Chapter XIV — Investment Under Uncertainty#

Note

This chapter is Chapter XIV of Thomas J. Sargent, Macroeconomic Theory, 2nd ed. (Academic Press, 1987). It builds directly on two chapters of this book: the Euler-equation and stable-roots-backward / unstable-roots-forward machinery of Chapter IX, and the prediction formulas of the time series part — in particular the geometric-lead formula (282) and the compact vector predictor (300). Some cross-references in the original (“Chapter I”, “Chapter III”, “Chapter XII”, “Chapter XVII”, “the next chapter”) point to other chapters of the 1987 textbook that are not part of this Jupyter book; they are retained as prose.

This chapter studies aspects of the capital accumulation process in setups where firms are uncertain about the future. Our first task is to extend our earlier study of quadratic dynamic optimization problems to the case in which there is uncertainty about future values of the exogenous processes facing agents. Then we shall present a simple version of Lucas and Prescott’s model of firms’ investment behavior in a competitive industry. In the process we shall be able to give a precise characterization of the concept of a rational expectations equilibrium.

1. Optimum Decision Rules Under a Quadratic Objective#

We consider the problem: maximize (at each point in time \(t\)) the discounted present value

(534)#\[v_t = E_t \sum_{j=0}^{\infty} b^j\, g(n_{t+j-1},\, n_{t+j},\, z_{t+j})\]

over stochastic processes for \(\{n_{t+j}\}_{j=0}^{\infty}\) subject to \(n_{t-1}=\bar n_{t-1}\) given. Here \(E_t(x)=E x\mid\Omega_t\), where \(E\) is the mathematical expectation operator and \(\Omega_t\) is an information set to be specified. We assume that the discount factor obeys \(0<b<1\) and that \(g(n_{t+j-1},n_{t+j},z_{t+j})\) is concave in \(n_{t+j-1},n_{t+j}\). Here \(z_{t+j}\) is a vector of random variables that are exogenous to the decision maker. At time \(t+j\) the decision maker will have available an information set \(\Omega_{t+j}\) on which to base his decision, with \(\Omega_t\supset\Omega_{t-1}\) for all \(t\). The decision maker chooses \(n_t\) and a strategy — contingency plans \(\tilde n_{t+1}(\cdot),\tilde n_{t+2}(\cdot),\ldots\) giving \(n_{t+1}=\tilde n_{t+1}(\Omega_{t+1})\), etc. — that make \(n_{t+j}\) a function of the information \(\Omega_{t+j}\) that will be available when it must be set.[1]

To match the notation of (534) with a problem that interests us, let

(535)#\[g(n_{t-1},n_t,z_t) = (f_0+a_t)n_t - \tfrac12 f_1 n_t^2 - w_t n_t - \tfrac12 d(n_t-n_{t-1})^2\]

where \(f_0,f_1,d>0\), \(w_t\) is the real wage, \(n_t\) is employment, and \(a_t\) is a random shock to the productivity of labor. When \(w_t\) and \(a_t\) are stochastic processes, the solution to (534) becomes a stochastic version of the demand for labor studied in Chapter IX.

Equating to zero the derivative of \(v_t\) in (534) with respect to \(n_t\) gives one first-order necessary condition,

(536)#\[g_2(n_{t-1},n_t,z_t) + b E_t g_1(n_t,n_{t+1},z_{t+1}) = 0 .\]

Here \(n_{t-1}\) and \(z_t\) are known at \(t\) while \(n_{t+1}\) and \(z_{t+1}\) are still random. At \(t+1\) the decision maker faces a problem of the same form as (534), whose first-order condition is \(g_2(n_t,n_{t+1},z_{t+1}) + b E_{t+1}g_1(n_{t+1},n_{t+2},z_{t+2}) = 0\). Continuing in this way, the plan must satisfy the system of stochastic difference equations — the stochastic Euler equations

(537)#\[g_2(n_{t+j-1},n_{t+j},z_{t+j}) + b E_{t+j}g_1(n_{t+j},n_{t+j+1},z_{t+j+1}) = 0, \qquad j=0,1,2,\ldots\]

The transversality condition is obtained, as in Chapter IX, by taking the finite-\(T\) first-order condition for \(n_{t+T}\) and letting \(T\to\infty\):

(538)#\[\lim_{T\to\infty} b^T E_t\, g_2(n_{T-1},n_T,z_T)\, n_T = 0 .\]

The labor-demand example. With \(g\) given by (535) the firm maximizes

\[ v_t = E_t \sum_{j=0}^{\infty} b^j\Big\{(f_0+a_{t+j}-w_{t+j})n_{t+j} - \tfrac12 f_1 n_{t+j}^2 - \tfrac12 d(n_{t+j}-n_{t+j-1})^2\Big\} \]

subject to \(n_{t-1}\) given. Assume the exogenous processes \(\{a_{t+j}\},\{w_{t+j}\}\) are of exponential order less than \(1/\sqrt b\): for some \(K>0\) and \(1\le x<1/\sqrt b\), \(|E_t w_{t+j}|<K(x)^{j+t}\) and \(|E_t a_{t+j}|<K(x)^{j+t}\). The Euler equations are \(f_0+a_{t+j}-w_{t+j}-f_1 n_{t+j}-d(n_{t+j}-n_{t+j-1})+dbE_{t+j}(n_{t+j+1}-n_{t+j})=0\), or

(539)#\[b E_{t+j}n_{t+j+1} + \phi\, n_{t+j} + n_{t+j-1} = d^{-1}(w_{t+j}-a_{t+j}-f_0), \qquad \phi = -\big[f_1 d^{-1}+(1+b)\big],\quad j=0,1,2,\ldots\]

with transversality condition

(540)#\[\lim_{T\to\infty} E_t b^T\big\{f_0+a_{t+T}-w_{t+T}-f_1 n_{t+T}-d(n_{t+T}-n_{t+T-1})\big\}n_{t+T}=0 .\]

Equations (539)(540) generalize the Euler equation and transversality condition of the nonstochastic problem of Chapter IX, since in the nonstochastic case \(E_{t+j}n_{t+j+1}=n_{t+j+1}\).

