Case studies of how (poorly designed) insurance creates moral hazard

During yesterday’s class meeting, we discussed (among other things) how contract designs and pricing strategies can “fix” the moral hazard that insurance might otherwise create. Insurance is “good” to the extent that it enables firms and individuals to manage the risks that they face. However, we also saw insurance has a potential “dark side.” The dark side is that too much insurance and/or incorrectly priced insurance can create moral hazard by insulating firms and individuals from the financial consequences of their decision-making. Thus, in real world insurance markets, we commonly observe partial rather than full insurance coverage. Partial insurance ensures that policyholders have incentives to mitigate risk. Furthermore, real world insurance markets are characterized by pricing strategies such as loss-sensitive premiums (commonly referred to as “experience rated” premiums), as well as premiums that are contingent upon the extent to which policyholders invest in safety.

In competitively structured private insurance markets, we expect that the market price for insurance will (on average) be greater than or equal to its actuarially fair value. Under normal circumstances, one does not expect to observe negative premium loadings in the real world. Negative premium loadings are incompatible with the survival of a private insurance market, since this would imply that insurers are not able to cover capital costs and would, therefore, have incentives not to supply such a market.

Which brings us to the National Flood Insurance Program (NFIP). The NFIP is a federal government insurance program managed by the Federal Emergency Management Agency (also known as “FEMA”). According to Cato senior fellow Doug Bandow’s blog posting entitled “Congress against Budget Reform: Voting to Hike Subsidies for People Who Build in Flood Plains”,

“…the federal government keeps insurance premiums low for people who choose to build where they otherwise wouldn’t. The Congressional Research Service figured that the government charges about one-third of the market rate for flood insurance. The second cost is environmental: Washington essentially pays participants to build on environmentally-fragile lands that tend to flood.”

Thus, the NFIP provides us with a fascinating case study concerning how subsidized flood insurance exacerbates moral hazard (i.e., makes moral hazard even worse) rather than mitigates moral hazard. It does this by encouraging property owners to take risks (in this case, building on environmentally fragile lands that tend to flood) which they otherwise would not be inclined to take if they had to pay the full expected cost of such risks.

There are many other examples of moral hazard created by insurance subsidies. Consider the case of crop insurance provided to farmers by the U.S. Department of Agriculture.  The effective premium loading on federally provided crop insurance is typically quite negative (often in excess of -60%), thus putting crop insurance on a similar footing to flood insurance in terms of cost compared with actuarially fair value. Just as mis-priced flood insurance effectively encourages property owners to build in flood plains, mis-priced crop insurance incentivizes farmers to cultivate acreage which may not even be particularly fertile.

I could go on (probably for several hundred more pages – there are innumerable other egregious examples which I could cite), but I think I will stop for now…

Moral Hazard

During yesterday’s class meeting, we discussed the concept of moral hazard. In finance, the moral hazard problem is commonly referred to as the “agency” problem. Many, if not most real-world contracts involve two parties – a “principal” and an “agent”. Contracts formed by principals and agents also usually have two key features: 1) the principal delegates some decision-making authority to the agent and 2) the principal and agent decide upon the extent to which they share risk.

The principal has good reason to be concerned whether the agent is likely to take actions that may not be in her best interests. Consequently, the principal has strong incentives to monitor the agent’s actions. However, since it is costly to closely monitor and enforce contracts, some actions can be “hidden” from the principal in the sense that she is not willing to expend the resources necessary to discover them since the costs of discovery may exceed the benefits of obtaining this information. Thus, moral hazard is often described as a problem of “hidden action”.

Since it is not economically feasible to perfectly monitor all of the agent’s actions, the principal needs to be concerned about whether the agent’s incentives line up, or are compatible with the principal’s objectives. This concern quickly becomes reflected in the contract terms defining the formal relationship between the principal and the agent. A contract is said to be incentive compatible if it causes principal and agent incentives to coincide. In other words, actions taken by the agent usually also benefit the principal. In practice, contracts typically scale agent compensation to the benefit received by the principal. Thus, in insurance markets, insurers are not willing to offer full coverage contracts; instead, they offer partial insurance coverage which exposes policyholders to some of the risk that they wish to transfer. In turn, partial coverage reinforces incentives for policyholders to prevent/mitigate loss.

