3.7. Presentación de resultados
3.7.1. Resultados de encuestas
1 : Suppose is an RE and de…ne Bi : R+ Htd!R by
Bi( i;t; yi;tjhtw) =E (1 ) i;t+ Uijhtd; i;t; yi;t : (6) Consider an on-path history ht
0. Then jht0 is a Perfect Bayesian Equilibrium of the con- tinuation game. In particular, for each on-pathDt; t; dt, and t, agent iis willing to choose
ei;t only if
ei;t 2argmax ei2R+
E (1 ) i;t+ Uijhtd; i;t; ei;t =ei (1 )c(ei): (7) We can rewrite (7) as
ei;t 2argmax ei2R+
E E (1 ) i;t + Uijhtd; i;t; ei;t =ei; yi;t jhtd; i;t; ei;t =ei (1 )c(ei):
Conditional on yi;t,ei;t does not a¤ect continuation play. So this constraint implies(2): Suppose Bi i;t; yi;tjhtd < E Ui h0t+1 htd; i;t; yi;t at some on-path historyhty. Then agent i may pro…tably deviate by choosing i;t = 0 and earning no less than Ui ht0+1 in the continuation game. So the left-hand side of (3) must hold in any RE. Suppose that the principal refuses to pay i;t 0 following hty. Following this deviation, the re…nement to RE has no bite. We claim the principal’s continuation payo¤ after this deviation is bounded from below by E " ; ht0+1 1 X t0=0
(1 ) t0(yi;t+t0 wi;t+t0 i;t+t0) hty
#
: (8)
To prove this claim, consider the following strategy for the principal following a deviation in i;t. Agent i observes this deviation, but no other agents do. Denote all variables that are observed by at least one agent j 6=i by [j6=i j ht
0 0 . In each period ht 0 0 2 Ht 0 0, the principal plays according to j [j6=i i ht 0
0 , with the sole exception that wi;t0 = i;t0 = 0 for alli2I. This strategy is identical to except for transfer payments. Transfer payments do not a¤ect the continuation game, so this strategy is feasible. Moreover, this strategy and are indistinguishable for every agent j 6=i. The principal’s payo¤ from this strategy equals (8), so this strategy bounds the principal’s payo¤ from below.
The principal is willing to pay i;t 0 only if
(1 )E i;tjhty E
" 1 X
t0=1
(1 ) t0(yi;t+t0 wi;t+t0 i;t+t0) hty
#
: (9)
Adding Ui to both sides of this expression yields
E (1 ) i;t+ Ui( ; ht0+1)jh t
y E Si( ; ht0+1)jh t y :
This inequality must hold a fortiori in expectation. Because htd; i;t, and yi;t are elements of
ht
y, iterating expectations and applying the de…nition of Bi yields the right-hand inequality of (3).
2: We construct a RE from . Recursively de…ne as follows:
1. Ift= 0, thenht;0 =ht0 =;, the unique null history. Otherwise, begin withht0; ht;0 2 Ht0
such that (i) ht
0 is on-path for , (ii) h t;
0 is on-path for , and (iii) ht0 and h t; 0 induce identical continuation games. De…ne jht;0 as follows, where starred variables represent actions played in and unstarred variables represent actions played in .
(a) Following the realization of t and Dt, the principal chooses history hte 2 Het using distribution j fht
0; t; Dtg. The principal chooses dt as in hte. For each agenti2 f1; : : : ; Ng, the principal pays
wi;t =E yi;t 1
1 Bi i;t; yi;tjh t
d Si ; ht0+1 htd; i;t . Notewi;t 0by (3). The principal sends a message to agent i:
mi;t =nht;0 ; ai;t; ei;t; Bi i;t; yi;tjhtd E Si ; ht0+1 jh t
d; i;t; yi;t y i;t 0
o
:
(b) Agent i chooses ai;t; ei;t as in mi;t.
(c) If output is yt, then for each agenti2 f1; : : : ; Ng, (1 ) i;t =Bi i;t; yi;tjh
t
d E Si ; ht0+1 jh t
d; i;t; yi;t
Note that i;t 0 by (3). Given mi;t and the realized yi;t, agent i can perfectly infer i;t.