For convenience let \(z_{t+j}=d^{-1}(w_{t+j}-a_{t+j}-f_0)\). In the nonstochastic case the solution was \(n_{t+j}=\lambda_1 n_{t+j-1}-\lambda_1\sum_{i=0}^\infty(1/\lambda_2)^i z_{t+j+i}\), where \(\lambda_1<1<\lambda_2\) solve \(1+\frac{\phi}{b}L+\frac1b L^2=(1-\lambda_1 L)(1-\lambda_2 L)\). It is natural to guess that a solution to (539)(540) is

(541)#\[n_{t+j} = \lambda_1 n_{t+j-1} - \lambda_1 \sum_{i=0}^{\infty}\Big(\tfrac{1}{\lambda_2}\Big)^i E_{t+j}\, z_{t+j+i}, \qquad j=0,1,2,\ldots\]

That (541) solves the Euler equation can be verified directly. Shifting forward one period and using the law of iterated expectations to condition on \(\Omega_{t+j}\) gives

(542)#\[E_{t+j}n_{t+j+1} = \lambda_1 n_{t+j} - \lambda_1 \sum_{i=0}^{\infty}\Big(\tfrac{1}{\lambda_2}\Big)^i E_{t+j}\, z_{t+j+1+i} .\]

Substituting (541) and (542) into (539) and using \(-\phi=b(\lambda_1+\lambda_2)\), \(b\lambda_2=1/\lambda_1\), so that \(\{b\lambda_1+\phi\}\lambda_1=-1\), collapses the identity to \(-\sum_{i=0}^\infty(1/\lambda_2)^{i+1}E_{t+j}z_{t+j+i+1}+\sum_{i=0}^\infty(1/\lambda_2)^i E_{t+j}z_{t+j+i}=z_{t+j}\), which holds identically. Under the assumption that \(z_t\) is of exponential order less than \(1/\sqrt b\), the transversality condition holds as in Chapter IX.

A constructive derivation. There is no need to guess. Write the Euler equation as

(543)#\[b E_s n_{s+1} + \phi E_s n_s + E_s n_{s-1} = E_s z_s ,\]

or \((bB^{-2}+\phi B^{-1}+1)E_s n_{s-1}=E_s z_s\), where the operator \(B\) is defined by \(B^{-j}E_{s-1}n_s=E_{s-1}n_{s+j}\) (it advances the date being forecast but leaves the information set fixed).[2] This can be written

(544)#\[b(\lambda_1-B^{-1})(\lambda_2-B^{-1})E_s n_{s-1} = E_s z_s ,\]

with \(|\lambda_1|<1\) and \(\lambda_2=1/\lambda_1 b\). Operating on both sides with \([(\lambda_2-B^{-1})]^{-1}\) — the only legitimate forward inverse[3] — and setting the constant multiplying \(\lambda_2^s\) to zero to satisfy transversality, then using \(\lambda_2^{-1}=\lambda_1 b\), gives

(545)#\[n_s - \lambda_1 n_{s-1} = -\lambda_1 \sum_{i=0}^{\infty}(\lambda_1 b)^i E_s z_{s+i} .\]

Certainty equivalence. The solution depends only on the conditional means \(E_t z_{t+j}\), not on higher moments — the “certainty equivalence” or “separation” principle possessed by quadratic \(g\)’s. The problem separates into two stages: first form the forecasts \(E_t z_{t+j}\); second, solve the nonstochastic problem

\[ \max_{n_t,n_{t+1},\ldots} V_t = \sum_{j=0}^{\infty} b^j\, g(n_{t+j-1},n_{t+j},E_t z_{t+j}), \qquad n_{t-1}\ \text{fixed}. \]

This separation of forecasting from optimization explains why quadratic objectives are assumed in much applied work; for general \(g\) the two problems do not separate.[4]

2. Optimal Linear Policies#

The decision rule (541) sets \(n_t\) as a linear function of \(n_{t-1}\) and the conditional expectations \(E_t w_{t+i}, E_t a_{t+i}\). In general these conditional expectations are nonlinear functions of the information in \(\Omega_t\), so the optimal decision rule is nonlinear.

Suppose instead we restrict ourselves to decision rules that express \(n_t\) as linear functions of \(n_{t-1}\) and the information in \(\Omega_t\). When \(g\) is quadratic, the optimal linear rule is obtained by replacing the conditional mathematical expectations in (541) with the corresponding linear least squares projections on \(\Omega_t\) (the regressions of Chapter X). In the special case where \(\{z_{t+j}\}\) is a multivariate normal process, conditional expectations are linear and equal the projections; then linear least squares policies are optimal among all decision rules. With Gaussian \(\{w_t\},\{a_t\}\) the rule can, with positive probability, call for negative employment; it is usual to assume the shock variances are small relative to the constants so that \(n_t<0\) occurs with negligible probability.

