Information Sharing between Vertical Hierarchies · ) Interaction between information exchange...
Transcript of Information Sharing between Vertical Hierarchies · ) Interaction between information exchange...
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Information Sharingbetween Vertical Hierarchies
Marco Pagnozzi Salvatore Piccolo
September 2012
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Introduction
Why do competitors share private information?
The essence of an organization is based on the trade-o¤between the costs and bene�ts of communication amongits members (Arrow)
Information sharing agreements are widespread:
Banks exchange information about borrowersSellers share information about consumers�demandFirms disclose information about management�s performance
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Introduction
Why do competitors share private information?
The essence of an organization is based on the trade-o¤between the costs and bene�ts of communication amongits members (Arrow)
Information sharing agreements are widespread:
Banks exchange information about borrowersSellers share information about consumers�demandFirms disclose information about management�s performance
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Introduction
Why do competitors share private information?
The essence of an organization is based on the trade-o¤between the costs and bene�ts of communication amongits members (Arrow)
Information sharing agreements are widespread:
Banks exchange information about borrowers
Sellers share information about consumers�demandFirms disclose information about management�s performance
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Introduction
Why do competitors share private information?
The essence of an organization is based on the trade-o¤between the costs and bene�ts of communication amongits members (Arrow)
Information sharing agreements are widespread:
Banks exchange information about borrowersSellers share information about consumers�demand
Firms disclose information about management�s performance
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Introduction
Why do competitors share private information?
The essence of an organization is based on the trade-o¤between the costs and bene�ts of communication amongits members (Arrow)
Information sharing agreements are widespread:
Banks exchange information about borrowersSellers share information about consumers�demandFirms disclose information about management�s performance
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Literature
IO: information sharing can increase e¢ ciency or reducecompetition � e.g., Novshek and Sonnenschein (1982),Clarke (1983), Vives (1984) ...
Banking: lenders exchange information to screen investmentprojects or discipline borrowers � e.g., Pagano and Jappelli(1993) ...
Consumers�privacy: sellers use information on consumers toprice discriminate � Acquisti and Varian (2005), Taylor(2004)
... But these papers neglect the source of information
Ganuza and Jansen (2012) show how information sharinga¤ects �rms�incentive to acquire signals about their costs
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Literature
IO: information sharing can increase e¢ ciency or reducecompetition � e.g., Novshek and Sonnenschein (1982),Clarke (1983), Vives (1984) ...
Banking: lenders exchange information to screen investmentprojects or discipline borrowers � e.g., Pagano and Jappelli(1993) ...
Consumers�privacy: sellers use information on consumers toprice discriminate � Acquisti and Varian (2005), Taylor(2004)
... But these papers neglect the source of information
Ganuza and Jansen (2012) show how information sharinga¤ects �rms�incentive to acquire signals about their costs
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Literature
IO: information sharing can increase e¢ ciency or reducecompetition � e.g., Novshek and Sonnenschein (1982),Clarke (1983), Vives (1984) ...
Banking: lenders exchange information to screen investmentprojects or discipline borrowers � e.g., Pagano and Jappelli(1993) ...
Consumers�privacy: sellers use information on consumers toprice discriminate � Acquisti and Varian (2005), Taylor(2004)
... But these papers neglect the source of information
Ganuza and Jansen (2012) show how information sharinga¤ects �rms�incentive to acquire signals about their costs
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Literature
IO: information sharing can increase e¢ ciency or reducecompetition � e.g., Novshek and Sonnenschein (1982),Clarke (1983), Vives (1984) ...
Banking: lenders exchange information to screen investmentprojects or discipline borrowers � e.g., Pagano and Jappelli(1993) ...
Consumers�privacy: sellers use information on consumers toprice discriminate � Acquisti and Varian (2005), Taylor(2004)
... But these papers neglect the source of information
Ganuza and Jansen (2012) show how information sharinga¤ects �rms�incentive to acquire signals about their costs
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Literature
IO: information sharing can increase e¢ ciency or reducecompetition � e.g., Novshek and Sonnenschein (1982),Clarke (1983), Vives (1984) ...
Banking: lenders exchange information to screen investmentprojects or discipline borrowers � e.g., Pagano and Jappelli(1993) ...
Consumers�privacy: sellers use information on consumers toprice discriminate � Acquisti and Varian (2005), Taylor(2004)
... But these papers neglect the source of information
Ganuza and Jansen (2012) show how information sharinga¤ects �rms�incentive to acquire signals about their costs
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Contribution
Principals obtain information through contracting withexclusive and privately informed agents
Principals compete and may share information
) Interaction between information exchange acrossorganizations and agency con�icts within organizations
e.g.: Two competing manufacturers that sell through privatelyinformed retailers and may share the information obtainedfrom them
Link between information sharing and vertical contracting:Calzolari and Pavan (2006) analyze information transmissionin sequential common agency
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Contribution
Principals obtain information through contracting withexclusive and privately informed agents
Principals compete and may share information
) Interaction between information exchange acrossorganizations and agency con�icts within organizations
e.g.: Two competing manufacturers that sell through privatelyinformed retailers and may share the information obtainedfrom them
Link between information sharing and vertical contracting:Calzolari and Pavan (2006) analyze information transmissionin sequential common agency
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Contribution
Principals obtain information through contracting withexclusive and privately informed agents
Principals compete and may share information
) Interaction between information exchange acrossorganizations and agency con�icts within organizations
e.g.: Two competing manufacturers that sell through privatelyinformed retailers and may share the information obtainedfrom them
