Monte Carlo evidence on consumer surplus estimation using parametric and semiparametric models

Abstract

Monte Carlo experiments are conducted to investigate consumer surplus derived from parametric and semiparametric models. The experiments are designed to simulate censored and truncated recreation demand data under different error distributions, sample sizes, and censorship rates. Within each experiment, consumer surplus is estimated using observed and predicted demand. Summary statistics are used to determine the direction and magnitude of the error introduced into consumer surplus due to model misspecification, and to examine the robustness of fconsumer surplus derived from semiparametric models. Various implications are clearly evident from the Monte Carlo experiments using Tobit models. First, non-normality (in general) leads to overestimation of consumer surplus when using either observed or predicted demand. Second, heteroscedasticity leads to 20% overestimation of consumer surplus when using predicted demand, but only leads to 3% overestimation when using observed demand. The overall implication of these results stresses the importance of testing for model misspecification when using the Tobit model to estimate consumer surplus. If misspecification is present, researchers need to correct for it in order to arrive at accurate and reliable consumer surplus estimates. Otherwise, the use of a semiparateric model is a viable alternative. Various results are also evident for the Monte Carlo experiments using semiparametric models. First, consumer surplus derived from the symmetrically trimmed least squares model tends to overestimate the simulated consumer surplus in all cases except for two sample designs. Second, consumer surplus computed from the Buckley and James model are nearly identical to the simulated consumer surplus in all cases. The overall implication of this finding supports the use of the Buckley and James model to compute consumer surplus. Finally, results from the Monte Carlo experiments using truncated parametric and semiparametric models are very pessimistic. The results indicate that consumer surplus derived from truncated parametric and semiparametric models tend to greatly overestimate the simulated consumer surplus (i.e., by greater than 70%). The implications of this finding are that researchers either need to cease using on-site sample surveys to estimate consumer surplus, or develop alternative truncated models.