3. Lucas’s Critique#

In §20 we derived a formula for the linear least squares prediction of a geometric lead of the kind in (545). Suppose \(z_t\) has the autoregressive representation

(546)#\[a(L)z_t = \epsilon_t, \qquad a(L)=1-a_1 L-\cdots-a_r L^r,\quad \epsilon_t=z_t-P[z_t\mid z_{t-1},\ldots].\]

Then formula (282) of §20 asserts that

(547)#\[E_t \sum_{i=0}^{\infty}\lambda_2^{-i} z_{t+i} = \frac{1-\lambda_1 b\, a(\lambda_1)^{-1} a(L) L^{-1}}{1-\lambda_1 b L^{-1}}\, z_t = a(\lambda_1 b)^{-1}\Big[1 + \sum_{j=1}^{r-1}\Big(\sum_{k=j+1}^{r}(\lambda_1 b)^{k-j}a_k\Big)L^j\Big] z_t ,\]

where \(\lambda_1 b=\lambda_2^{-1}\). Substituting (547) into (545) gives the pair of equations

(548)#\[n_t = \lambda_1 n_{t-1} - \lambda_1\, \frac{1-\lambda_1 b\, a(\lambda_1)^{-1}a(L)L^{-1}}{1-\lambda_1 b L^{-1}}\, z_t, \qquad a(L)z_t=\epsilon_t .\]

Equation (548) is an optimal linear decision rule for employment, expressing \(n_t\) as a linear function of \(n_{t-1}\) and \(z_t,z_{t-1},\ldots,z_{t-r+1}\),

(549)#\[n_t = h_1 n_{t-1} + h_2 z_t + h_3 z_{t-1} + \cdots + h_r z_{t-r+1} .\]

The form of dependence on the \(z\)’s depends on the parameters of \(a(L)\), because the \(z\)’s appear only insofar as they help predict the geometric sum \(\sum_{j\ge0}(\lambda_1 b)^j z_{t+j}\). Thus the optimal decision rule inherits parameters from the stochastic process for \(\{z_t\}\): the coefficients \(h_2,\ldots,h_r\) depend on the law of motion (546). It is fruitless to search for a single decision rule \(n_t=h(n_{t-1},\Omega_t)\) that is invariant across hypothetical environments (laws of motion for \(z_t\)). This principle, stated for labor supply by Gordon and Hynes (1970), underlies Robert E. Lucas’s (1976) critique of the econometric policy-evaluation procedures that existed in 1973, which treated decision rules like (549) as structural (invariant under interventions in the process for \(z_t\)). Policy evaluation should instead take into account the dependence of private decision rules on the government’s choice of a policy rule — a theme pursued in the cross-equation restrictions of rational expectations models and exact linear rational expectations models.[5]

4. Investment#

We now apply these methods to Lucas and Prescott’s (1971) model of investment under uncertainty, which gives a precise illustration of a rational expectations equilibrium. We first study a “perfect foresight” (nonstochastic) model and then move to the stochastic setting.

Consider an industry of \(n\) identical competitive firms using capital \(k_t\) to produce output \(f_0 k_t\), with \(f_0>1\). The industry demand curve is

(550)#\[p_t = A_0 - A_1 Y_t + u_t, \qquad A_0,A_1>0,\]

where \(Y_t=nf_0 k_t\) is industry output and \(u_t\) is a demand shock. The representative firm is a price-taker with respect to output prices \(\{p_{t+j}\}\) and prices of capital \(\{J_{t+j}\}\). In the nonstochastic case, taking known sequences \(\{p_{t+j}\},\{J_{t+j}\}\) of exponential order less than \(1/\sqrt b\), the firm chooses \(\{k_{t+j}\}\) to maximize

(551)#\[v_t = \sum_{j=0}^{\infty} b^j\Big\{ p_{t+j}(f_0 k_{t+j}) - J_{t+j}(k_{t+j}-k_{t-1+j}) - \tfrac{d}{2}(k_{t+j}-k_{t-1+j})^2 \Big\}\]

subject to \(k_{t-1}\) given, where \(d>0\) is an adjustment-cost coefficient. The Euler equation is

(552)#\[p_{t+j}f_0 - J_{t+j} + bJ_{t+j+1} - d(1+b)k_{t+j} + b\,d\,k_{t+j+1} + d\,k_{t+j-1} = 0, \qquad j=0,1,\ldots\]

which we rewrite as \(b\big(1-\tfrac1b L\big)(1-L)k_{t+j+1}=\tfrac1d(J_{t+j}-bJ_{t+j+1}-p_{t+j}f_0)\). The solution satisfying the transversality condition[6] is

(553)#\[(1-L)k_{t+j+1} = \frac{-d^{-1}}{1-bL^{-1}}\big(J_{t+j+1}-bJ_{t+j+2}-p_{t+j+1}f_0\big),\]

giving the firm’s rate of investment as a function of future values of the output price and the price of capital.

While each firm perceives \(p_t\) as independent of its own decisions, the price is determined by all firms together through \(p_t=A_0-A_1 nf_0 k_t+u_t\). We seek an equilibrium pair of sequences \(\{\bar p_{t+j}\},\{\bar k_{t+j}\}\) satisfying:

  • (i) Given \(\{\bar k_{t+j}\}\), prices clear the market: \(\bar p_{t+j}=A_0-A_1 nf_0\bar k_{t+j}+u_{t+j}\).

  • (ii) Facing \(\{\bar p_{t+j}\}\) as a price-taker, \(\{\bar k_{t+j}\}\) maximizes present value (551).

