Nicolas Klein The Importance of Being Honest – 1
The Importance of Being Honest
Nicolas Klein University of Bonn
April 11, 2012
Motivation
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Motivation
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 2
US$105 bn spent on the “War on Cancer” from 1971 till 2009.
Motivation
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
US$105 bn spent on the “War on Cancer” from 1971 till 2009.
Too much funding devoted to low-risk, low-yield projects?
(NY Times, June 28, 2009)
Motivation
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 2
US$105 bn spent on the “War on Cancer” from 1971 till 2009.
Too much funding devoted to low-risk, low-yield projects?
(NY Times, June 28, 2009)
Question: Optimal way of giving incentives so that scientists themselves would choose high-risk, high-yield projects?
The Ingredients
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
The Ingredients
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 3
Agent can
● kick back and relax (“shirk”),
● go low-risk (“cheat”),
● investigate an uncertain hypothesis (“be honest”).
The Ingredients
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Agent can
● kick back and relax (“shirk”),
● go low-risk (“cheat”),
● investigate an uncertain hypothesis (“be honest”).
● If honest, agent learns something about the
hypothesis/technology he is supposed to investigate.
The Ingredients
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 3
Agent can
● kick back and relax (“shirk”),
● go low-risk (“cheat”),
● investigate an uncertain hypothesis (“be honest”).
● If honest, agent learns something about the
hypothesis/technology he is supposed to investigate.
Principal
● only observes events agent produces;
The Ingredients
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Agent can
● kick back and relax (“shirk”),
● go low-risk (“cheat”),
● investigate an uncertain hypothesis (“be honest”).
● If honest, agent learns something about the
hypothesis/technology he is supposed to investigate.
Principal
● only observes events agent produces;
● only cares about the first honestly produced event;
The Ingredients
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 3
Agent can
● kick back and relax (“shirk”),
● go low-risk (“cheat”),
● investigate an uncertain hypothesis (“be honest”).
● If honest, agent learns something about the
hypothesis/technology he is supposed to investigate.
Principal
● only observes events agent produces;
● only cares about the first honestly produced event;
● can pay the agent non-negative amounts (limited liability!), conditional on the history he observes.
Alternative Interpretation
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Alternative Interpretation
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 4
Scientist/worker is hired to test a new hypothesis/production method.
Yet, he could manipulate data/secretly use old method.
Alternative Interpretation
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Scientist/worker is hired to test a new hypothesis/production method.
Yet, he could manipulate data/secretly use old method.
Fanelli (2009):
Alternative Interpretation
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 4
Scientist/worker is hired to test a new hypothesis/production method.
Yet, he could manipulate data/secretly use old method.
Fanelli (2009):
● One out of 7 scientists reports their colleagues have falsified data at least once.
Alternative Interpretation
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Scientist/worker is hired to test a new hypothesis/production method.
Yet, he could manipulate data/secretly use old method.
Fanelli (2009):
● One out of 7 scientists reports their colleagues have falsified data at least once.
● Only
< 2%
admit to having done so themselves.Alternative Interpretation
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 4
Scientist/worker is hired to test a new hypothesis/production method.
Yet, he could manipulate data/secretly use old method.
Fanelli (2009):
● One out of 7 scientists reports their colleagues have falsified data at least once.
● Only
< 2%
admit to having done so themselves.● One out of three (72%) admit to lesser distortion of knowledge (by their colleagues).
Alternative Interpretation
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Scientist/worker is hired to test a new hypothesis/production method.
Yet, he could manipulate data/secretly use old method.
Fanelli (2009):
● One out of 7 scientists reports their colleagues have falsified data at least once.
● Only
< 2%
admit to having done so themselves.● One out of three (72%) admit to lesser distortion of knowledge (by their colleagues).
● Note that this is all self-reported survey data!
Alternative Interpretation
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 4
Scientist/worker is hired to test a new hypothesis/production method.
Yet, he could manipulate data/secretly use old method.
Fanelli (2009):
● One out of 7 scientists reports their colleagues have falsified data at least once.
● Only
< 2%
admit to having done so themselves.● One out of three (72%) admit to lesser distortion of knowledge (by their colleagues).
● Note that this is all self-reported survey data!
⇒
Include the option of cheating into models of incentives.Overview of Main Results
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Overview of Main Results
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 5
– Either cheating option makes implementation of honesty impossible, or it leads to no distortions at all.
Overview of Main Results
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
– Either cheating option makes implementation of honesty impossible, or it leads to no distortions at all.
