Research " THE EFFECTS OF MACROECONOMIC UNCERTAINTY ON IRREVERSIBLE INVESTMENT " pptx

First, I develop a methodology for analyzing the effects of demand and supply uncertainty on the level of investment where sunk costs interact with the both uncertainty measures in the e

THE EFFECTS OF MACROECONOMIC UNCERTAINTY

ON IRREVERSIBLE INVESTMENT

A Dissertation submitted to the Faculty of the Graduate School of Arts and Sciences

of Georgetown University

in partial fulfillment of the requirements for the

degree of Doctor of Philosophy

in Economics

Ayse E Sile, M.A

Washington, DC July 2, 2003

ert

ee ey " 5) — cf oof

UMI Number: 3107363

Copyright 2003 by Sile, Ayse Esin

All rights reserved

INFORMATION TO USERS

The quality of this reproduction is dependent upon the quality of the copy submitted Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignment can adversely affect reproduction

In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted Also, if unauthorized copyright material had to be removed, a note will indicate the deletion

®

UMI

UMI Microform 3107363 Copyright 2004 by ProQuest Information and Learning Company

All rights reserved This microform edition is protected against unauthorized copying under Title 17, United States Code

ProQuest Information and Learning Company

300 North Zeeb Road P.O Box 1346 Ann Arbor, MI 48106-1346

Copyright 2003 by Ayse E Sile All Rights Reserved

il

submitted to the faculty of Economics

in partial fulfillment of the requirements for the degree of

Doctor of Philosophy

This dissertation is complete and satisfactory in all respects, and any and all revisions required

by the final examining committee have been made

Director of Graduate Studies ‘ Signatufe

M Daniel Westbrook " ⁄ 27 Dawrik 2z ool,

THE EFFECTS OF MACROECONOMIC UNCERTAINTY

ON IRREVERSIBLE INVESTMENT

Ayse E Sile, M.A

Thesis Advisor: Susan M Collins, Ph.D

ABSTRACT

I study the effects of demand and supply uncertainty on irreversible investment My objective is two fold First, I investigate whether demand and supply uncertainty have similar effects on investment Second, I study the effects of sunk costs on investment

I make three contributions to the existing literature First, I develop a methodology for analyzing the effects of demand and supply uncertainty on the level of investment where sunk costs interact with the both uncertainty measures in the estimation equation Second, I explore the extent to which the uncertainty-investment relationship is affected by the magnitude of sunk costs Third, I address the econometric issues raised by estimating a regression equation in which generated regressors appear

I use two different data sets in my empirical analyses In Chapters 1 and 2, I evaluate the effects of industry-wide uncertainty and sunk costs on the level of investment in 2-digit Turkish manufacturing industries In Chapter 3, I investigate how demand and supply uncertainty, and sunk costs affect the aggregate manufacturing investment in thirteen OECD countries

In Chapters 1 and 2, I find that the effect of supply uncertainty on investment is negative in full sample estimates and in the sample of industries where demand is price elastic The estimated impact, however, is positive in the sample of industries where demand is price

ili

inelastic I also find that investment is negatively associated both with sunk costs and the demand uncertainty

In Chapter 3, I find that the negative quantitative impact of the demand uncertainty on investment is larger than the impact of the supply uncertainty measure and that the estimated effects of the sunk cost proxy and the demand uncertainty measure are negative and statistically significant The results suggest that the methodology provides quite a good fit for the data on the OECD countries Therefore, the empirical results suggest that interaction variables are key

in studying sign of the uncertainty-investment relationship

These empirical findings could have important implications for countries They suggest that policy makers should take into consideration the interactions between different types of uncertainty and sunk costs

IV

Acknowledgments

I would like to thank Prof Susan M Collins for her support and guidance throughout this research She was always encouraging and considerate She helped me overcome all the difficulties I had during my research

I am especially grateful to my committee members Their supervision was enjoyable and ultimately informative I would like to thank Prof Robert Hussey, S.J for his diligent efforts in reviewing my work and for his thoughtful comments and suggestions I would also like to thank Prof Arik Levinson for his invaluable comments and for the interesting discussions

Last but not least, my innermost appreciation is to my family- my parents Aydan and Halit, and my husband Ihsan My parents provided me with the support, understanding, and motivation I almost always needed throughout this journey They have been always standing by

me, with their unconditional love and cheerful spirit As for my husband, I am thankful for him for his invaluable help, continuous moral support, and distinct understanding I owe a lot to you all Thank you

Dedication

I dedicate this dissertation to my grandmother, Muyesser Turkmenoglu and to my uncle, Osman Turkmenoglu, both of whom passed away during the course of this research I miss you both dearly

vi

Table of Contents

InfrOdUCfiOT SH TT Ki Đo B n n n BE BE nh tà ng tà tàn tàn ng 8 Chapter 1: The Impact of Macroeconomic Uncertainty and Sunk Costs on Investment

1.1: Introducfion -. ni HH KH nh nh nh nh nh hà hy

1.3: Theoretical ffamewWOFK nh nh nh nh ra 1.4: Methodology - cà nh nh nà nh nà Km nh TH bi tà ky 1.5: Data and preliminary sfafISEÍCS ccQQ nn HH nhe nh kia 1.6: Estimation r€SUÏfS - TH SH HT KT Ki kh kh By 1.7: CONCIUSION 00 ccc ecc cece cseeseseeeeseeevsuuesereeuseesrbeeeesaeesereaeeseuesees Chapter 2: Macroeconomic Uncertainty, Sunk Costs, and Investment: Time-Series and Cross-Sectional Evidence From Turkish Manufacturing c cà cv