Link between information sharing and vertical contracting:Calzolari and Pavan (2006) analyze information transmissionin sequential common agency
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Contribution
Principals obtain information through contracting withexclusive and privately informed agents
Principals compete and may share information
) Interaction between information exchange acrossorganizations and agency con�icts within organizations
e.g.: Two competing manufacturers that sell through privatelyinformed retailers and may share the information obtainedfrom them
Link between information sharing and vertical contracting:Calzolari and Pavan (2006) analyze information transmissionin sequential common agency
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Contribution
Principals obtain information through contracting withexclusive and privately informed agents
Principals compete and may share information
) Interaction between information exchange acrossorganizations and agency con�icts within organizations
e.g.: Two competing manufacturers that sell through privatelyinformed retailers and may share the information obtainedfrom them
Link between information sharing and vertical contracting:Calzolari and Pavan (2006) analyze information transmissionin sequential common agency
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Result
Information sharing a¤ects principals�incentiveto distort agents�output to extract rents
The choice to share information only depends on:
the nature of competition between principals andthe correlation of agents�information
) Principals share information i¤ externalities and correlationhave the same sign
This e¤ect is of �rst-order relative to thosewith complete information
Principals face a prisoner�s dilemmawhen they do not share information
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Result
Information sharing a¤ects principals�incentiveto distort agents�output to extract rents
The choice to share information only depends on:
the nature of competition between principals andthe correlation of agents�information
) Principals share information i¤ externalities and correlationhave the same sign
This e¤ect is of �rst-order relative to thosewith complete information
Principals face a prisoner�s dilemmawhen they do not share information
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Result
Information sharing a¤ects principals�incentiveto distort agents�output to extract rents
The choice to share information only depends on:
the nature of competition between principals and
the correlation of agents�information
) Principals share information i¤ externalities and correlationhave the same sign
This e¤ect is of �rst-order relative to thosewith complete information
Principals face a prisoner�s dilemmawhen they do not share information
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Result
Information sharing a¤ects principals�incentiveto distort agents�output to extract rents
The choice to share information only depends on:
the nature of competition between principals andthe correlation of agents�information
) Principals share information i¤ externalities and correlationhave the same sign
This e¤ect is of �rst-order relative to thosewith complete information
Principals face a prisoner�s dilemmawhen they do not share information
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Result
Information sharing a¤ects principals�incentiveto distort agents�output to extract rents
The choice to share information only depends on:
the nature of competition between principals andthe correlation of agents�information
) Principals share information i¤ externalities and correlationhave the same sign
This e¤ect is of �rst-order relative to thosewith complete information
Principals face a prisoner�s dilemmawhen they do not share information
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Result
Information sharing a¤ects principals�incentiveto distort agents�output to extract rents
The choice to share information only depends on:
the nature of competition between principals andthe correlation of agents�information
) Principals share information i¤ externalities and correlationhave the same sign
This e¤ect is of �rst-order relative to thosewith complete information
Principals face a prisoner�s dilemmawhen they do not share information
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Result
Information sharing a¤ects principals�incentiveto distort agents�output to extract rents
The choice to share information only depends on:
the nature of competition between principals andthe correlation of agents�information
) Principals share information i¤ externalities and correlationhave the same sign
This e¤ect is of �rst-order relative to thosewith complete information
Principals face a prisoner�s dilemmawhen they do not share information
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Model
Two principals: P1 and P2
Two exclusive agents: A1 and A2Ai is privately informed about marginal cost θi 2
�θ, θ, with:
Pr(θ, θ) = ν2 + α; Pr(θ, θ) = (1� ν)2 + αPr(θ, θ) = Pr(θ, θ) = ν (1� ν)� α
) α measures correlation between θ1 and θ2;Pr(θ) = ν; Pr(θ) = 1� ν;
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Model
Two principals: P1 and P2Two exclusive agents: A1 and A2
Ai is privately informed about marginal cost θi 2�
θ, θ, with:
Pr(θ, θ) = ν2 + α; Pr(θ, θ) = (1� ν)2 + αPr(θ, θ) = Pr(θ, θ) = ν (1� ν)� α
) α measures correlation between θ1 and θ2;Pr(θ) = ν; Pr(θ) = 1� ν;
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Model
Two principals: P1 and P2Two exclusive agents: A1 and A2Ai is privately informed about marginal cost θi 2
�θ, θ, with:
Pr(θ, θ) = ν2 + α; Pr(θ, θ) = (1� ν)2 + αPr(θ, θ) = Pr(θ, θ) = ν (1� ν)� α
) α measures correlation between θ1 and θ2;Pr(θ) = ν; Pr(θ) = 1� ν;
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Model
Two principals: P1 and P2Two exclusive agents: A1 and A2Ai is privately informed about marginal cost θi 2
�θ, θ, with:
Pr(θ, θ) = ν2 + α; Pr(θ, θ) = (1� ν)2 + α
Pr(θ, θ) = Pr(θ, θ) = ν (1� ν)� α
) α measures correlation between θ1 and θ2;Pr(θ) = ν; Pr(θ) = 1� ν;
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Model
Two principals: P1 and P2Two exclusive agents: A1 and A2Ai is privately informed about marginal cost θi 2
�θ, θ, with:
Pr(θ, θ) = ν2 + α; Pr(θ, θ) = (1� ν)2 + αPr(θ, θ) = Pr(θ, θ) = ν (1� ν)� α
) α measures correlation between θ1 and θ2;Pr(θ) = ν; Pr(θ) = 1� ν;
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Model
Two principals: P1 and P2Two exclusive agents: A1 and A2Ai is privately informed about marginal cost θi 2
�θ, θ, with:
Pr(θ, θ) = ν2 + α; Pr(θ, θ) = (1� ν)2 + αPr(θ, θ) = Pr(θ, θ) = ν (1� ν)� α
) α measures correlation between θ1 and θ2;Pr(θ) = ν; Pr(θ) = 1� ν;
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Utilities
Pi pays ti to Ai , and Ai produces qi
Risk-neutral players:
Ai : Ui = ti � θiqi
Pi : S i (qi , qj )| {z }�κ+βqi�q2i +δqiqj
� ti
) δ measures production externalities:
strategic complementarity (δ > 0) or substitutability (δ < 0)
We assume δ small and compute expected pro�ts throughTaylor expansions
Limited liability for agents
