eig ht s c hemas a re s pec ific , it a llow s to perform s imula tions
Impact on the rank of an improvement 0.1 in one dimension (here, for North African countries)
C FAR -m as a dec is ion s upport s olution
27
Case study 1 : Computing a CFAR-m Human Development Index
Education indexGDP index
Life expectancy index
As w eig ht s c hemas a re s pec ific , it a llow s to perform s imula tions
Impact on the rank of an improvement 0.1 in one dimension (here, for North African countries)
C FAR -m as a dec is ion s upport s olution
Number of ranks gained in overall ranking
Cliquez pour modifier le style des sous-titres du masque
28
Case study 2 : Computing a CFAR-m Governance Index
(comparison with the MINEFI-France aggregation methodology using weights based on statistical quality of data)
29
The " Ins titutiona l profiles " databas e
It gathers a whole set of indicators characterizing the institutions of 85 developed and emerging countries
Each variable is weighted according to its standard deviation
132 variables
Information ag g reg a tion
proc es s
9 governance indicators
Case study 2 : Computing a CFAR-m Governance Index
30
9 g overnanc e indic a tors
1 : Political institutions
2 : Public order
3 :Perfomance of Administration
4 :Efficiency of free markets
5 :Prospective and planning
6 : Security of transactions
7 : Regulation
8 : Foreign openness
9 : Social cohesion
Information ag g reg ation
proc es s
132 variables
85 c
ount
ries
Case study 2 : Computing a CFAR-m Governance Index
The " Ins titutiona l profiles " databas e
Gathers a whole set of indicators characterizing the institutions of 85 developed and emerging countries
31
1st dimension's case : ″political institutions″
Case study 2 : Computing a CFAR-m Governance Index
S tag e 1 : C ountry ranking
32
Components of the 1st dimension re. political institutions
R eminder : in the M IN E FI 's method, the w eig ht of one variable c omes from its s tandard devia tion
The component that weighs the most in the computation
The component that weighs the least in the computation
H ow leg itimate a re thos e w eig hting s ? And w hat about the fac t tha t they apply to a ll c ountries ?
Case study 2 : Computing a CFAR-m Governance Index
S tag e 2 : G enerating s pec ific w eig hts w ith C FAR -m
33
Case study 2 : Computing a CFAR-m Governance Index
S tag e 2 : G enerating s pec ific w eig hting s w ith C FAR -m
There is no empiric a l manipula tion. Weig hting s a re proc es s ed us ing the s ole information embedded in the variables .
Kuwait
Components of the 1st dimension re. political institutions
34
Case study 2 : Computing a CFAR-m Governance Index
S tag e 2 : G enerating s pec ific w eig hting s w ith C FAR -m
There is no empiric a l manipula tion. Weig hting s a re proc es s ed us ing the s ole information embedded in the variables .
Kuwait
Components of the 1st dimension re. political institutions
35
1st Dimension: "Political Institutions"Countries
topping the list CFAR-m ranking
MINEFI ranking
Ranking spread
Sweden 1 1 0France 2 3 -1New Zeland 3 2 1Spain 4 6 -2Canada 5 4 1Germany 6 5 1Norway 7 7 0USA 8 15 -7Italy 9 12 -3India 10 9 1Czech Rep. 11 8 3Ireland 12 11 1Senegal 13 16 -3Brazil 14 18 -4Israel 15 21 -6Hong Kong 16 26 -10Greece 17 10 7Hungary 18 14 4Argentine 19 19 0
Case study 2 : Computing a CFAR-m Governance Index
S tag e 2 : G enerating s pec ific w eig hting s w ith C FAR -m
S tag e 3 : C omputing a C FAR -m G overnanc e Index and ranking c ountries
36
Ranking according to CFAR-m Governance IndexCountries
topping the listCountries
closing the list
1 Sweden 76 Nigeria
2 Ierland 77 Cameroon
3 Israel 78 Yemen
4 Spain 79 Ouzbekistan
5 Canada 80 Mauritanie
6 Norway 81 Egypt
7 Italy 82 Syria
8 Germany 83 Iran
9 Portugal 84 Ivory Coast
10 Hungary 85 Chad
Onc e a ll dimens ions of the ins titutiona l profile have been c omputed w e have proc es s ed w ith the fina l a g g reg a tion, produc ed a C FAR -m indic a tor for eac h c ountry and then a g loba l rank ing , whic h the M IN E FI c ould not c omplete !
Case study 2 : Computing a CFAR-m Governance Index
Ra
nks
gain
ed in
the
wor
ld ra
nkin
g
37
C FAR -m is a va luable dec is ion s upport
Dimensions of the "institutional profile" when affected with a 10% increase
This is the dimens ion tha t a llow s to
prog res s the quic ker in the ranking
Case study 2 : Computing a CFAR-m Governance Index
Publ
ic o
rder
Perf
. of
Adm
in.
Perf
. of f
ree
mar
kets
Pros
pect
ive
& p
lann
ing
Secu
rity
of
tran
sact
.
