*ATTac-2001* makes flight bidding decisions based on a cost-benefit
analysis: in particular, *ATTac-2001* computes the incremental cost of
postponing bidding for a particular flight versus the value of
delaying commitment. In this section, we describe the determination of
the cost of postponing the bid.

Due to difficulties that compounded with more sophisticated
approaches, *ATTac-2001* used the following very simple model for estimating
the price of a flight ticket at a given future time. It is evident from
the formulation that--given --the expected
price increase from time to time was very nearly
of the form for some
. It was also clear that as long as the price did not hit the
artificial boundaries at $150 and $800, the constant must depend
linearly on . This linear dependence coefficient was then
estimated from several hundred flight price evolutions during the
qualifying round. Thus, for this constant , the expected price
increase from time to time was
. When a price
prediction was needed, this formula was first used for the first and
most recent actual price observations to obtain a guess for , and
then this was used in the formula again to estimate the future
price. No change was predicted if the formula yielded a price decrease.

This approach suffers from systemic biases of various kinds (mainly due to the fact that the variance of price changes gets relatively smaller over longer periods of time), but was thought to be accurate enough for its use, which was to predict whether or not the ticket can be expected to get significantly more expensive over the next few minutes.

In practice, during TAC-01, *ATTac-2001* started with the *flight-lookahead* parameter set to 3 (*i.e.*, cost of postponing is the
average of the predicted flight costs 1, 2, and 3 minutes in the
future). However, this parameter was changed to 2 by the end of the
finals in order to cause *ATTac-2001* to delay its flight commitments
further.