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Utilities
Pi pays ti to Ai , and Ai produces qiRisk-neutral players:
Ai : Ui = ti � θiqi
Pi : S i (qi , qj )| {z }�κ+βqi�q2i +δqiqj
� ti
) δ measures production externalities:
strategic complementarity (δ > 0) or substitutability (δ < 0)
We assume δ small and compute expected pro�ts throughTaylor expansions
Limited liability for agents
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Utilities
Pi pays ti to Ai , and Ai produces qiRisk-neutral players:
Ai : Ui = ti � θiqi
Pi : S i (qi , qj )| {z }�κ+βqi�q2i +δqiqj
� ti
) δ measures production externalities:
strategic complementarity (δ > 0) or substitutability (δ < 0)
We assume δ small and compute expected pro�ts throughTaylor expansions
Limited liability for agents
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Utilities
Pi pays ti to Ai , and Ai produces qiRisk-neutral players:
Ai : Ui = ti � θiqi
Pi : S i (qi , qj )| {z }�κ+βqi�q2i +δqiqj
� ti
) δ measures production externalities:
strategic complementarity (δ > 0) or substitutability (δ < 0)
We assume δ small and compute expected pro�ts throughTaylor expansions
Limited liability for agents
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Utilities
Pi pays ti to Ai , and Ai produces qiRisk-neutral players:
Ai : Ui = ti � θiqi
Pi : S i (qi , qj )| {z }�κ+βqi�q2i +δqiqj
� ti
) δ measures production externalities:
strategic complementarity (δ > 0) or substitutability (δ < 0)
We assume δ small and compute expected pro�ts throughTaylor expansions
Limited liability for agents
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Utilities
Pi pays ti to Ai , and Ai produces qiRisk-neutral players:
Ai : Ui = ti � θiqi
Pi : S i (qi , qj )| {z }�κ+βqi�q2i +δqiqj
� ti
) δ measures production externalities:
strategic complementarity (δ > 0) or substitutability (δ < 0)
We assume δ small and compute expected pro�ts throughTaylor expansions
Limited liability for agents
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Contracts
θi can be learned by Pi through private contractingand by Pj through information sharing
Pi o¤ers a direct mechanism: Ai report his cost θi and
without information sharing:
fti (θi ) , qi (θi )g
with information sharing:�ti�θi , θj
�, qi�θi , θj
�
Ai�s cost can be credibly transmitted by Pi to Pj/Aj
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Contracts
θi can be learned by Pi through private contractingand by Pj through information sharing
Pi o¤ers a direct mechanism: Ai report his cost θi and
without information sharing:
fti (θi ) , qi (θi )g
with information sharing:�ti�θi , θj
�, qi�θi , θj
�
Ai�s cost can be credibly transmitted by Pi to Pj/Aj
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Contracts
θi can be learned by Pi through private contractingand by Pj through information sharing
Pi o¤ers a direct mechanism: Ai report his cost θi and
without information sharing:
fti (θi ) , qi (θi )g
with information sharing:�ti�θi , θj
�, qi�θi , θj
�
Ai�s cost can be credibly transmitted by Pi to Pj/Aj
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Contracts
θi can be learned by Pi through private contractingand by Pj through information sharing
Pi o¤ers a direct mechanism: Ai report his cost θi and
without information sharing:
fti (θi ) , qi (θi )g
with information sharing:�ti�θi , θj
�, qi�θi , θj
�
Ai�s cost can be credibly transmitted by Pi to Pj/Aj
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Contracts
θi can be learned by Pi through private contractingand by Pj through information sharing
Pi o¤ers a direct mechanism: Ai report his cost θi and
without information sharing:
fti (θi ) , qi (θi )g
with information sharing:�ti�θi , θj
�, qi�θi , θj
�
Ai�s cost can be credibly transmitted by Pi to Pj/Aj
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Timing
1 Principals simultaneously choose whether to commit to shareinformation
2 Ai learns θi3 Principals contract with agents4 Pi discloses her information about Ai�s cost if she hascommitted to do so
5 Agents produce and payments are made
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Timing
1 Principals simultaneously choose whether to commit to shareinformation
2 Ai learns θi
3 Principals contract with agents4 Pi discloses her information about Ai�s cost if she hascommitted to do so
5 Agents produce and payments are made
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Timing
1 Principals simultaneously choose whether to commit to shareinformation
2 Ai learns θi3 Principals contract with agents
4 Pi discloses her information about Ai�s cost if she hascommitted to do so
5 Agents produce and payments are made
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Timing
1 Principals simultaneously choose whether to commit to shareinformation
2 Ai learns θi3 Principals contract with agents4 Pi discloses her information about Ai�s cost if she hascommitted to do so
5 Agents produce and payments are made
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Timing
1 Principals simultaneously choose whether to commit to shareinformation
2 Ai learns θi3 Principals contract with agents4 Pi discloses her information about Ai�s cost if she hascommitted to do so
5 Agents produce and payments are made
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Complete information within each hierarchy
Standard duopoly where �rms share cost information(Shapiro, 1986)
Let si = (θi , θj ) if Pj shares information, si = θi if Pj does not
Lemma: Pi�s expected equilibrium pro�t is
V �i = κ+(Esi [q�i (si ) jθi ]| {z }
average of q�i (si )
)2+Esi [q�i (si )�Esi [q
�i (si ) jθi )jθi ]]
2| {z }variance of q�i (si )
.
and expected output is the same regardless of principals�communication decisions
) Principals maximize output volatility
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Complete information within each hierarchy
Standard duopoly where �rms share cost information(Shapiro, 1986)
Let si = (θi , θj ) if Pj shares information, si = θi if Pj does not
Lemma: Pi�s expected equilibrium pro�t is
V �i = κ+(Esi [q�i (si ) jθi ]| {z }
average of q�i (si )
)2+Esi [q�i (si )�Esi [q
�i (si ) jθi )jθi ]]
2| {z }variance of q�i (si )
.
and expected output is the same regardless of principals�communication decisions
) Principals maximize output volatility
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Complete information within each hierarchy
Standard duopoly where �rms share cost information(Shapiro, 1986)
Let si = (θi , θj ) if Pj shares information, si = θi if Pj does not
Lemma: Pi�s expected equilibrium pro�t is
V �i = κ+(Esi [q�i (si ) jθi ]| {z }
average of q�i (si )
)2+Esi [q�i (si )�Esi [q
�i (si ) jθi )jθi ]]
2| {z }variance of q�i (si )
.
and expected output is the same regardless of principals�communication decisions
) Principals maximize output volatility
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Complete information within each hierarchy
Standard duopoly where �rms share cost information(Shapiro, 1986)
Let si = (θi , θj ) if Pj shares information, si = θi if Pj does not
Lemma: Pi�s expected equilibrium pro�t is
V �i = κ+(Esi [q�i (si ) jθi ]| {z }
average of q�i (si )
)2+Esi [q�i (si )�Esi [q
�i (si ) jθi )jθi ]]
2| {z }variance of q�i (si )
.
