Guest Lecture Risk Analysis of Investment-Based Models Xiaoji Lin London School of Economics and FMG University of Minnesota November 19, 2010 Lin Risk Analysis & Investment-Based Models November 19, 2010 1 / 32 Road Map 1 Dynamic programing and value function iteration 2 Benchmark investment-based model 3 Risk analysis Lin Risk Analysis & Investment-Based Models November 19, 2010 2 / 32 Dynamic Programing: Deterministic Case V (k ) s.t. k 0 = max0 ff (k, k 0 ) + βV (k 0 )g i ,k = (1 δ )k + i where k’means capital in the next period. If we de…ne the Bellman operator B (V ) which updates a value function V using the Bellman equation above, we know the following properties: 1 2 V such that B (V ) = V exists. V = limt !∞ B t (V 0 ) for any continuous function V 0 . In addition, B t (V 0 ) converges to V monotonically. In other words, if we supply an initial guess V 0 and keep applying the Bellman operator, we can asymptotically get to the solution V of the Bellman equation. Value function iteration is the solution method which uses the properties. Lin Risk Analysis & Investment-Based Models November 19, 2010 3 / 32 Algorithm: Deterministic Case 1 2 3 4 5 6 7 8 Set nk number of grid points, upper and lower bounds of k, and tolerance level ε Set grid points, usually equi-spaced. Set an initial value V 0 Update value function and obtain V 1 . For each grid point ki , choose the optimal kj that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 4. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 4 / 32 Algorithm: Deterministic Case 1 2 3 4 5 6 7 8 Set nk number of grid points, upper and lower bounds of k, and tolerance level ε Set grid points, usually equi-spaced. Set an initial value V 0 Update value function and obtain V 1 . For each grid point ki , choose the optimal kj that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 4. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 4 / 32 Algorithm: Deterministic Case 1 2 3 4 5 6 7 8 Set nk number of grid points, upper and lower bounds of k, and tolerance level ε Set grid points, usually equi-spaced. Set an initial value V 0 Update value function and obtain V 1 . For each grid point ki , choose the optimal kj that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 4. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 4 / 32 Algorithm: Deterministic Case 1 2 3 4 5 6 7 8 Set nk number of grid points, upper and lower bounds of k, and tolerance level ε Set grid points, usually equi-spaced. Set an initial value V 0 Update value function and obtain V 1 . For each grid point ki , choose the optimal kj that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 4. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 4 / 32 Algorithm: Deterministic Case 1 2 3 4 5 6 7 8 Set nk number of grid points, upper and lower bounds of k, and tolerance level ε Set grid points, usually equi-spaced. Set an initial value V 0 Update value function and obtain V 1 . For each grid point ki , choose the optimal kj that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 4. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 4 / 32 Algorithm: Deterministic Case 1 2 3 4 5 6 7 8 Set nk number of grid points, upper and lower bounds of k, and tolerance level ε Set grid points, usually equi-spaced. Set an initial value V 0 Update value function and obtain V 1 . For each grid point ki , choose the optimal kj that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 4. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 4 / 32 Algorithm: Deterministic Case 1 2 3 4 5 6 7 8 Set nk number of grid points, upper and lower bounds of k, and tolerance level ε Set grid points, usually equi-spaced. Set an initial value V 0 Update value function and obtain V 1 . For each grid point ki , choose the optimal kj that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 4. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 4 / 32 Algorithm: Deterministic Case 1 2 3 4 5 6 7 8 Set nk number of grid points, upper and lower bounds of k, and tolerance level ε Set grid points, usually equi-spaced. Set an initial value V 0 Update value function and obtain V 1 . For each grid point ki , choose the optimal kj that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 4. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 4 / 32 Dynamic Programing: Stochastic Case V (k, x ) s.t. k 0 Z = max0 ff (k, k 0 , x ) + β V (k 0 , x 0 )dx 0 jx g i ,k = (1 δ )k + i where x 0 is stochastic. For instance, x 0 is an AR(1) process: x 0 = (1 ρ)x̄ + ρx + σε0 There are many ways to compute the conditional expectation. Here we will use quadrature method to approximate the continuous x’process. 1 2 Tauchen (1986) and Tauchen and Hussey (1991) - don’t work well when x 0 is highly persistent Rouwenhorst (1995) can handle highly persistent AR (1) process. Lin Risk Analysis & Investment-Based Models November 19, 2010 5 / 32 Algorithm: Stochastic Case 1 2 3 4 5 6 7 8 9 Set nk number of grid points for k and nx number for x, upper and lower bounds of k, and tolerance level ε. Set grid points for k, usually equi-spaced. Apply Rouwenhorst (1995) to discretize x to get grid points for x and the transition matrix. Set an initial value V 0 . Note it is a two-dimensional object now. Update conditional expected value and update value function and obtain V 1 . For each grid point xi and kj , choose the optimal kh that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 5. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 6 / 32 Algorithm: Stochastic Case 1 2 3 4 5 6 7 8 9 Set nk number of grid points for k and nx number for x, upper and lower bounds of k, and tolerance level ε. Set grid points for k, usually equi-spaced. Apply Rouwenhorst (1995) to discretize x to get grid points for x and the transition matrix. Set an initial value V 0 . Note it is a two-dimensional object now. Update conditional expected value and update value function and obtain V 1 . For each grid point xi and kj , choose the optimal kh that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 5. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 6 / 32 Algorithm: Stochastic Case 1 2 3 4 5 6 7 8 9 Set nk number of grid points for k and nx number for x, upper and lower bounds of k, and tolerance level ε. Set grid points for k, usually equi-spaced. Apply Rouwenhorst (1995) to discretize x to get grid points for x and the transition matrix. Set an initial value V 0 . Note it is a two-dimensional object now. Update conditional expected value and update value function and obtain V 1 . For each grid point xi and kj , choose the optimal kh that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 5. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 6 / 32 Algorithm: Stochastic Case 1 2 3 4 5 6 7 8 9 Set nk number of grid points for k and nx number for x, upper and lower bounds of k, and tolerance level ε. Set grid points for k, usually equi-spaced. Apply Rouwenhorst (1995) to discretize x to get grid points for x and the transition matrix. Set an initial value V 0 . Note it is a two-dimensional object now. Update conditional expected value and update value function and obtain V 1 . For each grid point xi and kj , choose the optimal kh that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 5. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 6 / 32 Algorithm: Stochastic Case 1 2 3 4 5 6 7 8 9 Set nk number of grid points for k and nx number for x, upper and lower bounds of k, and tolerance level ε. Set grid points for k, usually equi-spaced. Apply Rouwenhorst (1995) to discretize x to get grid points for x and the transition matrix. Set an initial value V 0 . Note it is a two-dimensional object now. Update conditional expected value and update value function and obtain V 1 . For each grid point xi and kj , choose the optimal kh that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 5. