How stable and predictable are welfare estimates using recreation demand models?

Economic analysis of environmental policy projects typically use pre-existing welfare estimates that are then transferred over time to the policy relevant periods. Understanding how stable and predictable these welfare estimates are over time is important for applying these estimates in policy. Yet, revealed preference models of recreation demand have received few temporal stability assessments compared to other non-market valuation methods. We use a large administrative dataset on campground reservations covering ten years to study temporal stability and predictability of recreation demand welfare estimates of lake water quality changes. Based on single-year models, our findings suggest welfare estimates are temporally stable across years in around 50% of the comparisons. Using an event study design, we find evidence that welfare estimates are stable within a year, that is, for weeks after a change in water quality. Our findings further reveal that having two years of data for predicting welfare estimates in subsequent years improves the prediction accuracy by 22% relative to using a single year of data, but further improvements in the prediction accuracy are modest when including additional years of data. Predictions of welfare estimates are not necessarily improved when using data closer in time to the prediction year. We discuss the implications of our results for using revealed preference studies in policy analysis.