and expected output is the same regardless of principals�communication decisions
) Principals maximize output volatility
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Complete information
Proposition: With complete information: both principalsshare information if α > �ν (1� ν); no principal shareinformation if α < �ν (1� ν)
Information sharing has a direct positive e¤ect on outputvolatility (since contracts are conditioned on more states) butIf α < 0 information sharing reduces volatility because states(θ, θ) and (θ, θ) are more likely:
e.g., if δ > 0 outputs are more similar in those states
Proposition: Principals�information sharing decisions alwaysmaximize their pro�ts
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Complete information
Proposition: With complete information: both principalsshare information if α > �ν (1� ν); no principal shareinformation if α < �ν (1� ν)
Information sharing has a direct positive e¤ect on outputvolatility (since contracts are conditioned on more states) but
If α < 0 information sharing reduces volatility because states(θ, θ) and (θ, θ) are more likely:
e.g., if δ > 0 outputs are more similar in those states
Proposition: Principals�information sharing decisions alwaysmaximize their pro�ts
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Complete information
Proposition: With complete information: both principalsshare information if α > �ν (1� ν); no principal shareinformation if α < �ν (1� ν)
Information sharing has a direct positive e¤ect on outputvolatility (since contracts are conditioned on more states) butIf α < 0 information sharing reduces volatility because states(θ, θ) and (θ, θ) are more likely:
e.g., if δ > 0 outputs are more similar in those states
Proposition: Principals�information sharing decisions alwaysmaximize their pro�ts
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Complete information
Proposition: With complete information: both principalsshare information if α > �ν (1� ν); no principal shareinformation if α < �ν (1� ν)
Information sharing has a direct positive e¤ect on outputvolatility (since contracts are conditioned on more states) butIf α < 0 information sharing reduces volatility because states(θ, θ) and (θ, θ) are more likely:
e.g., if δ > 0 outputs are more similar in those states
Proposition: Principals�information sharing decisions alwaysmaximize their pro�ts
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Complete information
Proposition: With complete information: both principalsshare information if α > �ν (1� ν); no principal shareinformation if α < �ν (1� ν)
Information sharing has a direct positive e¤ect on outputvolatility (since contracts are conditioned on more states) butIf α < 0 information sharing reduces volatility because states(θ, θ) and (θ, θ) are more likely:
e.g., if δ > 0 outputs are more similar in those states
Proposition: Principals�information sharing decisions alwaysmaximize their pro�ts
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Asymmetric information
Since principals learn their agents�costs through contracting,agents earn an information rent
Principals want to distort outputs to minimize rent
Principals want to a¤ect rival�s output to increase pro�t(because of externality)
3 subgames:
1 No communication2 Bilateral information sharing3 Unilateral information sharing
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Asymmetric information
Since principals learn their agents�costs through contracting,agents earn an information rent
Principals want to distort outputs to minimize rent
Principals want to a¤ect rival�s output to increase pro�t(because of externality)
3 subgames:
1 No communication2 Bilateral information sharing3 Unilateral information sharing
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Asymmetric information
Since principals learn their agents�costs through contracting,agents earn an information rent
Principals want to distort outputs to minimize rent
Principals want to a¤ect rival�s output to increase pro�t(because of externality)
3 subgames:
1 No communication2 Bilateral information sharing3 Unilateral information sharing
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Asymmetric information
Since principals learn their agents�costs through contracting,agents earn an information rent
Principals want to distort outputs to minimize rent
Principals want to a¤ect rival�s output to increase pro�t(because of externality)
3 subgames:
1 No communication2 Bilateral information sharing3 Unilateral information sharing
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Asymmetric information
Since principals learn their agents�costs through contracting,agents earn an information rent
Principals want to distort outputs to minimize rent
Principals want to a¤ect rival�s output to increase pro�t(because of externality)
3 subgames:1 No communication
2 Bilateral information sharing3 Unilateral information sharing
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Asymmetric information
Since principals learn their agents�costs through contracting,agents earn an information rent
Principals want to distort outputs to minimize rent
Principals want to a¤ect rival�s output to increase pro�t(because of externality)
3 subgames:1 No communication2 Bilateral information sharing
3 Unilateral information sharing
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Asymmetric information
Since principals learn their agents�costs through contracting,agents earn an information rent
Principals want to distort outputs to minimize rent
Principals want to a¤ect rival�s output to increase pro�t(because of externality)
3 subgames:1 No communication2 Bilateral information sharing3 Unilateral information sharing
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
No communication
Pi maximizes
EθjEθi [S (qi (θi ), qe (θj ))� θiqi (θi ) jθi ]� ν∆θqi (θ)
First-order conditions are
Eθ [S1 (qe (θ), qe (θj )) jθ] = θ,
Eθ
�S1(qe (θ), qe (θj ))jθ
�= θ + ν
1�ν ∆θ.
) Type θ produces the output that equalizes marginal bene�tto marginal costType θ�s output is downward distorted to reduce agents�rent
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
No communication
Pi maximizes
EθjEθi [S (qi (θi ), qe (θj ))� θiqi (θi ) jθi ]� ν∆θqi (θ)
First-order conditions are
Eθ [S1 (qe (θ), qe (θj )) jθ] = θ,
Eθ
�S1(qe (θ), qe (θj ))jθ
�= θ + ν
1�ν ∆θ.
) Type θ produces the output that equalizes marginal bene�tto marginal costType θ�s output is downward distorted to reduce agents�rent
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
No communication
Pi maximizes
EθjEθi [S (qi (θi ), qe (θj ))� θiqi (θi ) jθi ]� ν∆θqi (θ)
First-order conditions are
Eθ [S1 (qe (θ), qe (θj )) jθ] = θ,
Eθ
�S1(qe (θ), qe (θj ))jθ
�= θ + ν
1�ν ∆θ.