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 6 / 32 Algorithm: Stochastic Case 1 2 3 4 5 6 7 8 9 Set nk number of grid points for k and nx number for x, upper and lower bounds of k, and tolerance level ε. Set grid points for k, usually equi-spaced. Apply Rouwenhorst (1995) to discretize x to get grid points for x and the transition matrix. Set an initial value V 0 . Note it is a two-dimensional object now. Update conditional expected value and update value function and obtain V 1 . For each grid point xi and kj , choose the optimal kh that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 5. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 6 / 32 Algorithm: Stochastic Case 1 2 3 4 5 6 7 8 9 Set nk number of grid points for k and nx number for x, upper and lower bounds of k, and tolerance level ε. Set grid points for k, usually equi-spaced. Apply Rouwenhorst (1995) to discretize x to get grid points for x and the transition matrix. Set an initial value V 0 . Note it is a two-dimensional object now. Update conditional expected value and update value function and obtain V 1 . For each grid point xi and kj , choose the optimal kh that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 5. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 6 / 32 Algorithm: Stochastic Case 1 2 3 4 5 6 7 8 9 Set nk number of grid points for k and nx number for x, upper and lower bounds of k, and tolerance level ε. Set grid points for k, usually equi-spaced. Apply Rouwenhorst (1995) to discretize x to get grid points for x and the transition matrix. Set an initial value V 0 . Note it is a two-dimensional object now. Update conditional expected value and update value function and obtain V 1 . For each grid point xi and kj , choose the optimal kh that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 5. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 6 / 32 Algorithm: Stochastic Case 1 2 3 4 5 6 7 8 9 Set nk number of grid points for k and nx number for x, upper and lower bounds of k, and tolerance level ε. Set grid points for k, usually equi-spaced. Apply Rouwenhorst (1995) to discretize x to get grid points for x and the transition matrix. Set an initial value V 0 . Note it is a two-dimensional object now. Update conditional expected value and update value function and obtain V 1 . For each grid point xi and kj , choose the optimal kh that gives highest Vi1,j . Compare V 0 and V 1 and compute the distance d. If d > ε, update value function V 0 = V 1 and go back to step 5. If d < ε, then we …nd the approximated optimal value function. Check if the bounds of state space is not binding. Make sure that ε is small enough. Make sure nk is large enough. Lin Risk Analysis & Investment-Based Models November 19, 2010 6 / 32 Tips for Numerical Work: Tony Smith 1 2 3 4 5 6 7 8 Start with the simplest possible model, preferably one with an analytical solution Add features incrementally. Never add another feature until you are con…dent of your current results. Use the simplest possible methods. Accuracy is more important than speed or elegance. When you learn (or develop) a new method, test it on the simplest possible problem, preferably one with an analytical solution. Don’t program when you are tired. Don’t program too quickly. Hamming’s motto: The Goal of Computing is Insight, Not Numbers. Lin Risk Analysis & Investment-Based Models November 19, 2010 7 / 32 Tips for Numerical Work: Tony Smith 1 2 3 4 5 6 7 8 Start with the simplest possible model, preferably one with an analytical solution Add features incrementally. Never add another feature until you are con…dent of your current results. Use the simplest possible methods. Accuracy is more important than speed or elegance. When you learn (or develop) a new method, test it on the simplest possible problem, preferably one with an analytical solution. Don’t program