Production-Based Asset Pricing

Asset Pricing Theory

Juan F. Imbet

Paris Dauphine University-PSL

Overview & Agenda

Focus: Production economies where firms make investment decisions

Key Insight: Asset prices reflect both consumption and investment opportunities

Topics:

  1. Production Economy Setup
  2. Household SDF and Firm Optimality
  3. q-Theory and the Return on Capital
  4. Cochrane’s Production-Based Model
  5. Investment-Specific Technology (IST)
  6. Equilibrium & Empirical Implications

Production Economy Framework

Objects and Notation

  • Time is discrete, \(t=0,1,2,\dots\).
  • A single consumption good is the numeraire. Let \(P_t^C \equiv 1\).
  • Firms use capital \(K_t\) and labor \(L_t\) to produce output \(Y_t\).
  • Capital accumulates with depreciation \(\delta \in (0,1)\) and possibly convex adjustment costs.

Technology

\[ Y_t = F(K_t, Z_t L_t),\qquad F_K>0,\;F_L>0,\;F_{KK}<0,\;F_{LL}<0, \] \[ K_{t+1} = (1-\delta)K_t + \Phi(I_t,K_t),\qquad \Phi_I>0,\;\Phi_{II}\le 0, \]

where \(\Phi(I,K)\) converts investment goods \(I_t\) into installed capital; with no adjustment costs, \(\Phi(I,K)=I\).

Household Problem

Preferences: \[ \max_{\{C_t,L_t,\{x^i_t\}\}} E_0 \sum_{t=0}^{\infty} \beta^t u(C_t,1-L_t),\qquad 0<\beta<1, \] with \(u_C>0\), \(u_{CC}<0\), \(u_{1-L}>0\), \(u_{(1-L)(1-L)}<0\).

Budget constraint in units of consumption: \[ C_t + \sum_i p_t^i x_t^i \le w_t L_t + \sum_i x_{t-1}^i (d_t^i+p_t^i) + \Pi_t, \]

Mathematical Note

  • Law of iterated expectations: \[ E_0[X] = E_0[E_t[X]],\quad \text{for any } t\ge 0. \]
  • Then, when computing Lagrangians which are computed using \(E_0\) we can use this propery to add a \(E_t\) operator inside the expectation. Then the first order conditions can be simplified by conditioning on the information set at time \(t\). Since the F.O.C. determine the optimal variables at time \(t\) it is not correct to condition on the time zero information set.

Stochastic Discount Factor (SDF)

\[ 1 = E_t\big[M_{t+1} R_{t+1}^i\big],\qquad M_{t+1} \equiv \beta\,\frac{u_C(C_{t+1},1-L_{t+1})}{u_C(C_t,1-L_t)}. \] Proof: The Lagrangian for the household problem includes: (since utilities are increasing on consumption, the budget constraint is always binding) \[ \begin{aligned} \mathcal{L} &= E_0 \sum_{t=0}^{\infty} \beta^t \left[u(C_t,1-L_t) + \mu_t\left(w_t L_t + \sum_i x_{t-1}^i(p_t^i+d_t^i) + \Pi_t - C_t - \sum_i p_t^i x_t^i\right)\right] \\ &= E_0 \sum_{t=0}^{\infty} \color{blue}E_t\color{black} \beta^t \left[u(C_t,1-L_t) + \mu_t\left(w_t L_t + \sum_i x_{t-1}^i(p_t^i+d_t^i) + \Pi_t - C_t - \sum_i p_t^i x_t^i\right)\right] \end{aligned} \] First-order condition (FOC) w.r.t. \(x_t^i\): \[ -\beta^t u_C(C_t,\cdot)\,p_t^i + \beta^{t+1} E_t\!\left[u_C(C_{t+1},\cdot)(p_{t+1}^i+d_{t+1}^i)\right]=0. \] Dividing by \(\beta^t u_C(C_t,\cdot)p_t^i\): \[ 1 = E_t\left[\beta\frac{u_C(C_{t+1},\cdot)}{u_C(C_t,\cdot)}\frac{p_{t+1}^i+d_{t+1}^i}{p_t^i}\right] = E_t[M_{t+1}R_{t+1}^i]. \]

Firm Problem: Setup

Competitive Firm

The representative firm maximizes its value (priced by \(M_{t+1}\)): \[ \max_{\{I_t,L_t\}} E_0 \sum_{t=0}^{\infty} \left[\prod_{s=0}^{t-1} M_{s+1}\right] D_t, \] with dividends

\[ D_t \equiv Y_t - I_t - \mathcal{G}(I_t,K_t) - w_t L_t, \]

where \(\mathcal{G}\) captures (convex) adjustment costs in units of consumption, typically \(\mathcal{G}(I,K)=\frac{\kappa}{2}\left(\frac{I}{K}-\delta\right)^2 K\).

