Eduardo de Rezende Francisco

1
Electricity Consumption asa Predictor of
Household Incomean Spatial Statistics approach
Eduardo de Rezende Francisco Francisco
Aranha Felipe Zambaldi Rafael Goldszmidt FGV-EAES
P
November 21th , 2006 Campos de Jordão, São Paulo,
Brazil
2
Topics

• Introduction
• Income and Economic Classification
• Brazilian Criterion of Economic Classification
• Electricity Consumption
• Objectives
• Research Methodology
• Adopted Model and Postulation of Hypotheses
• Selected Databases and Methodology
• Results
• Conclusions

3
Income and Economic Classification

• Income
• Indicator usually adopted in studies of Poverty,
  Living Conditions and Market
• Difficulty in the collection of accurate data on
  such a variable (BUSSAB FERREIRA, 1999)
• altered declaration, seasonal changes, refusal
  etc.

• (Social and) Economic Classification or
  Purchasing Power based on indicators
• Ownership of goods and the head of the familys
  educational level
• Supply of durable goods indicates the comfort
  level achieved by the family throughout the
  lifetime
• Social Status ? Economic Status ?
  Social-Economic Status
• Bottom of Pyramid X D and E Classes

INTRO
METHODS
RESULTS
CONCLUSION
4
Brazilian Criterion

• Brazil
• ABA Criterion (1970), ABA-ABIPEME (1982), Almeida
  and Wickerhausers Proposal (1991)
• CCEB Brazilian Economic Classification
  Criterion
• Created by ANEP in 1996 and supported by ABEP
  since 2004
• Estimates purchasing power of urban people and
  families
• Economic Classes from a point accumulation system

INTRO
METHODS
RESULTS
CONCLUSION
Source MATTAR, 1996 ABEP, 2004
5
Brazilian Criterion

• Brazil
• ABA Criterion (1970), ABA-ABIPEME (1982), Almeida
  and Wickerhausers Proposal (1991)
• CCEB Brazilian Economic Classification
  Criterion
• Created by ANEP in 1996 and supported by ABEP
  since 2004
• Estimates purchasing power of urban people and
  families
• Economic Classes from a point accumulation system
• Use of variables and indicators that dont have
  stability throughout the time and not well
  discriminate population strata (PEREIRA, 2004)
• It is not suitable for characterizing families
  which lie on the extremes of the income
  distribution (MATTAR, 1996 SILVA, 2004)
• Deeper studies need specializations and
  adjustments of Brazilian Criterion
• Inclusion of high coverage and capillarity
  indicators or variables with no need of constant
  update can be useful

INTRO
METHODS
RESULTS
CONCLUSION
6
Consumption of Electric Energy

• Consumption of Electric Energy can be a good
  indicator to better assist process of
  characterize customers
• Essential Utility
• Wide-ranging and Coverage
• 97.0 of Brazilian households (99.6 in urban
  areas)
• 99.9 in São Paulo municipality
• High Capillarity
• Higher than other utilities (sewer water,
  telecom, gas)
• A to E Customers
• Precision and History
• Address, customer geographic location
• Monthly collected
• History of billing and collection (bad debt
  management)
• Fulfill fundamental part in residential
  households day-by-day high influence in
  welfare of families
• Better characterization of target families (in
  social-economic terms and purchasing power)

INTRO
METHODS
RESULTS
CONCLUSION
Source FRANCISCO, 2002 IBGE, 2003, 2005
ABRADEE, 2003
7
Household Income Electricity Consumption

• OBJ Analyze the relationship between
  Residential Electricity Consumption and Household
  Income in the city of São Paulo
• Evaluate the potential benefits of
• Adding electricity consumption to the Brazilian
  Economic Classification Criteria
• Creating an electricity consumption criteria
• Level of Investigation
• Territorial 456 Weighted Areas (set of census
  tracts) in São Paulo city
• Demographic Census 2000 and Electric distribution
  company households database
• Methodology
• income-predicting models (spatial regression
  models)

INTRO
METHODS
RESULTS
CONCLUSION
8
Research Model and Postulation of Hypotheses
Electric Energy Consumption
Household Income
H2
H3
H4




H1
Ownership of goods
Posse de Bens
Posse de Bens
Posse de Bens
Posse de Bens
BrazilianEconomicStatus
Head of FamilysEducational Level

• H1 The higher the score in the Brazilian
  Criterion (Economic Classification), the higher
  the Household Income, in the city of São Paulo
• H2 The higher the consumption of Electric
  Energy, the higher the Household Income, in the
  city of São Paulo
• H3 There is a spatial dependence pattern of
  Household Income in the city of São Paulo, with
  decreasing income in direction Center-Suburbs
• H4 There is a spatial dependence pattern of
  Electric Energy Consumption in the city of São
  Paulo, with decreasing income in direction
  Center-Suburbs

INTRO
METHODS
RESULTS
CONCLUSION
9
Methodology

• Demographic Census Energy Consumption
• Analysis unit Weighted Areas
• 303,669 sampled households (representing
  3,032,095)
• 3,037,992 residential consumers of AES
  Eletropaulo