Similarly, in a completely different setting, consider the principal/agent relationship which exists between the owner and manager of a business. If the manager’s effort level is high, then the owner may earn higher profits compared with when the manager’s effort level is low. However, if managerial pay consists of a fixed salary and lacks any form of incentive compensation (e.g., bonuses based upon meeting or beating specific earnings targets), then the manager may be inclined to not exert extra effort, which results in less corporate profit. Thus, compensation contracts can be made more incentive compatible by including performance-based pay in addition to a fixed salary. This way, the owner and manager are both better off because incentives are better aligned.

Empirical evidence of risk aversion in the real world: Small stocks, Large stocks, Government bonds, and Treasury bills, 1926-2017

Although there may be various social contexts in which people stray from risk averse behavior (e.g., the risk loving behavior which is on display whenever people place bets on gambles with unfair odds), in other (more economically consequential settings), it does appear that risk averse behavior is more the rule rather than the exception. Indeed, risk aversion is what motivates people to buy insurance and diversify risk in their asset holdings.

The financial markets provide us with a superb example of risk averse behavior writ large. Historically, here are the long run (1926-2017) compound annual returns on stocks, bonds, and bills that are traded in U.S. financial markets (source: page 9 of http://bit.ly/sbbi2018):

Risk for these various asset classes is lowest for Treasury bills, a bit higher for Government bonds, a bit higher yet for Large stocks, and highest for Small stocks. If you are risk averse, then if one asset has higher risk than another, you are not willing to invest in the riskier asset unless you can reasonably expect that on average, you’ll be compensated for bearing the extra risk in the form of a higher expected return, and it turns out that this is exactly what happens in the real world. If investors were to act in a risk neutral fashion, then the average returns wouldn’t be all that different from each other. Finally if investors were to act in a risk loving fashion, they’d pay more for risky assets than for safe; this would cause risky assets to be bid up in value relative to safe assets, which in turn would imply lower average returns for risky than for safe assets.

Recap of Analytic and Numerical Proofs of the Bernoulli Principle and the Mossin Theorem

Yesterday, we discussed three particularly important insurance economics concepts: 1) the Bernoulli Principle,  2) the Mossin Theorem, and 3) the Arrow Theorem.  The Bernoulli Principle implies that if an actuarially fair full coverage insurance policy is offered, then an arbitrarily risk averse individual will purchase it. The Mossin Theorem implies that if insurance is actuarially unfair, then an arbitrarily risk averse individual will prefer partial insurance coverage over full insurance coverage.  Finally, the Arrow Theorem implies an arbitrarily risk averse individual will select deductible insurance from the menu of partial insurance choices that were considered during class yesterday.

In what follows, I present a recap of the analytic and numerical proofs of the Bernoulli Principle and Mossin Theorem which took up most of our attention in class yesterday.

The Bernoulli Principle is graphically illustrated in the following figure (taken from p. 4 in the Insurance Economics lecture note):

Here, the consumer has initial wealth of $120 and there is a 25% chance that a fire will occur which (in the absence of insurance) will reduce her wealth from$120 to $20. The expected loss is E(L) = .25(100) =$25, which is the actuarially fair price for a full coverage insurance policy. Thus if she remained uninsured, she would have E(W) = .25(20) + .75(120) = $95, and her utility would be EU. Note that the (95, EU) wealth/utility pair corresponds to point C in the figure above. Now suppose the consumer purchases an actuarially fair full coverage insurance policy. Since she is fully insured, she is no longer exposed to any risk; her net worth is$95 irrespective of whether the loss occurs, her utility is U(95), and her wealth/utility pair is located at point D in the figure above. Since U(95) > EU, she will fully insure. More generally, since this decision is optimal for an arbitrarily risk averse consumer, it is also optimal for all risk averse consumers, irrespective of the degree to which they are risk averse.