Capital accumulation: \[ K_{t+1}=(1-\delta)K_t+\Phi(I_t,K_t). \]

Let \(\lambda_t\) denote the Lagrange multiplier on capital accumulation in units of consumption (the marginal value of installed capital).

Firm Problem: First-Order Conditions

Define the lagrangian: \[ \mathcal{L} = E_0 \sum_{t=0}^{\infty} \left[\prod_{s=0}^{t-1} M_{s+1}\right] \left[D_t - \lambda_t\,(K_{t+1}-(1-\delta)K_t-\Phi(I_t,K_t))\right]. \]

Define \(S_t \equiv \prod_{s=0}^{t-1} M_{s+1}\) for notational convenience. The Lagrangian becomes

\[ \mathcal{L} = E_0 \sum_{t=0}^{\infty} S_t \left[D_t - \lambda_t\,(K_{t+1}-(1-\delta)K_t-\Phi(I_t,K_t))\right]. \]

F.O.C. conditions

\(I_t\)-FOC: \[ -1 - \mathcal{G}_I(I_t,K_t) + \lambda_t \Phi_I(I_t,K_t)=0 \quad \Rightarrow \quad \lambda_t = \frac{1+\mathcal{G}_I(I_t,K_t)}{\Phi_I(I_t,K_t)} \equiv q_t. \]

\(L_t\)-FOC: \[ F_L(K_t,Z_t L_t)\,Z_t = w_t. \]

F.O.C. w.r.t \(K_{t+1}\)

Envelope condition:

\[ \begin{aligned} \frac{\partial \mathcal{L}}{\partial K_{t+1}} &= E_0\left[S_t \left( - \lambda_t\right) + S_{t+1}\left(\frac{\partial D_{t+1}}{\partial K_{t+1}} + \lambda_{t+1}(1-\delta) + \lambda_{t+1}\Phi_K(I_{t+1},K_{t+1})\right)\right] = 0. \end{aligned} \]

Since \(K_{t+1}\) is chosen at time \(t\), we can condition on \(\mathcal{F}_t\)

\[ \begin{aligned} 0 &= E_t\left[-\lambda_t + M_{t+1}\left(\frac{\partial D_{t+1}}{\partial K_{t+1}} + \lambda_{t+1}(1-\delta+\Phi_K(I_{t+1},K_{t+1}))\right)\right]. \end{aligned} \]

Noting that \(\frac{\partial D_{t+1}}{\partial K_{t+1}} = F_K(K_{t+1},Z_{t+1}L_{t+1}) - \mathcal{G}_K(I_{t+1},K_{t+1})\):

\[ \lambda_t = E_t\left[M_{t+1}\left(F_K(K_{t+1},Z_{t+1}L_{t+1}) - \mathcal{G}_K(I_{t+1},K_{t+1}) + \lambda_{t+1}(1-\delta+\Phi_K(I_{t+1},K_{t+1}))\right)\right]. \]

The Return on Installed Capital

Consider the payoff of one marginal unit of installed capital at \(t\) (priced at \(q_t\)). Holding this unit from \(t\) to \(t+1\) delivers:

  • a cash flow (dividend) equal to the marginal contribution to profits: \[ d^K_{t+1} = F_K(K_{t+1},Z_{t+1}L_{t+1}) - \frac{\partial \mathcal{G}(I_{t+1},K_{t+1})}{\partial K_{t+1}}, \]

  • plus a resale value equal to its continuation marginal value after depreciation and spillovers: \[ q_{t+1}\cdot\big(1-\delta+\Phi_K(I_{t+1},K_{t+1})\big). \]

Thus, the gross return on installed capital is \[ R^K_{t+1} \equiv \frac{d^K_{t+1} + q_{t+1}\big(1-\delta+\Phi_K(I_{t+1},K_{t+1})\big)}{q_t}. \]

By the firm FOC and no-arbitrage: \[ 1=E_t\big[M_{t+1} R^K_{t+1}\big]. \]

q-Theory Link: Investment FOC

From the \(I_t\)-FOC: \[ q_t = \frac{1+\mathcal{G}_I(I_t,K_t)}{\Phi_I(I_t,K_t)}. \]

  • With no adjustment costs and linear installation \(\Phi(I,K)=I\), we get \(q_t=1\) (price of a new unit equals one unit of the consumption good).
  • With convex adjustment, \(q_t\) varies with \(I_t/K_t\): high investment raises \(q_t\).