ĐH ¡c(0)01)19:10).aaaAIaiadđaiaiiaaaaắẳẢẮ

2.3: Theoretical model and methodology - chưa 2.4: Data and preliminary sfatiSHICS - HH HH nhe nhe vn 2.5: Estimation r€SUÏfS - cu nu HH TT ng ky nh và

2.6: Conc[UsiOn cuc 0n ng HH HH TK Km TK n nh ky xa Chapter 3: Macroeconomic Uncertainty, Sunk Costs, and Aggregate Investment:

Empirical Evidence From the OECD Countries - c.c 2<:

vi

3.4: Data and preliminary s†afISfICS c HH n HH nh nh 74

3.6: Conclusion nọ nọ HH Ki Ki Ki Kế nộ nà ĐK nh vết 84 Appendix A: Data description and sources for Chapter Í - «cà ceằ 95 Appendix B: Computation of coefficients and standard errors for marginal effects in

Tables 5, 7, 11, 13, and 2( cọ HH HT HT HT nà Kì nh HH 98 Appendix C: Theoretical model for Chapfer 2 -. HH nh nhe ha 99 Appendix D: Data description and sources for Chapter 2 - -.- Ặ 2c se 101 Appendix E: Specification tests for the uncertainty measures 1n Chapter 2 103 Appendix F: Correction method for øenerafed T€ĐT€SSOTS uc S snnnnesằ 105 Appendix G: Data description and sources for Chapter 3 cẰc.Ằ 109 Appendix H: Specification tests for the uncertainty measures Chapter 3 111 Appendix I: Summary of existing literature on sunk cOsf DfOX1€S - 112

List of Figures

Figure 1: Real GDP Growth Rates (percent) in Turkey, (1970-2000) 27 Figure 2: Annual Inflation, WPI (percent) in Turkey, (1970-2000) ceceseceeceeeeneeees 27 Figure 3: Investment in Aggregate Manufacturing 1n Turkey, (1970-2000) 28 Figure 4: Real Exchange Rate Index 1n Turkey, (1980-2000) cccccẰ 3 28 List of Tables

Table 1: Descriptive statistics on investment in sixteen Turkish manufacturing

1TUSẨFI€S Q cọ HH TH nh ng KT Ki KT nọ kh Ki Ki Ki Ki KH ti Ki ke ky 29 Table 2: Descriptive statistics on the supply and demand uncertainty measures

Vill

and sunk cost measure across sixteen Turkish manufacturing Industrles 30 Table 3: Descriptive statistics on the supply and demand uncertainty measures

across sixteen Turkish manufacturing 1ndusfr1€s .‹ ‹-c-<c-<c< << 31 Table 4: Panel estimation results for the investment-value added ratio as a function of uncerfainfy measures, sunk cost measure, and OufpUt - nền 32 Table 5: Net effects of change in the sunk cost measure, demand uncertainty measure, and supply uncertainty measure on the investmenft-value added ratio - - -‹- 33 Table 6: Panel estimation results for the investment-value added ratio as a function of uncertainty measure, sunk cost measure, and oufput in industry ØTOUDS - 34 Table 7: Net effects of change in the sunk cost measure, demand uncertainty measure, and supply uncertainty measure on the investment-value added ratio in industry groups 35 Table 8: Descriptive statistics on labor productivity and inventory series in sixteen Turkish Manufacturing INGUSUFICS cece cece eee ceca eae e cece ease ncaa eetea seen ee eatateeetee pense ned 57 Table 9: Descriptive statistics on the demand and supply uncertainty measures and

sunk cost mneasure 1n sixteen Turkish manufacturing Industri€s - 38 Table 10: Panel estimation results for the investment-value added ratio as a function of uncertainty measures, sunk cosf me€asure, and Oufpuf ‹ ‹-.‹-c cv s<+2 59

Table 11: Net effects of change in the sunk cost measure, demand uncertainty measure, and supply uncertainty measure on the investment-value added ratio -.-.- - 60 Table 12: Panel estimation results for the investment-value added ratio as a function of the uncertainty measures, sunk cost measure, and outpuf 1n Industry øT0ups 61 Table 13: Net effects of change in the sunk cost measure, demand uncertainty measure, and supply uncertainty measure on the investment-value added ratio in elastic and inelastic industry

ix

Table 14: Aggregate empirical studies on the sign of uncertainty-investment relationship 87 Table 15: Disaggregate empirical studies on the sign of uncertainty-investment relationship 89 Table 16: Effects of different uncertainty measures that are used in empirical analyses in Tables

Table 17: Descriptive statistics on investment and output in aggregate manufacturing in thirteen 60); 00v000(0ý1- T6) 92) ch khttŸŸẢẢ4ẢẢ 91 Table 18: Descriptive statistics on the demand and supply uncertainty measures and the sunk cost proxy in thirteen OECD Countries ccceceee eee e cence eee sent eee neenae tee eneneenaeeaeaee es 92 Table 19: Panel estimation results for the investment-value added ratio as a function of uncertainty measures, sunk cost proxy and output in thirteen OECD countries 93 Table 20: Net effects of change in the sunk cost proxy, demand uncertainty measure, and supply uncertainty measure on the investment-value added ratio in thirteen OECD countries 94

