Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in...
Transcript of Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in...
![Page 1: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/1.jpg)
1/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Time inconsistent stochastic control
Dylan Possamaı, joint with Camilo Hernandez
Columbia University, NY
Probability and statistics seminar, University of Southern California,Los Angeles, November 11th, 2019
Dylan Possamaı Time–inconsistent stochastic control
![Page 2: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/2.jpg)
2/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Motivating exampleTime–inconsistencyThree approaches
Outline
1 The big pictureMotivating exampleTime–inconsistencyThree approaches
2 What is an equilibrium?In discrete–time, (almost) all is wellNot so much in continuous–time...
3 Main resultsAn extended DPPThe characterising BSDE systemVerification theoremExtensions
4 An application in contract theory
Dylan Possamaı Time–inconsistent stochastic control
![Page 3: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/3.jpg)
3/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Motivating exampleTime–inconsistencyThree approaches
Motivating example
• You receive an invitation to your first talk on a new subject a month fromnow.
When receiving the invitation email: plenty of time left, will prepare theslides in two weeks.
After two weeks: too many things to do actually, will prepare them in aweek.
After three weeks: still one week left, why start now?
Morning of the day of the talk: guess I should be starting!
What we should take home from this: human beings have time–depending andslightly inconsistent preferences.
Dylan Possamaı Time–inconsistent stochastic control
![Page 4: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/4.jpg)
3/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Motivating exampleTime–inconsistencyThree approaches
Motivating example
• You receive an invitation to your first talk on a new subject a month fromnow.
When receiving the invitation email: plenty of time left, will prepare theslides in two weeks.
After two weeks: too many things to do actually, will prepare them in aweek.
After three weeks: still one week left, why start now?
Morning of the day of the talk: guess I should be starting!
What we should take home from this: human beings have time–depending andslightly inconsistent preferences.
Dylan Possamaı Time–inconsistent stochastic control
![Page 5: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/5.jpg)
3/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Motivating exampleTime–inconsistencyThree approaches
Motivating example
• You receive an invitation to your first talk on a new subject a month fromnow.
When receiving the invitation email: plenty of time left, will prepare theslides in two weeks.
After two weeks: too many things to do actually, will prepare them in aweek.
After three weeks: still one week left, why start now?
Morning of the day of the talk: guess I should be starting!
What we should take home from this: human beings have time–depending andslightly inconsistent preferences.
Dylan Possamaı Time–inconsistent stochastic control
![Page 6: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/6.jpg)
3/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Motivating exampleTime–inconsistencyThree approaches
Motivating example
• You receive an invitation to your first talk on a new subject a month fromnow.
When receiving the invitation email: plenty of time left, will prepare theslides in two weeks.
After two weeks: too many things to do actually, will prepare them in aweek.
After three weeks: still one week left, why start now?
Morning of the day of the talk: guess I should be starting!
What we should take home from this: human beings have time–depending andslightly inconsistent preferences.
Dylan Possamaı Time–inconsistent stochastic control
![Page 7: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/7.jpg)
3/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Motivating exampleTime–inconsistencyThree approaches
Motivating example
• You receive an invitation to your first talk on a new subject a month fromnow.
When receiving the invitation email: plenty of time left, will prepare theslides in two weeks.
After two weeks: too many things to do actually, will prepare them in aweek.
After three weeks: still one week left, why start now?
Morning of the day of the talk: guess I should be starting!
What we should take home from this: human beings have time–depending andslightly inconsistent preferences.
Dylan Possamaı Time–inconsistent stochastic control
![Page 8: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/8.jpg)
3/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Motivating exampleTime–inconsistencyThree approaches
Motivating example
• You receive an invitation to your first talk on a new subject a month fromnow.
When receiving the invitation email: plenty of time left, will prepare theslides in two weeks.
After two weeks: too many things to do actually, will prepare them in aweek.
After three weeks: still one week left, why start now?
Morning of the day of the talk: guess I should be starting!
What we should take home from this: human beings have time–depending andslightly inconsistent preferences.
Dylan Possamaı Time–inconsistent stochastic control
![Page 9: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/9.jpg)
4/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Motivating exampleTime–inconsistencyThree approaches
The basic problem
On space (Ω,F), let Pν be a weak solution to the controlled SDE
Xt = x +
∫ t
0
σr (Xr∧·)(br (Xr∧·, νr )dr + dWr
), t ∈ [0,T ].
The reward functional
v(t, ν) := J(t, t, ν) = EPν[ ∫ T
t
fr (t,Xr∧·, νr )dr + F (t,XT∧·)
∣∣∣∣Ft
].
Control problem: because of the dependence in t, classical dynamic pro-gramming arguments fail (unless for exponential discounting of course).What to do?
Dylan Possamaı Time–inconsistent stochastic control
![Page 10: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/10.jpg)
4/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Motivating exampleTime–inconsistencyThree approaches
The basic problem
On space (Ω,F), let Pν be a weak solution to the controlled SDE
Xt = x +
∫ t
0
σr (Xr∧·)(br (Xr∧·, νr )dr + dWr
), t ∈ [0,T ].