Next, we introduced coinsurance. A coinsurance contract calls for proportional risk sharing between the consumer and the insurer. The consumer selects a coinsurance rate $\alpha$, where $\alpha$ represents the proportion of loss covered by the insurer. By definition, $\alpha$ is bounded from below at 0 and from above at 1. Thus, if the consumer selects $\alpha$ = 0, this means that she does not purchase any insurance (i.e., she self-insures). If she selects $\alpha$ = 1, this implies that she obtains full coverage. Furthermore, the price of a coinsurance contract is equal to the price of a full coverage insurance contract (P) multiplied by $\alpha$. On pp. 6-7 of the Insurance Economics lecture note, I analytically (via the calculus) confirm the Bernoulli principle by showing that if the consumer’s utility is U = W.5 and insurance is actuarially fair, then the value for $\alpha$ which maximizes expected utility is $\alpha$ = 1. An Excel spreadsheet called the “Bernoulli and Mossin Spreadsheet” is available from the course website which enables students to work this same problem using Solver. I recommend that you download this spreadsheet and use Solver in order to validate the results for a premium loading percentage (β) equal to 0 (which implies a full coverage (actuarially fair) premium P = E(L)(1+β) =$25 x (1+0) =$25) and β = 0.6 (which implies an actuarially unfair premium of $40, the analytic solution for which is presented on p. 8 in the Insurance Economics lecture note). Let’s use Solver to determine what the optimal coinsurance rate ($\alpha$) is for U = W.5 when β = 0 (i.e., when insurance is actuarially fair and full coverage can be purchased for the actuarially fair premium of$25). To do this, open up the Bernoulli and Mossin Spreadsheet and invoke Solver by selecting “Data – Solver”; here’s how your screen will look at this point:

This spreadsheet is based upon the so-called “power utility” function U = Wn, where is 0 < n < 1. Since we are interested in determining the optimal coinsurance rate for a consumer with U = W.5, we set cell B2 (labeled “exponent value”) equal to 0.500. With no insurance coverage (i.e., when $\alpha$ = 0) , we find that E(W) = $95 and E(U(W)) = 9.334. Furthermore, the standard deviation (σ) of wealth is$43.30. This makes sense since the only source of risk in the model is the risk related to the potentially insurable loss.

Since we are interested in finding the value for $\alpha$ which maximizes E(U(W)), Solver’s “Set Objective” field must be set to cell B1 (which is a cell in which the calculated value for E(U(W)) gets stored) and Solver’s “By Changing Variable Cells” field must be set to cell B3 (which is the cell in which the coinsurance rate $\alpha$ gets stored). Since insurance is actuarially fair (β = 0 in cell B4), the Bernoulli Principle implies that the optimal value for $\alpha$ is 1.0. You can confirm this by clicking on Solver’s “Solve” button:

Not only is $\alpha$ = 1.0, but we also find that E(W) = $95, σ = 0 and E(U(W)) = 9.75. Utility is higher because this risk averse individual receives the same expected value of wealth as before ($95) without having to bear any risk (since σ = 0 when $\alpha$ = 1.0).

Next, let’s determine what the optimal coinsurance rate ($\alpha$) is for U = W.5 when β = 0.60 (i.e., when insurance is actuarially unfair). As noted earlier, this implies that the insurance premium for a full coverage policy is $25(1.60) =$40. Furthermore, the insurance premium for partial coverage is $$\alpha$40. Reset $\alpha$ ’s value in cell B3 back to 0, β’s value in cell B4 equal to 0.60, and invoke Solver once again: On p. 8 in the Insurance Economics lecture note, we showed analytically that the optimal coinsurance rate is 1/7, and this value for $\alpha$ is indicated by clicking on the “Solve” button: Since $\alpha$ = 1/7, this implies that the insurance premium is (1/7)40 =$5.71 and the uninsured loss in the Fire state is (6/7)(100) = $85.71. Thus in the Fire state, state-contingent wealth is equal to initial wealth of$120 minus $5.71 for the insurance premium minus for$85.71 for the uninsured loss, or $28.57, and in the No Fire state, state-contingent wealth is equal to initial wealth of$120 minus $5.71 for the insurance premium, or$114.29. Since insurance has become very expensive, this diminishes the benefit of insurance in a utility sense, so in this case, only a very limited amount of coverage is demanded. Note that E(U(W)) = 9.354 compared with E(U(W)) = 9.334 if no insurance is purchased (as an exercise, try increasing β to 100%; you’ll find in that case that no insurance is demanded (i.e., $\alpha$ = 0).