Substituting \(A_0-A_1 f_0 nk_{t+j}+u_{t+j}\) for \(p_{t+j}\) in the Euler equation (552)after differentiating, so that the firm acts as a price taker[7] — gives

(554)#\[A_0 f_0 + f_0 u_{t+j} - J_{t+j} + bJ_{t+j+1} - \{d(1+b)+A_1 f_0^2 n\}k_{t+j} + d\,k_{t+j-1} + d\,b\,k_{t+j+1} = 0 .\]

Writing this as \(b\,k_{t+j+1}+\phi\,k_{t+j}+k_{t+j-1}=d^{-1}\{J_{t+j}-bJ_{t+j+1}-f_0 u_{t+j}-A_0 f_0\}\) with \(\phi=-\big[(1+b)+\tfrac{A_1 f_0^2 n}{d}\big]\), and factoring \(1+\tfrac\phi b L+\tfrac1b L^2=(1-\lambda_1 L)(1-\lambda_2 L)\) with \(\lambda_1<\tfrac1b<\lambda_2\), the solution satisfying the firm’s transversality condition is

(555)#\[k_{t+j+1} = \lambda_1 k_{t+j} - \frac{\lambda_1 d^{-1}}{1-\lambda_2^{-1}L^{-1}} \big\{J_{t+j+1}-bJ_{t+j+2}-f_0 u_{t+j+1}-A_0 f_0\big\}, \qquad j=-1,0,1,\ldots\]

and the equilibrium output price is

(556)#\[p_{t+j} = A_0 - A_1 nf_0 k_{t+j} + u_{t+j} .\]

5. A Digression on the Relation Between Equilibrium and Optimality#

Is the equilibrium difference equation (554) the Euler equation of an interesting maximum problem? Consider maximizing

(557)#\[W_t = \sum_{j=0}^{\infty} b^j\Big\{\big[A_0 f_0 nk_{t+j}-\tfrac12 A_1(f_0^2 n^2 k_{t+j}^2)+f_0 nu_{t+j}k_{t+j}\big] - nJ_{t+j}(k_{t+j}-k_{t+j-1}) - \tfrac12 dn(k_{t+j}-k_{t+j-1})^2\Big\}\]

subject to \(k_{t-1}\) given. Its Euler equation is exactly (554). The bracketed term is the area under the demand curve,

\[ \int_0^{Y_t}(A_0-A_1 x+u_t)\,dx = A_0 Y_t - \tfrac12 A_1 Y_t^2 + Y_t u_t = A_0 f_0 nk_t - \tfrac12 A_1 f_0^2 n^2 k_t^2 + f_0 nk_t u_t . \]

Thus the equilibrium implicitly maximizes the social welfare criterion (557), which equals discounted consumer surplus minus producer surplus. Lucas and Prescott used the observation that an equilibrium implicitly solves a social-welfare problem to characterize equilibria in settings where the direct method is unavailable: their device replaces a “fixed point” problem with a maximization problem.

6. Investment Under Uncertainty#

Now take \(\{u_{t+j}\},\{J_{t+j}\}\) to be exogenous stochastic processes of exponential order less than \(1/\sqrt b\). The firm, a price-taker with respect to \(\{J_{t+j}\}\) and the equilibrium price process \(\{p_{t+j}\}\), chooses \(\{k_{t+j}\}\) to maximize

(558)#\[v_t = E_t \sum_{j=0}^{\infty} b^j\Big\{p_{t+j}f_0 k_{t+j}-J_{t+j}(k_{t+j}-k_{t+j-1}) - \tfrac{d}{2}(k_{t+j}-k_{t+j-1})^2\Big\} .\]

Paralleling Section 1, the firm’s optimum plan is

(559)#\[k_{t+j} = k_{t+j-1} - \frac1d \sum_{i=0}^{\infty} b^i E_{t+j}\big\{J_{t+j+i}-bJ_{t+j+1+i}-f_0 p_{t+j+i}\big\},\]

which agrees with (553) in the nonstochastic case. We seek an equilibrium pair of stochastic processes \(\{\bar p_{t+j}\},\{\bar k_{t+j}\}\) such that (i) given the firm’s plan, \(\bar p_{t+j}=A_0-A_1 nf_0\bar k_{t+j}+u_{t+j}\) clears the market; and (ii) facing \(\{\bar p_{t+j}\}\) as a price-taker, \(\{\bar k_{t+j}\}\) maximizes (558). Such an equilibrium is a rational expectations equilibrium: firms form the forecasts of future prices appearing in (558) by taking conditional expectations with respect to the stochastic process that actually governs prices. Proceeding in complete analogy with the nonstochastic analysis,

(560)#\[k_{t+j+1} = \lambda_1 k_{t+j} - \frac{\lambda_1}{d}\sum_{i=0}^{\infty}\Big(\tfrac1{\lambda_2}\Big)^i E_{t+j+1}\big\{J_{t+j+1+i}-bJ_{t+j+2+i}-f_0 u_{t+j+1+i}-A_0 f_0\big\},\]
(561)#\[p_{t+j} = A_0 - A_1 nf_0 k_{t+j} + u_{t+j} .\]

Now suppose \(J_t,u_t\) are governed by

(562)#\[a(L)J_t = \epsilon_{Jt}, \qquad g(L)u_t = \epsilon_{ut},\]

with \(a(L)=1-a_1 L-\cdots-a_r L^r\), \(g(L)=1-g_1 L-\cdots-g_r L^r\), and innovations \(\epsilon_{Jt}=J_t-P[J_t\mid J_{t-1},u_{t-1},\ldots]\), \(\epsilon_{ut}=u_t-P[u_t\mid\ldots]\), where the zeros of \(a(z),g(z)\) exceed unity in modulus. Using formula (282) of §20, (560) can be represented as the equilibrium motion of capital

