manuscriptpreparedforpublicationin! …web.uvic.ca/~egugl/dresden_files/tu.pdf · ! 4!...

9
1 Manuscript prepared for publication in Studies in Microeconomics May 2014 Transferable Utility in the Case of Many Private and Many Public Goods Elisabeth Gugl 1 Abstract It is possible to have income effects on more than one good in utility profiles that lead to Transferable Utility and in the presence of many private and many public goods. Assuming that the utility functions are of the Generalized Quasilinear form is not necessary for TU to hold. I present a much broader class of utility profiles generating TU in which GQL emerges as a special case. 1 Department of Economics, University of Victoria, P.O. Box 1700 STN CSC, Victoria, BC, V8W 2Y2, Canada. Email: [email protected] I would like to thank Ted Bergstrom, PierreAndré Chiappori, Somdeb Lahiri, an anonymous referee, and Linda Welling for their helpful comments.

Upload: ngoxuyen

Post on 10-May-2018

216 views

Category:

Documents


1 download

TRANSCRIPT

  1  

Manuscript  prepared  for  publication  in  

Studies  in  Microeconomics  

May  2014  

Transferable  Utility  in  the  Case  of  Many  Private  and  Many  Public  Goods  

Elisabeth  Gugl1  

Abstract  

It  is  possible  to  have  income  effects  on  more  than  one  good  in  utility  profiles  that  lead  to  Transferable  Utility  and  in  the  presence  of  many  private  and  many  public  goods.  Assuming  that  the  utility  functions  are  of  the  Generalized  Quasi-­‐linear  form  is  not  necessary  for  TU  to  hold.  I  present  a  much  broader  class  of  utility  profiles  generating  TU  in  which  GQL  emerges  as  a  special  case.  

                                                                                                               1  Department  of  Economics,  University  of  Victoria,  P.O.  Box  1700  STN  CSC,  Victoria,  BC,  V8W  2Y2,  Canada.  Email:  [email protected]  I  would  like  to  thank  Ted  Bergstrom,  Pierre-­‐André  Chiappori,  Somdeb  Lahiri,  an  anonymous  referee,  and  Linda  Welling  for  their  helpful  comments.  

  2  

1.  Introduction  

Chiappori  (2010)  derives  testable  implications  of  Transferable  Utility  (TU)  by  stating  conditions  on  demand  that  have  to  be  satisfied.  One  condition  is  that  there  can  be  only  one  private  good  which  has  a  strictly  positive  income  effect.  This  condition  relies  on  the  assumption  that  TU  requires  utility  functions  of  the  Generalized  Quasi-­‐linear  (GQL)  form.  In  this  article  I  point  out  that  while  TU  is  satisfied  if  utility  functions  are  of  GQL  form,  there  are  other  utility  functions  for  which  TU  is  satisfied  and  they  do  allow  for  strictly  positive  income  effects  on  more  than  one  private  good.  In  fact,  Bergstrom  has  made  this  argument  repeatedly  Bergstrom  (1989,  1999)  and  it  goes  back  to  Bergstrom  and  Varian  (1985).  This  article  therefore  accomplishes  two  goals.  To  recall  the  exhaustive  list  of  indirect  utility  functions  that  lead  to  TU  in  the  presence  of  private  and  public  goods  and  to  give  examples  of  a  broader  range  of  utility  functions  satisfying  TU  than  the  subclass  considered  by  Chiappori.            

Chiappori  identifies  testable  implications  of  an  important  subclass  of  utility  functions  that  lead  to  TU,  but  they  are  by  no  means  necessary  for  TU  to  hold.  Hence  Chiappori's  results  in  his  2010  article  should  be  read  as  testable  implications  of  GQL  utility  functions,  rather  than  TU  itself.            

In  the  next  section  I  introduce  the  model  following  Chiappori's  notation  as  much  as  possible.  I  give  examples  of  utility  profiles  satisfying  TU  with  and  without  the  consideration  of  public  goods.  Then  I  present  a  broad  class  of  utility  profiles  satisfying  TU  when  many  private  and  public  goods  are  considered.  As  noted  this  class  is  considerably  larger  than  utility  profiles  generated  by  GQL  utility  functions.  