The reward functional
v(t, ν) := J(t, t, ν) = EPν[ ∫ T
t
fr (t,Xr∧·, νr )dr + F (t,XT∧·)
∣∣∣∣Ft
].
Control problem: because of the dependence in t, classical dynamic pro-gramming arguments fail (unless for exponential discounting of course).What to do?
Dylan Possamaı Time–inconsistent stochastic control
![Page 11: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/11.jpg)
4/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Motivating exampleTime–inconsistencyThree approaches
The basic problem
On space (Ω,F), let Pν be a weak solution to the controlled SDE
Xt = x +
∫ t
0
σr (Xr∧·)(br (Xr∧·, νr )dr + dWr
), t ∈ [0,T ].
The reward functional
v(t, ν) := J(t, t, ν) = EPν[ ∫ T
t
fr (t,Xr∧·, νr )dr + F (t,XT∧·)
∣∣∣∣Ft
].
Control problem: because of the dependence in t, classical dynamic pro-gramming arguments fail (unless for exponential discounting of course).What to do?
Dylan Possamaı Time–inconsistent stochastic control
![Page 12: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/12.jpg)
5/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Motivating exampleTime–inconsistencyThree approaches
The three approaches
Dismiss the problem altogether: non–fully rational economic agents are toocumbersome.
Assume precomittment: solve the maximisation problem
supν
J(0, 0, ν),
and obtain an ”optimal” action ν?. However, in general ν? will fail to beoptimal if one maximises J(t, t, ν) for t > 0 (Karnam, Ma, Zhang, Zhou,...)
Take time–inconsistency seriously: consider a non–cooperative game, wherethe agent plays against future versions of himself, and look for sub–gameperfect Nash equilibria (Barro, Czichowsky, Ekeland, Laibson, Lazrak, Pol-lak, Privu, Strotz, Zhou,...)
Dylan Possamaı Time–inconsistent stochastic control
![Page 13: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/13.jpg)
5/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Motivating exampleTime–inconsistencyThree approaches
The three approaches
Dismiss the problem altogether: non–fully rational economic agents are toocumbersome.
Assume precomittment: solve the maximisation problem
supν
J(0, 0, ν),
and obtain an ”optimal” action ν?. However, in general ν? will fail to beoptimal if one maximises J(t, t, ν) for t > 0 (Karnam, Ma, Zhang, Zhou,...)
Take time–inconsistency seriously: consider a non–cooperative game, wherethe agent plays against future versions of himself, and look for sub–gameperfect Nash equilibria (Barro, Czichowsky, Ekeland, Laibson, Lazrak, Pol-lak, Privu, Strotz, Zhou,...)
Dylan Possamaı Time–inconsistent stochastic control
![Page 14: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/14.jpg)
5/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Motivating exampleTime–inconsistencyThree approaches
The three approaches
Dismiss the problem altogether: non–fully rational economic agents are toocumbersome.
Assume precomittment: solve the maximisation problem
supν
J(0, 0, ν),
and obtain an ”optimal” action ν?. However, in general ν? will fail to beoptimal if one maximises J(t, t, ν) for t > 0 (Karnam, Ma, Zhang, Zhou,...)
Take time–inconsistency seriously: consider a non–cooperative game, wherethe agent plays against future versions of himself, and look for sub–gameperfect Nash equilibria (Barro, Czichowsky, Ekeland, Laibson, Lazrak, Pol-lak, Privu, Strotz, Zhou,...)
Dylan Possamaı Time–inconsistent stochastic control
![Page 15: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/15.jpg)
6/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Outline
1 The big pictureMotivating exampleTime–inconsistencyThree approaches
2 What is an equilibrium?In discrete–time, (almost) all is wellNot so much in continuous–time...
3 Main resultsAn extended DPPThe characterising BSDE systemVerification theoremExtensions
4 An application in contract theory
Dylan Possamaı Time–inconsistent stochastic control
![Page 16: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/16.jpg)
7/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Equilibria in discrete–time
Perfectly understandable situation. If (ti )0≤i≤n is a partition of [0,T ]
Given actions played by Player 0,..., and Player tn−2, Player tn−1 faces astandard optimisation problem.
Player tn−2 can solve the problem faced by Player tn−1 and obtains a Stack-elberg equilibrium, using the optimal response of Player tn−1 in his ownoptimisation.
Repeat backwardly.
Unfortunately, optimisation problems often lose concavity after a few iter-ations =⇒ equilibria do not exist in general.
Dylan Possamaı Time–inconsistent stochastic control
![Page 17: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/17.jpg)
7/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Equilibria in discrete–time
Perfectly understandable situation. If (ti )0≤i≤n is a partition of [0,T ]
Given actions played by Player 0,..., and Player tn−2, Player tn−1 faces astandard optimisation problem.
Player tn−2 can solve the problem faced by Player tn−1 and obtains a Stack-elberg equilibrium, using the optimal response of Player tn−1 in his ownoptimisation.
Repeat backwardly.