I highly recommend that students conduct sensitivity analysis by making the consumer poorer or richer (by reducing or increasing cell B5 from its initial value of 120) and more or less risk averse (by lowering or increasing cell B2 from its initial value of 0.500). Other obvious candidates for sensitivity analysis include changing the probability of Fire (note: I have coded the spreadsheet so that any changes in the probability of Fire are also automatically reflected by corresponding changes in the probability of No Fire) as well as experimenting with changes in loss severity (by changing cell C8).

Arrow-Pratt method vis–à–vis the “exact” method for calculating risk premiums

I received an email from a Finance 4335 student earlier today asking for further clarification of the two methods for calculating risk premiums which we covered in class last Thursday. Under the so-called “exact” method, one 1) calculates expected utility, 2) sets expected utility equal to the utility of the certainty-equivalent of wealth, 3) solves for the certainty-equivalent of wealth, and 4) obtains the risk premium by calculating the difference between expected wealth and the certainty-equivalent of wealth.  On the other hand, the Arrow-Pratt method is an alternative method for calculating the risk premium which is based upon Taylor series approximations of expected utility of wealth and the utility of the certainty equivalent of wealth (the derivation for which appears on pp. 16-18 of http://fin4335.garven.com/fall2018/lecture6.pdf). Both of these approaches for calculating risk premiums are perfectly acceptable for purposes of Finance 4335.

The value added of Arrow-Pratt is that it analytically demonstrates how risk premiums depend upon two factors: 1) the magnitude of the risk itself (as indicated by variance), and 2) the degree to which the decision-maker is risk averse. For example, we showed in class on Thursday that the Arrow-Pratt coefficient for the logarithmic investor (for whom U(W) = ln W) is twice as large as the Arrow-Pratt coefficient for the square root investor (for whom U(W) = W.5); 1/W for the logarithmic investor compared with .5/W for the square root investor. Thus, the logarithmic investor behaves in a more risk averse than the square root investor; other things equal, the logarithmic investor will prefer to allocate less of her wealth to risky assets and buy more insurance than the square root investor. Another important insight yielded by Arrow-Pratt (at least for the utility functions considered so far in Finance 4335) is the notion of decreasing absolute risk aversion (DARA). Other things equal,  an investor with DARA preferences become less (more) risk averse as wealth increases (decreases).  Furthermore, such an investor increases (reduces) the dollar amount that she is willing to put at risk as she becomes wealthier (poorer).

How Do Energy Companies Measure the Temperature? Not in Fahrenheit or Celsius

Instead of Fahrenheit or Celsius, a metric called “degree days” is used to capture variability in temperature. The risk management lesson here is that this metric makes it possible to create risk indices which companies can rely upon for pricing and hedging weather-related risks with weather derivatives.

Fewer Households in Hurricane Florence’s Path Have Flood Insurance Than in 2013

Not only is Hurricane Florence a highly destructive natural catastrophe; it will also apparently be a financial catastrophe for the many thousands of households who, for whatever reason, lack flood insurance coverage.

How Hurricane Florence Could Move Insurance Markets

Hurricane Florence provides a particularly timely and compelling case study of the economic consequences of natural catastrophes; specifically, the nexus of direct and indirect effects upon property insurance markets, reinsurance markets, alternative risk markets (e.g., catastrophe bonds), and public policy.

Some hurricanes are worse than others — both for people in the way and the insurance industry that tries to understand storms and put a price on their risks.

Confirmation bias in the form of “information avoidance”

This article from the Wall Street Journal provides an interesting follow-up to yesterday’s behavioral finance discussion. “Information avoidance” represents a particularly strong (and potentially deadly) form of confirmation bias!

Getting past information avoidance to deal with health issues, financial difficulties and other worries.