(d) Letht0+1; be the realized history at the end of periodt. The principal drawsht0+1 2
Ht0+1 from j fhte; ytg. Continuation play jht0+1; is constructed by repeating steps (a)-(d) usinght0+1 and ht0+1; .
2. If agent i observes a deviation, then he takes his outside option and pays no transfers in this and every subsequent period. If the principal observes a deviation, thenmj;t0 =
wj;t0 = j;t0 = 0for each agentj 2 f1; : : : ; Ngin each future period. If agentideviates, the principal chooses dt to min-max agent i. Otherwise, dt is chosen uniformly at random following a deviation.
First, note that past e¤ort choices do not a¤ect current play. Moreover, agents know the true history at the start of a period ht;0 whenever they take actions in that period. Therefore, if we can show that no player has a pro…table deviation from , it immediately follows that jht;0 is a PBE for every on-pathht;0 and hence is a RE. Furthermore, both total continuation surplus andi-dyad surplus for every i2 f1; :::; Ng are identical in jht;0
and jht0 by construction.
First, consider the principal. For any on-path ht;d and each agent i 2 f1; : : : ; Ng, the distribution overyi;t is the same as in jht
d. So
E yi;t wi;t i;t ht;d = 0
and ; ht;d = 0. If the principal deviates in dt, wi;t, or mi;t, then each agent i either knows that this action is a deviation or not. If agent i knows these actions are a deviation, then the principal earns0from agent ibecause that agent rejects production in this and all future periods. If agent i does not know these actions are a deviation, then the principal must announce some on-path history ~htd such that i(~htd) = i(h
t;
d ). But then E [yi;t0
wi;t0 i;t0j~htd] = 0 in every t0 t because h~td is on-path. So the principal cannot pro…tably deviate indt; wi;t, ormi;t. The principal likewise has no pro…table deviation from i;t because
i;t 0. So the principal has no pro…table deviation on the equilibrium path.
Suppose the principal deviates o¤ the equilibrium path. As with the argument in the previous period, if this deviation is detected by agent i then the principal earns 0 from agent i. If this deviation is not detected by agent i, then it must be in dt; wi;t, or mi;t, since i;t >0is always detected as a deviation. But the principal earns payo¤0 following a deviation dt; wi;t, ormi;t by the argument above. So she has no pro…table deviation o¤ the equilibrium path.
Consider agent i. At each on-path history ht;0 , E ui;tjht;0 = E yi;t c(ei;t)jht;0 : SoUi ; ht;0 =Si ; ht;0 . By construction of ,Si ; ht;0 =Si( ; ht0). Sincewi;t 0, agent i has no pro…table deviation from wi;t. After agent i observes an on-path wi;t; mi;t
and chooses ei;t, he earns
E (1 ) i;t+ Si( ; ht0+1; ) h t;
d ; wi;t; mi;t; ei;t c(ei;t);
since he can perfectly infer ht;d from Dt; t; dt; and mi;t. Plugging in the de…nition of i;t yields E hBi i;t; yi;tjh t d E Si ; ht0+1 jh t d; wi;t; yi;t + Si( ;~ht0+1) h t;
d ; wi;t; mi;t; ei;t
i
c(ei;t):
Now, E Bi( i;t; yi;tjhtd)jh t;
d ; wi;t; mi;t; ei;t = E Bi( i;t; yi;tjhtd)jhtd; ei;t because t; dt are the same inht
dandh t;
d . Similarly,E Si( ; ht0+1)jhtd; i;t; yi;t =E Si( ; ht0+1; )jh t;
d ; wi;t; mi;t; yi;t by construction. Therefore, the agent is willing to choose ei;t so long as
ei;t 2arg max ei
E Bi i;t; yi;tjhtd jhtd; ei c(ei):
E¤ort ei;t satis…es this constraint because (2) holds. O¤ the equilibrium path, continuation play is independent of ei;t and so the agent optimally chooses ai;t = 0.