São Paulo13.278 Tracts
São Paulo456 Areas
10
Methodology

• Demographic Census Energy Consumption
• Analysis unit Weighted Areas

• Geographic overlay and Spatial Junction

AES Eletropaulo consumers Database
Weighted Areas (IBGE)
Average INCOMEper Weighted Area
ENERGY CONSUMPTIONper Consumer
Spatial Join
INCOME andENERGY CONSUMPTIONper Weighted Areas
11
Methodology

• Demographic Census Energy Consumption
• Analysis unit Weighted Areas

• Geographic overlay and Spatial Junction

• Creation of Adjusted Brazilian Criteria based on
  Demographic Census 2000

12
Results Traditional Correlation and Regression

• Similar behavior between various representatives
  of Household Income construct and Electric Energy
  Consumption construct
• High correlation and determination coefficient
  (R2) between Household Income, Electric Energy
  Consumption and Brazilian Economic Criteria, it
  grows down for low income territories

y Household Income (R) xLUZ Electric Energy
Consumption (US)
y Household Income (R) xCBA Brazilian Economic
Criteria
Household Income (R)
Household Income (R)
INTRO
Electric Energy Consumption (kWh)
Brazilian Economic Status
METHODS
Kolmogorov-Smirnov test of Normality 0.129
Kolmogorov-Smirnov test of Normality 0.171
RESULTS

• Non-normality of the residuals

CONCLUSION
13
Neighborhood Graphs

• For different neighborhood matrix analyzed,
  Morans I showed high values (0.78)
• It suggests high influence of neighborhood in
  Household Income behavior
• LISA maps Increase of income concentration in
  direction Suburbs-Center. The same for
  Electricity consumption

14
Results Spatial Statistics
Spatial Auto-regressive Model

• Data set electric energy
• Spatial Weight areaqueen1.GAL (Queen
  Graph)
• Dependent Variable LNINCOME Number of
  Observations 456
• Mean dependent var 7.46738 Number of
  Variables 3
• S.D. dependent var 0.633242 Degrees of
  Freedom 453
• Lag coeff. (Rho) 0.607507
• R-squared 0.936675 Log likelihood
  171.909
• Sq. Correlation - Akaike info
  criterion -337.818
• Sigma-square 0.0253932 Schwarz
  criterion -325.451
• S.E of regression 0.159352

Morans I 0.07(almost 0)
INTRO
METHODS

• Use of Neperian Logarithms of dependent and
  independent variables
• Residual error of this model assumed normal
  distribution pattern and homoskedasticity -
  Absence of spatial dependence in residuals

RESULTS
CONCLUSION
15
Conclusions

• Use of the mean household electricity
  consumption, at a territorial aggregated level,
  is an excellent regional indicator of income
  concentration in the city of São Paulo

INTRO
METHODS
BrazilianEconomicStatus
Electric Energy Consumption
Household Income
RESULTS
CONCLUSION
16
Managerial Implications
Census tracts
Households
Concentric circles (progressive radius of 125 m)
As it is an easily available, flexible and
monthly updated information, the electric energy
consumption indicators, when published widely by
energy distribution companies, can be useful for
strategy formulation and decision making which
use data of household income classification,
concentration analysis and prediction.
Quadricules (1 square kilometer)
17
Household Income Electricity Consumption

• Conclusions
• Energy consumption alone cannot substitute for
  the Brazilian Criteria
• Nevertheless, household income forecasts can be
  enhanced when the electricity bill and the
  number of residents are included in a regression
  model of household income against the Brazilian
  Criteria
• Among low income households, the level of
  association between income and electricity
  consumption was very weak
• Use of the mean household electricity
  consumption, at a territorial aggregated level,
  is an excellent regional indicator of income
  concentration in the city of São Paulo
  (coefficient of determination R2 reached more
  than 0,90)

INTRO
METHODS
RESULTS
CONCLUSION
18
Household Income Electricity Consumption

• Next Steps (Future researchs)
• Investigation of other statistical models
• Geostatistics, Spatial Econometrics and
  Hierarchical methods (spatial regression)
• To handle heterokedasticity and non-normality in
  some regression models
• Support for Low Income Microcredit Programs
• Inclusion of Household electricity monthly bill
  in Discriminant analysis models
• Replacement of declared Household Income by Mean
  electricity consumption of region that locates
  household of tomador de crédito
• Validation of territorial results with more
  updated data, when and if it is available
• Replication in other regions (inside and outside
  Brazil)
• Comparative studies (Europe, Brazil Latin
  America)

BrazilianEconomicStatus
Electric Energy Consumption
Household Income
INTRO
METHODS
RESULTS
CONCLUSION
19
Thank You !!!
Electricity Consumption as a Predictor of
Household Incomean Spatial Statistics approach
Eduardo de Rezende Francisco, Francisco
Aranha,Felipe Zambaldi, Rafael Goldszmidt FGV
EAESP November 21th 2006 , Campos de Jordão,
SP, Brazil