) Type θ produces the output that equalizes marginal bene�tto marginal costType θ�s output is downward distorted to reduce agents�rent
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
No communication
Proposition. When principals do not share information:
� qe (θ) < q�(θ)
� qe (θ) > q�(θ) i¤ δ < 0
Because of production externalities, the low-cost agent�soutput is distorted since principals expect rivals to produceless to reduce information rents:
If goods are substitutes (complements), Aj�s lower outputinduces Ai to produce more (less)
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
No communication
Proposition. When principals do not share information:� qe (θ) < q�(θ)
� qe (θ) > q�(θ) i¤ δ < 0
Because of production externalities, the low-cost agent�soutput is distorted since principals expect rivals to produceless to reduce information rents:
If goods are substitutes (complements), Aj�s lower outputinduces Ai to produce more (less)
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
No communication
Proposition. When principals do not share information:� qe (θ) < q�(θ)
� qe (θ) > q�(θ) i¤ δ < 0
Because of production externalities, the low-cost agent�soutput is distorted since principals expect rivals to produceless to reduce information rents:
If goods are substitutes (complements), Aj�s lower outputinduces Ai to produce more (less)
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
No communication
Proposition. When principals do not share information:� qe (θ) < q�(θ)
� qe (θ) > q�(θ) i¤ δ < 0
Because of production externalities, the low-cost agent�soutput is distorted since principals expect rivals to produceless to reduce information rents:
If goods are substitutes (complements), Aj�s lower outputinduces Ai to produce more (less)
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
No communication
Proposition. When principals do not share information:� qe (θ) < q�(θ)
� qe (θ) > q�(θ) i¤ δ < 0
Because of production externalities, the low-cost agent�soutput is distorted since principals expect rivals to produceless to reduce information rents:
If goods are substitutes (complements), Aj�s lower outputinduces Ai to produce more (less)
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Contracts specify qi and ti contingent on (θi , θj )
Relevant constraints:
Ui�θ, θj
�= ti
�θ, θj
�� θqi
�θ, θj
�� 0 8θj ,
Eθj [Ui (θ, θj ) jθ] � Eθj
�ti�θ, θj
�� θqi
�θ, θj
�jθ�
) Pi maximizes:
EθjEθi [S (qi (θi , θj ) , qe (θj , θi ))� θiqi (θi , θj ) jθi ]
� ν∆θEθj
�qi (θ, θj )jθ
�(No full surplus extraction due to limited liability)
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Contracts specify qi and ti contingent on (θi , θj )
Relevant constraints:
Ui�θ, θj
�= ti
�θ, θj
�� θqi
�θ, θj
�� 0 8θj ,
Eθj [Ui (θ, θj ) jθ] � Eθj
�ti�θ, θj
�� θqi
�θ, θj
�jθ�
) Pi maximizes:
EθjEθi [S (qi (θi , θj ) , qe (θj , θi ))� θiqi (θi , θj ) jθi ]
� ν∆θEθj
�qi (θ, θj )jθ
�(No full surplus extraction due to limited liability)
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Contracts specify qi and ti contingent on (θi , θj )
Relevant constraints:
Ui�θ, θj
�= ti
�θ, θj
�� θqi
�θ, θj
�� 0 8θj ,
Eθj [Ui (θ, θj ) jθ] � Eθj
�ti�θ, θj
�� θqi
�θ, θj
�jθ�
) Pi maximizes:
EθjEθi [S (qi (θi , θj ) , qe (θj , θi ))� θiqi (θi , θj ) jθi ]
� ν∆θEθj
�qi (θ, θj )jθ
�
(No full surplus extraction due to limited liability)
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Contracts specify qi and ti contingent on (θi , θj )
Relevant constraints:
Ui�θ, θj
�= ti
�θ, θj
�� θqi
�θ, θj
�� 0 8θj ,
Eθj [Ui (θ, θj ) jθ] � Eθj
�ti�θ, θj
�� θqi
�θ, θj
�jθ�
) Pi maximizes:
EθjEθi [S (qi (θi , θj ) , qe (θj , θi ))� θiqi (θi , θj ) jθi ]
� ν∆θEθj
�qi (θ, θj )jθ
�(No full surplus extraction due to limited liability)
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Necessary and su¢ cient �rst-order conditions:
S1 (qe (θ, θj ), qe (θj , θ)) = θ 8θj ,
S1(qe (θ, θj ), qe (θj , θ)) = θ + ν1�ν
Pr(θj jθ)Pr(θj jθ)
∆θ 8θj
) No distortion for θ and downward distortionfor θ to reduce information rentsDistortion increases with Pr(θj jθ)
Pr(θj jθ)since:
Pr�θj jθ
�measures how often Pi pays rent to type θ
Pr(θj jθ) measures how often output is ine¢ cient
Pr(θjθ)Pr(θjθ) >
Pr(θjθ)Pr(θjθ) , α > 0,
) if costs are positively correlated, the distortion of type θ�soutput is larger when his opponent has a low rather than ahigh cost, since this is more likely
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Necessary and su¢ cient �rst-order conditions:
S1 (qe (θ, θj ), qe (θj , θ)) = θ 8θj ,
S1(qe (θ, θj ), qe (θj , θ)) = θ + ν1�ν
Pr(θj jθ)Pr(θj jθ)
∆θ 8θj
) No distortion for θ and downward distortionfor θ to reduce information rents
Distortion increases with Pr(θj jθ)Pr(θj jθ)
since:
Pr�θj jθ
�measures how often Pi pays rent to type θ
Pr(θj jθ) measures how often output is ine¢ cient
Pr(θjθ)Pr(θjθ) >
Pr(θjθ)Pr(θjθ) , α > 0,
) if costs are positively correlated, the distortion of type θ�soutput is larger when his opponent has a low rather than ahigh cost, since this is more likely
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Necessary and su¢ cient �rst-order conditions:
S1 (qe (θ, θj ), qe (θj , θ)) = θ 8θj ,
S1(qe (θ, θj ), qe (θj , θ)) = θ + ν1�ν
Pr(θj jθ)Pr(θj jθ)
∆θ 8θj
