Estimating discount rates using referendum-style choice experiments: An analysis of multiple methodologies

There is a growing literature that uses stated preference surveys to estimate discount rates. A review of the literature reveals large variation both in the discount rate estimates coming from different stated preference surveys and in the specific empirical methodologies used to estimate discount rates. While most use similar theory and logic in deriving discount rate estimates, it remains an open question how much of the variation seen in the literature is due to differences in empirical methodology. Using a single data set, we estimate annual discount rates using six different methodologies, including endogenous and ex-post estimation techniques as well as a variety of parametric and nonparametric choice models. We find that most of our estimates are tightly clustered between 14.5 and 31% while one methodology yields an outlier value of 200%. We also use multiple metrics to determine which methodology yields the “right” discount rate and find that the methodology with the best “goodness-of-fit” using information statistics does not always yield the highest predictive accuracy. Our findings suggest that, while methods grounded in similar theory often produce comparable methods, caution and robustness checks are critical.