Unfortunately, optimisation problems often lose concavity after a few iter-ations =⇒ equilibria do not exist in general.
Dylan Possamaı Time–inconsistent stochastic control
![Page 18: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/18.jpg)
7/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Equilibria in discrete–time
Perfectly understandable situation. If (ti )0≤i≤n is a partition of [0,T ]
Given actions played by Player 0,..., and Player tn−2, Player tn−1 faces astandard optimisation problem.
Player tn−2 can solve the problem faced by Player tn−1 and obtains a Stack-elberg equilibrium, using the optimal response of Player tn−1 in his ownoptimisation.
Repeat backwardly.
Unfortunately, optimisation problems often lose concavity after a few iter-ations =⇒ equilibria do not exist in general.
Dylan Possamaı Time–inconsistent stochastic control
![Page 19: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/19.jpg)
7/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Equilibria in discrete–time
Perfectly understandable situation. If (ti )0≤i≤n is a partition of [0,T ]
Given actions played by Player 0,..., and Player tn−2, Player tn−1 faces astandard optimisation problem.
Player tn−2 can solve the problem faced by Player tn−1 and obtains a Stack-elberg equilibrium, using the optimal response of Player tn−1 in his ownoptimisation.
Repeat backwardly.
Unfortunately, optimisation problems often lose concavity after a few iter-ations =⇒ equilibria do not exist in general.
Dylan Possamaı Time–inconsistent stochastic control
![Page 20: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/20.jpg)
8/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Equilibria in continuous–time
Definition [Ekeland, Lazrak (2008)]
ν? is an equilibrium if for any (t, ν) ∈ [0,T )× V
liminf`→0
J(t, t, ν?)− J(t, t, ν`)
`≥ 0,
where for ` ∈ [0,T − t], ν` is given by
ν`r := 1t≤r<t+`νr + 1t+`≤r≤Tν
?r .
Definition used in most of the literature in continuous–time.
Not completely satisfying, when liminf is 0.
Not a ”local” property.
Dylan Possamaı Time–inconsistent stochastic control
![Page 21: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/21.jpg)
8/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Equilibria in continuous–time
Definition [Ekeland, Lazrak (2008)]
ν? is an equilibrium if for any (t, ν) ∈ [0,T )× V
liminf`→0
J(t, t, ν?)− J(t, t, ν`)
`≥ 0,
where for ` ∈ [0,T − t], ν` is given by
ν`r := 1t≤r<t+`νr + 1t+`≤r≤Tν
?r .
Definition used in most of the literature in continuous–time.
Not completely satisfying, when liminf is 0.
Not a ”local” property.
Dylan Possamaı Time–inconsistent stochastic control
![Page 22: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/22.jpg)
8/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Equilibria in continuous–time
Definition [Ekeland, Lazrak (2008)]
ν? is an equilibrium if for any (t, ν) ∈ [0,T )× V
liminf`→0
J(t, t, ν?)− J(t, t, ν`)
`≥ 0,
where for ` ∈ [0,T − t], ν` is given by
ν`r := 1t≤r<t+`νr + 1t+`≤r≤Tν
?r .
Definition used in most of the literature in continuous–time.
Not completely satisfying, when liminf is 0.
Not a ”local” property.
Dylan Possamaı Time–inconsistent stochastic control
![Page 23: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/23.jpg)
8/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Equilibria in continuous–time
Definition [Ekeland, Lazrak (2008)]
ν? is an equilibrium if for any (t, ν) ∈ [0,T )× V
liminf`→0
J(t, t, ν?)− J(t, t, ν`)
`≥ 0,
where for ` ∈ [0,T − t], ν` is given by
ν`r := 1t≤r<t+`νr + 1t+`≤r≤Tν
?r .
Definition used in most of the literature in continuous–time.
Not completely satisfying, when liminf is 0.
Not a ”local” property.
Dylan Possamaı Time–inconsistent stochastic control
![Page 24: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/24.jpg)
9/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Equilibria in continuous–time (2)
Definition [Hernandez, P. (2019)]
ν? is an equilibrium if for any ε > 0 there exists `ε > 0 such that for any (t, ν, τ) ∈ [0,T )×V ×Tt,t+`ε
J(t, t, ν?) ≥ J(t, t, ντ )− ε`ε.
If `ε can be taken independent of ε, ν? is called a strict equilibrium, for which
J(t, t, ν?) ≥ J(t, t, ντ ).
First definition is roughly speaking the same as before: ε−equilibrium.
Strict equilibria are the ones we want, but (much) harder to prove existence(recent contribution of Huang and Zhou (2018) for CTMC).
Dylan Possamaı Time–inconsistent stochastic control
![Page 25: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/25.jpg)
9/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Equilibria in continuous–time (2)
Definition [Hernandez, P. (2019)]
ν? is an equilibrium if for any ε > 0 there exists `ε > 0 such that for any (t, ν, τ) ∈ [0,T )×V ×Tt,t+`ε
J(t, t, ν?) ≥ J(t, t, ντ )− ε`ε.
If `ε can be taken independent of ε, ν? is called a strict equilibrium, for which
J(t, t, ν?) ≥ J(t, t, ντ ).
