Firm Market Value and Investment: The Role of Firm’s Market Power and Different Types of Adjustment Costs August 2005 Nihal Bayraktar Penn State – Harrisburg and World Bank Plutarchos Sakellaris Athens University of Economics and Business, and IMOP
• Introduction • Models • Data • Indirect inference
Introduction
Neoclassical investment models with convex adjustment costs … • Empirical specifications are based on two simplifying assumptions – Profit function: Homogenous of degree one in capital – Capital adjustment cost function: Homogenous of degree one in capital and investment • As a result – Expected marginal Q (not observable) = Average Q (observable) – Coefficients of investment regressions give information about the parameters of the convex adjustment cost • Initially disappointing empirical results – Fundamentals (Tobin’s Q) cannot explain investment – Unreasonably high adjustment costs
One response: Introduction of nonlinearities in the investment process Example: Barnett and Sakellaris (1999) • Tobin’s Q and investment are nonlinearly related when a higher order, linearly homogenous, convex capital adjustment cost is used • Lessons: 1. Fundamentals (Tobin's Q) are informative for investment once nonlinearities are allowed. 2. Lower adjustment capital cost • Possible misspecification problem? – Their empirical specification is based on the simplifying assumptions
Recent studies question the validity of simplifying assumptions … 1) Constant returns to scale profit function or perfectly competitive product market may not be correct. Monopoly power or decreasing returns to scale : • Cooper and Haltiwanger (2003) at the plant level • Bayraktar (2002), and Cooper and Ejarque (2003a and 2003b) in COMPUSTAT data at the firm level • Bayraktar, Sakellaris, and Vermeulen (2004) with German firm-level data.
Recent studies question the validity of simplifying assumptions … 2) Linearly homogenous convex adjustment costs may not be sufficient to capture different types of investment costs. Non-convex adjustment costs : • Cooper and Haltiwanger (2003) • Bayraktar (2002) • Bayraktar, Sakellaris, and Vermeulen (2004)
Our paper … • Can a dynamic investment model with more realistic assumptions replicate the nonlinear relationship between investment and Tobin's Q? – Non-convex adjustment cost function (fixed cost) – Profit function with decreasing returns to scale • Focus is on Barnett and Sakellaris (1999)
MODELS
Model used in Barnett and Sakellaris (1999) Value maximization V A it , K it max A it , K it − C K it , I it E A it 1 ∣ A it V A it 1 , K it 1 , I it subject to I it K it 1 − 1 − K it Π (·): homogenous degree one in K
Higher order, linearly homogenous, convex cost function 2 K it C K it , I it pI it 1 I it 2 I it 2 K it 3 K it 4 4 K it t I it i I it 3 I it I it 3 4 K it K it Barnett and Sakellaris (1999)’s empirical specification • First order condition produces … 2 4 i it 3 p t i − it 1 Q it 1 1 2 i it 3 i it
Alternative model Profit function A it , K it A it K it where θ is the profitability parameter. If θ < 1, decreasing returns to scale.
Adjustment Costs • Convex costs 2 K it . I it 2 K it • Fixed costs FK it .
Value maximization V ∗ A it , K it max V a A it , K it , V na A it , K it K it 1 A it , K it − C j K it , I it E A it 1 ∣ A it V ∗ A it 1 , K it 1 V j A it , K it max Investment cost in case of capital adjustment 2 K it FK it C a K it , I it pI it I it 2 K it Investment cost in case of no capital adjustment C na K it , I it 0
• Set of structural parameters { , r , , , , F } • Transition matrix P A it 1 | A it
Data
Data Set • Unbalanced panel: 1561 U.S. manufacturing firms • Period: 1960-1987 • 23,207 observations. • COMPUSTAT database (Hall, 1990).