(563)#\[\begin{split}\begin{aligned} k_{t+j+1} = \lambda_1 k_{t+j} - \lambda_1 d^{-1}\Big\{ &\Big[\tfrac{1-\lambda_1 b\, a(\lambda_1)^{-1}a(L)L^{-1}}{1-\lambda_1 b L^{-1}}\Big]J_{t+j+1} - b\Big[\tfrac{L^{-1}I-L^{-1}a(b\lambda_1)^{-1}a(L)}{1-\lambda_1 b L^{-1}}\Big]J_{t+j+1} \\ &- f_0\Big[\tfrac{1-\lambda_1 b\, g(\lambda_1)^{-1}g(L)L^{-1}}{1-\lambda_1 b L^{-1}}\Big]u_{t+j+1} - A_0 f_0/(1-\lambda_1 b)\Big\} . \end{aligned}\end{split}\]

7. Supply, Demand and Identification#

Equations (560) and (563) can be used to derive a dynamic “supply curve.” For convenience specialize (562) to \(g(L)=I\) (so \(u_t\) is white noise) and \(a(L)=(1-\rho L)\), \(|\rho|<1\) (so \(J_t\) is first-order autoregressive). The firm’s stock obeys (559):

(564)#\[k_t = k_{t-1} - d^{-1}\sum_{j=0}^{\infty} b^j E_t\{J_{t+j}-bJ_{t+j+1}\} + d^{-1}f_0\sum_{j=0}^{\infty} b^j E_t p_{t+j} .\]

Let \(K_t=nk_t\) be aggregate capital, \(\bar f_0=f_0\lambda_1 d^{-1}n\), \(\bar A_0=A_0/(1-\lambda_1 b)\). The aggregate capital stock follows

(565)#\[K_t = \lambda_1 K_{t-1} - \lambda_1 d^{-1}n\Big\{\tfrac{1-b\rho}{1-\lambda_1 b\rho}\Big\}J_t + \bar f_0(u_t+\bar A_0) .\]

Using (565) and \(J_t=\rho J_{t-1}+\epsilon_{Jt}\), the law of motion of the vector \((K_t,J_t,1)\) is

(566)#\[\begin{split}\begin{pmatrix} K_{t+1}\\ J_{t+1}\\ 1\end{pmatrix} =\begin{pmatrix} \lambda_1 & \dfrac{-\lambda_1\rho d^{-1}n(1-b\rho)}{1-\lambda_1 b\rho} & \bar f_0\bar A_0\\[2ex] 0 & \rho & 0\\[1ex] 0 & 0 & 1 \end{pmatrix} \begin{pmatrix} K_t\\ J_t\\ 1\end{pmatrix} +\begin{pmatrix} \dfrac{-\lambda_1 d^{-1}n(1-b\rho)}{1-\lambda_1 b\rho}\,\epsilon_{Jt+1}+\bar f_0 u_{t+1}\\[2ex] \epsilon_{Jt+1}\\[1ex] 0 \end{pmatrix}\end{split}\]

or \(x_{t+1}=A x_t+\epsilon_{t+1}\), where \(x_t=(K_t,J_t,1)^T\). From §24 we have the compact predictor (300),

(567)#\[E_t \sum_{j=0}^{\infty}\mu^j x_{t+j} = (I-\mu A)^{-1}x_t ,\]

valid for any scalar \(\mu\) whose product with the modulus of the largest eigenvalue of \(A\) is less than one. Applying (567) to (566) gives

(568)#\[\begin{split}E_t \sum_{j=0}^{\infty}\mu^j \begin{pmatrix}K_{t+j}\\ J_{t+j}\\ 1\end{pmatrix} =\begin{pmatrix} \dfrac{1}{1-\mu\lambda_1}K_t - \dfrac{\mu n\lambda_1\rho d^{-1}(1-b\rho)}{(1-\lambda_1 b\rho)(1-\mu\lambda_1)(1-\rho\mu)}J_t + \dfrac{\bar f_0\bar A_0}{(1-\mu\lambda_1)(1-\mu)}\\[2.5ex] \dfrac{1}{1-\rho\mu}J_t\\[2ex] \dfrac{1}{1-\mu} \end{pmatrix} .\end{split}\]

Returning to (564) with \(a(L)=(1-\rho L)\) and substituting the demand curve \(p_{t+j}=A_0-A_1 f_0 K_{t+j}+u_{t+j}\) for \(p_{t+j}\),

(569)#\[k_t = k_{t-1} - d^{-1}J_t + \frac{d^{-1}f_0 A_0}{1-b} + d^{-1}f_0 u_t - d^{-1}f_0^2 A_1 \sum_{j=0}^{\infty} b^j E_t K_{t+j} .\]

Using (568) to evaluate the geometric sum, eliminating \(K_t\) via the demand curve, and using \(y_t=f_0 k_t\), gives the dynamic supply curve