First definition is roughly speaking the same as before: ε−equilibrium.
Strict equilibria are the ones we want, but (much) harder to prove existence(recent contribution of Huang and Zhou (2018) for CTMC).
Dylan Possamaı Time–inconsistent stochastic control
![Page 26: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/26.jpg)
9/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Equilibria in continuous–time (2)
Definition [Hernandez, P. (2019)]
ν? is an equilibrium if for any ε > 0 there exists `ε > 0 such that for any (t, ν, τ) ∈ [0,T )×V ×Tt,t+`ε
J(t, t, ν?) ≥ J(t, t, ντ )− ε`ε.
If `ε can be taken independent of ε, ν? is called a strict equilibrium, for which
J(t, t, ν?) ≥ J(t, t, ντ ).
First definition is roughly speaking the same as before: ε−equilibrium.
Strict equilibria are the ones we want, but (much) harder to prove existence(recent contribution of Huang and Zhou (2018) for CTMC).
Dylan Possamaı Time–inconsistent stochastic control
![Page 27: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/27.jpg)
10/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Literature review
Large literature in discrete–time. In continuous–time, mostly specific mod-els, solved on a case–by–case basis.
Ekeland, Lazrak and Pirvu introduced an extended HJB equation allowingfor verification–type results: any smooth solution to the equation allows toconstruct an equilibrium.
Extended by Bjork, Khapko, and Murgoci to general Markovian diffusionmodels.
However, extended HJB equation only justified formally by passing to thelimit in discrete–time models.
Different from classical control problems, where HJB equation and optimalcontrols are intimately linked.
Dylan Possamaı Time–inconsistent stochastic control
![Page 28: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/28.jpg)
10/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Literature review
Large literature in discrete–time. In continuous–time, mostly specific mod-els, solved on a case–by–case basis.
Ekeland, Lazrak and Pirvu introduced an extended HJB equation allowingfor verification–type results: any smooth solution to the equation allows toconstruct an equilibrium.
Extended by Bjork, Khapko, and Murgoci to general Markovian diffusionmodels.
However, extended HJB equation only justified formally by passing to thelimit in discrete–time models.
Different from classical control problems, where HJB equation and optimalcontrols are intimately linked.
Dylan Possamaı Time–inconsistent stochastic control
![Page 29: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/29.jpg)
10/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Literature review
Large literature in discrete–time. In continuous–time, mostly specific mod-els, solved on a case–by–case basis.
Ekeland, Lazrak and Pirvu introduced an extended HJB equation allowingfor verification–type results: any smooth solution to the equation allows toconstruct an equilibrium.
Extended by Bjork, Khapko, and Murgoci to general Markovian diffusionmodels.
However, extended HJB equation only justified formally by passing to thelimit in discrete–time models.
Different from classical control problems, where HJB equation and optimalcontrols are intimately linked.
Dylan Possamaı Time–inconsistent stochastic control
![Page 30: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/30.jpg)
10/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Literature review
Large literature in discrete–time. In continuous–time, mostly specific mod-els, solved on a case–by–case basis.
Ekeland, Lazrak and Pirvu introduced an extended HJB equation allowingfor verification–type results: any smooth solution to the equation allows toconstruct an equilibrium.
Extended by Bjork, Khapko, and Murgoci to general Markovian diffusionmodels.
However, extended HJB equation only justified formally by passing to thelimit in discrete–time models.
Different from classical control problems, where HJB equation and optimalcontrols are intimately linked.
Dylan Possamaı Time–inconsistent stochastic control
![Page 31: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/31.jpg)
10/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
In discrete–time, (almost) all is wellNot so much in continuous–time...
Literature review
Large literature in discrete–time. In continuous–time, mostly specific mod-els, solved on a case–by–case basis.
Ekeland, Lazrak and Pirvu introduced an extended HJB equation allowingfor verification–type results: any smooth solution to the equation allows toconstruct an equilibrium.
Extended by Bjork, Khapko, and Murgoci to general Markovian diffusionmodels.
However, extended HJB equation only justified formally by passing to thelimit in discrete–time models.
Different from classical control problems, where HJB equation and optimalcontrols are intimately linked.
Dylan Possamaı Time–inconsistent stochastic control
![Page 32: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/32.jpg)
11/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An extended DPPThe characterising BSDE systemVerification theoremExtensions
Outline
1 The big pictureMotivating exampleTime–inconsistencyThree approaches
2 What is an equilibrium?In discrete–time, (almost) all is wellNot so much in continuous–time...
3 Main resultsAn extended DPPThe characterising BSDE systemVerification theoremExtensions
4 An application in contract theory
Dylan Possamaı Time–inconsistent stochastic control
![Page 33: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/33.jpg)
12/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An extended DPPThe characterising BSDE systemVerification theoremExtensions
An extended DPP
Though classical DPP does not hold, can obtain
Lemma [Hernandez, P. (2019)]
If ν? is an equilibrium, then, for any ε > 0, τ ∈ T0,`ε
v(0, ν?) ≤ supν∈V
J(0, ν),
v(0, ν?) ≥ EPν [ ∫ τ
0fr (0, Xr∧·, νr )dr + v(τ, ν?) + J(τ, 0, ν?) − J(τ, τ, ν?)
