C2922 Economics Utility Functions
T.C. Johnson October 30, 2007

1

Introduction

“Utility” refers to the perceived value of a good and utility theory spans mathematics, economics and psychology. For example, if you prefer vanilla ice-cream to chocolate, you would assign greater utility to vanilla ice-cream than to the same quantity of chocolate ice-cream. The fact that different agents have different utilities for goods is the basis of all markets. In the context of actuarial science, the focus is on the utility of money, the study of the utility of money started in the early 1700’s with the St Petersburg Paradox. 1.0.1 The St Petersburg Paradox A player pays some fee to a casino in order to play a game where a fair coin will be tossed repeatedly until a ”tail” first appears, ending the game. The payoff of the game is 2N −1 where N is the number of times until the first tail appears. How much should the casino charge to play this game, how much should a player expect to play the game? This game was played in 18th century St Petersburg and numerous mathematicians, including Bernoulli were interested in it, since the expected payoff of the game is infinite
∞

E[X] = k=1 2

k−1

1 = 2k

∞

k=1

1 2

1

Utility functions In 1728, a Swiss mathematician, Gabriel Cramer, observed that “the mathematicians estimate money in proportion to its quantity, and men of good sense in proportion to the usage that they may make of it”.

2

Modern psychology explains utility in terms of how we perceive wealth, for example • A beggar will value a £1 much more than a millionaire. A pound may double the wealth of the beggar where as a millionaire would never notice the loss of a pound. • The expectation of an uncertain outcome is worth less than a certain outcome. • Future earnings are perceived as being worth less than current earnings, you do not know what your future situation will be (i.e. dead). • Money in your pocket is worth more than money invested. Money in your pocket has immediate utility, it can be spent. Money invested cannot be spent immediately. The third of these examples are usually handled by discounting (time value of money), if the investment in the forth example is risk free, it would be handled simply by discounting as well. In actuarial mathematics we are particularly interested in uncertain outcomes, which treats utility of wealth (the perceived value of wealth) as a random variable.

2

Utility functions

Mathematically we need a function to map between the physical measure of money and the perceived value of money. Such functions are called utility functions, and in the context of wealth being a random variable on a probability space, they need to be measurable functions on that space, and hence, utility functions are random variables. Bernoulli was the first to suggest a utility function in 1738 as an solution to the St Petersburg Paradox. The theory was developed in its modern form by von Neumann and Morgenstern in 1944. They developed the axioms underlying utility theory, in a synthesis of economics and probability, as

Utility functions

3

Independence of different utility functions (associated with the fact that utility functions are random variables). Completeness all outcomes are assigned a utility. Transitivity if A is preferred to B, and B is preferred to C, then A is preferred to C. Continuity of utility (if wealth is continuous). However, the basic attribute of a utility function is that it an increasing function, everyone would value more money over less money, so u (x) > 0. The fact that u (x) = 0 means that there is non-satiation, i.e. the agent never becomes completely satisfied and will always prefer more to less. In the example of the beggar and the millionaire, we can see that as wealth increases, each additional £1 has a lower perceived value. This is not surprising and is known as decreasing marginal utility, that is u (x) ≤ 0. Consider Figure 1. u(x) u(x + ∆x)

u(x)

u(x − ∆x)

∆x

∆x

x

Figure 1. It is clear that if u (x) ≤ 0 then u(x + ∆x) − u(x) ≤ u(x) − u(x − ∆x), and so each additional unit of wealth increases utility by a smaller amount, the marginal utility, u(x + ∆x) − u(x) decreases. A consequence of decreasing marginal utility is risk aversion, and an agent with decreasing marginal utility would value certain outcomes over uncertain ones, this effect will be demonstrated in the section on Risk Aversion.

