both LHS expressions
must be upward-sloping curves at the respective optima to ensure minimization of W, which
means that the higher curve is associated with the smaller L solution. On the one hand, the
presence of the term −θˆf(L), which is negative by Proposition 1, tends to make the secondbest
curve from (32) lower than the first-best curve, which in turn tends to make ˆL larger
than the L∗. The intuitive explanation comes from recalling that the conditional sequential
frequency solution is inefficiently low, with ˆf(L) < f∗(L). Given ˆf(L) > 0, it follows that
the planner can use his second-best L choice to raise f from its inefficiently low level, an
intervention that would require setting L above its first-best value. On the other hand, the
fact that ˆf(L) < f∗(L) makes the initial term in the second-best expression from (32) larger
(less negative) than the corresponding term in the above first-best expression, which tends to
make the second-best curve higher than the first-best curve.
Despite this apparent indeterminacy, once the f∗ and ˆf functions are substituted into (29)
and (32) and appropriate manipulations are made, it can be shown that ˆL < L ∗ holds (see
the appendix). This conclusion means that the latter effect discussed above (which depresses
ˆL
) dominates the goal of using ˆL to boost the airlines’ inefficiently low flight frequency. With
20
ˆL < L∗, it follows from (15) and (28) that ˆf(ˆL) < f∗(L∗), so that the second-best flight
frequency under sequential choice is inefficiently low. In addition, using ˆL < L∗, it can be
shown that n∗ < ˆn = ˆL/ˆf(ˆL), where ˆn is the second-best optimal n value (see the appendix).
Summarizing yields26
Proposition 5. The second-best cumulative noise limit under sequential choice is
less than the first-best limit (ˆL < L
∗
). In addition, the associated second-best flight
frequency is lower than the first-best level, and the second-best per-aircraft noise level
is higher than the first-best level.