]− ε`ε.
If ν? is a strict equilibrium then
v(0, ν?) = supν∈V
EPν [ ∫ τ
0fr (0, Xr∧·, νr )dr + v(τ, ν?) + J(τ, 0, ν?) − J(τ, τ, ν?)
].
Extends to arbitrary t ∈ [0,T ],
but not to intervals of length longer than `ε.
Dylan Possamaı Time–inconsistent stochastic control
![Page 34: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/34.jpg)
12/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An extended DPPThe characterising BSDE systemVerification theoremExtensions
An extended DPP
Though classical DPP does not hold, can obtain
Lemma [Hernandez, P. (2019)]
If ν? is an equilibrium, then, for any ε > 0, τ ∈ T0,`ε
v(0, ν?) ≤ supν∈V
J(0, ν),
v(0, ν?) ≥ EPν [ ∫ τ
0fr (0, Xr∧·, νr )dr + v(τ, ν?) + J(τ, 0, ν?) − J(τ, τ, ν?)
]− ε`ε.
If ν? is a strict equilibrium then
v(0, ν?) = supν∈V
EPν [ ∫ τ
0fr (0, Xr∧·, νr )dr + v(τ, ν?) + J(τ, 0, ν?) − J(τ, τ, ν?)
].
Extends to arbitrary t ∈ [0,T ],
but not to intervals of length longer than `ε.
Dylan Possamaı Time–inconsistent stochastic control
![Page 35: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/35.jpg)
12/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An extended DPPThe characterising BSDE systemVerification theoremExtensions
An extended DPP
Though classical DPP does not hold, can obtain
Lemma [Hernandez, P. (2019)]
If ν? is an equilibrium, then, for any ε > 0, τ ∈ T0,`ε
v(0, ν?) ≤ supν∈V
J(0, ν),
v(0, ν?) ≥ EPν [ ∫ τ
0fr (0, Xr∧·, νr )dr + v(τ, ν?) + J(τ, 0, ν?) − J(τ, τ, ν?)
]− ε`ε.
If ν? is a strict equilibrium then
v(0, ν?) = supν∈V
EPν [ ∫ τ
0fr (0, Xr∧·, νr )dr + v(τ, ν?) + J(τ, 0, ν?) − J(τ, τ, ν?)
].
Extends to arbitrary t ∈ [0,T ],
but not to intervals of length longer than `ε.
Dylan Possamaı Time–inconsistent stochastic control
![Page 36: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/36.jpg)
13/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An extended DPPThe characterising BSDE systemVerification theoremExtensions
An extended DPP (2)
Iterating the previous DPP for arbitrary partitions of [0,T ] with mesh smallerthan `ε, and passing to the limit
Theorem [Hernandez, P. (2019)]
If ν? is an equilibrium then
v(0, ν?) = supν∈V
EPν [
v(τ, ντ ) +
∫ τ0
(f (r, r, Xr∧·, νr ) −
∂F
∂s(r) −
∫ T
r
∂f
∂s(r, u, Xr∧·, ν
?u )du
)dr
].
Dylan Possamaı Time–inconsistent stochastic control
![Page 37: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/37.jpg)
14/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An extended DPPThe characterising BSDE systemVerification theoremExtensions
HJB BSDE system
Define the Hamiltonian
Ht(x , z) := supa∈A
ht(t, x , z), ht(s, x , z , a) := ft(s, x , a) + bt(x , a) · σt(x)>z ,
ν?t (x , z) := argmaxa∈Aht(t, x , z , a).
Extended DPP relates value of the agent at equilibrium toYt = F (T ,X·∧T ) +
∫ T
t
(Hr (X ,Zr )− ∂Y r
r )dr −∫ T
t
Zr · dXr ,
Y st = F (s,X·∧T ) +
∫ T
t
hr (s,X , ν?r (Xr∧·,Zr ),Z
sr )dr −
∫ T
t
Z sr · dXr ,
where
∂Y st :=
∂
∂sY s
t .
Dylan Possamaı Time–inconsistent stochastic control
![Page 38: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/38.jpg)
14/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An extended DPPThe characterising BSDE systemVerification theoremExtensions
HJB BSDE system
Define the Hamiltonian
Ht(x , z) := supa∈A
ht(t, x , z), ht(s, x , z , a) := ft(s, x , a) + bt(x , a) · σt(x)>z ,
ν?t (x , z) := argmaxa∈Aht(t, x , z , a).
Extended DPP relates value of the agent at equilibrium toYt = F (T ,X·∧T ) +
∫ T
t
(Hr (X ,Zr )− ∂Y r
r )dr −∫ T
t
Zr · dXr ,
Y st = F (s,X·∧T ) +
∫ T
t
hr (s,X , ν?r (Xr∧·,Zr ),Z
sr )dr −
∫ T
t
Z sr · dXr ,
where
∂Y st :=
∂
∂sY s
t .