Introduction

Given the importance of investment for a nation’s economic performance, it comes as

no surprise that analysis of investment remains a subject of key theoretical and empirical concern to economists Within the broad corpus of work on investment, the real options approach to investment literature focuses on how uncertainty affects investment when investment expenditures are irreversible and when there is uncertainty over the future returns from the investment The theoretical work in this literature shows that greater uncertainty may have a negative or a positive impact on irreversible investment depending on the choice of parameters of the underlying model Although most empirical work finds a negative association between uncertainty and investment, empirical results are far from establishing a consensus about the sign of the uncertainty-investment relationship

In this work, I utilize a theoretical framework from the real options approach to investment literature to study the effects of demand and supply uncertainty on the level of investment I investigate whether or not demand and supply uncertainty have similar effects on investment I also study the extent to which the uncertainty-investment relationship is affected

by sunk costs I make three contributions to the existing literature First, I develop and implement a methodology for analyzing the effects of both demand and supply uncertainty on the level of investment Second, I explore the extent to which the uncertainty-investment relationship is affected by the magnitude of sunk costs Third, I address the econometric issues raised by estimating a regression equation in which generated regressors appear

I study the impact of uncertainty and sunk costs on investment using a theoretical model developed by Caballero and Pindyck (1996) In the model, the average value of a firm’s output fluctuates stochastically Firms invest when the average value of industry output hits an

optimally derived investment trigger The investment trigger is a function of sunk costs, demand and supply uncertainty, and the price elasticity of demand There is a negative relationship between the trigger and the level of investment

Theoretically, I show that all else equal (a) An increase in sunk costs decreases the level

of investment, (b) An increase in demand uncertainty decreases the level of investment, and (c)

An increase in supply uncertainty decreases (increases) the level of investment if demand is price elastic (inelastic) I then build on these theoretical predictions and construct an estimation equation for the level of investment According to this methodology, sunk costs interact with the demand and supply uncertainty measures in the estimation equation

I test the theoretical results using two different data sets In Chapters 1 and 2, I evaluate the effects of industry-wide uncertainty and sunk costs on the level of investment in 2-digit Turkish manufacturing industries using panel data over the years 1977-2000 In Chapter 3, I investigate how demand and supply uncertainty, as well as sunk costs affect the aggregate manufacturing investment in thirteen OECD countries using panel data over the years 1975-

2001

I construct measures for both demand and supply uncertainty in my empirical analysis

in Chapter 1 Both of these measures are industry-specific I use data on industry output and hourly labor input to construct an index of labor productivity The sample variance of this index

is the supply uncertainty measure I use inventory data to construct a series of annual change in inventories The sample variance of this series is the demand uncertainty measure Using industry-specific data, I construct a ratio of fixed assets to net wealth, which is the sunk cost

measure.

In Chapter 1, I use time-invariant uncertainty measures However, the assumption that uncertainty measures are time-invariant is difficult to credit based on my sample In addition, almost all empirical work uses conditional variance Therefore in Chapter 2, I study the effects

of industry-wide uncertainty and sunk costs on the level of Turkish investment using time- variant uncertainty measures I use a univariate ARCH (1) method to model both uncertainty measures, Consequently, I exploit both the cross-sectional and time-series variation in the data The data set is the same as in Chapter 1 In order to make sound inferences from the estimation results, I make the necessary allowances in calculating the standard errors of the estimated coefficients using a methodology introduced by Pagan (1984) I extend his methodology to accommodate multiple generated regressors and interaction variables

My empirical findings in both Chapters 1 and 2 show that the effect of supply uncertainty measure on investment is negative in full sample estimates and in the sample of industries where demand is price elastic On the other hand, the estimated impact of supply uncertainty on investment is positive in the sample of industries where demand is price inelastic In addition, I find that investment is negatively associated both with the sunk cost measure and the demand uncertainty measure Finally, the empirical results show that the interaction variables explain much of the variation in the level of investment across industries in the data This is an important finding because the impact of interaction variables on the level of investment has not been fully studied in the existing empirical literature

In Chapter 3, I apply the methodology developed in Chapter 1 to panel data on aggregate manufacturing investment in thirteen OECD countries I use the conditional variance

of change in inventories as the demand uncertainty measure and the conditional variance of an index of labor productivity as the supply uncertainty measure Similar to Chapter 1, I use a

univariate ARCH (1) method to model both uncertainty measures I use Turkish industry- specific sunk cost proxy and the country-specific industry distribution in the manufacturing sector across the countries in my sample to construct the country-specific sunk cost proxy All data is aggregate and country-specific

The key empirical findings are (1) The negative quantitative impact of the demand uncertainty on investment is larger than the impact of the supply uncertainty measure Estimated

at sample means, the investment ratio decreases by 1.7 percent for one percent increase in the demand uncertainty measure, whereas it decreases only by 0.5 percent for one percent increase

in the supply uncertainty measure, and (2) The estimated effects of the sunk cost proxy and the demand uncertainty measure are negative and statistically significant Consequently, I find that greater demand uncertainty and higher sunk costs lower the level of investment across countries

