Insurance for Low Probability Risks: Experimental Evidence

1
Insurance for Low Probability Risks Experimental
Evidence

• Presentation to CESIR, Stanford University
• May 12, 1999
• by Philip Ganderton
• Economics Department, UNM

2
Outline

• Introduction
• Experiment design
• Model
• Data
• Analysis
• Discussion
• Conclusions

3
Introduction

• Economically non-rational behavior observed
• Bi-modal distribution (McClelland, Schulze, et
  al.)
• Theory is weak
• Experiments obvious source of data
• Objectives
• investigate EU in experiments
• estimate WTP(insurance)

4
Extensive Form of Game
5
Event Probabilities
6
Experiment Design

• Private Good
• Session
• Treatment
• Period (receive income, decide to buy)
• Round (experience event, loss)

7
Empirical Model

• buy insurance if C lt EL RP
• C cost
• EL expected loss f(event pr., loss pr., loss
  amt)
• RP risk premium
• Pr(buy) f(C, EL, R)
• R risk attitudes

8
Data

• 13,179 individual decisions
• 449 subjects
• (extra information on 149 subjects)
• 90 treatment combinations
• subject makes 4.3 decisions each treatment
• dependent var. binary buy decision
• indep. vars include loss and prob. values

9
Analysis

• Logit for all observations
• Logit for sub-sample of 149 subjects
• Alternative Specifications
• no random effects
• individual risk measure a fixed effect
• Model Performance
• Table 3

10
Discussion

• Table 4 - Predicted Probabilities of Buying
  Insurance
• No difference across losses (columns)
• Increasing high prob. event likelihood lowers
  buying likelihood by 28.
• Increasing both event likelihoods increases
  buying likelihood by 80

11
Willingness to Pay for Insurance

• Cameron (1988) method
• Mean WTP of 47, median WTP of 4.4
• Can derive Bid/Eloss distributions (Figure 3)
• No bi-modality.

12
Conclusions

• Some evidence in support of EUT
• Support cost , loss , wealth -
• Failure exposure to events -, prob. more
  important than loss amount
• Willingness to Pay calculated
• no bi-modality