An Introduction to Experimental Economics and Insurance Experiments J. Todd Swarthout
One possible way of figuring out economic laws... is by controlled experiments.... Economists (unfortunately )... cannot perform the controlled experiments of chemists or biologists because they cannot easily control other important factors. Like astronomers or meteorologists, they generally must be content largely to observe. (Samuelson and Nordhaus, 1985, p. 8)
Experimental vs. Non experimental Common view that some disciplines are inherently experimental, and some are not History does not support this (e.g., Physics in Aristotle s time vs. Galileo s time) A discipline becomes experimental when innovators develop appropriate techniques for experimentation
Early Economics Experiments Individual choice (Thurstone 1931) Experimentally determined individuals indifference curves (Allais 1953) Experimentally observed violations of expected utility theory Game Theory (Flood 1952) repeated Prisoner s dilemma showed subjects do not always play Nash equilibrium 1952 conference The design of Experiments in Decision Processes Industrial Organization and Markets (Chamberlain 1948) decentralized bilateral exchange did not result in competitive outcomes (Smith 1962) assessed whether competitive outcomes would be observed by changing the price making rules in Chamberlain s basic design
Growth of Experimental Economics Experiments now common in many areas of Economics (individual choice, markets, game theory, social preferences, etc.) Increasing publications (surveys by Holt 2006, Noussair 2011) 2002 Nobel Prize in Economics awarded to experimentalists
Data sources Laboratory Happenstance Experimental Field e.g., rate of inflation e.g., income maintenance experiments e.g., discovery of penicillin e.g., laboratory asset market experiments
Data sources Laboratory Happenstance Experimental Field e.g., rate of inflation e.g., income maintenance experiments e.g., discovery of penicillin e.g., laboratory asset market experiments
Purposes of Experiments Testing (and refining) theories Elicitation of preferences e.g., goods, risk, time, fairness Establish empirical regularities as a basis for new theories Theory free comparison of institutions Wind tunnel experiments Exploring boundedly rational behavior Teaching experiments
Realism Initial instinct may be to think experiment must closely resemble the field Alternatively, some may think experiment must exactly replicate a formal model Both approaches are effectively impossible, and further are not needed Judge experiment by impact on our understanding, not by strict adherence to reality or formal model
Control & induced value Sufficient conditions for experimental control: 1. Nonsatiation (more of reward is better) 2. Saliency (payoff differences between alternatives) 3. Dominance (rewards dominate any subjective participation costs)
Validity of results Internal validity Do the data permit correct causal inferences? External validity Can we generalize inferences from lab to field? Parallelism Lab results apply to nonlab settings where similar ceteris paribus conditions hold
Let s now run our own experiment
Insurance for Low Probability Events
Underinsurance of low probability high loss events Several field studies report that people underinsure against low probability, high loss events such as: Floods Fires Earthquakes Windstorms
Suggested reasons Lack of information Consumers lack information (e.g., probability of loss, average loss amounts, load factors, etc.) Transactions costs Costly effort required to obtain optimal insurance Faith in the government Belief that the government will step in and provide coverage in the event of a natural disaster Decision making heuristics Low probabilities are processed as essentially zero Insurance price is too high
Laboratory evidence Slovic, Fischhoff, Corrigan, and Combs (1977) Journal of Risk and Insurance Subjects were more willing to insure as probability of loss increased (with constant EV) Cited hundreds of times Used hypothetical (non salient) rewards Laury, McInnes, and Swarthout (2009) Journal of Risk and Uncertainty re examined this issue and found some interesting results
Slovic et al. replication 80 70 Pecent of Insurance Purchase 60 50 40 30 20 SFLCC Replication 10 0 0.0001 0.001 0.005 0.01 0.05 0.1 0.25 0.5 Probability of Loss (holding expected value of loss as one)
New experiment Changes Salient rewards Earned monetary endowment Non abstract wording Results Insurance purchase behavior was the opposite of Slovic et al. Hypothetical treatment resulted in no trend
A Laboratory Investigation of Index Insurance
Index insurance Index insurance pays according to an index value and not an actual loss suffered by an insured person The index is Objectively verifiable Non manipulative highly correlated with the insured risk Overcomes traditional transactions costs and informational problems
Examples of index insurance Index insurance programs have been introduced to cover many different weather risks, including: Drought Excess rainfall Flooding Hurricanes Humidity Area yield Price fluctuation Livestock mortality Hail Crop disease Frost earthquake 22 Source: International Fund for Agricultural Development and World Food Programme. 2010. Potential for scale and sustainability in weather index insurance for agriculture and rural livelihoods, by P. Hazell, J. Anderson, N. Balzer, A. Hastrup Clemmensen, U. Hess and F. Rispoli. Rome.
Claims of low take up rates Demand for index insurance appears low, and many reasons are offered in the literature: Financial illiteracy Price (Cole et al. 2011) Lack of trust (Cole et al. 2011) Basis risk Pre existing informal risk sharing (Mobarak and Rosenzweig 2012) Form of utility and degree of risk aversion (Clarke 2011) 23
Making sense of it all There is no general consensus in the literature for why index insurance purchase is low Further, there is debate whether this low demand is a rational response, or instead an irrational behavior to correct How can we say what a low take up rate is, without an appropriate baseline for comparison? 24
A lab experiment a relatively inexpensive tool for investigating this behavior Idiosyncratic field influences can be avoided, allowing more controlled investigation of the basic behavioral phenomenon Will not replace field experiments, rather inform and guide subsequent field experiments
Results Data do not support the stylized fact of lower rates of index insurance purchase. Subjects were more likely to purchase index insurance in presence of basis risk relative to the control treatment of no basis risk A better understanding of this lab behavior should ultimately lead to a better understanding of field behavior
Let s now look at your results 1 EV=.15 1 EV=.30 1 EV=.60 0.9 0.9 0.9 0.8 0.8 0.8 Proportion of Insurance Purchases 0.7 0.6 0.5 0.4 0.3 0.2 0.7 0.6 0.5 0.4 0.3 0.2 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.1 0.1 0 p=.01 p=.1 0 p=.01 p=.1 0 p=.01 p=.1