This connects marginal q (a valuation object) to average q and investment rates, delivering testable implications.

Cochrane’s Production-Based Model

Core Pricing Condition

For any asset \(i\): \[ E_t[R_{t+1}^i] - R_{t+1}^f = -\frac{\text{Cov}_t(R_{t+1}^i,M_{t+1})}{E_t[M_{t+1}]},\qquad R_{t+1}^f \equiv \frac{1}{E_t[M_{t+1}]}. \]

For installed capital: \[ 1=E_t[M_{t+1}R^K_{t+1}]\quad \Longrightarrow \quad E_t[R^K_{t+1}]-R_{t+1}^f = -\frac{\text{Cov}_t(R^K_{t+1},M_{t+1})}{E_t[M_{t+1}]}. \]

Production-Based SDF Proxies

In equilibrium \(M_{t+1}=\beta \dfrac{u_C(C_{t+1},1-L_{t+1})}{u_C(C_t,1-L_t)}\).
Cochrane shows that innovations in investment opportunities—measured by \(q_t\), investment-to-capital \(I_t/K_t\), or related marginal conditions—serve as simple linear proxies for the SDF in the data: \[ M_{t+1}\approx a - b \cdot \Delta \log q_{t+1} \quad \text{or} \quad M_{t+1}\approx a' - b'\cdot \frac{I_{t}}{K_{t}} + \varepsilon_{t+1}, \] yielding the investment CAPM: assets that covary positively with good investment states (high \(q\), high \(I/K\)) have low expected returns.

Intuition and Investment CAPM

Intuition: Good times for investment (high \(q\)) are times when the marginal utility of consumption is low (low \(M\)). Assets that pay off in those states hedge poorly and must offer higher premia.

This is the key insight of production-based asset pricing: investment opportunities serve as a state variable that proxies for the SDF.

Start from \(1=E_t[M_{t+1}R^K_{t+1}]\).

Using the standard asset pricing equation, for any asset \(i\): \[ 1 = E_t[M_{t+1}R_{t+1}^i] = E_t[M_{t+1}]E_t[R_{t+1}^i] + \text{Cov}_t(M_{t+1}, R_{t+1}^i). \]

Rearranging: \[ E_t[R_{t+1}^i] = R_{t+1}^f - \frac{\text{Cov}_t(M_{t+1}, R_{t+1}^i)}{E_t[M_{t+1}]}. \]

If we use the return on capital as a factor proxy (linear SDF approximation), \[ M_{t+1}\approx a - b\,R^K_{t+1}, \] then \[ E_t[R_{t+1}^i]-R_{t+1}^f \approx b\;\text{Cov}_t(R_{t+1}^i,R^K_{t+1}), \] which is the investment-based CAPM form.

Investment-Specific Technology (IST): Setup

Suppose the investment good has a relative price \(Q_t\) in consumption units, or—equivalently—an efficiency \(\mu_t \equiv 1/Q_t\) so that one unit of consumption buys \(\mu_t\) units of effective investment:

\[ K_{t+1}=(1-\delta)K_t+\mu_t\,\Phi(I_t,K_t), \qquad \mu_t>0. \]

Output is still in consumption units: \[ Y_t=F(K_t,Z_t L_t). \]

Dividends (in consumption units): \[ D_t = Y_t - I_t - \mathcal{G}(I_t,K_t) - w_t L_t. \]

Let \(q_t\) again denote the marginal value of installed capital in consumption units.