The empirical results also show that the methodology developed in this Chapter provides quite a good fit for the data on the OECD countries In fact, the adjusted RỂ is higher when compared to similar empirical studies on the investment-uncertainty relationship across the OECD countries In addition, the estimation equation is a relatively simple model with a parsimonious fit Therefore, the empirical results suggest that interaction variables are key in studying sign of the uncertainty-investment relationship

As is the case in almost all-empirical work, further research is desirable in certain areas First, it is possible that the demand and supply uncertainty measures are endogenous Endogeneity needs to be taken into consideration in interpreting the empirical evidence I was not able to find good instruments for the uncertainty measures due to data limitations In future work, I hope to obtain data to solve this issue Second, the sunk cost measure used in Chapter 3

is not a perfect proxy It is only partially based on individual country data Therefore, it is desirable to construct a sunk cost measure based on each individual country’s data on net wealth and fixed assets in future work Another possibility is to construct alternative proxies for sunk costs, which could detect advertising or R&D expenditures

These empirical findings could have important implications for both developed and developing countries Having a stable and growing investment record is always considered a key strategy for achieving accelerated and sustained growth objectives in an economy My empirical results suggest that it is important to differentiate between demand and supply uncertainty The sign of the uncertainty-investment relationship not only depends on the type of uncertainty but also on the magnitude of the price elasticity of demand The empirical results also suggest that policy makers should take into consideration the interactions between different types of uncertainty and sunk costs These results may also help to explain why some industries

do better than others when they all face uncertainty over the future returns

Chapter 1: The Impact of Macroeconomic Uncertainty and Sunk Costs on Investment 1.1 Introduction

Although the uncertainty-investment relationship has received a great deal of theoretical attention in recent years, there is little consensus about the sign of this relationship Different theories emphasize different channels, some pointing to a positive and some to a negative relationship The real options approach to investment literature focuses on how uncertainty affects investment when investment expenditures are irreversible This literature shows that when investment is irreversible and can be delayed, it becomes very sensitive to uncertainty over future payoffs These models show that the effect of greater uncertainty on investment depends on the underlying parameters of the model The empirical literature is scant compared

to its theoretical counterpart and is mostly confined to a few single-country studies focusing on the UK and the US The empirical results indicate that investment may increase or decrease when there is greater uncertainty

I analyze the impact of macroeconomic uncertainty on the level of investment in 2-digit Turkish manufacturing industries using panel data for the years 1977-2000 In addition, I explore the extent to which the uncertainty-investment relationship is affected by the degree of industry-wide irreversibility, an issue that has not received much attention in existing empirical work,

I make three contributions to the existing literature First, I theoretically show that the sign of the supply uncertainty and investment relationship depends on the magnitude of the price elasticity of demand If demand is price elastic (inelastic) the relationship is negative (positive) Demand uncertainty, on the other hand, is always negatively associated with investment Second, I develop and empirically implement a methodology for analyzing the

effects of both demand and supply uncertainty on investment across industries Third, I examine the importance of industry-specific sunk costs as a determinant of investment According to the empirical model I develop, sunk costs interact with supply and demand uncertainty In other words, the effect of uncertainty on the level of investment varies with how high sunk costs are The empirical results indicate that these interaction variables are important in understanding investment behavior across industries in Turkey

In the next section, I review the theoretical and empirical results in the real option approach to investment literature Section 1.3 introduces the theoretical model Section 1.4 lays out the methodology Section 1.5 describes the data and explains the construction of the demand and supply uncertainty measures Section 1.6 presents the empirical results and Section 1.7 concludes

1.2 Literature review

Economists have been trying to develop models that would explain and predict changes

in investment spending for some time The most frequently used specifications for the analysis

of investment spending have been the neoclassical model pioneered by Jorgenson (1963) and the q-theory of investment developed by Tobin (1969)' The basic foundation of these models is the net present value rule According to this rule, investment is undertaken when the present value of expected cash flows is at least as large as the costs The net present value rule, however, is silent on the irreversible nature of investment projects’ Most investment expenditures, however, are at least in part irreversible In other words, sunk costs cannot be recovered should market conditions change adversely

' Chirinko (1993) offers a review of the literature on investment

* One can always redefine the net present value rule by subtracting from the conventional calculation the opportunity cost of exercising the option to invest, and state the rule as “invest if NPV is positive.” Such a terminology is fine as long as the definition includes all relevant option value characteristics of investment

7

The real options approach to investment literature (McDonald and Siegel (1986), Bertola (1988), Pindyck (1988), Bertola and Caballero (1994), and Dixit and Pindyck (1994), Dixit, Pindyck, and Sodal (1999)) shows that the ability to delay an irreversible investment can profoundly affect the decision to invest According to this literature, a firm with an opportunity

to invest is holding an option analogous to a financial option When a firm makes an irreversible investment, it kills its option to invest It gives up the possibility of waiting for new information

to arrive that might affect the desirability or timing of investment This literature also shows that the opportunity cost is highly sensitive to macroeconomic uncertainty over future project values, so that changing market conditions that affect the riskiness of future cash flows can have

a large impact on investment spending These models focus on the threshold that triggers investment and explore whether it depends on measures of uncertainty Investment is undertaken only when projects yield an expected return that hits or exceeds this optimally derived trigger’ Dixit and Pindyck (1994) show that an increase in the investment trigger decreases the level of investment

