next up previous
Next: Entertainment Tickets Up: Bidding Strategy Previous: Flights


When in the passive mode, ATTac-2000 bids in the hotel auctions either to try to win hotels cheaply should the auction close early, or to try to prevent the hotel auctions from closing early. It might be advantageous to prevent a hotel auction from closing if no rooms are currently desired in order to keep open the option of switching to that hotel should future market prices warrant it.

For each hotel room of type i (such as ``Grand Hotel, night 3''), let Hi be the number of rooms of type i needed for G*. Based on the current price of i, Pi, ATTac-2000 tries to acquire n rooms where

\begin{displaymath}n = \left\{\begin{array}{cl}
8 & \mbox{if\ } P_i = 0 \mbox{\ ...
...0 \\
\max(H_i,1) & \mbox{if\ } P_i \leq 50.

If ATTac-2000's outstanding bids would already win n rooms should the auction close at the current price, then ATTac-2000 does nothing: should the auction close prematurely, ATTac-2000 wins the nrooms cheaply, and competitors lose the opportunity to get any rooms of type i later in the game. Otherwise, ATTac-2000 bids for n rooms at $1 above the current ask price. The formula for computing n was selected so as to risk wasting up to $40-$50 per room type for the benefit of maintaining flexibility later in the game. The exact parameters here were chosen in an ad-hoc fashion without detailed experimentation. Our intuition is that ATTac-2000's performance is not very sensitive to their exact values.

In the active mode, ATTac-2000 bids on hotel rooms based on their marginal value within allocation G*. Let V(G*) be the value of G* (the income from all clients, minus the cost of the yet-to-be-acquired goods). Let ${G^*}_c^\prime$ be the optimal allocation should client c fail to get its hotel rooms. Note that ${G^*}_c^\prime$ might differ from G* in the distribution of entertainment tickets as well as in the flights and hotels of client c. ATTac-2000 bids for the hotel rooms assigned to client c in G* at a price of $V({G^*})-
V({G^*}_c^\prime)$. Since at this point flights are a sunk cost, this price tends to be more than $1000.

Notice that ATTac-2000 bids the full marginal utility for each hotel room required by the client's travel package. An alternative would have been to divide the marginal utility over the number of rooms in the package, which would have eliminated the risk of spending more on hotels than the itinerary is worth. On the other hand, failing to win a single hotel room is enough to invalidate the entire itinerary. ATTac-2000 bids the full marginal utility to maximize the chance that valid itineraries are obtained for all clients. In a combinatorial auction, the bidder would be able to be place a bid for the conjunction of the desired rooms and would therefore not need to choose between these two alternatives.

next up previous
Next: Entertainment Tickets Up: Bidding Strategy Previous: Flights
Peter Stone