(570)#\[\begin{split}\begin{aligned} y_t = y_{t-1} &- d^{-1}f_0\Big\{1-\tfrac{f_0^2 A_1 b\lambda_1\rho d^{-1}n}{(1-\lambda_1 b\rho)(1-b\lambda_1)}\Big\}J_t + \Big[\tfrac{d^{-1}f_0^2 A_0}{1-b}-\tfrac{d^{-1}f_0^3 A_1 b\bar f_0\bar A_0 n}{(1-b\lambda_1)(1-b)} -\tfrac{d^{-1}f_0^2 A_0}{1-b\lambda_1}\Big]\\ &- \tfrac{d^{-1}f_0^2 b\lambda_1}{1-b\lambda_1}u_t + \tfrac{d^{-1}f_0^2}{1-b\lambda_1}p_t , \end{aligned}\end{split}\]

expressing the firm’s output as a function of lagged output and current factor and output prices. All of the demand-curve parameters appear in the supply curve — directly and through their influence on \(\lambda_1\) — and the demand disturbance \(u_t\) appears on the right side. Any variable that helps predict future prices appears in firms’ supply curve as an “information variable.” This subverts the exclusion restrictions ordinarily relied upon to identify a supply curve. The identifying restrictions that are available come from the cross-equation restrictions linking the equilibrium law of motion (563) to the laws of motion (562) — the common appearance of \(a(L),g(L)\) in both — exploited by Taylor (1979, 1980) and Hansen and Sargent (1980).

8. Investment Under Uncertainty and an Externality#

Now let the technology exhibit an externality,

(571)#\[y_t = (f_0 k_t + f_1 K_t), \qquad f_1\neq 0,\]

where \(K_t=nk_t\) is the aggregate capital stock (a linear version of Romer 1983). Define the one-period rental on capital \(w_t=J_t-bE_t J_{t+1}\), equivalently \(J_t=\sum_{j=0}^\infty b^j E_t w_{t+j}\).[8] The representative firm maximizes \(E_0\sum_{t=0}^\infty b^t\{p_t(f_0 k_t+f_1 K_t)-w_t k_t-\tfrac d2(k_t-k_{t-1})^2\}\) subject to \(p_t=A_0-A_1(f_0+nf_1)K_t+u_t\) and the laws of motion for \(w_t,u_t\). Its Euler equation is \(E_t\{p_t f_0-w_t-d(k_t-k_{t-1})+bd(k_{t+1}-k_t)\}=0\). Substituting the demand curve and multiplying by \(n\) gives the equilibrium difference equation

(572)#\[-bE_t K_{t+1} + \big[(1+b)+A_1 nd^{-1}(f_0^2+nf_1 f_0)\big]K_t - K_{t-1} = d^{-1}\big[A_0 f_0 n + f_0 nu_t - nw_t\big] .\]

The social planning problem maximizes expected discounted consumer plus producer surplus,

(573)#\[E_0 \sum_{t=0}^{\infty} b^t\Big\{A_0(f_0+nf_1)K_t - \tfrac{A_1}{2}[(f_0+nf_1)K_t]^2 + u_t(f_0+nf_1)K_t - w_t K_t - \tfrac{d}{2n}(K_t-K_{t-1})^2\Big\},\]

whose Euler equation is

(574)#\[-bE_t K_{t+1} + \Big[(1+b)+\tfrac{A_1 n(f_0+nf_1)^2}{d}\Big]K_t - K_{t-1} = d^{-1}\big[nA_0(f_0+nf_1)+n(f_0+nf_1)u_t-nw_t\big] .\]

Comparing (572) and (574), when \(f_1\neq 0\) the competitive equilibrium is not optimal relative to the welfare criterion (573): the externality drives a wedge.

Correcting the externality with a tax. Let the government levy \(\tau_t k_t\) with

(575)#\[\tau_t = \alpha_0 + \alpha_1 u_t + \alpha_2 w_t + \alpha_3 k_t .\]

Under this schedule the firm’s Euler equation, after substituting the demand curve and multiplying by \(n\), becomes

(576)#\[-bE_t K_{t+1} + \big[(1+b)+A_1 nd^{-1}(f_0^2+nf_1 f_0)+2\alpha_3 d^{-1}\big]K_t - K_{t-1} = d^{-1}\big[(nA_0 f_0-n\alpha_0)+(nf_0-n\alpha_1)u_t-(n+\alpha_2 n)w_t\big] .\]

Equating (576) to the planner’s (574) requires the government to set

(577)#\[\alpha_0 = -A_0 nf_1, \quad \alpha_1 = -nf_1, \quad \alpha_2 = 0, \quad \alpha_3 = \tfrac{A_1 n}{2}\big[(f_0+nf_1)^2-(f_0^2+nf_1 f_0)\big] .\]

The optimal tax (577) requires the government to know the parameters of preferences \((A_0,A_1)\), technology \((f_0,f_1)\), and industry structure \((n)\). The reason rational expectations econometrics aims to estimate the “deep” parameters of preferences and technologies is precisely that these are the parameters needed to derive optimal policy interventions.

9. Conclusions#

The rational expectations competitive equilibrium has the attractive property that firms in the industry forecast the output price optimally: firms forecast prices as well as the economist modelling them. The output price is endogenous, influenced by firms’ behavior in light of their forecasts. There is a mapping from firms’ perceived law of motion for the output price to the actual law of motion; a rational expectations equilibrium is a fixed point of this mapping. Linear Lucas–Prescott models have served as the basis for econometric implementation (Sargent 1981; Hansen and Sargent 1980), multi-factor extensions (Hansen and Sargent 1981; Eichenbaum 1983), and studies of government interventions and exhaustible resources (Eckstein and Eichenbaum 1985a,b; Hansen, Epple, and Roberds 1985; Townsend 1983).