Utility functions

4

2.1

Examples α > 0 (or if you want positive utility

The most common utility functions are • Exponential u(x) = −e−αx , u(x) = 1 − e−αx , α > 0. • Log, u(x) = log x • Power, u(x) = • Iso-elastic u(x) = x1−ρ , 1−ρ ρ XA with probability (1−pA ). To participate ¯ in this investment the investor would have to pay X = pA XA + (1 − pA )XB , i.e. the investment opportunity is ‘fair’. If the utility function is concave, ¯ Figure 2 shows that the utility of X is greater than the utility of the expected value of the investment, pA u(XA ) + (1 − pA )u(XB ) u(x) u(XB ) ¯ u(X) pA u(XA ) +(1 − pA )u(XB ) u(XA )

XA

¯ X

XB

x

Figure 2 ¯ This means that the investor would prefer to keep X rather than the expected pay-off of the investment. This result is a consequence of Jensen’s inequality.

Utility functions

9

Definition 4.1 (Jensen’s inequlity) For any random variable X and any function g that is strictly concave, that is g (x) < 0, then E[g(X)] ≤ g(E[X]) Equality holds if and only if X is constant. Jensen’s inequality tells us that, if the utility is concave, the expected utility is less than the utility of the expected value. If the gamble is fair, that is the cost of participating in the gamble is equal to its expected outcome, the utility associated with this cost is greater than the expected utility of the winnings. Therefore the loss of utility, associated with buying the gamble, is more than the expected gain in utility provided by the possible winning.

4.1

Measures of Risk Aversion

From Figure 2 you may be able to see that the curvature of a utility function, the way the marginal utility changes, is important since it determines how an agent acts in the face of uncertainty, and is known as risk aversion. There are two ways to measure the level of risk aversion, absolute risk aversion (ARA) Ra (x) = − and relative risk aversion, (RRA) Rr (x) = −x u (x) . u (x) u (x) u (x)

Absolute risk aversion measures risk aversion to a loss in absolute terms where as relative risk aversion measures aversion to a loss relative to agents wealth. An agent’s allocation of their wealth to risky assets depends on the risk aversion characteristics of their utility function in the following way. • If an agent has increasing absolute risk aversion (IARA), then as wealth increases they will hold fewer pounds in risky assets. • If an agent has constant absolute risk aversion (CARA), as wealth increases they will the same number of pounds in risky assets.

Utility functions

10

• If an agent has decreasing absolute risk aversion (DARA), then as wealth increases they will hold more pounds in risky assets. • If an agent has increasing relative risk aversion (IRRA), then as wealth increases they will hold a lower percentage of their wealth in risky assets. • If an agent has constant relative risk aversion (CRRA), as wealth increases they will the same percentage of their wealth in risky assets. • If an agent has decreasing relative risk aversion (DRRA), then as wealth increases they will hold a higher percentage of their wealth in risky assets. CRRA matches empirical observations more than CARA, for example as your wealth increases would you expect to invest the same absolute value or the same proportion in the risky asset? Generally, we would expect rational agent’s to invest a greater absolute amount in the risky asset as they become richer. Exponential utility is unique is exhibiting constant absolute risk aversion (CARA), while log utility exhibits DARA, decreasing absolute risk aversion (which is consistent with experimental data). You could also construct functions that exhibit increasing absolute risk aversion. The ”iso-elastic” utility function exhibits constant relative risk aversion (CRRA) with Rr (x) = ρ. If ρ = 1, then the iso-elastic utility function is equivalent to log utility function. Traditionally risk aversion was considered to be always rational, however Kahneman and Tversky, in 1979, showed that when faced with losses a riskseeking utility function would be rational. To see this, observe that when faced with a choice of being given £100 with certainty or a lottery to win £1000 with a probability of 0.1 and nothing with probability 0.9, it is rational to take the £100. However, when faced with a loss of £100 with certainty or a lottery to lose £1000 with a probability of 0.1 and nothing with probability 0.9, it is rational to take the gamble. This is consistent with increasing marginal utility, a loss of £1, 001 is not that different from a loss of £1, 000, but a loss of £2 is very different to a loss of £1. There are now cognitive explanations for increasing and decreasing marginal utility of wealth, which will lead to risk tolerant or risk seeking behaviour.

Utility functions

11

4.2

Summary

Utility functions map the “real world” value of money onto an agent’s perceived value. They are increasing functions, and if you are considering gains, the marginal utility is decreasing.