Reg
ulat
ion
Soci
al
cohe
sion
Fore
ign
open
ness
Polit
ical
in
stitu
tions
38
Case study 3 Computing a CFAR-m Country Risk Index
(comparison with the PRS Group aggregation methodology based on expert opinion)
39
Proc es s :
E c onomic variables
Information ag g reg a tion
proc es s
R is k indic ator
"B lac k box"Generally, there is no indication about the computation
method
Case study 3 : Computing a CFAR-m Country Risk Index
40
E ac h s ub-indic a tor is c ompos ed w ith s evera l fac tors too :
Applic a tion to thehe PR S G roup's International Country Risk Guide
The IC R G breaks the c ountry ris k into 3 s ub-c las s es :
C ompos ite indic a tor :
C ountry-ris k indic ator
S ub-indic a tor #1 :
Politic a l ris k
S ub-indic a tor #2 :
E c onomic ris k
S ub-indic a tor #3 :
Financ ia l ris k
Case study 3 : Computing a CFAR-m Country Risk Index
41
E very s ub-indic ator is a c ompos ite its elf :
S ub-indic a tor #1 :
Politic a l R is k
12 factors Score (max)
A Government's stability 12B Social and Economic environment 12C Investment environment 12D Internal conflicts 12E External conflicts 12F Corruption 6G Military's influence on policy 6H Influence of religions on policy 6
I Law and regulation 6J Ethnic lobbying 6K Democratic responsibility 6M Administration and stability of the
institutions4
Total 100
Applic a tion to thehe PR S G roup's International Country Risk Guide
Case study 3 : Computing a CFAR-m Country Risk Index
42
5 factors Score (max)
A GDP per capita 5B GDP growth 10C Inflation rate 10D Balance of payments (% of GDP) 10E Current account (% of GDP) 15
Total 50
S ub-indic a tor #2 :
E c onomic ris k
Applic a tion to thehe PR S G roup's International Country Risk Guide
Case study 3 : Computing a CFAR-m Country Risk Index
43
S ub-indic a tor #3 :
Financ ia l ris k5 factors Score (max)
A External debt (% of GDP) 10B Cost of external debt (% of GDP) 10C Current account (% of goods and
services exports)15
D International net liquidity (months of import funding)
5
E Exchange rate stability 10Total 50
Applic a tion to thehe PR S G roup's International Country Risk Guide
Case study 3 : Computing a CFAR-m Country Risk Index
44
M eas uring the politic a l-ris k fac tor for year 2006Country Govern
ment's stability
Social and
Economic
environment
Investment
environment
Internal conflicts
External conflicts
Corruption
Military's influence on policy
Influence of
religions on policy
Law and regulation
Ethnic lobbying
Democratic
responsibility
Administration
and stability of the institutions
Albania 8.5 5.5 8.0 10.0 11.0 1.0 5.0 5.0 2.5 4.5 5.0 2.0
Algeria 9.6 5.8 9.1 8.9 10.0 1.5 3.0 2.5 3.0 3.5 4.5 2.0
Angola 9.6 2.0 7.9 9.3 11.0 2.0 2.0 4.0 3.0 3.0 2.0 1.0
Argentina 10.2 5.2 6.6 10.0 10.0 2.5 4.5 6.0 2.5 6.0 4.5 3.0
Armenia 8.4 4.0 8.0 8.6 7.6 1.5 3.5 5.0 3.0 5.5 3.0 1.0
Australia 10.3 9.7 12.0 9.3 9.6 4.6 6.0 6.0 5.5 4.0 6.0 4.0
……….
C ountry-ris k indic a tor
Politic a l ris k E c onomic ris k Financ ia l ris k
Case study 3 : Computing a CFAR-m Country Risk Index
45
S tag e 1 : C ountry ranking
Case study 3 : Computing a CFAR-m Country Risk Index
46
R eminder : PR S G roup methodolog y Weig hting of eac h variable defined by experts Weig hting s a re the s ame w hatever the c ountry
1st dimension factors re. political institutions
12 factors
V1 Government's stability V7 Military's influence on policy
V2 Social and Economic environment
V8 Influence of religions on policy
V3 Investment environment V9 Law and regulation
V4 Internal conflicts V10 Ethnic lobbying
V5 External conflicts V11 Democratic responsibility
V6 Corruption V12 Administration and stability of the institutions
Case study 3 : Computing a CFAR-m Country Risk Index
S tag e 2 : defining s pec ific w eig hting s w ith C FAR -m
47
N ot a ll c ountries have the s ame w eig hting s : it s how s that C FAR -m is a non-linear proc es s
Case study 3 : Computing a CFAR-m Country Risk Index
1st dimension factors re. political institutions
48
N ot a ll c ountries have the s ame w eig hting s : it s how s that C FAR -m is a non-linear proc es s
Case study 3 : Computing a CFAR-m Country Risk Index
S tag e 2 : defining s pec ific w eig hting s w ith C FAR -m
1st dimension factors re. political institutions
49
M eas uring the country-ris k fac tor for year 2006
CFAR-m results PRS Group results
Countries topping the
list Ranking Country Ranking Country Spread
1 Finland 1 Finland 0
2 Iceland 2 Luxembourg 1
3 Luxembourg 3 Iceland -1
4 Sweden 4 Ireland 1
5 Ireland 5 Sweden -1
Case study 3 : Computing a CFAR-m Country Risk Index
50
M eas uring the c ountry-ris k fac tor for year 2006
CFAR-m results PRS Group results
Countries in the
middle of the list
Ranking Country Ranking Country Spread
……… ……… ……… ………
68 Saudi Arabia 68 Saudi Arabia 0
69 El Salvador 74 El Salvador -5
70 Guatemala 80 Guatemala -10
71 Ghana 67 Ghana 4
72 Brazil 76 Brazil -4
……… ……… ……… ………
Case study 3 : Computing a CFAR-m Country Risk Index
51
M eas uring the country-ris k fac tor for year 2006
CFAR-m results PRS Group results
Countries closing the list
Ranking Country Ranking Country Spread
135 Haiti 135 Ivory Coast 1
136 Ivory Coast 136 Haiti -1
137 Serbia 137 Congo, RD 3
138 Montenegro 138 Iraq 1
139 Iraq 139 Serbia -2
140 Congo, RD 140 Montenegro -2
141 Somalia 141 Somalia 0
Case study 3 : Computing a CFAR-m Country Risk Index
Cliquez pour modifier le style des sous-titres du masque
Questions & Answers
Going forward...