Dylan Possamaı Time–inconsistent stochastic control
![Page 39: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/39.jpg)
15/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An extended DPPThe characterising BSDE systemVerification theoremExtensions
HJB BSDE system (2)
For exponential discounting fs(t, x , a) = e−δ(s−t) f (s, x , a), we have ∂Y st =
−δYs =⇒ no coupling and classical HJB BSDE.
This is the non–Markovian version of the non–local PDE system derivedfirst by Ekeland and Lazrak (2008).
Earlier contributions argued by passing informally to the limit from discrete–time. Thanks to our extended DPP
Theorem [Hernandez, P. (2019)]
If ν? is an equilibrium, then there exists a solution to the BSDE system, andnecessarily ν? = ν?(X ,Z).
Dylan Possamaı Time–inconsistent stochastic control
![Page 40: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/40.jpg)
15/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An extended DPPThe characterising BSDE systemVerification theoremExtensions
HJB BSDE system (2)
For exponential discounting fs(t, x , a) = e−δ(s−t) f (s, x , a), we have ∂Y st =
−δYs =⇒ no coupling and classical HJB BSDE.
This is the non–Markovian version of the non–local PDE system derivedfirst by Ekeland and Lazrak (2008).
Earlier contributions argued by passing informally to the limit from discrete–time. Thanks to our extended DPP
Theorem [Hernandez, P. (2019)]
If ν? is an equilibrium, then there exists a solution to the BSDE system, andnecessarily ν? = ν?(X ,Z).
Dylan Possamaı Time–inconsistent stochastic control
![Page 41: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/41.jpg)
15/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An extended DPPThe characterising BSDE systemVerification theoremExtensions
HJB BSDE system (2)
For exponential discounting fs(t, x , a) = e−δ(s−t) f (s, x , a), we have ∂Y st =
−δYs =⇒ no coupling and classical HJB BSDE.
This is the non–Markovian version of the non–local PDE system derivedfirst by Ekeland and Lazrak (2008).
Earlier contributions argued by passing informally to the limit from discrete–time. Thanks to our extended DPP
Theorem [Hernandez, P. (2019)]
If ν? is an equilibrium, then there exists a solution to the BSDE system, andnecessarily ν? = ν?(X ,Z).
Dylan Possamaı Time–inconsistent stochastic control
![Page 42: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/42.jpg)
16/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An extended DPPThe characterising BSDE systemVerification theoremExtensions
Verification theorem
Theorem [Hernandez, P. (2019)]
If there exists a sufficiently integrable solution to the BSDE system, thenν?(X ,Z) is an equilibrium,
Y st = J(s, t, ν?), and Yt = v(t, ν?).
Non–Markovian extension of earlier results (Ekeland and Lazrak, Bjork etal., Wei et al.).
The equilibrium is not necessarily strict.
Under mild conditions (Lipschitz + integrability) can prove wellposednessof the BSDE system
Dylan Possamaı Time–inconsistent stochastic control
![Page 43: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/43.jpg)
16/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An extended DPPThe characterising BSDE systemVerification theoremExtensions
Verification theorem
Theorem [Hernandez, P. (2019)]
If there exists a sufficiently integrable solution to the BSDE system, thenν?(X ,Z) is an equilibrium,
Y st = J(s, t, ν?), and Yt = v(t, ν?).
Non–Markovian extension of earlier results (Ekeland and Lazrak, Bjork etal., Wei et al.).
The equilibrium is not necessarily strict.
Under mild conditions (Lipschitz + integrability) can prove wellposednessof the BSDE system
Dylan Possamaı Time–inconsistent stochastic control
![Page 44: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/44.jpg)
16/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An extended DPPThe characterising BSDE systemVerification theoremExtensions
Verification theorem
Theorem [Hernandez, P. (2019)]
If there exists a sufficiently integrable solution to the BSDE system, thenν?(X ,Z) is an equilibrium,
Y st = J(s, t, ν?), and Yt = v(t, ν?).
Non–Markovian extension of earlier results (Ekeland and Lazrak, Bjork etal., Wei et al.).
The equilibrium is not necessarily strict.
Under mild conditions (Lipschitz + integrability) can prove wellposednessof the BSDE system
Dylan Possamaı Time–inconsistent stochastic control
![Page 45: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/45.jpg)
16/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An extended DPPThe characterising BSDE systemVerification theoremExtensions
Verification theorem
Theorem [Hernandez, P. (2019)]
If there exists a sufficiently integrable solution to the BSDE system, thenν?(X ,Z) is an equilibrium,
Y st = J(s, t, ν?), and Yt = v(t, ν?).
Non–Markovian extension of earlier results (Ekeland and Lazrak, Bjork etal., Wei et al.).
The equilibrium is not necessarily strict.
Under mild conditions (Lipschitz + integrability) can prove wellposednessof the BSDE system
Dylan Possamaı Time–inconsistent stochastic control
![Page 46: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/46.jpg)
17/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An extended DPPThe characterising BSDE systemVerification theoremExtensions
Extensions
Extension to controlled volatility: 2BSDE system instead of BSDE system.Verification, and DPP still hold. But need more regularity to produceequilibria.