when you are tired. Don’t program too quickly. Hamming’s motto: The Goal of Computing is Insight, Not Numbers. Lin Risk Analysis & Investment-Based Models November 19, 2010 7 / 32 Tips for Numerical Work: Tony Smith 1 2 3 4 5 6 7 8 Start with the simplest possible model, preferably one with an analytical solution Add features incrementally. Never add another feature until you are con…dent of your current results. Use the simplest possible methods. Accuracy is more important than speed or elegance. When you learn (or develop) a new method, test it on the simplest possible problem, preferably one with an analytical solution. Don’t program when you are tired. Don’t program too quickly. Hamming’s motto: The Goal of Computing is Insight, Not Numbers. Lin Risk Analysis & Investment-Based Models November 19, 2010 7 / 32 Tips for Numerical Work: Tony Smith 1 2 3 4 5 6 7 8 Start with the simplest possible model, preferably one with an analytical solution Add features incrementally. Never add another feature until you are con…dent of your current results. Use the simplest possible methods. Accuracy is more important than speed or elegance. When you learn (or develop) a new method, test it on the simplest possible problem, preferably one with an analytical solution. Don’t program when you are tired. Don’t program too quickly. Hamming’s motto: The Goal of Computing is Insight, Not Numbers. Lin Risk Analysis & Investment-Based Models November 19, 2010 7 / 32 Tips for Numerical Work: Tony Smith 1 2 3 4 5 6 7 8 Start with the simplest possible model, preferably one with an analytical solution Add features incrementally. Never add another feature until you are con…dent of your current results. Use the simplest possible methods. Accuracy is more important than speed or elegance. When you learn (or develop) a new method, test it on the simplest possible problem, preferably one with an analytical solution. Don’t program when you are tired. Don’t program too quickly. Hamming’s motto: The Goal of Computing is Insight, Not Numbers. Lin Risk Analysis & Investment-Based Models November 19, 2010 7 / 32 Tips for Numerical Work: Tony Smith 1 2 3 4 5 6 7 8 Start with the simplest possible model, preferably one with an analytical solution Add features incrementally. Never add another feature until you are con…dent of your current results. Use the simplest possible methods. Accuracy is more important than speed or elegance. When you learn (or develop) a new method, test it on the simplest possible problem, preferably one with an analytical solution. Don’t program when you are tired. Don’t program too quickly. Hamming’s motto: The Goal of Computing is Insight, Not Numbers. Lin Risk Analysis & Investment-Based Models November 19, 2010 7 / 32 Tips for Numerical Work: Tony Smith 1 2 3 4 5 6 7 8 Start with the simplest possible model, preferably one with an analytical solution Add features incrementally. Never add another feature until you are con…dent of your current results. Use the simplest possible methods. Accuracy is more important than speed or elegance. When you learn (or develop) a new method, test it on the simplest possible problem, preferably one with an analytical solution. Don’t program when you are tired. Don’t program too quickly. Hamming’s motto: The Goal of Computing is Insight, Not Numbers. Lin Risk Analysis & Investment-Based Models November 19, 2010 7 / 32 Tips for Numerical Work: Tony Smith 1 2 3 4 5 6 7 8 Start with the simplest possible model, preferably one with an analytical solution Add features incrementally. Never add another feature until you are con…dent of your current results. Use the simplest possible methods. Accuracy is more important than speed or elegance. When you learn (or develop) a new method, test it on the simplest possible problem, preferably one with an analytical solution. Don’t program when you are tired. Don’t program too