Following any deviation in i;t <0, agent iearns continuation surplus Ui h t+1;
0 . Agent
i observes or infersht;d ; wi;t; mi;t; yi;t in . So agent iis willing to pay i;t <0if (1 ) i;t E Ui ; h0t+1; Ui ht0+1; h
t;
d ; wi;t; mi;t; yi;t Recall that Ui( ; ht0+1; ) =Si( ; ht0+1; ) by construction. Moreover,
E Si ; ht0+1; jh t;
d ; wi;t; mi;t; yi;t =E Si( ; ht0+1)jh t
d; i;t; yi;t
because continuation dyad-surplus in following ht;d ; wi;t; mi;t; yi;t is drawn is drawn ac- cording to j(ht
e; yt). Furthermore, E Ui ht0+1; jh t;
d ; wi;t; mi;t; yi;t = E Ui(ht0+1)jhtd because ht
d and h t;
d induce identical continuation games in periods t+ 1 onwards, and all other variables do not a¤ect the continuation game. Therefore, agent i is willing to pay so long as (1 ) i;t E Si ; ht0+1 h t d; i;t; yi;t E Ui ht0+1 h t d :
i;t depends only on variables that agent i has observed or can perfectly infer from mi;t. Plugging in i;t, agent i is willing to pay so long as
Bi i;t; yi;tjh t d E Si ; ht0+1 h t d; i;t; yi;t E Si ; ht0+1 h t d; i;t; yi;t E Ui ht0+1 h t d
or Bi i;t; yi;tjhtd E Ui ht0+1 htd . This inequality holds by (3). O¤ the equilibrium path, agent i’s payo¤ is independent of i;t and so he chooses i;t = 0. So agent i has no pro…table deviation from i;t, regardless of his beliefs about the true history.
We conclude that is an RE with the desired properties.
Proof of Proposition 1
7.0.1 Proof of Statement 2
De…nition 1. De…ne the transformation
Gi yij ; di;d~i; ei;e~i =Fi 1 Fi yij ;e~i;d~i ; ei; di :
The distribution over outcomesyi has full support, soFi is strictly increasing and henceFi 1 is a function. Fi 1 is continuously di¤erentiable, because Fi is continuously di¤erentiable. Claim 1. The distribution over Gi yij ; di;d~i; ei;~ei induced by ;d~i;e~i is identical to the distribution overyi induced by ( ; di; ei): Gi(yi)j ;d~i;e~i
d
= yij ; di; ei. Proof of Claim 1. To prove the claim, it su¢ ces to show that for every yi,
F yij ;d~i;~ei =Fi Gi yij ; di;d~i; ei;~ei ; di; ei . This is true by de…nition of Gi.
De…nition 2. Fix a distribution overi-dyad surplus. De…ne
ei( ; di; ) = argmax ei
E[yij ; di; ei] c(ei)
subject to: there exists a mapping Si : [0;1) ! R and a reward scheme Bi : [0;1) ! R satisfying
1. E¤ort IC: ei 2argmaxeifE[Bi(yi)j ; di; ei] c(ei)g
2. Dynamic enforcement: for all yi 2[0;1), Ui( ) Bi(yi) Si(yi) 3. Distribution-matching: Sij ; di; ei
d = .
De…nition 3. For monotonically increasing Si : [0;1) ! R, de…ne e^i ; di;d~i; ei Si implicitly by c0(^ei) = Z yi( ;di;ei) 0 Ui( ) @fi @ei yij ;d~i;e^i dyi (1) + Z 1 yi( ;di;ei) Si Gi yij ; di;d~i; ei;e^i @fi @ei yij ;d~i;^ei dyi.
Claim 1. Suppose ; di;d~i; ei are such that di = ~di ande^i( ; di; di; eijSi) =ei. Then^ei is di¤erentiable on a neighborhood about that point.