– In the latter case, only reward the
m + 1
-st observable event (m
appropriately large).Overview of Main Results
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 5
– Either cheating option makes implementation of honesty impossible, or it leads to no distortions at all.
– In the latter case, only reward the
m + 1
-st observable event (m
appropriately large).– Only honest agent believes he can produce many future events.
Overview of Main Results
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
– Either cheating option makes implementation of honesty impossible, or it leads to no distortions at all.
– In the latter case, only reward the
m + 1
-st observable event (m
appropriately large).– Only honest agent believes he can produce many future events.
– Hence, principal will construct a continuation phase so that agent will only ever want to enter after an honest success.
Some Stylized Facts on Innovation Performance
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 6
Some Stylized Facts on Innovation Performance
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
⇒
Focussing incentives on the long term helps innovation.Some Stylized Facts on Innovation Performance
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 6
⇒
Focussing incentives on the long term helps innovation.Effort is a necessary condition for innovative breakthroughs.
But is it sufficient?
Some Stylized Facts on Innovation Performance
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
⇒
Focussing incentives on the long term helps innovation.Effort is a necessary condition for innovative breakthroughs.
But is it sufficient?
Francis, Hasan, Sharma, 2009: No impact of performance sensitivity of CEO pay on firm’s innovation performance.
Some Stylized Facts on Innovation Performance
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 6
⇒
Focussing incentives on the long term helps innovation.Effort is a necessary condition for innovative breakthroughs.
But is it sufficient?
Francis, Hasan, Sharma, 2009: No impact of performance sensitivity of CEO pay on firm’s innovation performance.
But: Skewing incentives toward the long term does have a positive and significant impact on the number of patents and citations to patents (Francis, Hasan, Sharma, 2009, and
Lerner & Wulf, 2007).
Some Stylized Facts on Innovation Performance
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
⇒
Focussing incentives on the long term helps innovation.Effort is a necessary condition for innovative breakthroughs.
But is it sufficient?
Francis, Hasan, Sharma, 2009: No impact of performance sensitivity of CEO pay on firm’s innovation performance.
But: Skewing incentives toward the long term does have a positive and significant impact on the number of patents and citations to patents (Francis, Hasan, Sharma, 2009, and
Lerner & Wulf, 2007).
Rewarding long-term performance seems crucial in spurring innovation.
Literature Overview
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature
Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 7
Literature Overview
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature
Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
● Holmstr ¨om & Milgrom (1991): multi-tasking agent;
different tasks can be monitored more (less) accurately;
no learning.
Literature Overview
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature
Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 7
● Holmstr ¨om & Milgrom (1991): multi-tasking agent;
different tasks can be monitored more (less) accurately;
no learning.
● Bergemann & Hege (1998, 2005), H ¨orner & Samuelson (2009): venture-capital financing; no “faking” of success.
Literature Overview
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature
Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
● Holmstr ¨om & Milgrom (1991): multi-tasking agent;
different tasks can be monitored more (less) accurately;
no learning.
● Bergemann & Hege (1998, 2005), H ¨orner & Samuelson (2009): venture-capital financing; no “faking” of success.
● Fong (2009): Optimal Scoring Rules for Surgeons;
“cheating” is part of the model but learning is not.
(Surgeons know their type perfectly.)
Literature Overview
Introduction
●Introduction
●Ingredients
●Interpretation
●Overview
●Some Stylized Facts
●Literature
Setup
Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 7
● Holmstr ¨om & Milgrom (1991): multi-tasking agent;
different tasks can be monitored more (less) accurately;
no learning.
● Bergemann & Hege (1998, 2005), H ¨orner & Samuelson (2009): venture-capital financing; no “faking” of success.
● Fong (2009): Optimal Scoring Rules for Surgeons;
“cheating” is part of the model but learning is not.
(Surgeons know their type perfectly.)
● Manso (2011): Two-period binomial model.
Here: Can make additional statements on structure of incentive scheme, and optimal stopping.
The Basic Setup
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
The Basic Setup
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 8
One principal, one agent (both risk neutral).
The Basic Setup
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
One principal, one agent (both risk neutral).
Time continuous; finite horizon
T < ∞
; end date if noobserved event
T < T
(at first exogenous, to be endogenized later); common discount rater > 0
.The Basic Setup
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 8
One principal, one agent (both risk neutral).
Time continuous; finite horizon
T < ∞
; end date if noobserved event
T < T
(at first exogenous, to be endogenized later); common discount rater > 0
.Agent operates a bandit machine, with
The Basic Setup
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
One principal, one agent (both risk neutral).