4.3

Exercises
(a) exponential utility exhibits CARA and IRRA. (b) log utility exhibits DARA and CRRA. What utility function would be better at reflecting normal behaviour?

1. Show that

2. What type of absolute risk aversion do the utility functions u(x) = exhibit. 2α 1 − 2αx and u(x) = βx

5

Insurance

Consider a risk averse agent (that is the utility function is concave), with initial wealth x, who faces a random loss L, which will occur with probability pL , and they can insure against this loss by buying insurance for a premium P < L. Under what conditions will the risk averse agent purchase insurance? The agent should base their decision on expected utility, in particular if the expected utility of paying for insurance and not suffering the loss is more than the expected utility of bearing any losses, they should insure, E[U (x − p)] > E[U (x − L)] and since x and p are deterministic, this reduces to U (x − p) > E[U (x − L)].

Utility functions u(x) u(X − P )

12

The expected utility without insurance will lie on this line. u(X − L)

X −L

X −P

X

W ealth

Figure 3 The expected utility without insurance, E[U (x − L)], will lie along the line indicated in Figure 3, its position being determined by pL (the higher pL the closer expected utility without insurance will be to U (x−L)). The maximum premium the agent will pay, pmax is given by U (x − pmax ) = E[U (x − L)] Given Jensen’s inequality(the expected utility is less than the utility of the expected value), for the rhs we have E[U (x − L)] ≤ U (E[x − L]) and so, given x is deterministic, U (x − pmax ) ≤ U (x − E[L]). Since u is increasing, this implies that pmax ≥ E[L]. Remark 5.1 A risk averse agent will always be happy to pay an insurance premium greater than the expected loss covered by the insurance. To calculate pmax you first find the expected utility without insurance, E[U (x − L)]. This gives you U (x − pmax ), and inverting you have x − pmax , this procedure is shown in Figure 4.

Utility functions u(x) 13

Agent’s utility function

E[u(X − L)]

X − L X − Pmax E[X − L]

X

W ealth

Figure 4 (note that since pmax and x are known and deterministic E[x − pmax ] = x − pmax ) The insurer, with wealth w, faces a similar problem. They calculate the minimum premium, pmin that they would accept as E[U (w + pmin − L)] = U (w). Using Jensen’s inequality (the expected utility is less than the utility of the expected value), we calculate that U (w) = E[U (w + pmin − L)] ≤ U (w + pmin − E[L]) ⇒ U (w + pmin − E[L]) ≥ U (w) Since u is strictly increasing, this implies that pmin − E[L] ≥ 0, and so pmin ≥ E[L]. Competition in insurance markets implies that insures should charge the minimum principle, this is called the zero utility principle. Example 5.1 A risk averse agent, whose utility is given by U (x) = ln x and wealth is £50, 000 is faced with a potential loss of £10, 000 with a probability of pl = 0.1. What is the maximum premium they would be willing to pay to protect themselves against this loss? What is the minimum premium that an insurer, with the same utility function and with wealth £1, 000, 000, be willing to charge to cover this loss?

Utility functions The agent’s expected wealth without insurance is E[WL ] = (50, 000 × 0.9) + (40, 000 × 0.1) = 49, 000 while their expected utility of wealth without insurance is E[U (WI )] = (ln(50, 000) × 0.9) + (ln(40, 000) × 0.1) = 10.797463... To identify the maximum premium we need to equate these two, ln(50, 000 − pmax ) = 10.797463.. or pmax = = = = 50, 000 − e10.797463 50, 000 − e10.797463 50, 000 − 48, 869.64 1, 103.39

14

Note that pmax = 1, 103.39 > 1, 000, the expected loss. The insurer will set pmin by equating U (w) = E[U (w + pmin − L)] ln(1, 000, 000) = 0.9 × ln(1, 000, 000 + pmin ) + 0.1 × ln(1, 000, 000 + pmin − 10, 000) ⇒ pmin ≈ 1, 004.52 Observe that the insurer had initial wealth £10, 000, 000 then pmin ≈ 1, 000.45, showing that the wealthier the insurer, the better they are able to compete for business.