Exercises#

Exercise 59

Assume that

\[ J_{t+1}=\alpha J_t+\eta_{t+1},\ \ |\alpha|<1/b, \qquad u_{t+1}=\beta u_t+\varepsilon_{t+1},\ \ |\beta|<1/b, \]

where \(E_t\eta_{t+1}=\bar\eta\) and \(E_t\varepsilon_{t+1}=\bar\varepsilon\). Use this information and (560) to calculate a “reduced form” for investment of the form

\[ k_{t+j+1}-\lambda_1 k_{t+j}=\gamma_0+\gamma_1 J_{t+1+j}+\gamma_2 u_{t+1+j}, \]

giving explicit formulas for \(\gamma_0,\gamma_1,\gamma_2\).

Exercise 60

(Certainty equivalence principle.) Let \(x\) be a random variable with density \(g(x)\), and let \(\alpha\) be a parameter set by a decision-maker. Let \(f(x,\alpha)\) be concave and twice continuously differentiable. Consider Problem 1: choose \(\alpha\) to maximize \(E f(x,\alpha)=\int f(x,\alpha)g(x)\,dx\).

  A. Find the first-order condition for choosing \(\alpha\).

  B. Suppose \(f(x,\alpha)=(x,\alpha)A(x,\alpha)'+(x,\alpha)B\), where \(B\) is \(2\times1\), \((x,\alpha)\) is \(1\times2\), and \(A\) is a \(2\times2\) negative definite matrix. Prove that in this case choosing \(\alpha\) to solve Problem 1 gives the same \(\alpha\) as Problem 2: choose \(\alpha\) to maximize \(f(Ex,\alpha)\).

Exercise 61

(Eckstein 1983.) A large number \(n\) of identical farms produce corn; each maximizes

\[ E_0\sum_{t=0}^{\infty} b^t\Big\{p_{t+1}y_{t+1}-c_0 a_t-\tfrac{c_1}{2}a_t^2-c_2 a_t a_{t-1}-w_t a_t\Big\} \]

subject to \(a_{-1}\) given, with \(c_0,c_1,c_2>0\), \(0<b<1\), output \(y_{t+1}=fa_t\) (\(f>0\)), and demand \(p_{t+1}=\beta_0-\beta_1 Y_{t+1}+u_{t+1}\) where \(Y_{t+1}=ny_{t+1}\) and \(u_t=\rho u_{t-1}+\varepsilon_t\), \(|\rho|<1\). Fertilizer cost \(w_t=d_0\eta_t+d_1\eta_{t-1}\) (\(\eta\) white noise). At \(t\) the farm sees \(w_t,\ldots,u_t,\ldots,a_{t-1}\) and \(A_{t-1}=na_{t-1}\).

  A. Carefully define a rational expectations equilibrium.

  B. Describe how to compute it; get as far as you can for \(\beta_1=10^{-6}\), \(n=10^6\), \(f=1\), \(b=1\), \(c_1=16\), \(c_2=4\).

  C. Describe the effect on the equilibrium law of motion for \(A_t\) of changing the fertilizer-cost process to \(w_t=g_0\varepsilon_t+g_1\varepsilon_{t-1}+g_2\varepsilon_{t-2}\).

Exercise 62

A small country produces bananas competitively with free entry. The world price \(p_t\) is exogenous with Wold representation \(p_t=c(L)\varepsilon_t\) (\(c(L)^{-1}\) one-sided, square-summable). Output \(y_t=f(L)n_t\), \(f(L)=\sum_{j\ge0}f_j L^j\), \(n_t\) employment. Each firm takes \(w_t\) parametrically, but the country wage obeys \(w_t=\beta_0+\beta_1 N_t\) (\(N_t\) total employment). The representative firm faces \(p_t\) and \(w_t\) parametrically and solves \(\max E_0\sum_{t=0}^{\infty}\{p_t f(L)n_t-w_t n_t\}\) (constant returns to scale).

  A. Find the “marginal expected present value” of employing an additional worker at \(t\).

  B. Impose free entry (zero expected present value) and derive the equilibrium condition \(w_t=E_t\sum_{j=0}^{\infty}h_j p_{t+j}\), giving the \(h_j\).

  C. With \(f(L)=1/(1-\lambda L)\) and \(c(L)=1/(1-\rho_1 L-\rho_2 L^2)\), derive \(w_t=\sum_{j=0}^{\infty}g_j p_{t-j}\), giving \(g_j\) in terms of \(\lambda,\rho_1,\rho_2\).

  D. How would this change if a banana cartel changed \(c(L)\) to \(c(L)=1+0.99L\)? How does this illustrate Lucas’s critique?

Exercise 63

(Stabilizing prices vs. quantities.) An industry of \(n\) identical firms has \(y_t=fk_t\), \(f>0\). The representative firm maximizes \(E_0\sum_{t=0}^{\infty}\beta^t\{p_t y_t-(w_t+\tau_t)k_t-\tfrac{d}{2}(k_t-k_{t-1})^2\}\), \(0<\beta<1\), with demand \(p_t=A_0-A_1 Y_t+u_t\), \(Y_t=fK_t\), \(\tau_t\) a tax-subsidy on capital, \(w_t=\bar\omega+\varepsilon_{wt}\) (\(\varepsilon_{wt}\) serially independent, mean zero) and \(u_t\) serially independent mean zero.

  A. With \(\tau_t\equiv0\), define a rational expectations competitive equilibrium and display the equilibrium law of motion for \(K_t\).

  B. Let the government set \(\tau_t\) as a linear function of \(\{u_{t-1},\ldots,\varepsilon_{w,t-1},\ldots\}\) (zero constant). Find the feedback rule that minimizes the stationary variance of \(Y_t\); then the rule that minimizes the stationary variance of \(p_t\).

