I use the Delta Method implemented with the Wald command in LIMDEP to estimate confidence intervals for ratios of regression coefficients (i.e., willingness-to-pay):

fn1=-b1/b2$

Most others seem to use the more difficult to program Krinsky-Robb (K-R) simulations (10,000 draws and all that) and I've always felt a persistent sense of inadequacy or a tendency to self-diminishment, sometimes resulting in excessive aggressiveness through overcompensation (reference). So, one day, I asked a colleague (whose name I'll withhold to protect his identity) to compare the Delta method with the K-R. Here is the resulting email message (slightly edited to correct my unnamed colleague's numerous and substantive errors):

Per trip mean = $747

95% CI = $559 - $1,032

Using the Wald Command, I get the following:

[Per trip mean =] $729

... This gives 95% CI $508 to $951, so narrower confidence intervals than the Krinsky-Robb procedure.