Summary Statistics Mean median St. 25 th 75 th dev percentile percentile I it / K it-1 0.20 0.15 0.24 0.09 0.023 1.77 1.23 2.12 0.84 1.94 Q it 1.68 1.20 1.83 0.83 1.86 Q it+1 β Q it+1 - P 0.60 0.14 1.76 -0.22 0.77
Distribution of the Investment Rate 16.00 14.00 12.00 10.00 8.00 6.00 4.00 2.00 0.00 3 6 9 2 5 8 1 4 7 3 6 9 2 3 > 0 0 0 1 1 1 2 2 2 . 3 3 3 4 2 0 . . . . . . . . . . . . . 4 0 0 0 0 0 0 0 0 0 0 0 0 0 . = 0 <
Profitability and Shocks Estimation of the profit function and the profit shocks θ is estimated at 0.87, with a standard error of 0.07.
Calculation and decomposition of the profit shocks A it , K it c ∗ w it L it A it / c w it L it / K it
A it / c • Decompose into a fixed component, and time varying component by regression the log of on (a A it / c constant and) fixed firm effects. – Residuals of the regression = (total profit shocks) a it • They are estimates of the time varying part of the profit a t a it shock (in logs): • Split to obtain estimates of the aggregate and the a it idiosyncratic components a t and a it by regressing a it on time dummies. – Residuals of the regression = (idiosyncratic shocks) a it – Time dummies = (aggregate shocks) a t
Features of the firm demeaned profit shocks (in logs): a it Min: -0.82 Max: 0.71 a it Std. dev. : 0.24 a t Std. dev. : 0.078 a it Autocorrelation : 0.67
Tobin’s Q Numerator = market value of common stock + liquidating value of preferred stock + market value of long-term debt + book value of short-term debt Denominator = Replacement value of fixed capital + Replacement value of inventories
The relationship between investment and Tobin’s Q 3 p t i − it 1 2 4 i it Q it 1 1 2 i it 3 i it Q it 1 = present discounted value of end-of- period average Q p = relative price of new investment
Table 3: Auxiliary Regression Barnett and Sakellaris (1999) (Table 4 on page 257) 1.44* (0.08) i it i it 2 -0.36* (0.04) i it 3 0.023* (0.003) 0.65 Adjusted R-sq Note: The dependent variable is Q it 1 − P t . * significant at the 1% level.
Structural Estimation Indirect Inference
Structural Estimation r = 0.0413, β = 1/(1+ r ) = 0.96, δ = 0.1, p = 0.978, θ = 0.87. Structural parameters to be estimated Θ ≡ , F
Indirect inference: a) Fix Θ . Solve dynamic program. Generate corresponding optimal policy functions. b) Use these policy functions and arbitrary initial conditions to generate simulated data (10 panels × 1000 firms × 27 years). c) Use simulated data to calculate the model analogues of coefficients Θ J Θ d − s Θ ′ W d − s Θ d) min where W is a weighting matrix (3x3).
Definition of Tobin’s Q in the model E Ait 1 ∣ Ait V ∗ A it 1 , K it 1 Q it 1 pK it 1 E A it 1 ∣ A it V ∗ A it 1 , K it 1 = present discounted future value of the firm K it+1 = end-of-period capital stock
Estimates of the structural parameters Parameter Estimate γ 0.020 F 0.045
Auxilliary Regression Coefficients Actual vs simulated data Dependent variable: Q it 1 − p Coefficient Data Std. error Model Std. error 1.44 (0.08) 0.312 (0.016) i it -0.36 (0.04) -0.067 (0.010) 2 i it 3 0.023 (0.003) 0.008 (0.002) i it
Comparing moments DATA MODEL corr( β Q it+1 -p , i it ) 0.23 0.24 mean( β Q it+1 -p ) 0.60 0.89 stdev( β Q it+1 -p ) 1.76 0.82 mean( i it ) 0.20 0.20 stdev( i it ) 0.24 0.64
Initial Result • Even in the absence of homogeneity assumptions of the conventional Q-theory, the nonlinear and significant responsiveness of investment to changes in average Q can be explained.
THE END
Features of the distribution of the investment rate I it /K it-1 < 0.01 0.004 I it /K it-1 < 0.025 0.014 I it /K it-1 > 0.2 0.319 I it /K it-1 > 0.25 0.216 Corr( I it /K it-1 , I it-1 /K it-2 ) 0.310
Estimates of the structural parameters Parameter Estimate Standard error γ 0.020 F 0.045
Recommend
More recommend