06.06.2013 Views

STOCHASTIC

STOCHASTIC

STOCHASTIC

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

some qualitative characteristics of the optimal first-period decision in many<br />

dynamic models. However, the equivalent static program cannot generally be<br />

determined in explicit analytic form; hence quantitative properties of the<br />

optimal first-period decision cannot generally be ascertained.<br />

In certain special cases, one can determine an explicit functional form for<br />

the derived utility function. Exercise CR-2 illustrates such a calculation when<br />

the utility function is quadratic and there are two risky assets. Conditions are<br />

developed that lead to optimal investment entirely in one of the risky assets.<br />

Exercise ME-2 illustrates the classic Simon (1956)-Theil (1957) certainty<br />

equivalence results. Preferences for state and/or decision variables are assumed<br />

to be quadratic. The constraints that link the state and decision variables are<br />

linear functions whose uncertainty enters additively. It is also assumed that<br />

all relevant maxima and expectations exist and that the maxima occur at<br />

interior points of any additional constraints. The backward induction process<br />

then yields a sequence of induced quadratic programs. It also develops that<br />

one can replace uncertain random variables by their conditional means to<br />

obtain a certainty equivalent. The certainty equivalent is an explicit deterministic<br />

quadratic program involving only variables from period 1 and whose<br />

optimal solution is an optimal first-period decision for the multiperiod<br />

problem. Exercise ME-3 develops similar results when it is not assumed that<br />

the constraints on the decision variables are nonbinding. The discussion<br />

proceeds by considering the following questions: (1) When is the optimal<br />

decision in each period independent of all random vectors and all other<br />

decisions; (2) when is the optimal decision in each period independent of all<br />

other decisions; and (3) when is it possible to develop an explicit deterministic<br />

static program whose solution provides an optimal first-period decision?<br />

Conditions are presented that provide answers to each of these questions.<br />

They generally involve the assumption that the state vectors appear linearly<br />

in the preference function and in the period-by-period linkage constraints.<br />

For the first two questions, it is also necessary to assume that certain intertemporal<br />

separability conditions are satisfied by the decision and random<br />

vectors. All of the conditions provide for the optimality of zero-order decision<br />

rules so that optimal decisions for each period can be determined before any<br />

random variables are observed. The results are used to construct a multiperiod<br />

stochastic capital budgeting model in Exercise ME-4.<br />

Some dynamic investment problems have the property that they possess<br />

simple optimal policies that are stationary in time. Exercise CR-5 considers<br />

the dynamic portfolio problem when the utility function is logarithmic and<br />

the investment choice in each period is between a risk-free asset and a lognormally<br />

distributed asset. Conditions are developed such that it is optimal in<br />

each period to invest totally in the risky asset. A similar result obtains if the<br />

utility function is a power function (Exercise CR-6) or if the investor receives<br />

INTRODUCTION 369

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!