 

2.  The  Model  

Consider  a  group  of  

S  individuals  with  preferences  over  

L  private  goods  and  over  

N  public  goods.  The  public  goods  are  public  within  the  group  only.  The  group  purchases  these  goods  in  the  market  at  prices  

p,P( )  where  

p = p1,..., pL( )  is  the  price  vector  for  private  goods  and  

P = P1,...,PN( )  is  the  price  vector  for  public  goods.  Each  individual  

m ∈S  consumes  a  commodity  vector  

xm ,X( )  where  

xm = xm1 ,...,xm

L( )  is  the  vector  of  private  goods  consumption  and  

X = X1,...,XN( )  is  the  vector  of  public  goods  consumption.  Each  individual  has  a  fixed  income  of  

ym  and  the  group's  income  is  given  by  

y = ym∑ .  The  group's  budget  constraint  is  given  by      

pxm + PX = ym∈S∑         (1)  

Denote  by  

um xm,X( )  the  utility  function  of  group  member  

m .  In  what  follows  it  will  be  useful  to  distinguish  between  the  group  expenditure  on  public  goods  and  

  3  

individual  expenditure  on  private  goods.  For  any  fixed  

X ,  the  disposable  income  of  group  

S  for  private  consumption  goods  (in  short  disposable  income)  is  

I = y − PX .  

Denote  by  

Im  individual  the  disposable  income  of  member  

m ,  such  that  

Im = I∑ .            

The  indirect  utility  function  conditional  on  given  amounts  of  the  public  goods  is  given  by  

vm X, p,Im( ) .  It  is  the  value  function  of  the  maximization  problem  

maxxm u xm,X( ) s. t. pxm = Im  

We  next  recall  the  definition  of  TU  following  Chiappori  (2010).      

Definition  1    The  framework  defined  above  satisfies  the  transferable  utility  (TU)  property  on  some  open  set  

Ο  of  price-­‐income  bundles  if  the  following  property  is  satisfied:  for  each  agent  

m ,  there  exists  a  cardinal  representation  

um  of  

m ′s  utility  such  that  for  each  

p,P,y( ) ∈Ο  there  exists  a  

λ ∈ℜ  such  that  the  utility  possibility  frontier  defined  by  the  budget  constraint  (1)  is  the  hyperplane  defined  by  

um = λm∈S∑ .  Because  

λ ∈ℜ  depends  on  

p,P,y( )  and  on  the  utility  profile  

u ,  we  denote  it  by  

λ p,P,y,u( ) .        

3.  Results              

The  discrepancy  between  Bergstrom's  and  Chiappori's  interpretation  of  what  is  required  for  TU  seems  to  stem  from  a  fundamental  disagreement  of  what  is  required  for  TU  in  exchange  economies  with  private  goods  and  convex  preferences.  Chiappori  (2010,  p.  1305)  notes,  "When  all  commodities  are  privately  consumed,  then  TU  is  known  to  be  equivalent  to  quasi-­‐linear  utilities,  which  requires  zero  income  elasticity  for  all  demands  but  one."  Contrary  to  this  claim,  Bergstrom  and  Varian  (1985)  established  that  in  an  exchange  economy  with  private  goods  only,  the  indirect  utility  functions  of  any  individual  

m ∈S  must  be  of  the  Gorman  polar  form  for  there  to  be  TU.2        

Proposition  1  Bergstrom  and  Varian  (1985).  With  

L ≥ 2  and  

N = 0 ,  there  is  TU  if  and  only  if  

vm p,Im( ) = α(p)Im + βm (p)       (2)  

where  

α(p)  and  

βm (p)  are  functions  of  the  price  vector  only.      

Proof  of  Proposition  1  To  see  these  indirect  utility  functions  yield  TU,  note  that  without  public  goods,  the  UPF  gives  us  all  the  combinations  of  agents'  utilities  that  

                                                                                                               2  Bergstrom  and  Varian  (1985)  prove  their  result  while  Chiappori's  statement  is  based  on  his  reading  of  the  literature.  

  4  

arise  from  different  distributions  of  

I .  It  follows  that  if  individuals'  indirect  utility  functions  are  of  the  Gorman  form,  then  the  sums  resulting  from  different  distributions  of  

I  are  the  same      

vm p,Im( ) = α(p)I +m∈S∑ βm (p)m∈S∑  

A  change  in  the  distribution  of  

I  has  therefore  no  impact  on  the  sum  of  indirect  utilities  and  hence  the  UPF  for  each  

(p,y)∈Ο  is  given  by      

um = α(p)I +m∈S∑ βm (p)m∈S∑  

Transferable  utility  holds.    QED.            