Time continuous; finite horizon
T < ∞
; end date if noobserved event
T < T
(at first exogenous, to be endogenized later); common discount rater > 0
.Agent operates a bandit machine, with
● a safe arm; private benefit flow of
s > 0
;The Basic Setup
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 8
One principal, one agent (both risk neutral).
Time continuous; finite horizon
T < ∞
; end date if noobserved event
T < T
(at first exogenous, to be endogenized later); common discount rater > 0
.Agent operates a bandit machine, with
● a safe arm; private benefit flow of
s > 0
;● arm 0 (“cheating”); “successes” after exponentially
distributed times according to a known parameter
λ
0k
0,t;The Basic Setup
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
One principal, one agent (both risk neutral).
Time continuous; finite horizon
T < ∞
; end date if noobserved event
T < T
(at first exogenous, to be endogenized later); common discount rater > 0
.Agent operates a bandit machine, with
● a safe arm; private benefit flow of
s > 0
;● arm 0 (“cheating”); “successes” after exponentially
distributed times according to a known parameter
λ
0k
0,t;● arm 1 (“honesty”); successes after exponentially
distributed times according to the parameter
θλ
1k
1,t;λ
1 known,θ ∈ { 0, 1 }
initially unknown state of the world.The Basic Setup
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 8
One principal, one agent (both risk neutral).
Time continuous; finite horizon
T < ∞
; end date if noobserved event
T < T
(at first exogenous, to be endogenized later); common discount rater > 0
.Agent operates a bandit machine, with
● a safe arm; private benefit flow of
s > 0
;● arm 0 (“cheating”); “successes” after exponentially
distributed times according to a known parameter
λ
0k
0,t;● arm 1 (“honesty”); successes after exponentially
distributed times according to the parameter
θλ
1k
1,t;λ
1 known,θ ∈ { 0, 1 }
initially unknown state of the world.Principal wants to implement use of arm 1 until the first breakthrough (possibly all the way up to
T
).Learning
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Learning
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 9
Whenever using arm 1, agent updates his private belief according to Bayes’ rule (conditional on no success):
ˆ
p
t= p
0e
−λ1∫0tk1,τ dτp
0e
−λ1∫0tk1,τ dτ+ 1 − p
0,
Learning
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Whenever using arm 1, agent updates his private belief according to Bayes’ rule (conditional on no success):
ˆ
p
t= p
0e
−λ1∫0tk1,τ dτp
0e
−λ1∫0tk1,τ dτ+ 1 − p
0,
or, equivalently,
p ˙ˆ
t= −λ
1k
1,tp ˆ
t( 1 − p ˆ
t)
.Learning
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 9
Whenever using arm 1, agent updates his private belief according to Bayes’ rule (conditional on no success):
ˆ
p
t= p
0e
−λ1∫0tk1,τ dτp
0e
−λ1∫0tk1,τ dτ+ 1 − p
0,
or, equivalently,
p ˙ˆ
t= −λ
1k
1,tp ˆ
t( 1 − p ˆ
t)
.In equilibrium, the principal knows the agent’s belief:
p
t= p ˆ
t= p
0e
−λ1tp
0e
−λ1t+ 1 − p
0.
Learning
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Whenever using arm 1, agent updates his private belief according to Bayes’ rule (conditional on no success):
ˆ
p
t= p
0e
−λ1∫0tk1,τ dτp
0e
−λ1∫0tk1,τ dτ+ 1 − p
0,
or, equivalently,
p ˙ˆ
t= −λ
1k
1,tp ˆ
t( 1 − p ˆ
t)
.In equilibrium, the principal knows the agent’s belief:
p
t= p ˆ
t= p
0e
−λ1tp
0e
−λ1t+ 1 − p
0.
