Definition
4: unassigned bidders and the bottom slot
Adjacent allocated-rank no-envy → also no bottom-slot envy
by every unassigned bidder: require
for every unassigned bidder
,
where
and
are the bottom position’s click-through rate and per-click price, and
is that bidder’s value. This restricts the equilibria covered by Lemma 5
and Theorem 7’s revenue comparison.
Counterexample to the printed condition: one
position with click rate
,
values
,
and bids
is a GSP Nash equilibrium with vacuous adjacent-rank no-envy. The winner
pays zero; the loser cannot profitably outbid
,
but values the position at its recorded zero price. Thus the assignment
is unstable, and its revenue
is below the VCG revenue
.
The added inequality excludes it. Checking only the first unassigned
bidder would not control lower bidders whose values need not follow bid
order.
Lemma 6: positive winner
surplus
Every stable assignment → stable assignments with strictly
positive utility for every winner. This added domain condition
makes the Appendix’s constructed bids strictly decreasing, preserving
the allocation and winner payments under strict bid order.
Obstruction without strict surplus: take click
rates
,
values
,
and assign the first two bidders at per-click prices
.
Utilities are
:
the first bidder gets only
in the lower position, the second is indifferent to the upper position,
and the third loses from either. The assignment is stable, but exact GSP
payment realization forces both the second and third bids to equal
.
Under random tie ordering, the
value-
bidder prefers bidding slightly less to risking a loss. This
demonstrates the strict-bid obstruction.
Unspecified bid lower bound → real-valued bids.
With multiple unassigned bidders, the strictly decreasing auxiliary bid
tail can be negative. A nonnegative-bid version would need a
positive-tail condition; it is outside this implementation result.
Theorem 8: symmetric
continuation plans
Bidder-specific continuous strategies
→ one common continuation plan instantiated at every bidder.
Here
is the number of remaining bidders,
the auction history, and
bidder
’s
value. Effective dropout-action uniqueness is proved only among
equilibria in this symmetric class. Uniqueness over arbitrary
bidder-specific profiles remains unproved; the scope restriction
supplies no counterexample to it.