quickly. Hamming’s motto: The Goal of Computing is Insight, Not Numbers. Lin Risk Analysis & Investment-Based Models November 19, 2010 7 / 32 Benchmark Model a la Zhang (2005) Pro…t: π jt ktj +1 j = e xt +zt ktj = itj + (1 α f δ)ktj SDF: log Mt +1 γt = log β + γt (xt xt +1 ) = γ 0 + γ 1 ( xt x ) Adjustment costs: ctj 8 > > > a+ ktj + b + itj + > > < c (itj , ktj ) = 0 > > > > j j > : a kt + b it + where a > a+ , b adjustment costs. Lin < b + , and c c+ 2 c 2 itj k tj itj k tj 2 2 ktj if itj > 0 if itj = 0 ktj if itj < 0 > c + capture the nonconvexities of Risk Analysis & Investment-Based Models November 19, 2010 8 / 32 Benchmark Model Dividend: dtj v (ktj , xt , ztj ) = max fitj ,k tj +1 g dtj + ZZ π jt ctj Mt +1 v (ktj +1 , xt +1 , ztj +1 ) Qx (dxt +1 jxt ) Qz dztj +1 jztj Risk: Et [rtj +1 ] = rft + βjt λMt in which rft 1/Et [M t +1 ] is the real interest rate. And βjt is risk de…ned as: βjt Covt [rtj +1 , Mt +1 ] Vart [Mt +1 ] and λMt is the price of risk de…ned as λMt Lin Vart [M t +1 ]/Et [M t +1 ]. Risk Analysis & Investment-Based Models November 19, 2010 9 / 32 Calibration Parameters Lin α δ 0.60 0.10 z }| { Output ρx σx 4 0.014 0.98 z }| { Agg Shock Risk Analysis & Investment-Based Models ρz σz 0.70 0.35 z }| { Idio Shock November 19, 2010 10 / 32 Frictionless Model The production function is given by: j π jt = e xt +zt ktj 0.6 0 Adjustment costs: ctj Lin 8 > > > 0ktj + 1itj + 0 > > < c (itj , ktj ) = 0 > > > > j j > : 0kt + 1it + 0 itj ktj itj ktj Risk Analysis & Investment-Based Models 2 2 ktj if itj > 0 if itj = 0 ktj if itj < 0 November 19, 2010 11 / 32 Frictionless Model Value Functions 70 low z mid z high z 60 v 50 40 30 20 10 0 5 10 15 20 25 30 35 40 45 50 k Lin Risk Analysis & Investment-Based Models November 19, 2010 12 / 32 Frictionless Model Policy Functions 50 low z mid z high z 40 30 20 i 10 0 -10 -20 -30 -40 0 5 10 15 20 25 30 35 40 45 50 k Lin Risk Analysis & Investment-Based Models November 19, 2010 13 / 32 Frictionless Model Conditional Beta 0.45 low z mid z high z 0.4 0.35 β 0.3 0.25 0.2 0.15 0.1 0 5 10 15 20 25 30 35 40 45 50 k Lin Risk Analysis & Investment-Based Models November 19, 2010 14 / 32 Frictionless Model Quantitative Results: 5 productivity portfolios Z Y Low -0.68 0.43 2 -0.17 1.19 3 0.00 1.35 4 0.17 2.56 High 0.68 4.32 Lin I -1.69 -0.94 -0.69 1.46 7.85 Cost Div 0.00 2.11 0.00 2.06 0.00 2.04 0.00 0.64 0.00 -3.54 K 2.78 7.32 8.84 13.74 28.25 P 14.66 21.00 22.35 31.58 44.90 Risk Analysis & Investment-Based Models MB 4.21 2.30 2.18 1.69 1.47 Beta Ret 0.22 8.89 0.12 6.35 0.10 5.82 0.07 4.75 0.01 3.13 November 19, 2010 15 / 32 Convex Costs and Fixed Costs Quantitative Results: 5 BM portfolios Growth 2 3 4 Value Z -0.68 -0.17 0.00 0.17 0.68 Lin Y 0.43 1.19 1.35 2.56 4.32 I -1.33 -0.97 -0.26 0.91 7.63 Cost Div 0.00 1.73 0.00 2.08 0.00 1.58 0.00 1.21 0.00 -3.29 K 2.78 7.32 8.84 13.74 28.25 Risk Analysis & Investment-Based Models P 14.66 21.00 22.35 31.58 44.90 MB 4.21 2.30 2.18 1.69 1.47 Beta 0.22 0.12 0.10 0.07 0.01 November 19, 2010 Ret 8.89 6.34 5.82 4.75 3.13 16 / 32 Convex Costs The production function is given by: j π jt = e xt +zt ktj α 0 Adjustment costs: ctj Lin 8 > > > 0ktj + 1itj + > > < c (itj , ktj ) = 0 > > > > j j > : 0kt + 1it + 1 2 10 2 itj k tj 2 itj k tj Risk Analysis & Investment-Based Models ktj 2 if itj > 0 if itj = 0 ktj if itj < 0 November 19, 2010 17 / 32 Convex Costs Policy Functions b + = 1; b = 10; 1.2 low z mid z high z 1 0.8 i 0.6 0.4 0.2 0 -0.2 Lin 0 1 2 3 4 5 6 Risk Analysis & Investment-Based Models k 7 8 9 10 November 19, 2010 18 / 32 Convex Costs Policy Functions b + = 1; b = 10; 7 low z mid z high z 6 5 ik 4 3 2 1 0 -1 Lin 0 1 2 3 4 5 6 Risk Analysis & Investment-Based Models k 7 8 9 10 November 19, 2010 19 / 32 Convex Costs Conditional Beta b + = 1; b = 10; 0.65 low z mid z high z 0.6 0.55 β 0.5 0.45 0.4 0.35 0.3 0.25 Lin 0 1 2 3 4 5 6 Risk Analysis & Investment-Based Models k 7 8 9 10 