IST: First-Order Conditions

\(L_t\)-FOC (unchanged): \[ F_L(K_t,Z_tL_t)\,Z_t = w_t. \]

\(I_t\)-FOC: \[ -1 - \mathcal{G}_I(I_t,K_t) + q_t\,\mu_t\,\Phi_I(I_t,K_t)=0 \quad \Rightarrow \quad q_t = \frac{1+\mathcal{G}_I(I_t,K_t)}{\mu_t\,\Phi_I(I_t,K_t)}. \]

Envelope / shadow value: \[ q_t = F_K(K_t,Z_t L_t) - \mathcal{G}_K(I_t,K_t) + E_t\!\left[M_{t+1}\,q_{t+1}\left(1-\delta+\mu_{t+1}\Phi_K(I_{t+1},K_{t+1})\right)\right]. \]

IST: Return on Capital

Return on installed capital: \[ R^K_{t+1} \equiv \frac{d^K_{t+1} + q_{t+1}\left(1-\delta+\mu_{t+1}\Phi_K(I_{t+1},K_{t+1})\right)}{q_t}, \] with \[ d^K_{t+1} \equiv F_K(K_{t+1},Z_{t+1}L_{t+1}) - \mathcal{G}_K(I_{t+1},K_{t+1}). \]

Pricing: \[ 1=E_t\!\left[M_{t+1} R^K_{t+1}\right]. \]

Key IST implication: shocks to \(\mu_t\) (or \(Q_t=1/\mu_t\)) move \(q_t\) and \(R^K_{t+1}\), altering investment opportunities and thereby risk premia through \(1=E_t[M_{t+1}R^K_{t+1}]\).

IST and the SDF

In competitive complete-markets equilibrium the exact SDF remains the MRS: \[ M_{t+1}=\beta\,\frac{u_C(C_{t+1},1-L_{t+1})}{u_C(C_t,1-L_t)}. \]

IST does not change the MRS definition; it changes returns via \(\mu_t\) (i.e., it changes what is being priced by the same \(M_{t+1}\)). A helpful linear proxy is: \[ M_{t+1}\approx a - b_1 \cdot \Delta \log q_{t+1} - b_2 \cdot \Delta \log \mu_{t+1}, \] so that \[ E_t[R_{t+1}^i]-R_{t+1}^f \approx b_1 \,\text{Cov}_t(R_{t+1}^i,\Delta \log q_{t+1}) + b_2 \,\text{Cov}_t(R_{t+1}^i,\Delta \log \mu_{t+1}). \]

Interpretation: assets that pay off when IST improves (high \(\mu_{t+1}\), cheap/effective investment) hedge low-\(M\) states poorly and earn higher premia.

Equilibrium: Market Clearing

Goods: \[ C_t + I_t + \mathcal{G}(I_t,K_t) = Y_t. \]

Labor and asset markets: \[ L_t \in [0,1], \qquad \text{and all asset holdings clear with net supply given by firms}. \]

Equilibrium Definition

An equilibrium is \(\{C_t,L_t,I_t,K_{t+1},q_t,w_t\}_{t\ge 0}\) and prices \(\{M_{t+1},R_{t+1}^i\}\) such that:

  1. Household chooses \(\{C_t,L_t,\{x_t^i\}\}\) to maximize utility given prices.
  2. Firm chooses \(\{I_t,L_t\}\) to maximize value given prices.
  3. Markets clear and constraints hold.
  4. Prices satisfy \(1=E_t[M_{t+1}R_{t+1}^i]\) for all traded assets \(i\).

Key Results (1): Euler for Capital

\[ 1=E_t\!\left[M_{t+1} \frac{d^K_{t+1} + q_{t+1}\left(1-\delta+\Phi_K^{\text{(IST)}}\right)}{q_t}\right], \] with \(\Phi_K^{\text{(IST)}} \equiv \mu_{t+1}\Phi_K(I_{t+1},K_{t+1})\) (equals \(\Phi_K\) without IST).

Proof of Result (1)

With IST, the capital accumulation constraint is: \[ K_{t+1}=(1-\delta)K_t+\mu_t\,\Phi(I_t,K_t). \]

The firm’s FOC w.r.t. \(I_t\) gives: \[ q_t = \frac{1+\mathcal{G}_I(I_t,K_t)}{\mu_t\,\Phi_I(I_t,K_t)}. \]

The envelope condition w.r.t. \(K_{t+1}\) gives: \[ q_t = E_t\left[M_{t+1}\left(F_K(K_{t+1},Z_{t+1}L_{t+1}) - \mathcal{G}_K(I_{t+1},K_{t+1}) + q_{t+1}(1-\delta+\mu_{t+1}\Phi_K(I_{t+1},K_{t+1}))\right)\right]. \]