Although the real options approach to investment literature has received a great deal of theoretical attention in recent years, there is little consensus on how uncertainty affects irreversible investment One body of this literature (Bertola 1988, Dixit 1989, Bertola and Caballero 1994) shows that greater uncertainty about the expected future returns of investment projects increases the investment trigger Consequently, greater uncertainty leads to less willingness to invest This result occurs because the firms anticipate that the irreversibility constraint may bind in the future and thus are more reluctant to invest today Abel and Eberly

* Empirical work shows that the investment trigger is typically three or four times higher than the Jorgensonian user cost of capital Abel and Eberly (1995) discuss this issue in detail

8

(1995), however, question these theoretical results They show that whether or not uncertainty implies a lower level of investment depends on the parameters of the model

Caballero and Pindyck (1992) and Dixit and Pindyck (1994) show that, when competitive industry equilibrium is considered, there is a negative relationship between industry investment and industry-wide demand uncertainty When imperfect competition is introduced, the intuition behind the above argument remains intact (Dixit and Pindyck (1994)) Even in oligopolistic industries, firms have an incentive to delay investment in the presence of uncertainty Thus, via the irreversibility channel, greater uncertainty is expected to decrease investment However, fear of preemption by a rival and the consequent need to act quickly may counteract the desire to wait Hence, without specifying the nature of a strategic interaction among oligopolists, it appears difficult to predict the sign of the investment-uncertainty relationship under imperfect competition

The theoretical models, therefore, imply that greater uncertainty may have a negative or

a positive effect on investment spending depending on the choice of the parameter values of the underlying model and the degree of market power Empirical results mirror these predictions Empirical evidence shows that investment may decrease, increase, or even remain unchanged when there is greater uncertainty

At the firm level, Leahy and Whited (1995) evaluate the empirical significance of various theoretical models on the relationship between investment and uncertainty Working with a panel of US manufacturing firms, the authors obtain a measure of uncertainty from the variance of the firms’ daily stock returns Leahy and Whited argue that this variance reflects higher demand or factor price volatility Their results show that uncertainty is negatively related

to firm investment in reduced-form panel regressions Patillo (2000) examines the effect of

demand uncertainty on a panel of Ghanaian manufacturing firms She uses a survey of Ghanaian manufacturing firms, which includes a question on firms’ perceptions of uncertainty

in the context of a volatile macroeconomic environment In the survey, firm owners report their probability distribution over future demand for the firm’s products Patillo constructs a variable representing the firm’s uncertainty about future demand conditions based on the survey results Empirical results in this paper support the prediction that firms wait to invest until the marginal revenue product of capital reaches a firm-specific hurdle-level The author also finds that greater demand uncertainty decreases the level of firm investment

At the industry level, Caballero and Pindyck (1996) test whether greater uncertainty increases the investment trigger using panel data on US manufacturing industries The standard deviation of the marginal product of capital is used as a proxy for uncertainty They find that the uncertainty proxy is positively correlated with the trigger and conclude that uncertainty makes investment less likely across US manufacturing industries Using the same data set, Ghosal and Loungani (1996) estimate the impact of price uncertainty on investment Pooling the data for all industries, they find that price uncertainty has no impact on investment However, for industries that have low levels of seller concentration and thus are likely to be highly competitive, the estimated impact is negative and statistically significant For industries with high levels of seller concentration, on the other hand, the estimated impact is always small and not significantly different from zero In a later study, Ghosal and Loungani (2000) study the impact of profit uncertainty on investment using data on US manufacturing Further, they analyze whether or not the sign of the relationship is different in industries that are dominated by small firms versus those that are dominated by relatively larger firms They find that the sign of the investment-

10

profit uncertainty relationship is negative and that the quantitative impact is substantially greater

in industries dominated by small firms

Only a few empirical papers analyze investment behavior in Turkey and they all study aggregate investment performance * To my knowledge, there is no empirical study on Turkey that analyzes the effects of industry-wide uncertainty on irreversible investment

1.3 Theoretical framework

I study the impact of uncertainty and sunk costs on industry-wide investment using a model developed by Caballero and Pindyck (1996) Before introducing the theoretical model it

is useful to summarize its underlying intuition

The model does not describe the level of investment per se, but rather the critical threshold required to trigger investment However, inferences can be made about the ways in which investment levels should respond to changes in uncertainty An increase in the investment trigger leads to less willingness to invest’ In the model, the average value of a firm’s output fluctuates stochastically Firms invest when the average value of industry output hits an optimally derived investment trigger’ I interpret this decision to invest by firms as entry

of new competitors in an industry The investment trigger, which is industry specific, is a function of sunk costs, demand and supply uncertainty, and other parameters

The model is an industry analysis and assumes that firms are risk-neutral, identical and have rational expectations about the underlying stochastic processes and the decision rules of

° In the model, all firms are identical and face the same industry threshold for investment

11

the other firms’ The variables in the model are all implicitly indexed by the industry subscript i Total industry output, Q(t), is given by

0

where 6 is the discount rate, and y is an exogenous rate at which firms exit the industry and will follow dIN(t)/dt= -yN Using Eq (1) and (2), the average value of output (B(t)) in an industry is given by the following