Bergstrom  (1999,  p.19),  interpreting  the  Bergstrom-­‐Varian  result,  points  out,  "The  Gorman  class  includes  preferences  represented  by  a  quasi-­‐linear  direct  utility  function.  [...]  But  the  Gorman  class  also  includes  identical  homothetic  preferences  and  more  generally,  preferences  which,  like  the  Stone-­‐Geary  utility  function,  are  represented  by  utility  functions  of  the  form  

[U(xm − em )] ,  where  U  is  a  homogenous  function  and  

[em ]  is  some  vector  that  "displaces  the  origin"  and  which  may  be  different  for  different  people.  Thus  it  is  possible  to  have  transferable  utility,  but  to  have  preferences  differing  among  individuals  and  to  have  income-­‐responsive  demand  for  all  goods."3        

Example  1  Consider  two  people  

m = 1,2  with  their  utility  functions  given  by    

um xm( ) = xm1( )a xm2( )1−a + xm

3( )bm  where  

0 < a,b1,b2 < 1.  The  indirect  utility  functions  are  then  of  the  form  given  by  (2)  with4  

α(p1, p2) =ap1

⎛ ⎝ ⎜

⎞ ⎠ ⎟ a 1− a

p2⎛ ⎝ ⎜

⎞ ⎠ ⎟ 1−a

β(p1, p2, p3) = p3α(p)( )bmbm −1

1bm

−1⎛ ⎝ ⎜

⎞ ⎠ ⎟ 1bm

⎛ ⎝ ⎜

⎞ ⎠ ⎟

1bm −1

 

From  the  indirect  utility  function  above,  it  is  clear  that  goods  1  and  2  have  an  income  effect  while  good  3  has  not.5    This  means  that  this  example  already  shows  

                                                                                                               3  Notation  changed  for  consistency  with  remainder  of  the  article.  4  It  is  clear  that  corner  solutions  are  possible.    It  should  be  noted  that  corner  solutions  also  occur  with  quasi-­‐linear  utility  functions  alone  and  hence  moving  to  other  forms  of  utility  functions  does  not  restrict  the  range  in  which  TU  holds  any  more  than  by  choosing  a  particular  quasi-­‐linear  specification.  These  non-­‐negativity  constraints  are  largely  ignored  in  the  literature.  As  this  article  is  mostly  a  comment  on  the  existing  literature  I  do  not  focus  on  corner  solutions  either.    

  5  

that  it  is  possible  to  generate  TU  and  observe  demands  for  more  than  one  good  that  change  with  income.            

Incorporating  public  goods  into  the  model  leads  to  specific  requirements  of  how  they  need  to  enter  utility  functions  in  order  to  preserve  TU.    For  example,  if  all  agents  have  identical  and  linearly  homogeneous  CES  utility  functions  in  the  case  of  all  private  goods,  we  have  TU,  but  making  one  of  the  goods  a  public  good  does  no  longer  lead  to  TU.        

Example  2    Assume  there  are  two  individuals  and  they  have  utility  functions  of  the  form  

um X,xm1 ,xm

2( ) = X a + xm1( )a + xm

2( )a( )1a  

where  

0 < a < 1.  

Given  

X ,  the  indirect  utility  function  for  each  individual  is  given  by  

vm p,Im ,X( ) = X a + α(p)Ima( )1a  

Given  

X ,  we  would  need  to  distribute  income  

I  in  different  ways  to  pick  up  all  the  different  distributions  on  the  UPF.  But  note,  the  UPF  will  no  longer  be  of  TU  form,  because  the  sum  of  indirect  utilities  changes  with  different  distributions  of  

I .  Letting  

X  vary,  could  we  still  get  TU?  The  answer  is  no.  To  examine  the  properties  of  the  UPF,  we  can  use  the  indirect  utility  functions  derived  above.  Let  

µ, φ  be  Lagrange  multipliers,  we  need        

maxX ,I1 ,µ,φ X a + α(p)Ima( )

1a + µ X a + α(p)(I − Im )

a( )1a − u2

⎛ ⎝ ⎜

⎞ ⎠ ⎟ + φ(y − PX − I)  

This  leads  to  the  Samuelson  Condition  of  efficient  provision  of  public  goods    

X a−1

α(p)I1a−1 +

X a−1

α(p)(y − PX − I1 )a−1 = P  

The  Samuelson  Condition  tells  us  that  the  efficient  amount  of  the  public  good  is  not  independent  of  the  distribution  of  disposable  income.  Bergstrom  and  Cornes  (1983)  proof  that  the  efficient  amount  of  the  public  good  must  be  independent  of  the  distribution  of  disposable  income  for  TU  to  hold  and  hence  the  utility  profile  in  Example  2  does  not  lead  to  TU.              

                                                                                                               

5  By  Roy's  identity,  the  demand  of  a  good  

xml  is  given  by  

xml p,Im( ) = −

∂vm /∂pl

∂vm /∂Im.  Since  the  

demands  for  goods  

l = 1,2  depend  on  

Im ,  they  have  an  income  effect.  