) No distortion for θ and downward distortionfor θ to reduce information rentsDistortion increases with Pr(θj jθ)
Pr(θj jθ)since:
Pr�θj jθ
�measures how often Pi pays rent to type θ
Pr(θj jθ) measures how often output is ine¢ cient
Pr(θjθ)Pr(θjθ) >
Pr(θjθ)Pr(θjθ) , α > 0,
) if costs are positively correlated, the distortion of type θ�soutput is larger when his opponent has a low rather than ahigh cost, since this is more likely
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Necessary and su¢ cient �rst-order conditions:
S1 (qe (θ, θj ), qe (θj , θ)) = θ 8θj ,
S1(qe (θ, θj ), qe (θj , θ)) = θ + ν1�ν
Pr(θj jθ)Pr(θj jθ)
∆θ 8θj
) No distortion for θ and downward distortionfor θ to reduce information rentsDistortion increases with Pr(θj jθ)
Pr(θj jθ)since:
Pr�θj jθ
�measures how often Pi pays rent to type θ
Pr(θj jθ) measures how often output is ine¢ cient
Pr(θjθ)Pr(θjθ) >
Pr(θjθ)Pr(θjθ) , α > 0,
) if costs are positively correlated, the distortion of type θ�soutput is larger when his opponent has a low rather than ahigh cost, since this is more likely
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Necessary and su¢ cient �rst-order conditions:
S1 (qe (θ, θj ), qe (θj , θ)) = θ 8θj ,
S1(qe (θ, θj ), qe (θj , θ)) = θ + ν1�ν
Pr(θj jθ)Pr(θj jθ)
∆θ 8θj
) No distortion for θ and downward distortionfor θ to reduce information rentsDistortion increases with Pr(θj jθ)
Pr(θj jθ)since:
Pr�θj jθ
�measures how often Pi pays rent to type θ
Pr(θj jθ) measures how often output is ine¢ cient
Pr(θjθ)Pr(θjθ) >
Pr(θjθ)Pr(θjθ) , α > 0,
) if costs are positively correlated, the distortion of type θ�soutput is larger when his opponent has a low rather than ahigh cost, since this is more likely
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Necessary and su¢ cient �rst-order conditions:
S1 (qe (θ, θj ), qe (θj , θ)) = θ 8θj ,
S1(qe (θ, θj ), qe (θj , θ)) = θ + ν1�ν
Pr(θj jθ)Pr(θj jθ)
∆θ 8θj
) No distortion for θ and downward distortionfor θ to reduce information rentsDistortion increases with Pr(θj jθ)
Pr(θj jθ)since:
Pr�θj jθ
�measures how often Pi pays rent to type θ
Pr(θj jθ) measures how often output is ine¢ cient
Pr(θjθ)Pr(θjθ) >
Pr(θjθ)Pr(θjθ) , α > 0,
) if costs are positively correlated, the distortion of type θ�soutput is larger when his opponent has a low rather than ahigh cost, since this is more likely
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Proposition. If both principals share information,
qe (θ, θ) = q�(θ, θ)qe (θ, θj ) < q�(θ, θj ) 8θjqe (θ, θ) > q�(θ, θ) i¤ δ < 0Expected output does not depend on principals�communication decision
The output of type θ is ine¢ ciently low
Because of production externality, this induces Pi to alsodistort the output of type θ when θj = θ:
if δ > 0, qe (θ, θ) is lower since Pi wants to produce lesswhen Pj produces lessif δ < 0, qe (θ, θ) is higher since Pi wants to produce morewhen Pj produces less
) Strategic linkage between Pi�s output and Aj�s cost
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Proposition. If both principals share information,
qe (θ, θ) = q�(θ, θ)
qe (θ, θj ) < q�(θ, θj ) 8θjqe (θ, θ) > q�(θ, θ) i¤ δ < 0Expected output does not depend on principals�communication decision
The output of type θ is ine¢ ciently low
Because of production externality, this induces Pi to alsodistort the output of type θ when θj = θ:
if δ > 0, qe (θ, θ) is lower since Pi wants to produce lesswhen Pj produces lessif δ < 0, qe (θ, θ) is higher since Pi wants to produce morewhen Pj produces less
) Strategic linkage between Pi�s output and Aj�s cost
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Proposition. If both principals share information,
qe (θ, θ) = q�(θ, θ)qe (θ, θj ) < q�(θ, θj ) 8θj
qe (θ, θ) > q�(θ, θ) i¤ δ < 0Expected output does not depend on principals�communication decision
The output of type θ is ine¢ ciently low
Because of production externality, this induces Pi to alsodistort the output of type θ when θj = θ:
if δ > 0, qe (θ, θ) is lower since Pi wants to produce lesswhen Pj produces lessif δ < 0, qe (θ, θ) is higher since Pi wants to produce morewhen Pj produces less
) Strategic linkage between Pi�s output and Aj�s cost
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Proposition. If both principals share information,
qe (θ, θ) = q�(θ, θ)qe (θ, θj ) < q�(θ, θj ) 8θjqe (θ, θ) > q�(θ, θ) i¤ δ < 0
Expected output does not depend on principals�communication decision
The output of type θ is ine¢ ciently low
Because of production externality, this induces Pi to alsodistort the output of type θ when θj = θ:
if δ > 0, qe (θ, θ) is lower since Pi wants to produce lesswhen Pj produces lessif δ < 0, qe (θ, θ) is higher since Pi wants to produce morewhen Pj produces less
) Strategic linkage between Pi�s output and Aj�s cost
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Proposition. If both principals share information,
qe (θ, θ) = q�(θ, θ)qe (θ, θj ) < q�(θ, θj ) 8θjqe (θ, θ) > q�(θ, θ) i¤ δ < 0Expected output does not depend on principals�communication decision
The output of type θ is ine¢ ciently low
Because of production externality, this induces Pi to alsodistort the output of type θ when θj = θ:
if δ > 0, qe (θ, θ) is lower since Pi wants to produce lesswhen Pj produces lessif δ < 0, qe (θ, θ) is higher since Pi wants to produce morewhen Pj produces less