12

diog D(t)=(a, -1/26,° jit + ,4z,(t) (5)

dlog A(t)=(c, -1/26,’ hit +o, dz, (t) (6)

where Og, &, and ont ; o are constant drift and volatility parameters, respectivelyỶ dz, and dz, are uncorrelated Weiner processes Then B(t) follows a simple regulated geometric Brownian motion with volatility:

2

¬]

om [act ) G

TỊ and with a rate of drift:

of the price elasticity of demand

* This model assumes that volatility is constant If volatility can change then, in principle, the process by which it changes should be part of the model However, models of financial option valuation in which volatility follows a stochastic process suggest that assign this complication would not change the basic results See Scott (1987) and Wiggins (1987)

13

This model is simple enough to find a closed form solution for the optimal investment rule using dynamic programming’:

* Caballero and Pindyck (1996) also derive (9)

'° All the firms face the same boundary U, which implies that the investment trigger U is the same for any firm in the industry

14

positive supply shock, which causes sales to increase and prices to decrease When demand is unit elastic, any sales effect is offset by the reduction in prices Consequently, supply-side shocks do not affect the volatility [(7)] or the rate of drift [(8)] of average industry output and

Case 2: Demand is price inelastic (n<1):

đU _êU dc do, do, <0 (13)

The effect of supply uncertainty on the investment trigger depends on the magnitude of price elasticity of demand If demand is price elastic, n>1, (inelastic, n<1) the relationship is positive (negative) When demand is price inelastic the sales effect overcomes the reduction in prices, which leads to an increase in revenues and causes investment to increase

A novelty of my theoretical analysis is that I evaluate the effects of both supply and demand uncertainty on the investment trigger By doing so, I show that the sign of the uncertainty-investment relationship not only depends on the magnitude of price elasticity of demand but also on the type of uncertainty Based on the analysis of (9), I summarize the results

as follows: All else equal, (a) Higher sunk costs are associated with a higher investment trigger, (b) An increase in demand uncertainty increases the investment trigger, (c) An increase in

15

supply uncertainty increases (decreases) the investment trigger if demand is price elastic (inelastic)

1.4 Methodology

The theoretical model states that the optimal investment rule in an industry is when the average value of output in an industry reaches the investment trigger For the empirical implementation, I interpret this condition more broadly to imply that given the industry-specific investment tirgger

Investment>0 if B,,>U, 1

Investment=0 otherwise

The level of investment in industry i at time ¢ is assumed to be a linear function of [B, -U, |" The level of investment is inversely related to the investment trigger, ie an increase in investment trigger implies lower level of investment I use data on industry value- added as a proxy for industry characteristics Assume the level of investment in industry i at

time f as

where VA, , is value added in industry i at time ¢ and 0, , is the error term’” Re-write (14) as

1.5 Data and preliminary statistics

The data set comprises sixteen 2-digit Turkish manufacturing industries over the years

1977 to 2000 To the best of our knowledge, the data on private gross fixed investment has not been used in any previous empirical investment study Appendix A describes all variable definitions in detail, including data construction and sources

Figures 1 and 2 show real growth rates for gross domestic product (GDP) and annual inflation rates in Turkey, respectively Although Turkey has averaged 4.0% growth since 1980, economic growth has been quite volatile as seen in Figure 1 Oil-price shocks in the 1970s and related balance of payments problems contributed an economic recession in the late 1970s After the introduction of a stabilization and liberalization program in 1980, the government was able to accelerate economic growth in the following years However, after 1986, the volatility of annual GDP growth rates increased Other events such as the 1900-1991 Persian Gulf crises, the

1994 Turkish financial crises, 1998 Russian crises and 1999 earthquakes contributed to output volatility in the economy

! The Taylor series approximation is available upon request

17

Volatile output growth has been accompanied by chronic inflation in Turkey As seen in Figure 2, Turkey experienced accelerations in whole price index (WPI) inflation in 1977-1978, 1979-1980, 1987-1988, and 1994 Major events associated with these increases in inflation rates are oil price shocks in 1978-1979, balance of payments crisis in 1977-1980, devaluation of 1980 and 1994 financial crisis

Figure 3 plots investment in manufacturing in 1985 US Dollars Investment, too, has been volatile in Turkey Events such as the 1990-1991 Persian Gulf crises, the 1994 Turkish financial crises, and 1999 earthquakes contributed to investment volatility in Turkey An IMF report, which analyzes the private investment behavior in Turkey, concludes, “Overall investment behavior in Turkey followed a modest upward trend but was never able to reach to the economy’s potential (due to macroeconomic instability and uncertainty)”

Table 1 presents the investment-output, investment-value added ratios, and industry investment as a share of aggregate investment in manufacturing The first two columns of the Table show that there is variation in the average values of both ratios across industries Wood products, professional, scientific and controlling equipment, and transport equipment industries are among the industries that have the highest average investment ratios The machinery industry, on the other hand, has the lowest average across industries Although not reported, the time-series behavior of the data on investment ratios does not show any obvious trends However, the time-series behavior of investment varies across industries'* The last column of Table 1 shows industry investment as a share of aggregate investment in Turkish

'* Cited from IMF Report #3452 on Turkey (1997), page 6

'S Only similar industries’ investment series move like each other over the years For example, the correlation between textiles and leather industries is 0.765 (highest correlation in the sample) The lowest correlation in the sample is 0.036 and is between machinery and textiles

