Presentation on theme: "1 Chiappori & Salanie on contracts, G604, Rasmusen, April 30, 2003."— Presentation transcript:
1 Chiappori & Salanie on contracts, G604, Rasmusen, April 30, 2003
2 RAND Health Insurance Experiment Cost $130 Million in 1984 dollars. 14 insurance plans, with lump sum payment to make all participants benefit Going from a free plan to 25% co-payment reduced outpatient services; bigger copayments had no further effect Manning et al
3 Ausubel (1999) Bank Credit Cards People who responded to bank credit card mailings were worse credit risks, as expected People who responded to worse offers (higher interest rates, etc.) were worse credit risks. This last finding persisted even after controlling for observable characteristics. Is this evidence for adverse selection? An alternative, not mentioned by CS, is that dumb people respond to the ads, especially the worst ones, and dumb people are bad credit risks. How could that hypothesis be tested?
4 Dionne & St. Michel (1991): Quebec Co- insurance The Quebec public insurance plan became more generous, exogenously. Number of days of compensation (days off work?) increased, but only for cases that are hard to diagnose (e.g., back pain)
5 Fortin et al. (1994): Canadian Workmens Compensation and Unemployment Insurance Workmens comp became more generous in Quebec. This led to longer-lasting injuries. Injuries are longer-lasting at the end of the construction season (when workers would become unemployed anyway). Unemployment insurance became less generous. This also led to longer-lasting injuries, but just of the kind that are hard to diagnose. Implication: injured workers become more reluctant to return to the labor market
6 Wolak (1994)– Mechanism Design Fancy Estimation The discussion here, on page 16, is bad. None of the notation is useful. All that CS are saying is that Wolak set up likelihood functions for a model with symmetric info and a model with asymmetric info, and tested for which was better. They do make some good points, including that this kind of brute-force maximum likelihood estimation is very vulnerable to specifying the model wrong. OLS is better because it is more like descriptive statistics--- you can interpret it as just conditional correlations.
7 Paarsch & Shearer (2000)– Tree planting piece rates Wages(piecerates) = BaseLabor + Utility cost of effort We want to estimate (Utility Cost of Effort + Efficiency Gains from Effort) Q(piecerates) = Conditions A + BaseLabor + Utility cost of effort + Efficiency Gains from Effort Q(flat wage) = ConditionsB + BaseLabor Difference = Conditions A - Conditions B +Utility cost of effort + Efficiency Gains from Effort Wages(piecerates) = BaseLabor + Utility cost of effort Wages(flat) = BaseLabor Difference = Utility cost of effort
8 Chiappori, Abbring, Heckman and Pinquet (2001) French car insurance data, from a company The price can rely on experience of the insured person, but only in a regulated way: it must change linearly in years without an accident and bump up a constant amount after each year with an accident. Two effects of having an accident: less moral hazard but more info that the person is a bad driver
9 Managerial Pay Jensen and Murphy (1990): For each $1000 increase in firm value,the CEO gets $3.25 more data. Hall-Liebman (1998): this goes up to $5.3 median for , and $25 average. Haubrich (1994) reasonable levels of risk aversion generate the JM result. Remember: $3.25/1000 results in huge swings in compensation Puzzle: Why not use pay based on performance of the firm relative to the industry? Puzzle: Bertrand and Mullainathan (2000) find that pay reacts as much to firm performance predictable from luck measures as from unpredictable changes that we would attribute to a manager.