3. ORGANIGRAMA GENERAL DE LA EMPRESA
4.1 FUNCIONES
4.1.3 CUANTIFICACIÓN Y REGISTRO DE OBRA
In games with complete information, it can be shown that ISR can be computed applying Dekel and Fudenberg’s (1990) S1W-procedure to the normal form of the game: That is, one round of deletion of weakly dominated strategies, followed by iter- ated strict dominance.18 As in Dekel and Fudenberg (1990, Section 5), it is instructive
to discuss how the procedure applies to the …nitely repeated prisoners dilemma: Let stage payo¤s be as in the following table, and the game be repeated a …nite number
T of times (players sum payo¤s over periods, with no discounting):
C S
C 10;10 0;11
S 11;0 1;1
For any T <1, all the strategies that prescribeC at the last stage are deleted at the …rst round, because they are not sequentially rational: for any conditional beliefs held at any history of lengthT, actionC is dominated, hence all strategies prescribing that behavior at some of these node are not sequentially rational. (In terms of the normal form, these are weakly dominated strategies.) If T = 2, after the …rst round of deletion, only two strategies survive for each player:
ISR1i =fhS;SSSSi;hC;SSSSig
It is immediate to see that at this point strategyhC;SSSSicannot be a sequential best response to any conjecture i such that i( )
2 ISR1i . (In terms of the
normal form, strategies hC;SSSSi are strictly dominated in ISR1.) Hence, at the second round, hCSSSSi is deleted, and ISRi =fhS;SSSSig: In the unique ISR-
outcome both players always Shirk.
18Proposition A.1 in Appendix A.3 shows that in private-values environments with payo¤s in generic position, ISR can be computed applying the S1W-procedure to the interim (reduced) normal form of the game.
Things change ifT 3: In this case, the deletion procedure stops before delivering the subgame perfect solution. Suppose thatT = 3. Let a1
ijaCCi aCSi aSCi aSSi ja3i denote
player i’s strategy that plays a1
i 2 fC; Sg in the …rst period, a a1
1a12
i 2 fC; Sg in the
second round if(a11a12)was played in the …rst period, and actiona3i 2 fC; Sgin the last period, irrespective of the history. As argued above, after the …rst round of deletion only strategies such thata3
i =S survive. Hence, after the …rst round, we have ISR1i = a1ijaiCCaCSi aSCi aSSi jS : a1i; aCCi ; aiCS; aSCi ; aSSi 2 fC; Sg
5
Applying the S1W-procedure, at the second round strategies that in the second pe-
riod “cooperate no-matter-what”(i.e. strategies a1
ijCCCCCSCSCCSSjS ) are deleted,
because they are strictly dominated by the strategy that only replaces the second pe- riod’s behavior with “shirk no-matter-what” (i.e. a1ijSCCSCSSSCSSSjS ). Hence, the reduced strategies that survive the second round of deletion are:
ISR21 = s 1 1 = C 1 jCCCSCS jS ;s21 = C1jSCCCCS jS ; s31 = C1jSCCSCS jS ;s41 = S1j CSCSSSjS ; s51 = S1j SSCCSSjS ;s61 = S1j SSCSSSjS ISR22 = s 1 2 = C 1 jCCC SSC jS ;s22 = C1jSCC CSC jS ; s32 = C1jSCC SSC jS ;s42 = S1j CCS SSSjS ; s52 = S1j SCS CSSjS ;s62 = S1j SCS SSSjS
The submatrix of the reduced normal form at this point of the procedure is the following 6 6matrix (payo¤s in bold corresponds to best responses):
s1 2 s22 s32 s42 s52 s62 s11 21;21 11;22 11;22 12;12 2;13 2;13 s21 22;11 12;12 12;12 11;22 1;23 1;23 s3 1 22;11 12;12 12;12 12;12 2;13 2;13 s4 1 12;12 22;11 12;12 3;3 13;2 3;3 s51 13;2 23;1 13;2 2;13 12;12 2;13 s6 1 13;2 23;1 13;2 3;3 13;2 3;3
It is easy to see that no further strategies are deleted at this point, because none of them is strictly dominated. Hence, ISRi = fs1i;s2i;si3;s4i;s5i;s6ig. Notice
that that if s4
i were deleted at this round, iterated deletion of strictly dominated
strategies would delete, in order, strategiess1i, thens2i ands3i, and …nallys5i, uniquely selecting the subgame perfect outcome (s6
1;s62). But s4i is not deleted, because it is
only weakly dominated, not strictly. To see how s4
i can be a sequential best response
to conjectures consistent with initial certainty of ISR2, consider 1’s conjectures 1
such that 1( ) [s6
2] = 1. Clearly,s41 is a best response to 1(h)at historiesh= ; SS
and at any h in the last stage: In particular, since SS is consistent with s6
2, and1’s
initial beliefs are concentrated on s6
2, also at SS player 1 is certain of s62. Now, to
see how s41 is a best response at history SC too (we don’t need to check history CS
because CS =2 H(s4
1)), notice that SC =2 H(s62), so receives zero probability under i( ). Hence,
ISR imposes no restrictions on i(SC) (in particular, i(SC) need
not be concentrated on ISR22). So, set i(SC) [s2] = 1 where s2 is a strategy that
prescribes that 2 plays C in the last stage if and only if 1 plays C in the second: Given these conditional beliefs at h = SC, action C at the second period followed by S in the last period is indeed a best response at h = SC for player 1. Notice though that s2 is a dominated strategy, that had been deleted in the …rst round of the procedure. But, as discussed above, once players are surprised (i.e. at unexpected histories), ISRallows them to believe that the opponents may play anything, even
dominated strategies.