  6  

Bergstrom  and  Cornes  (1983)  show  that  in  order  to  lead  to  TU,  utility  functions  with  one  private  good  and  many  public  goods  must  be  of  the  type  

   

um xm,X( ) = b(X)xm1 + Am (X)         (3)  

Chiappori  considers  many  private  goods  in  addition  to  many  public  goods  and  makes  the  following  generalization  to  (3).      

Definition  2    A  utility  function  is  of  the  Generalized  Quasi-­‐linear  form  if  the  function  can  be  written  as        

um xm,X( ) = b(X)xm1 + Am (xm

2 ,...,xmL ,X)  

Chiappori  (2010)  then  shows  that  if  

bm (X)  is  the  same  for  all  agents,  TU  holds.  6  

Taking  a  different  approach,  Bergstrom  (1999)  focuses  again  as  in  Berstrom  and  Varian  (1985)  on  indirect  utility  functions  and  notes  that  agents'  indirect  utility  functions  must  be  of  the  form        

vm X, p,Im( ) = α(p,X)Im + βm (p,X)       (4)  

for  TU  to  be  satisfied  in  the  case  of  many  private  and  many  public  goods.7  That  these  indirect  utility  functions  lead  to  TU  can  be  easily  seen  by  considering  the  Samuelson  Condition  again.  Given  (4),  the  condition  for  the  efficient  amount  of  any  public  good  

X n  is  given  by        

∂α∂Xn Im +

∂βm

∂Xn

α(p,X)m∈S∑ = Pn   or  equivalently    

∂α∂Xn I +

∂βm

∂Xnm∈S∑α(p,X)

= Pn  

This  means  that  the  efficient  amount  of  any  public  good  is  independent  of  the  distribution  of  

I .  It  implies  that  in  order  to  solve  for  the  UPF,  one  can  solve  the  following  problem        

                                                                                                               6  Chiappori  makes  the  following  statements  on  page  1305:  "Necessary  and  sufficient  conditions  for  transferability  have  been  known  for  some  time.  [...]  In  a  seminal  contribution,  [Bergstrom  and  Cornes  (1983)  and  Bergstrom  (1989)]  have  shown  that  transferable  utility  (TU)  requires  three  properties:    

1.  individual  utilities  are  of  the  generalized  quasi-­‐linear  form  with  respect  to  the  same  commodity,  say  1;    2.  the  

bm (X)  functions  are  identical  across  agents:  

b1(X) = ...= bS (X) = b(X) ;    3.  the  allocation  of  resources  between  members  is  such  that  each  member  has  a  positive  consumption  of  commodity  1."    

Note  that  Bergstrom  himself  does  not  claim  that  the  case  of  many  private  goods  and  many  public  goods  is  considered  in  Bergstrom  (1989)  or  Bergstrom  and  Cornes  (1983)  as  Chiappori  seems  to  state  in  above  quote.    7  Bergstrom  (1999)  does  not  formally  prove  this  statement  or  provides  a  reference  to  a  formal  proof  but  claims  that  the  result  is  obvious  by  combining  the  Bergstrom-­‐Varian  (many  private  goods)  and  the  Bergstrom-­‐Cornes  (many  public  goods,  one  private  good)  results.  

  7  

λ(p,P,y,u) = maxX ,I ,φ α(p,X)I + βm (p,X)m∈S∑ + φ(y − PX − I)  

and  the  UPF  is  then  given  by  

um = λm∈S∑ (p,P,y,u)  

We  now  ask  the  question  which  utility  functions  would  lead  to  such  indirect  utility  functions.  It  is  straightforward  to  verify  that  utility  functions  of  the  GQL  form  with  identical  

bm (X)  functions  lead  to  indirect  utility  functions  given  by  (4).    However,  GQL  is  not  necessary  for  TU  in  case  of  many  private  and  many  public  goods  just  as  quasi-­‐linear  utility  functions  are  not  necessary  in  the  case  of  all  private  goods.  Below  I  present  a  much  larger  class  of  utility  functions  that  lead  to  TU.        

Proposition  2  Partition  the  set  of  private  goods  into  

L1  and  

L2  such  that  

∅ ≠ L1 ⊆ L  and  

L2 ⊂ L .  Then  any  utility  functions  of  the  form  

   

um xm,X( ) = A(X)h xmk{ }k∈L1

− em( ) + Bm X, xml{ }l∈L2( )     (5)  

where  

h xmk{ }k∈L1( )  is  homogenous  of  degree  1  and  

em  is  some  vector  of  fixed  

quantities,  will  lead  to  TU.      