18

manufacturing Textiles, machinery, and iron and steel industries have relatively high shares compared to other industries

a Measuring uncertainty

The theoretical framework evaluates the effects of both demand and supply uncertainty

on investment The empirical analysis, too, uses measures for both demand and supply uncertainty Both of these measures are industry-specific I use data on industry output and hourly labor input to construct an index of labor productivity'® The sample variance of this index is the supply uncertainty measure Economic analyses over the years have shown that labor productivity gains are one of the most important sources of economic growth Consequently, volatility in labor productivity directly affects the volatility of output Therefore volatility in an index of labor productivity detects the uncertainty that face firms from the production side G; denotes the supply uncertainty measure

I use inventory data to construct a series of annual change in inventories The sample variance of this series is the demand uncertainty measure Volatility in labor productivity captures uncertainty faced by firms from their production side Among others Metzler (1941), Blinder and Maccini (1991), and Flood and Lowe (1993) have empirically shown that most of the volatility in inventories is representative of volatility coming from the demand side of the economy ơa denotes the demand uncertainty measure

Table 2 presents the uncertainty measures and shows that there is significant variation

in both measures of uncertainty across industries The demand uncertainty measure ranges from

a high of 14.78 in one industry to a low of 0.52 in another The supply uncertainty measure, on

'® T also used data on value added to construct an alternative index of labor productivity This index is quantitatively very similar to the index of labor productivity used in the empirical analysis Therefore, I use output data to construct an index of labor productivity instead of value added data

19

the other hand, ranges from a high of 31.90 to a low of 6.05 across industries Table 3 too presents some descriptive statistics on the uncertainty measures This table provides further evidence that there is considerable variation in these measures across industries Furthermore, the uncertainty measures are not highly correlated

The last column of Table 2 shows the sunk costs measure across industries Some industries require quite substantial amounts of sunk costs relative to others For example, the textiles, iron and steel, and motor vehicle industries are the industries with the highest values of sunk costs The leather, non-metallic and scientific equipment, and pottery industries, on the other hand, have the low sunk cost values

'8 Naturally, this sunk costs proxy does not include all aspects of sunk costs incurred by firms in an industry, such as advertising costs Given that the data is on manufacturing industries, this proxy provides

a sensible idea of the magnitude of the physical assets that firms use while entering in a given industry

20

However, these empirical studies do not consider the effects of the interaction variables on investment Therefore, I estimate the following alternate specification:

Before presenting the OLS estimation results for (17), I present the estimation results for the model in which uncertainty measures by themselves are added to (17) in the second column The third column shows the estimation results for (17) The estimation results show that the level of investment is positively associated with the level of output The estimated coefficient for the sunk cost measure is, though small in magnitude, negative as predicted by the theoretical model With the exception of ơøK , all interaction variables have negative estimated coefficients I then compute the F test where under the null hypothesis the coefficients for

co and om are zero in specification 2 The F-test and the corresponding p-value at 5 percent suggest that null hypothesis is not rejected and that the coefficients for o? and a4 are zero'”, In addition, a comparison of adjusted R’ values suggests that specification 3 provides a better fit for the investment data in Turkey than specification 1

' The F ratio is 0.83 At 5 percent, this F value is not statistically significant given that Fo 5(2, 384)=3.0

21

To check the robustness of my results, I drop two higher order terms [ơ¿K and ơø2K] from specification 1 and present the estimation results in the third column of Table 4°° The results show that the investment ratio is positively associated with the interaction of sunk costs with supply uncertainty but negatively associated with the interaction of sunk costs with demand uncertainty In this specification the estimated coefficient for sunk costs is negative and Statistically significant

In order to capture the marginal effects of sunk costs and the demand and supply

uncertainty measures, I calculate the net effects of change in each variable First, I evaluate the estimated coefficients from specification 3 in Table 4 at sample means Then, I evaluate the estimated coefficients from specification 3 in Table 4 at the iron and steel industry and at the pottery and earthenware industry sample values, so that I can better understand the net effect of change in sunk costs and the uncertainty measures for a range of sample values The iron and steel industry (pottery and earthenware industry) is representative of relatively high (low) sample values of sunk costs and the demand and supply uncertainty measures I report the coefficients for the marginal effects in Table 5”!

The results in Table 5 show that the net effect of sunk costs on the investment ratio is negative The investment ratio decreases by 6.7 percent for a one percent increase in sunk costs when all variables are evaluated at their sample means For both relatively high and low sample values, the net effect of sunk costs on the investment ratio is negative However, the effect is bigger in magnitude when there is greater uncertainty These results show that greater uncertainty deepens the negative association between sunk costs and the investment ratio

