40 3 Utility Maximisation on Finite Probability Spaces

proportionality relation fails to hold. Then one immediately deduces from a

marginal variation argument that the investment of the agent cannot be opti-
mal. Indeed, by investing a little more in the more favourable Arrow asset and
a little less in the less favourable one, the economic agent can strictly increase
her expected utility under the same budget constraint. Hence for the optimal
investment the proportionality must hold true. The above result also identifies
the proportionality factor as y = u (x), where x is the initial endowment of
the investor. This marginal utility of the indirect utility function u(x) also
allows for a straightforward economic interpretation.
Theorem 3.1.3 indicates an easy way to solve the utility maximisation
problem at hand: calculate v(y) using (3.19), which reduces to a simple one-
dimensional computation. Once we know v(y), the theorem provides easy
formulae to calculate all the other quantities of interest, e.g., X T (x), u(x),

u (x) etc.
Another message of the previous theorem is that the value function x →
u(x) may be viewed as a utility function as well, sharing all the qualitative
features of the original utility function U . This makes sense economically, as
the “indirect utility” function u(x) denotes the expected utility of an agent
with initial endowment x, when optimally investing in the financial market S.
Let us now give an economic interpretation of the formulae for u (x) in item
(iii) along these lines: suppose the initial endowment x is varied to x + h, for
some small real number h. The economic agent may use the additional endow-
ment h to finance, in addition to the optimal pay-off function X T (x), h units
of the numéraire asset, thus ending up with the pay-off function X T (x) + h at
time T . Comparing this investment strategy to the optimal one corresponding
to the initial endowment x + h, which is X T (x + h), we obtain

u(x + h) − u(x) E[U (XT (x + h)) − U (X

T (x))]
lim = lim (3.23)
h→0 h h→0 h
E[U (XT (x) + h) − U (X
T (x))]
≥ lim (3.24)
h→0 h
T (x))].
= E[U  (X

Using the fact that u is differentiable and that h may be positive as well
as negative, we must have equality in (3.24) and have therefore found another
proof of formula (3.21) for u (x); the economic interpretation of this proof
is that the economic agent, who is optimally investing, is indifferent of first
order towards a (small) additional investment into the numéraire asset.
Playing the same game as above, but using the additional endowment
h ∈ R to finance an additional investment into the optimal portfolio X T (x)
x+h 
(assuming, for simplicity, x = 0), we arrive at the pay-off function x XT (x).
Comparing this investment with X T (x + h), an analogous calculation as in

(3.23) leads to the formula for u (x) displayed in (3.22). The interpretation