After the first success,
p
t= 1
.The Principal’s Instruments
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 10
The Principal’s Instruments
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
The principal has two tools: Payments and choosing the end date
T ˇ ( t ) ∈ [ t, T )
conditional on the first event havingoccurred at time
t
. Formally,The Principal’s Instruments
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 10
The principal has two tools: Payments and choosing the end date
T ˇ ( t ) ∈ [ t, T )
conditional on the first event havingoccurred at time
t
. Formally,● Point processes
N
t1 andN
t0: number of events produced on arm 1 (0) up to, and including, timet
.The Principal’s Instruments
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
The principal has two tools: Payments and choosing the end date
T ˇ ( t ) ∈ [ t, T )
conditional on the first event havingoccurred at time
t
. Formally,● Point processes
N
t1 andN
t0: number of events produced on arm 1 (0) up to, and including, timet
.●
( N
t0, N
t1)
0≤t≤T induces filtrationF ∶= { F
t}
0≤t≤T.The Principal’s Instruments
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 10
The principal has two tools: Payments and choosing the end date
T ˇ ( t ) ∈ [ t, T )
conditional on the first event havingoccurred at time
t
. Formally,● Point processes
N
t1 andN
t0: number of events produced on arm 1 (0) up to, and including, timet
.●
( N
t0, N
t1)
0≤t≤T induces filtrationF ∶= { F
t}
0≤t≤T.●
{ N
t}
0≤t≤T (withN
t∶= N
t0+ N
t1) induces filtrationF
N∶= { F
Nt}
0≤t≤T.The Principal’s Instruments
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
The principal has two tools: Payments and choosing the end date
T ˇ ( t ) ∈ [ t, T )
conditional on the first event havingoccurred at time
t
. Formally,● Point processes
N
t1 andN
t0: number of events produced on arm 1 (0) up to, and including, timet
.●
( N
t0, N
t1)
0≤t≤T induces filtrationF ∶= { F
t}
0≤t≤T.●
{ N
t}
0≤t≤T (withN
t∶= N
t0+ N
t1) induces filtrationF
N∶= { F
Nt}
0≤t≤T.● At
t = 0
, the principal commits to a non-decreasing, non-negative,F
N-adapted process of payments{W
t}
0≤t≤T, and to a schedule of end dates{ T ˇ ( t )}
0≤t≤T,with
W
t: time-0 value of cumulated payments to agent up to timet
.A Simplification
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 11
A Simplification
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Clearly not a good idea to pay the agent in the absence of a event. Can simplify notation!
A Simplification
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 11
Clearly not a good idea to pay the agent in the absence of a event. Can simplify notation!
h
t∶= e
rt(W
t− lim
τ↑tW
τ)
: immediate reward for first event;A Simplification
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Clearly not a good idea to pay the agent in the absence of a event. Can simplify notation!
h
t∶= e
rt(W
t− lim
τ↑tW
τ)
: immediate reward for first event;w
t: timet
expected continuation value of an agent who has had a breakthrough on arm 1 at timet
;A Simplification
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 11
Clearly not a good idea to pay the agent in the absence of a event. Can simplify notation!
h
t∶= e
rt(W
t− lim
τ↑tW
τ)
: immediate reward for first event;w
t: timet
expected continuation value of an agent who has had a breakthrough on arm 1 at timet
;ω
t( p ˆ
t)
: timet
expected continuation value of an agent who has had a “breakthrough” on arm 0 at timet
while holding the (private) beliefp ˆ
t.A Simplification
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Clearly not a good idea to pay the agent in the absence of a event. Can simplify notation!
h
t∶= e
rt(W
t− lim
τ↑tW
τ)
: immediate reward for first event;w
t: timet
expected continuation value of an agent who has had a breakthrough on arm 1 at timet
;ω
t( p ˆ
t)
: timet
expected continuation value of an agent who has had a “breakthrough” on arm 0 at timet
while holding the (private) beliefp ˆ
t.The principal’s objective is to minimize
∫
0 Te
−rt−λ1∫0tpτ dτp
tλ
1( h
t+ w
t) dt
The Agent’s Strategies & Objective
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 12
The Agent’s Strategies & Objective
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
By choosing a strategy, the agent influences the distribution over
( N
t0, N
t1)
t and( N
t)
t.The Agent’s Strategies & Objective
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 12
By choosing a strategy, the agent influences the distribution over
( N
t0, N
t1)
t and( N
t)
t.The agent chooses a process
( k
0,t, k
1,t)
t that isF
-predictable, whereThe Agent’s Strategies & Objective
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
By choosing a strategy, the agent influences the distribution over
( N
t0, N
t1)
t and( N
t)
t.The agent chooses a process
( k
0,t, k
1,t)
t that isF
-predictable, where●
k
0,t≥ 0
,k
1,t≥ 0
, andk
0,t+ k
1,t≤ 1
;The Agent’s Strategies & Objective
Introduction Setup
●Setup
●Learning
●The Principal’s Tools
●Notation
●Strategies
●No Surprise Continuation Scheme Before the First Breakthrough Optimal Stopping Conclusion Appendix
Nicolas Klein The Importance of Being Honest – 12
By choosing a strategy, the agent influences the distribution over
( N
t0, N
t1)
t and( N
t)
t.The agent chooses a process
( k
0,t, k
1,t)
t that isF
-predictable, where●
k
0,t≥ 0
,k
1,t≥ 0
, andk
0,t+ k
1,t≤ 1
;●
k
i,t: fraction devoted to armi
at timet
;●