) Strategic linkage between Pi�s output and Aj�s cost
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Proposition. If both principals share information,
qe (θ, θ) = q�(θ, θ)qe (θ, θj ) < q�(θ, θj ) 8θjqe (θ, θ) > q�(θ, θ) i¤ δ < 0Expected output does not depend on principals�communication decision
The output of type θ is ine¢ ciently low
Because of production externality, this induces Pi to alsodistort the output of type θ when θj = θ:
if δ > 0, qe (θ, θ) is lower since Pi wants to produce lesswhen Pj produces lessif δ < 0, qe (θ, θ) is higher since Pi wants to produce morewhen Pj produces less
) Strategic linkage between Pi�s output and Aj�s cost
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Proposition. If both principals share information,
qe (θ, θ) = q�(θ, θ)qe (θ, θj ) < q�(θ, θj ) 8θjqe (θ, θ) > q�(θ, θ) i¤ δ < 0Expected output does not depend on principals�communication decision
The output of type θ is ine¢ ciently low
Because of production externality, this induces Pi to alsodistort the output of type θ when θj = θ:
if δ > 0, qe (θ, θ) is lower since Pi wants to produce lesswhen Pj produces lessif δ < 0, qe (θ, θ) is higher since Pi wants to produce morewhen Pj produces less
) Strategic linkage between Pi�s output and Aj�s cost
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Proposition. If both principals share information,
qe (θ, θ) = q�(θ, θ)qe (θ, θj ) < q�(θ, θj ) 8θjqe (θ, θ) > q�(θ, θ) i¤ δ < 0Expected output does not depend on principals�communication decision
The output of type θ is ine¢ ciently low
Because of production externality, this induces Pi to alsodistort the output of type θ when θj = θ:
if δ > 0, qe (θ, θ) is lower since Pi wants to produce lesswhen Pj produces less
if δ < 0, qe (θ, θ) is higher since Pi wants to produce morewhen Pj produces less
) Strategic linkage between Pi�s output and Aj�s cost
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Proposition. If both principals share information,
qe (θ, θ) = q�(θ, θ)qe (θ, θj ) < q�(θ, θj ) 8θjqe (θ, θ) > q�(θ, θ) i¤ δ < 0Expected output does not depend on principals�communication decision
The output of type θ is ine¢ ciently low
Because of production externality, this induces Pi to alsodistort the output of type θ when θj = θ:
if δ > 0, qe (θ, θ) is lower since Pi wants to produce lesswhen Pj produces lessif δ < 0, qe (θ, θ) is higher since Pi wants to produce morewhen Pj produces less
) Strategic linkage between Pi�s output and Aj�s cost
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Bilateral information sharing
Proposition. If both principals share information,
qe (θ, θ) = q�(θ, θ)qe (θ, θj ) < q�(θ, θj ) 8θjqe (θ, θ) > q�(θ, θ) i¤ δ < 0Expected output does not depend on principals�communication decision
The output of type θ is ine¢ ciently low
Because of production externality, this induces Pi to alsodistort the output of type θ when θj = θ:
if δ > 0, qe (θ, θ) is lower since Pi wants to produce lesswhen Pj produces lessif δ < 0, qe (θ, θ) is higher since Pi wants to produce morewhen Pj produces less
) Strategic linkage between Pi�s output and Aj�s cost
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Unilateral information sharing
Pi shares information, Pj does not
FOCs areEθj
�S1(qei (θ), q
ej (θj , θ))jθ
�= θ
Eθj
�S1(qei (θ), q
ej (θj , θ)jθ
�= θ + ν
1�ν ∆θ
S1(qej (θ, θi ), qei (θi )) = θ 8θi
S1(qej (θ, θi ), qei (θi )) = θ + ν
1�νPr(θi jθ)Pr(θi jθ)
∆θ 8θi
Pi conditions her contracts only on θi , while Pj conditions hercontract on θj and θi
) Pj has a competitive advantage relative to Pi since she canimpose a higher distortion in the less likely states
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Unilateral information sharing
Pi shares information, Pj does not
FOCs areEθj
�S1(qei (θ), q
ej (θj , θ))jθ
�= θ
Eθj
�S1(qei (θ), q
ej (θj , θ)jθ
�= θ + ν
1�ν ∆θ
S1(qej (θ, θi ), qei (θi )) = θ 8θi
S1(qej (θ, θi ), qei (θi )) = θ + ν
1�νPr(θi jθ)Pr(θi jθ)
∆θ 8θi
Pi conditions her contracts only on θi , while Pj conditions hercontract on θj and θi
) Pj has a competitive advantage relative to Pi since she canimpose a higher distortion in the less likely states
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Unilateral information sharing
Pi shares information, Pj does not
FOCs areEθj
�S1(qei (θ), q
ej (θj , θ))jθ
�= θ
Eθj
�S1(qei (θ), q
ej (θj , θ)jθ
�= θ + ν
1�ν ∆θ
S1(qej (θ, θi ), qei (θi )) = θ 8θi
S1(qej (θ, θi ), qei (θi )) = θ + ν
1�νPr(θi jθ)Pr(θi jθ)
∆θ 8θi
Pi conditions her contracts only on θi , while Pj conditions hercontract on θj and θi
) Pj has a competitive advantage relative to Pi since she canimpose a higher distortion in the less likely states
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Unilateral information sharing
Pi shares information, Pj does not
FOCs areEθj
�S1(qei (θ), q
ej (θj , θ))jθ
�= θ
Eθj
�S1(qei (θ), q
ej (θj , θ)jθ
�= θ + ν
1�ν ∆θ
S1(qej (θ, θi ), qei (θi )) = θ 8θi
S1(qej (θ, θi ), qei (θi )) = θ + ν
1�νPr(θi jθ)Pr(θi jθ)
∆θ 8θi
Pi conditions her contracts only on θi , while Pj conditions hercontract on θj and θi
) Pj has a competitive advantage relative to Pi since she canimpose a higher distortion in the less likely states
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Do principals share information?