Next, Table 5 presents the net effect of changes in the demand uncertainty and supply uncertainty measures, respectively The results show a negative relationship between the investment ratio and the demand uncertainty measure The investment ratio decreases by 8.7 percent for a one percent increase in the demand uncertainty measure, when coefficients are evaluated at sample values of the iron and steel industry These results confirm the theoretical findings and show that the negative impact of demand uncertainty on the investment ratio is quantitatively stronger when there is greater uncertainty and higher sunk costs In addition, all

of the coefficients are significant

The net effect of change in supply uncertainty, on the other hand, can be negative or positive depending on the magnitude of other variables Table 5 shows that when all variables are evaluated at their sample means, a one percent increase in the supply uncertainty measures causes an 8.7 percent decrease in the investment ratio This negative effect becomes positive, however, when the coefficients are evaluated at the sample values of pottery, chine, ceramic, and earthenware industry In fact, the investment ratio increases by 3.1 percent for one percent increase in the supply uncertainty measure These results suggest that when there is relatively low sunk costs and uncertainty, supply uncertainty does not necessarily decrease investment They provide further proof that the interaction variables are important in understanding the uncertainty-investment relationship

Since this is a pooled estimation, there is the possibility of parameter heterogeneity If the parameters differ across industries, then the pooled estimation imposes an invalid inequality restriction Such a restriction will lead to inconsistent estimates Although the ideal way to approach this problem is to estimate each industry separately, I was not able to do this due to

23

data limitations Therefore, this possibility is investigated by separating industries based on the magnitudes of price elasticity of demand

Unfortunately, there is little research on the magnitudes of price elasticities across industries in Turkey Consequently, these results should be viewed merely as suggestive The grouping of industries is based on preliminary research undertaken by the State Institute of Statistics of Turkey The study uses data available from the Consumer Expenditures Survey and surveys of published economic and marketing research Based on these studies, eleven of the industries are grouped as elastic and the remaining five as inelastic” I present the estimation results based on sub-samples in Table 6 The estimated coefficients for all interaction terms are negative in both groups However, the estimated coefficients are significant only in the elastic industry group The estimated impact of sunk costs on the investment ratio is quantitatively small in the inelastic industry group Industry output is positively associated with the investment ratio in both groups

I calculate the net effect of changes in the sunk cost measure and the demand and supply uncertainty measures using the estimated coefficients from specification 3 in Table 6 Table 7 reports the results The adjusted coefficients show that the net effect of a change in sunk costs on the investment ratio is negative for a range of sample values in both groups The net effect of a change in demand uncertainty is also negative in both groups As the theoretical model predicts, the net effect of change in supply uncertainty is negative in elastic industries In inelastic industries, the net effect of supply uncertainty is only positive when all variables are evaluated at their relatively low sample values This result may indicate that not only the price

* The following industries have inelastic demand: Man of food and beverage, Man of industrial chemicals, Man of iron and steel, Man of non-ferrous metal basic industries, and Man of transport equipment The remaining industries are assumed to have elastic demand

24

elasticity of demand but also relative values of demand uncertainty and sunk costs affect the investment supply uncertainty relationship

Overall, these empirical results confirm the theoretical predictions The results indicate that interaction of sunk costs with the uncertainty measures explain the differences in investment behavior more successfully than the uncertainty measures by themselves

1.7 Conclusion

The real options approach to investment literature emphasizes three important features

of investment projects that have been overlooked by previous models: the irreversibility of investment projects, uncertainty about expected future returns, and the timing of investment projects I extend the theoretical findings in this literature by showing that the effect of supply uncertainty on investment depends on the magnitude of the price elasticity of demand If demand is price elastic (inelastic) the relationship is negative (positive) I then empirically explore the relationship between investment, uncertainty and sunk costs using panel data on 2- digit Turkish manufacturing industries I find that the effect of uncertainty on the level of investment varies with how high sunk costs are

The key empirical findings are: (1) Investment is substantially more sensitive to the demand uncertainty measure than to the supply uncertainty measure, (2) The effect of supply uncertainty on investment is negative in full and elastic sample estimates The estimated impact

of supply uncertainty on investment is only positive in inelastic industries when both demand uncertainty and sunk costs are relatively low The estimated effect of demand uncertainty on investment, on the other hand, is always negative, and (3) Investment is negatively associated with the sunk costs measure In addition, the empirical results show that the interaction variables explain much of the variation in investment behavior across industries This is an

25

important finding because the impact of interaction variables on investment behavior has not received much attention in the empirical literature These results shed some light to different empirical findings on the sign of the investment-uncertainty relationship They underline the importance of identifying what type of uncertainty the uncertainty measures used in the empirical analysis represent My empirical results are robust to grouping the sample of industries based on the price elasticity of demand

While my results are encouraging, further research is desirable in certain areas First, the measures of demand and supply uncertainty used in this paper are measured as sample variances and are time-invariant The assumption that uncertainty measures are time-invariant is extremely difficult to credit I will explore this issue in future work Second, it would be desirable to obtain more accurate data on industry demand elasticities Finally, it is possible that the demand and supply uncertainty measures are endogenous Endogeneity needs to be taken into consideration in interpreting the empirical evidence I was not able to find good instruments for the uncertainty measures due to data limitations at the industry level Still I re-estimated (17) using the sample variance of wage rate (as an instrument for supply uncertainty) and the sample variance of the consumer price index (as an instrument for demand uncertainty) The results are not reported here but they are not very different from the previously reported results However, the precision is poor among the coefficients

26

Figurel: Real GDP Growth Rates (percent) in Turkey, (1970-2000)

Source: State Institute of Statistics, Turkey

Figure 2: Annual Inflation, WPI (percent) in Turkey, (1970-2000)

(1963=100), Source: State Institute of Statistics, Turkey

Figure 3: Investment in Aggregate Manufacturing in Turkey (1970-2000)