More general time–inconsistency
J(t, t,M) := EPν [ ∫ T
tf (t, r, Xr∧·, νr )dr + G(t, XT∧·)
∣∣∣∣Ft
]+ F
(t, EP
ν[h(XT∧·)|Ft ]
),
including mean–variance: system of 3 coupled (2)BSDEs (need to addEPν [h(XT∧·)|Ft ] as a state).
Dylan Possamaı Time–inconsistent stochastic control
![Page 47: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/47.jpg)
17/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An extended DPPThe characterising BSDE systemVerification theoremExtensions
Extensions
Extension to controlled volatility: 2BSDE system instead of BSDE system.Verification, and DPP still hold. But need more regularity to produceequilibria.
More general time–inconsistency
J(t, t,M) := EPν [ ∫ T
tf (t, r, Xr∧·, νr )dr + G(t, XT∧·)
∣∣∣∣Ft
]+ F
(t, EP
ν[h(XT∧·)|Ft ]
),
including mean–variance: system of 3 coupled (2)BSDEs (need to addEPν [h(XT∧·)|Ft ] as a state).
Dylan Possamaı Time–inconsistent stochastic control
![Page 48: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/48.jpg)
18/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Outline
1 The big pictureMotivating exampleTime–inconsistencyThree approaches
2 What is an equilibrium?In discrete–time, (almost) all is wellNot so much in continuous–time...
3 Main resultsAn extended DPPThe characterising BSDE systemVerification theoremExtensions
4 An application in contract theory
Dylan Possamaı Time–inconsistent stochastic control
![Page 49: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/49.jpg)
19/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
The setting
• Setting similar to Holmstrom and Milgrom (1987). Principal hires Agentto perform a task on his behalf. Agent chooses control α ∈ A, and inducesprobability measure Pα such that
Xt = x +
∫ t
0
αsds + σW αt , t ∈ [0,T ].
• Principal rewards Agent at time T with ξ(X·∧T ). Agent’s criterion is
J(t, α, ξ) := EPα[f (T − t)ξ(X·∧T )−
∫ T
t
f (s − t)α2s
2ds
],
with f (0) = 1
Dylan Possamaı Time–inconsistent stochastic control
![Page 50: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/50.jpg)
19/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
The setting
• Setting similar to Holmstrom and Milgrom (1987). Principal hires Agentto perform a task on his behalf. Agent chooses control α ∈ A, and inducesprobability measure Pα such that
Xt = x +
∫ t
0
αsds + σW αt , t ∈ [0,T ].
• Principal rewards Agent at time T with ξ(X·∧T ). Agent’s criterion is
J(t, α, ξ) := EPα[f (T − t)ξ(X·∧T )−
∫ T
t
f (s − t)α2s
2ds
],
with f (0) = 1
Dylan Possamaı Time–inconsistent stochastic control
![Page 51: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/51.jpg)
20/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
The setting
• Agent looks for equilibria, so that Principal can only offer ξ(X·∧T ) such that atleast one exists. By our results, equilibria exist if and only if we have a solutionto
Yt = ξ(X·∧T ) +
∫ T
t
(Z 2r
2− ∂Y r
r
)dr −
∫ T
t
Zr · dXr ,
Y st = f (T − s)ξ(X·∧T ) +
∫ T
t
ZrZsr − f (r − s)
Z 2r
2dr −
∫ T
t
Z sr · dXr ,
the equilibrium is ν?· := Z·, and J(t, α?, ξ) = Yt = Y tt .
Dylan Possamaı Time–inconsistent stochastic control
![Page 52: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/52.jpg)
21/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
The BSDE system
• We can directly check that
Y st =
f (T − s)
f (T )Y 0
t + f (T − s)EPν?[ ∫ T
t
(f (r)
f (T )− f (r − s)
f (T − s)
)Z 2r
2cdr
∣∣∣∣Ft
].
Notice that when f (t) := e−δt , the second term vanishes −→ this is the effectdue to time–inconsistency.
• This implies also that
Z st =
f (T − s)
f (T )Z 0t + f (T − s)Z s
t ,
where Z s appears in the following martingale representation
Mst := EPν
?[ ∫ T
0
(f (r)
f (T )− f (r − s)
f (T − s)
)Z 2r
2dr
∣∣∣∣Ft
]= Ms
0 +
∫ t
0
Z sr dXr .
Dylan Possamaı Time–inconsistent stochastic control
![Page 53: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/53.jpg)
21/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
The BSDE system
• We can directly check that
Y st =
f (T − s)
f (T )Y 0
t + f (T − s)EPν?[ ∫ T
t
(f (r)
f (T )− f (r − s)
f (T − s)
)Z 2r
2cdr
∣∣∣∣Ft
].
Notice that when f (t) := e−δt , the second term vanishes −→ this is the effectdue to time–inconsistency.