Proof  of  Proposition  2    To  proof  Proposition  2,  note  first  as  in  the  Example  1,  that  the  demand  for  any  private  good  

l ∈L2  depends  on  the  whole  price  vector  but  not  on  disposable  income.    Once  these  demands  are  found  -­‐-­‐  we  denote  them  by  

xml (p,X)  -­‐-­‐  

we  can  write  the  utility  maximization  problem  as            

max{xmk }k∈L1 A(X)h {xmk }k∈L1 − em( )

s. t. {pk}k∈L1 {xmk }k∈L1 = Im − {p

l}l∈L2 {xml }l∈L2

 

Since  this  problem  is  just  like  solving  for  an  indirect  utility  function  with  a  utility  function  that  is  of  the  Stone-­‐Geary  form,  the  indirect  utility  function  for  this  problem  -­‐-­‐  denoted  by  

ϖm X, p,Im( )  -­‐-­‐  is  given  by  

ϖm X, p,Im( ) = γ X,{pk}k∈L1( ) Im − {pl}l∈L2 {xml (X, p)}l∈L2( )  Assembling  all  the  partial  results,  we  end  up  with  an  indirect  utility  function  of  (5)  given  by      

vm X, p,Im( ) = γ X,{pk}k∈L1( )Im − γ X,{pk}k∈L1( ){pl}l∈L2 {xml (X, p)}l∈L2 + Bm X,{xml (X, p)}l∈L2( )

 

This  indirect  utility  function  is  of  the  desired  form  (4)  with  

  8  

α X, p( ) = γ X,{pk}k∈L1( )   and                    

βm X, p( ) = Bm X,{xml (X, p)}l∈L2( ) − γ X,{pk}k∈L1( ){pl}l∈L2 {xml (X, p)}l∈L2  

I  conclude  this  section  by  pointing  out  special  cases  of  (5).  Chiappori's  GQL  emerges  as  the  special  case  with  L₁=1.  Another  special  case  is  the  Generalized  Cobb-­‐Douglas  (GCD)  utility  function  given  by  

um xm,X( ) = A(X) xm1( )γ 1 × ...× xm

L( )γ L + Bm (X)  

where  

γ k = 1k∈L∑ .  Yet,  another  example  is  a  hybrid  of  the  GCD  and  GQL  forms  given  

by      

um xm,X( ) = A(X) xm1( )γ 1 × ...× xm

L1( )γ L1 + Bm (X,{xml }l∈L2 )  

where    

γ k = 1k∈L1∑ .      

 

4.  Conclusion              

In  many  theoretical  models  in  which  TU  is  invoked,  there  is  only  one  private  good  and  one  public  good,  so  the  Bergstrom-­‐Cornes  result  applies  which  can  be  considered  as  a  special  case  of  the  GQL  form  with  

L = 1  and  hence  the  discussion  in  this  article  is  irrelevant.  However,  if  the  focus  is  on  empirical  implications  of  TU  acknowledging  that  families  or  other  groups  of  people  consume  many  private  and  public  goods,  then  focusing  on  GQL  is  too  narrow.  As  I  have  shown  here,  utility  functions  outside  the  class  of  GQL  also  satisfy  TU.  While  any  of  the  examples  given  make  strong  assumptions  about  people's  preferences,  they  do  admit  for  more  than  one  private  good  having  a  positive  income  effect.  Developing  tests  for  the  larger  class  of  utility  functions  is  left  to  future  research.      

 

References  

 Bergstrom,  Th.  C.,  1989,  A  Fresh  Look  at  the  Rotten  Kid  Theorem  -­‐-­‐  and  Other  Household  Mysteries,  Journal  of  Political  Economy  97,  1138-­‐59.            

Bergstrom,  Th.  C.,  1999,  A  Survey  of  Theories  of  the  Family  in  Handbook  of  Population  and  Family  Economics.  Edited  by  M.R.  Rosenzweig  and  O.  Stark,  volume  1A,  22-­‐79.            

Bergstrom,  Th.  C.  and  R.  C.  Cornes,  1983,  Independence  of  Allocative  Efficiency  from  Distribution  in  the  Theory  of  Public  Goods,  Econometrica  51,  1753-­‐66.            

  9  

Bergstrom,  Th.  C.  and  H.  R.  Varian,  1985,  When  do  market  games  have  transferable  utility?  Journal  of  Economic  Theory  35,  222-­‐33.            

Chiappori,  PA.,  2010,  Testable  implications  of  transferable  utility,  Journal  of  Economic  Theory  145,  1302-­‐17.