November 19, 2010 20 / 32 Convex Costs Quantitative Results: 5 Productivity portfolios Z Y Low -0.63 0.30 2 -0.27 0.41 3 0.00 0.51 4 0.27 0.67 High 0.63 0.92 Lin I 0.11 0.18 0.25 0.35 0.51 Cost 0.03 0.03 0.04 0.05 0.07 Div 0.13 0.17 0.21 0.29 0.44 K 2.29 2.53 2.73 2.99 3.37 P 6.10 6.69 7.14 7.83 8.84 Risk Analysis & Investment-Based Models MB 2.66 2.63 2.62 2.61 2.60 Beta Ret 0.31 6.72 0.29 6.38 0.27 6.12 0.26 5.78 0.24 5.36 November 19, 2010 21 / 32 Convex Costs Quantitative Results: 5 BM portfolios Growth 2 3 4 Value Z -0.05 -0.01 0.02 0.04 0.00 Lin Y 0.48 0.54 0.58 0.63 0.68 I 0.30 0.30 0.29 0.28 0.24 Cost Div 0.06 0.11 0.05 0.18 0.05 0.23 0.04 0.30 0.03 0.41 K 2.19 2.50 2.74 3.01 3.46 P MB 6.62 2.97 7.05 2.75 7.37 2.61 7.73 2.48 8.21 2.29 Risk Analysis & Investment-Based Models Beta 0.29 0.28 0.27 0.26 0.25 Ret 6.33 6.13 5.99 5.88 5.76 November 19, 2010 22 / 32 Convex Costs and Operational Leverage The production function is given by: j π jt = e xt +zt ktj 0.6 0.07 Adjustment costs: ctj Lin 8 > > > 0ktj + 1itj + > > < c (itj , ktj ) = 0 > > > > j j > : 0kt + 1it + 1 2 10 2 itj k tj 2 itj k tj Risk Analysis & Investment-Based Models ktj 2 if itj > 0 if itj = 0 ktj if itj < 0 November 19, 2010 23 / 32 Convex Costs and Operational Leverage Quantitative Results: 5 Productivity portfolios Z Y Low -0.63 0.28 2 -0.27 0.38 3 0.00 0.46 4 0.27 0.60 High 0.63 0.83 Lin I 0.09 0.15 0.21 0.29 0.43 Cost Div 0.02 0.05 0.03 0.10 0.04 0.14 0.05 0.22 0.07 0.36 K 1.98 2.18 2.35 2.57 2.89 P 3.81 4.32 4.70 5.30 6.18 Risk Analysis & Investment-Based Models MB 1.79 1.85 1.89 1.95 2.02 Beta 0.29 0.26 0.24 0.22 0.19 Ret 9.33 7.91 7.30 6.61 5.94 November 19, 2010 24 / 32 Convex Costs and Operational Leverage Quantitative Results: 5 BM portfolios Growth 2 3 4 Value Z 0.34 0.13 0.01 -0.13 -0.34 Lin Y 0.62 0.56 0.53 0.51 0.46 I 0.34 0.27 0.23 0.20 0.13 Cost 0.06 0.05 0.04 0.03 0.02 Div 0.16 0.16 0.16 0.18 0.20 K 2.19 2.29 2.37 2.47 2.65 Risk Analysis & Investment-Based Models P 5.04 4.96 4.94 4.96 4.97 MB 2.16 1.99 1.89 1.80 1.66 Beta 0.23 0.23 0.23 0.24 0.25 Ret 6.26 6.72 7.19 7.71 9.25 November 19, 2010 25 / 32 Convex Costs and Operational Leverage Time Series Analysis: Firm Speci…c Productivity Firm-s pec ifc s hoc k 0.6 Low High 0.4 0.2 Lev el 0 -0.2 -0.4 -0.6 -0.8 50 100 150 200 250 300 350 400 450 500 Year Lin Risk Analysis & Investment-Based Models November 19, 2010 26 / 32 Convex Costs and Operational Leverage Time Series Analysis:Output Y 0.55 Low High 0.5 0.45 0.4 Lev el 0.35 0.3 0.25 0.2 0.15 0.1 0.05 50 100 150 200 250 300 350 400 450 500 Year Lin Risk Analysis & Investment-Based Models November 19, 2010 27 / 32 Convex Costs and Operational Leverage Time Series Analysis:IK IK 0.45 Low High 0.4 0.35 0.3 lev el 0.25 0.2 0.15 0.1 0.05 0 -0.05 50 100 150 200 250 300 350 400 450 500 Year Lin Risk Analysis & Investment-Based Models November 19, 2010 28 / 32 Convex Costs and Operational Leverage Time Series Analysis:Adj Costs Adj Costs 0.14 Low High 0.12 0.1 Lev el 0.08 0.06 0.04 0.02 0 50 100 150 200 250 300 350 400 450 500 Year Lin Risk Analysis & Investment-Based Models November 19, 2010 29 / 32 Convex Costs and Operational Leverage Time Series Analysis:Dividend Dividend 0.5 Low High 0.4 0.3 Lev el 0.2 0.1 0 -0.1 -0.2 50 100 150 200 250 300 350 400 450 500 Year Lin Risk Analysis & Investment-Based Models November 19, 2010 30 / 32 Convex Costs and Operational Leverage Time Series Analysis: Conditional Beta Beta 0.25 Low Low-High Zero High Z 0.2 Lev el 0.15 0.1 0.05 0 50 100 150 200 250 300 350 400 450 500 Year Lin Risk Analysis & Investment-Based Models November 19, 2010 31 / 32 Future Work 1 External …nancing: debt vs/& equity, credit risk, capital structure choices 2 Multipal capital goods: labor, intangible capital, etc 3 Additional aggregate risks 4 General equilibrium: aggregate implications Lin Risk Analysis & Investment-Based Models November 19, 2010 32 / 32