Proof of Claim 1. Take Si : [0;1)! R to be a monotonically increasing function. The equation (1) may be rewritten as H= 0, where
H Z yi( ;di;ei) 0 Ui( ) @fi @ei yij ;d~i;^ei dyi + Z 1 Gi(yi( ;di;ei)) Si(yi) @Gi 1 @yi @fi @ei Gi 1(yi) ;d~i;^ei dyi c0(^ei):
This expression is continuously di¤erentiable in both d~i and e^i. Therefore, by the Implicit Function Theorem, @^ei
@d~i exists on a neighborhood about ; di;
~
di; ei as long as @H@e^i 6= 0. To show that this is the case, we will bound H from above by a function H that is di¤erentiable in ^ei and strictly decreasing in e^i on a neighborhood about ; di;d~i; ei , and coincides withH ate^i =ei. For " >0, let
H Z yi( ;di;ei) 0 Ui( ) @fi @ei (yij ; di;e^i)dyi + Z yi( ;di;ei)+" yi( ;di;ei) Si(Gi(yij ; di; di; ei;e^i)) @fi @ei (yij ; di;^ei)dyi + Z 1 yi( ;di;ei)+" Si(yi) @fi @ei (yij ; di;e^i) c0(^ei).
At^ei =ei,Gi(yi) =yi, soH =H. For ^ei > ei su¢ ciently close, we claim thatH H. Note that Gi(yij ; di; di; ei;e^i) yi if e^i ei, because Fi is FOSD increasing in ei. Therefore,
Si(Gi(yij ; di; di; ei;^ei)) Si(yi), because Si is monotonically increasing. Further, for ^ei su¢ ciently close to ei, @f@eii (yij ; di;^ei) 0 for yi yi ( ; di; ei) +", because @f@eii is strictly monotonically increasing in yi and equals0 atyi ( ; di; ei) for decision di and e¤ortei. This proves thatH H.
If" = 0, then @H@e^
i <0by CDFC. @H
@^ei is continuous in", becauseGi is continuous inyi, so @H
@^ei <0 for " su¢ ciently close to 0. SoH is such that H =H atei = ^ei,H H for ^ei > ei
su¢ ciently close, and @H@e^
i <0. We conclude that @H @e^i <0.
Claim 2. Let be a surplus-maximizing equilibrium. Then for any on-path history ht d 2
Ht
d, and agent i, let i ( jhtd) be the distribution Si ht0+1 htd induced by . Then ei;t =
ei t; di;t; i ( jhtd) .
Proof of Claim 2. Suppose ei;t > ei t; di;t; i ( jhtd) . It is easy to see that (IC) is a necessary condition for a credible reward scheme to exist. So ei;t satis…es (2) and (3) by
Lemma 1 and induces a distribution over continuation dyad-surplus equal to i ( jht
d). So it must be thatei;t > eF B
i ( t; di;t)by de…nition ofei. By De…nition 2, there exists some mapping
G(yi;t) such that the distribution Si G(yi;t) htd; ei;t t; di;t; i ( jhtd) equals i ( jhtd). De…ne ~ as a strategy that is identical to except following history ht
d. At history
htd, agents j 6= i play as in . Agent i chooses e¤ort ei t; di;t; i ( jhtd) . Following the realization of output yt, agent i’s output yi;t is treated as output G(yi;t), but otherwise continuation play is identical to . Following ht0+1, ~ has a credible reward scheme, be- cause does. In period t, there exists a credible reward scheme that induces agent i to choose ei t; di;t; i ( jhtd) by De…nition 2. Agents j 6= i face marginal distributions over continuation payo¤s that are identical to , so they are willing to choose ej;t as in . At histories ht00 for t0 < t, Sj ~; ht 0 0 = Sj ; ht 0 0 for j 6= i, while Si ~; ht 0 0 Si ; ht 0 0 , becauseei t; di;t; i ( jhdt) leads to strictly higher surplus at history htdthan ei;t. So ~ has a credible reward scheme and so is payo¤-equivalent to an equilibrium, which contradicts being surplus-maximizing.
If ei;t < ei t; di;t; i ( jhdt) , then ei;t < eF Bi ( t; di;t) as well. Then the alternative equilibrium that is identical to except that agenti choosesei t; di;t; i ( jhtd) leads to a strictly higher total surplus, which is a contradiction. Thereforeei;t =ei t; di;t; i ( jhtd) . Claim 3. ei( t; di;t; i ( jhtd)) is weakly increasing indi;t.
Proof of Claim 3. Fix di and di > di. By assumption, for any di di, there exists a