Proposition: When agents are privately informedabout their marginal costs:
� If δα < 0, there is a unique equilibrium in dominant strategiesin which both principals share information
� If δα > 0, there is a unique equilibrium in dominant strategiesin which no principal shares information
� If δ = 0, principals are indi¤erent betweensharing information or not
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Intuition
Principals share information i¤ δα < 0
If δ 6= 0, a correlated distortion e¤ect induces principals tocoordinate outputs:
Suppose that α > 0If Pi shares information, Pj distorts the output of her high-costagent more (i.e. produces less) when θi = θ(since this is less likely than θi = θ)This increases Pi�s pro�ts with strategic substitutes (δ < 0)since Pi wants to produce more when Pj produces lessThis reduces Pi�s pro�ts with strategic complements (δ > 0)since Pi wants to produce less when Pj produces less
The e¤ect is of �rst-order: only the sign of δ mattersand not its magnitude
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Intuition
Principals share information i¤ δα < 0
If δ 6= 0, a correlated distortion e¤ect induces principals tocoordinate outputs:
Suppose that α > 0If Pi shares information, Pj distorts the output of her high-costagent more (i.e. produces less) when θi = θ(since this is less likely than θi = θ)This increases Pi�s pro�ts with strategic substitutes (δ < 0)since Pi wants to produce more when Pj produces lessThis reduces Pi�s pro�ts with strategic complements (δ > 0)since Pi wants to produce less when Pj produces less
The e¤ect is of �rst-order: only the sign of δ mattersand not its magnitude
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Intuition
Principals share information i¤ δα < 0
If δ 6= 0, a correlated distortion e¤ect induces principals tocoordinate outputs:
Suppose that α > 0
If Pi shares information, Pj distorts the output of her high-costagent more (i.e. produces less) when θi = θ(since this is less likely than θi = θ)This increases Pi�s pro�ts with strategic substitutes (δ < 0)since Pi wants to produce more when Pj produces lessThis reduces Pi�s pro�ts with strategic complements (δ > 0)since Pi wants to produce less when Pj produces less
The e¤ect is of �rst-order: only the sign of δ mattersand not its magnitude
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Intuition
Principals share information i¤ δα < 0
If δ 6= 0, a correlated distortion e¤ect induces principals tocoordinate outputs:
Suppose that α > 0If Pi shares information, Pj distorts the output of her high-costagent more (i.e. produces less) when θi = θ(since this is less likely than θi = θ)
This increases Pi�s pro�ts with strategic substitutes (δ < 0)since Pi wants to produce more when Pj produces lessThis reduces Pi�s pro�ts with strategic complements (δ > 0)since Pi wants to produce less when Pj produces less
The e¤ect is of �rst-order: only the sign of δ mattersand not its magnitude
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Intuition
Principals share information i¤ δα < 0
If δ 6= 0, a correlated distortion e¤ect induces principals tocoordinate outputs:
Suppose that α > 0If Pi shares information, Pj distorts the output of her high-costagent more (i.e. produces less) when θi = θ(since this is less likely than θi = θ)This increases Pi�s pro�ts with strategic substitutes (δ < 0)since Pi wants to produce more when Pj produces less
This reduces Pi�s pro�ts with strategic complements (δ > 0)since Pi wants to produce less when Pj produces less
The e¤ect is of �rst-order: only the sign of δ mattersand not its magnitude
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Intuition
Principals share information i¤ δα < 0
If δ 6= 0, a correlated distortion e¤ect induces principals tocoordinate outputs:
Suppose that α > 0If Pi shares information, Pj distorts the output of her high-costagent more (i.e. produces less) when θi = θ(since this is less likely than θi = θ)This increases Pi�s pro�ts with strategic substitutes (δ < 0)since Pi wants to produce more when Pj produces lessThis reduces Pi�s pro�ts with strategic complements (δ > 0)since Pi wants to produce less when Pj produces less
The e¤ect is of �rst-order: only the sign of δ mattersand not its magnitude
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Intuition
Principals share information i¤ δα < 0
If δ 6= 0, a correlated distortion e¤ect induces principals tocoordinate outputs:
Suppose that α > 0If Pi shares information, Pj distorts the output of her high-costagent more (i.e. produces less) when θi = θ(since this is less likely than θi = θ)This increases Pi�s pro�ts with strategic substitutes (δ < 0)since Pi wants to produce more when Pj produces lessThis reduces Pi�s pro�ts with strategic complements (δ > 0)since Pi wants to produce less when Pj produces less
The e¤ect is of �rst-order: only the sign of δ mattersand not its magnitude
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
The value of communication
Proposition. Principals�expected pro�ts are higher whenthey both share information than with no communication
Since costs are correlated, communication creates aninformational externality that reduces agents�rent
(For small externalities, this e¤ect is stronger thanthe strategic e¤ect due to correlated distortions)
Corollary. Principals�decision not to share information whenδα > 0 does not maximize their joint pro�ts
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
The value of communication
Proposition. Principals�expected pro�ts are higher whenthey both share information than with no communication
Since costs are correlated, communication creates aninformational externality that reduces agents�rent
(For small externalities, this e¤ect is stronger thanthe strategic e¤ect due to correlated distortions)
Corollary. Principals�decision not to share information whenδα > 0 does not maximize their joint pro�ts
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
The value of communication
Proposition. Principals�expected pro�ts are higher whenthey both share information than with no communication
Since costs are correlated, communication creates aninformational externality that reduces agents�rent
(For small externalities, this e¤ect is stronger thanthe strategic e¤ect due to correlated distortions)
Corollary. Principals�decision not to share information whenδα > 0 does not maximize their joint pro�ts
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
The value of communication
Proposition. Principals�expected pro�ts are higher whenthey both share information than with no communication
Since costs are correlated, communication creates aninformational externality that reduces agents�rent
(For small externalities, this e¤ect is stronger thanthe strategic e¤ect due to correlated distortions)
Corollary. Principals�decision not to share information whenδα > 0 does not maximize their joint pro�ts
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Conclusions
When do principals independently choose to share theinformation obtained from informed agents?
Principals want to:
a¤ect rivals�strategies because of externalitiesreduce agents�information rents
Incentive to share information only depend on the sign of δα
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Conclusions
When do principals independently choose to share theinformation obtained from informed agents?
Principals want to:
a¤ect rivals�strategies because of externalitiesreduce agents�information rents
Incentive to share information only depend on the sign of δα
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Conclusions
When do principals independently choose to share theinformation obtained from informed agents?
Principals want to:
a¤ect rivals�strategies because of externalities
reduce agents�information rents
Incentive to share information only depend on the sign of δα
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Conclusions
When do principals independently choose to share theinformation obtained from informed agents?
Principals want to:
a¤ect rivals�strategies because of externalitiesreduce agents�information rents
Incentive to share information only depend on the sign of δα
Introduction Model Complete information Asymmetric information Information sharing? Conclusions
Conclusions
When do principals independently choose to share theinformation obtained from informed agents?
Principals want to:
a¤ect rivals�strategies because of externalitiesreduce agents�information rents
Incentive to share information only depend on the sign of δα