• This implies also that
Z st =
f (T − s)
f (T )Z 0t + f (T − s)Z s
t ,
where Z s appears in the following martingale representation
Mst := EPν
?[ ∫ T
0
(f (r)
f (T )− f (r − s)
f (T − s)
)Z 2r
2dr
∣∣∣∣Ft
]= Ms
0 +
∫ t
0
Z sr dXr .
Dylan Possamaı Time–inconsistent stochastic control
![Page 54: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/54.jpg)
22/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An upper bound for the problem
• Notice that
ξ(X·∧T ) =Y0
f (T )+
∫ T
0
f (r)
f (T )
Z 2r
2dr +
∫ T
0
σZ 0r dW
α?
r .
• Principal wants to solve
supξ∈Ξ
EPα? [
XT − ξ(X·∧T )]
= x − Y0
f (T )+ supξ∈Ξ
EPα?[ ∫ T
0
(Zr −
f (r)
f (T )
Z 2r
2
)dr
]≤ x − Y0
f (T )+
1
2
∫ T
0
f (T )
f (r)dr ,
the optimal Z being Z?t := f (T )/f (t)
• Y0 is fixed to the reservation utility R of Agent.
Dylan Possamaı Time–inconsistent stochastic control
![Page 55: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/55.jpg)
22/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An upper bound for the problem
• Notice that
ξ(X·∧T ) =Y0
f (T )+
∫ T
0
f (r)
f (T )
Z 2r
2dr +
∫ T
0
σZ 0r dW
α?
r .
• Principal wants to solve
supξ∈Ξ
EPα? [
XT − ξ(X·∧T )]
= x − Y0
f (T )+ supξ∈Ξ
EPα?[ ∫ T
0
(Zr −
f (r)
f (T )
Z 2r
2
)dr
]≤ x − Y0
f (T )+
1
2
∫ T
0
f (T )
f (r)dr ,
the optimal Z being Z?t := f (T )/f (t)
• Y0 is fixed to the reservation utility R of Agent.
Dylan Possamaı Time–inconsistent stochastic control
![Page 56: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/56.jpg)
22/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
An upper bound for the problem
• Notice that
ξ(X·∧T ) =Y0
f (T )+
∫ T
0
f (r)
f (T )
Z 2r
2dr +
∫ T
0
σZ 0r dW
α?
r .
• Principal wants to solve
supξ∈Ξ
EPα? [
XT − ξ(X·∧T )]
= x − Y0
f (T )+ supξ∈Ξ
EPα?[ ∫ T
0
(Zr −
f (r)
f (T )
Z 2r
2
)dr
]≤ x − Y0
f (T )+
1
2
∫ T
0
f (T )
f (r)dr ,
the optimal Z being Z?t := f (T )/f (t)
• Y0 is fixed to the reservation utility R of Agent.
Dylan Possamaı Time–inconsistent stochastic control
![Page 57: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/57.jpg)
23/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Verification
• Since Z? is deterministic, Z s = 0, so that
Z 0r =
f (T )
f (T − r)Z?r =
f 2(T )
f (r)f (T − r).
• The previous reasoning gives us the following candidate optimal contract
ξ? :=R
f (T )+
∫ T
0
f (T )
2f (r)− f 2(T )
f 2(r)f (T − r)dr +
∫ T
0
f (T )
f (r)f (T − r)dXr .
• Can check directly that this is an admissible contract which attains the upperbound.
Dylan Possamaı Time–inconsistent stochastic control
![Page 58: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/58.jpg)
23/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Verification
• Since Z? is deterministic, Z s = 0, so that
Z 0r =
f (T )
f (T − r)Z?r =
f 2(T )
f (r)f (T − r).
• The previous reasoning gives us the following candidate optimal contract
ξ? :=R
f (T )+
∫ T
0
f (T )
2f (r)− f 2(T )
f 2(r)f (T − r)dr +
∫ T
0
f (T )
f (r)f (T − r)dXr .
• Can check directly that this is an admissible contract which attains the upperbound.
Dylan Possamaı Time–inconsistent stochastic control
![Page 59: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/59.jpg)
23/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Verification
• Since Z? is deterministic, Z s = 0, so that
Z 0r =
f (T )
f (T − r)Z?r =
f 2(T )
f (r)f (T − r).
• The previous reasoning gives us the following candidate optimal contract
ξ? :=R
f (T )+
∫ T
0
f (T )
2f (r)− f 2(T )
f 2(r)f (T − r)dr +
∫ T
0
f (T )
f (r)f (T − r)dXr .
• Can check directly that this is an admissible contract which attains the upperbound.
Dylan Possamaı Time–inconsistent stochastic control
![Page 60: Time inconsistent stochastic control · In discrete{time, (almost) all is well Not so much in continuous{time... 3 Main results An extended DPP The characterising BSDE system Veri](https://reader033.fdocuments.us/reader033/viewer/2022050100/5f3f607a9b173046827dd2f3/html5/thumbnails/60.jpg)
24/24
The big pictureWhat is an equilibrium?
Main resultsAn application in contract theory
Thank you for your attention!
Dylan Possamaı Time–inconsistent stochastic control