This can be written as: \[ 1 = E_t\left[M_{t+1}\frac{d^K_{t+1} + q_{t+1}(1-\delta+\mu_{t+1}\Phi_K(I_{t+1},K_{t+1}))}{q_t}\right] = E_t[M_{t+1}R^K_{t+1}]. \]

Key Results (2): Investment q-FOC

\[ q_t = \frac{1+\mathcal{G}_I(I_t,K_t)}{\Phi_I(I_t,K_t)}\quad\text{(no IST)}, \] \[ q_t = \frac{1+\mathcal{G}_I(I_t,K_t)}{\mu_t\,\Phi_I(I_t,K_t)}\quad\text{(with IST)}. \]

Proof of Result (2)

The firm’s Lagrangian (with IST) includes: \[ \mathcal{L} = E_0 \sum_{t=0}^{\infty} \left[\prod_{s=0}^{t-1} M_{s+1}\right] \left[D_t - \lambda_t\,(K_{t+1}-(1-\delta)K_t-\mu_t\Phi(I_t,K_t))\right], \] where \(D_t = Y_t - I_t - \mathcal{G}(I_t,K_t) - w_t L_t\).

FOC w.r.t. \(I_t\): \[ -1 - \mathcal{G}_I(I_t,K_t) + \lambda_t \mu_t\Phi_I(I_t,K_t)=0. \]

Solving for \(\lambda_t = q_t\): \[ q_t = \frac{1+\mathcal{G}_I(I_t,K_t)}{\mu_t\,\Phi_I(I_t,K_t)}. \]

Without IST, \(\mu_t=1\), so: \[ q_t = \frac{1+\mathcal{G}_I(I_t,K_t)}{\Phi_I(I_t,K_t)}. \]

Key Results (3): Risk Premia

For any asset \(i\), \[ E_t[R_{t+1}^i]-R_{t+1}^f = -\frac{\text{Cov}_t(R_{t+1}^i,M_{t+1})}{E_t[M_{t+1}]}, \] and using linear SDF proxies in \(q\) and IST shocks yields investment-based factor models.

Proof of Result (3)

Start from the fundamental pricing equation: \[ 1 = E_t[M_{t+1}R_{t+1}^i] = E_t[M_{t+1}]E_t[R_{t+1}^i] + \text{Cov}_t(M_{t+1},R_{t+1}^i). \]

Rearranging: \[ E_t[R_{t+1}^i] = \frac{1 - \text{Cov}_t(M_{t+1},R_{t+1}^i)}{E_t[M_{t+1}]}. \]

Since \(R_{t+1}^f = 1/E_t[M_{t+1}]\): \[ E_t[R_{t+1}^i] - R_{t+1}^f = -\frac{\text{Cov}_t(M_{t+1},R_{t+1}^i)}{E_t[M_{t+1}]}. \]

To obtain factor models, project \(M_{t+1}\) onto investment-related variables (e.g., \(\Delta \log q_{t+1}\), \(\Delta \log \mu_{t+1}\)): \[ M_{t+1} = a + \sum_k b_k f_{k,t+1} + \varepsilon_{t+1}, \] where \(f_{k,t+1}\) are factors and \(E_t[\varepsilon_{t+1}f_{j,t+1}]=0\). Substituting into the risk premium equation yields: \[ E_t[R_{t+1}^i] - R_{t+1}^f \approx -\sum_k \frac{b_k}{E_t[M_{t+1}]}\text{Cov}_t(R_{t+1}^i,f_{k,t+1}). \]

Empirical Implications

  1. Investment Growth Predictability: High \(I_t/K_t\) (or high \(q_t\)) signals low future returns, as states with good investment opportunities have low \(M_{t+1}\).
  2. Return–q Covariance: Cross-sectional premia align with \(\text{Cov}(R^i,\Delta \log q)\) or \(\text{Cov}(R^i, R^K)\).
  3. IST Channels: Improvements in \(\mu_t\) (declines in the relative price of investment) strengthen investment opportunities and are priced in both time series and cross section.

Connections and Extensions

  • Adjustment Costs (Jermann, 1998; Zhang, 2005): Quantitatively match volatility and premia via \(\mathcal{G}\) and \(\Phi\) curvature.
  • Labor Frictions (Belo, 2010): Labor market dynamics amplify variation in \(q_t\) and risk premia.
  • Multi-Sector (Belo, Lin, Bazdresch, 2014): Sectoral investment opportunities improve cross-sectional fit.