  C. Under what more general assumption about \(\{u_t\}\) would the two objectives give different rules for \(\tau_t\)?

  D. Is minimizing the stationary variance of price a good policy goal? Of quantity?

Exercise 64

An industry of \(n\) firms faces demand \(p_t=A_0-A_1 Y_t+u_t\) (\(u_t\) serially uncorrelated, mean zero) with representative-firm cost \(c_t=c_0+c_1 y_t+\tfrac{c_2}{2}y_t^2+\tfrac{c_3}{2}(y_t-y_{t-1})^2+\tfrac{c_4}{2}(Y_t-Y_{t-1})^2+J_t y_t+\tau_t y_t\), where \(Y_t=ny_t\), \(J_t=\lambda J_{t-1}+\varepsilon_t\), and the tax rule is \(\tau_t=\delta_0+\delta_1 y_t+\delta_2 y_{t-1}\) (each firm recognizes the dependence of \(\tau_t\) on its own \(y_t\)). The term \(\tfrac{c_4}{2}(Y_t-Y_{t-1})^2\) is an industry-wide adjustment cost — an externality. Firms maximize \(E\sum_{t=0}^{\infty}\beta^t\{p_t y_t-c_t\}\).

  A. Define a rational expectations competitive equilibrium.   B. Compute it.   C. For the “social planning” problem (maximize expected consumer minus producer surplus, omitting the transfer \(\tau_t y_t\)), does the competitive equilibrium solve it for arbitrary \((\delta_0,\delta_1,\delta_2)\)?   D. Find \((\delta_0,\delta_1,\delta_2)\) that make it do so.

Exercise 65

(Duck decoys.) A fixed number \(n\) of identical firms produce duck decoys, \(y_t=fk_t\) (\(f>0\)), aggregate \(Y_t=fK_t\), \(K_t=nk_t\). Demand \(p_t=A_0-A_1 Y_t+u_t\) (\(A_0,A_1>0\)) with \(u_t=\frac{1}{1-\rho L}\varepsilon_{ut}\), \(|\rho|<1\). The rental on capital obeys the upward-sloping industry supply \(w_t=B_0+B_1 K_t\) (\(B_0,B_1>0\)). Net cash flow \(\pi_t=p_t y_t-w_t k_t-\tfrac{d}{2}(k_t-k_{t-1})^2\), present value \(E\sum_{t=0}^{\infty}b^t\pi_t\).

  A. Define a rational expectations competitive equilibrium.   B. Compute it.   C. What social planning problem does it solve?   D. Define a monopolistic (collusive) rational expectations equilibrium.   E. Compute it.   F. Show that the feedback coefficient of \(K_t\) on \(K_{t-1}\) is smaller under monopoly than under competition.   G. Justice Department economists know \(f,n,b,d\) but not \((B_0,B_1,A_0,A_1)\); they observe \(\{(Y_s,K_s)\}\). Can they tell whether the industry is competitive or monopolistic?   H. Same, but they also observe \(\{(p_s,w_s)\}\). Can they now tell?

References#

[CNEck84]

Zvi Eckstein. A rational expectations model of agricultural supply. Journal of Political Economy, 92(1):1–19, 1984.

[CNEic83]

Martin S. Eichenbaum. A rational expectations equilibrium model of inventories of finished goods and employment. Journal of Monetary Economics, 12(2):259–277, 1983.

[CNGH70]

Donald F. Gordon and Allan Hynes. On the theory of price dynamics. In Edmund S. Phelps, editor, Microeconomic Foundations of Employment and Inflation Theory. Norton, New York, 1970.

[CNHS80]

Lars Peter Hansen and Thomas J. Sargent. Formulating and estimating dynamic linear rational expectations models. Journal of Economic Dynamics and Control, 2(1):7–46, 1980.

[CNHS81]

Lars Peter Hansen and Thomas J. Sargent. Linear rational expectations models for dynamically interrelated variables. In Jr. Lucas, Robert E. and Thomas J. Sargent, editors, Rational Expectations and Econometric Practice. University of Minnesota Press, Minneapolis, 1981.

[CNHMMS60]

Charles C. Holt, Franco Modigliani, John F. Muth, and Herbert A. Simon. Planning Production, Inventories, and Work Force. Prentice-Hall, Englewood Cliffs, NJ, 1960.

[CNKS72]

Huibert Kwakernaak and Raphael Sivan. Linear Optimal Control Systems. Wiley-Interscience, New York, 1972.

[CNLuc76]

Jr. Lucas, Robert E. Econometric policy evaluation: a critique. In Karl Brunner and Allan H. Meltzer, editors, The Phillips Curve and Labor Markets, volume 1 of Carnegie-Rochester Conference Series on Public Policy, pages 19–46. North-Holland, Amsterdam, 1976.

[CNLP71]

Jr. Lucas, Robert E. and Edward C. Prescott. Investment under uncertainty. Econometrica, 39(5):659–681, 1971.

[CNRom83]

Paul M. Romer. Externalities and increasing returns in dynamic competitive analysis. Working Paper, University of Rochester, 1983.

[CNSar81]

Thomas J. Sargent. Interpreting economic time series. Journal of Political Economy, 89(2):213–248, 1981.

[CNTay79]

John B. Taylor. Estimation and control of a macroeconomic model with rational expectations. Econometrica, 47(5):1267–1286, 1979.

[CNTG72]

Lester G. Telser and Robert L. Graves. Functional Analysis in Mathematical Economics. University of Chicago Press, Chicago, 1972.

[CNTow83]

Robert M. Townsend. Forecasting the forecasts of others. Journal of Political Economy, 91(4):546–588, 1983.