Working Paper: NBER ID: w2507
Authors: Richard Arnott; Joseph E. Stiglitz
Abstract: It is by now well-known that, in the presence of moral hazard or adverse selection, randomization of insurance premia and benefits may be Pareto efficient. This paper: i) provides a typology of the various forms that randomization may take; ii) derives necessary and/or sufficient conditions for the desirability of these various forms of randomization; iii) obtains some simple characterization theorems of the efficient random policies; iv) gives some intuition behind the results; and v) considers why randomization appears to occur less often in practice than the theory suggests it should.
Keywords: randomization; asymmetric information; moral hazard; adverse selection; insurance
JEL Codes: D82; G22
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| randomization of insurance premiums and benefits (G22) | Pareto efficiency (D61) |
| randomization mitigates moral hazard (D82) | individuals exert more effort in accident prevention (J28) |
| effort levels in accident prevention (J28) | overall welfare (I31) |
| desirability of randomization (C90) | balancing welfare gains from reduced moral hazard and welfare losses from increased risk (G52) |
| conditions for randomization to be desirable (C90) | convexity of the first-order condition for effort and degree of risk aversion (D11) |
| randomization as a self-selection device (C90) | insurers differentiate between high- and low-risk individuals (G52) |
| complexity of contracts, costs of enforcement, and lack of trust (D86) | rarity of randomization in practice (C90) |