Working Paper: NBER ID: w22369
Authors: Liran Einav; Amy Finkelstein; Paul Schrimpf
Abstract: A large literature in empirical public finance relies on “bunching” to identify a behavioral response to non-linear incentives and to translate this response into an economic object to be used counterfactually. We conduct this type of analysis in the context of prescription drug insurance for the elderly in Medicare Part D, where a kink in the individual’s budget set generates substantial bunching in annual drug expenditure around the famous “donut hole.” We show that different alternative economic models can match the basic bunching pattern, but have very different quantitative implications for the counterfactual spending response to alternative insurance contracts. These findings illustrate the importance of modeling choices in mapping a compelling reduced form pattern into an economic object of interest.
Keywords: Bunching; Health Insurance; Medicare Part D; Spending Responses
JEL Codes: D12; G22
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| kink in the Medicare Part D budget set (H51) | significant bunching in annual drug expenditures (H51) |
| significant bunching in annual drug expenditures (H51) | behavioral response to the nonlinear incentives (D91) |
| different economic models (E10) | vastly different quantitative implications for spending elasticity (D12) |
| static Saez-style model (C51) | elasticity estimates approximately five times lower than dynamic model (C51) |
| modeling choices (C52) | significantly affect counterfactual predictions and interpretation of elasticity estimates (C51) |