University of Southern California Dissertations and Theses

Workplace organization and asset pricing

(USC Thesis Other)

Workplace organization and asset pricing

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content 1

University of Southern California
Doctoral Dissertation

Workplace Organization and Asset Pricing
Author:
Yuan Wan
Finance and Business Economics
Marshall School of Business
University of Southern California

Advisors:
Prof. Selale Tuzel (Chair) Prof. Christopher Jones
Finance and Business Economics Finance and Business Economics
Marshall School of Business Marshall School of Business
University of Southern California University of Southern California

Prof. Joel David Prof. Fernando Zapatero
Economics Finance and Business Economics
College of Letters, Arts and Sciences Marshall School of Business
University of Southern California University of Southern California

A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy
in Business Administration

USC Graduate School
Marshall School of Business
Department of Finance and Business Economics
August 2019

2

Workplace Organization and Asset Pricing

Yuan Wan
*

USC Marshall School of Business

Abstract
Flexible workplace organization (WO) increases the efficiency of a firm’s future capital investment.
Hence, investment in WO positively predicts future capital investment and exposure to future aggregate
investment-specific technology (IST) shocks. As a result, firms that invest in WO become riskier and
systematically experience higher equity returns. This paper documents a robust positive correlation
between firms’ WO investments and future equity returns. Firms with higher industry-adjusted WO
investments have average returns that are 3.7% higher than those of firms with lower WO investments.
In addition, firms with high WO investments experience a high conditional IST beta in the future. I tie all
my empirical findings together in a production-based equilibrium model with closed-form solutions.

*Yuan Wan: yuanwan@usc.edu. I am grateful to my advisers Selale Tuzel, Miao (Ben) Zhang for their countless advices and guidance. I thank
Christopher Jones, Fernando Zapatero, Joel M. David, Miao (Ben) Zhang and Selale Tuzel, my committee members for their invaluable
support. I thank Cary Frydman, Juhani Linnainmaa, Oguzhan Ozbas, and other USC Marshal School of Business seminar participants for their
helpful comments and discussions. I am fully responsible for all errors in this paper.

3

Contents
Abstract 2
1 Introduction 7
2 Data and Measurement 12
2.1 Financial and accounting data 13
2.2 Monthly Stock Return Data 14
2.3 Macro Factor Price Data 14
2.4 Patent citation data 14
3 Empirical Findings and Portfolio Sorting 15
3.1 Panel regression and Fama-MacBeth regression 16
3.2 Organization capital and workplace organization investment 17
3.3 PDC source of variation in PDC 17
3.4 Portfolio sorting on PDC 18
3.5 Risk factor model regressions 20
3.6 Relation to macroeconomic shocks and investment 22
3.7 IST shocks and firm-level investment decisions 24
3.8 Firm future investment and PDC 26
4 Model 28
4.1 Firm production and capital accumulation 28
4.2 SDF 29
4.3 Firm optimization problem 30
5 Asset Pricing Mechanism 34
5.1 Proposition I-V 34
5.2 Mechanism summary 35
6 Conclusion 36
7 Appendix 54
4

7.1 A simple model explaining the relationship between IT and flexible WO 54
7.2 Proofs for Proposition I-V 56
7.3 Exploitation vs. exploration (external validation) 64

5

List of Tables
1 Hand-matched link table summary 40
2 Panel regression and Fama-MacBeth results 41
3 Site level panel regression of PDC on desktop and laptop growth 43
4 Firm characteristics summary 44
5 Portfolio sorting results 45
6 Factor model regressions (equal weighted) 46
7 Factor model regressions (value weighted) 47
8 Macroeconomic shock regressions 48
9 Conditional equity beta regressions 49
10 Conditional IMC beta regressions 50
11 Investment rate and PDC regression 51
12 Depth and Scope regression 52

6

List of Figures
1 Investment 53

7

1. Introduction
“… AT 3M, You can rework your schedule based on YOU. Here, we offer simple, individual arrangements
between each 3Mer and their Supervisor for a flexible work practice that maximizes professional and
personal productivity…”
---- From the 3M company website: https://www.3m.com/3M/en_US/careers-
us/culture/flexability/
In recent years, a new form of workplace organization (WO) that favors flexibility and communication
has emerged among US firms. This is a major shift from the early postwar era, during which rigidly
structured American mass-production-based organizations dominated global product markets.
Following the lead of American firms, this low-cost scheme has also been adopted by other countries.
More recently, in response to rising competition, firms that expect to expand and seek future
opportunities outside of their existing operations have started to adopt a new form of work organization
that emphasizes flexibility, independent judgment and communication (Smith 1997). Economists have
found evidence that flexible work organizations are often associated with firms in internationally
competitive product markets and firms following a “high road” strategy that emphasizes variety, service,
and quality rather than low cost (Osterman 1994).
The adoption of a new organizational structure requires firms to make specific investments in WO
capital, which is distinct from physical capital. Recent empirical research shows that investment in
flexible WO increases demand for IT investments (Bresnahan, Brynjolfsson and Hitt 2002) and is
associated with high measured firm-level productivity (Bresnahan, Brynjolfsson and Hitt 2002, Black and
Lynch 2001). Therefore, WO investment predicts future investment, especially IT investments.
8

In this paper, I study the asset pricing implications of WO. I construct an empirical proxy for a flexible
WO from a proprietary IT investment database, CiTBD, and find that firms that invest heavily in flexible
WO earn higher future equity returns. I propose a dynamic equilibrium model that generates the same
relationship between WO investment and expected firm returns. In my model, flexible WO increases the
efficiency of new capital investments. Therefore, firms that face promising capital investment
opportunities optimally build a flexible WO. Because capital investment is risky (i.e., subjects the firm to
aggregate investment-specific IST shocks), investors will demand higher expected returns to hold these
firms.
To study the empirical relationship between flexible WO investment and equity returns, I construct an
establishment-level measure of WO investment using the CiTBD database. I measure flexible WO
investment as the change in the ratio of laptop computers to the total number of laptop and desktop
computers (portable to desktop change, or PDC in short). This proxy is motivated by findings in the
technology and education literature (Cengiz Gulek and Demirtas 2005, Nicol and MacLeod 2005). While
desktops are commonly used in relatively closed office spaces, laptops facilitate open workspaces and
promote communication and flexibility. Previous research shows that computers are an advanced form
of information and communication technology that is closely related to WO investments (Bresnahan,
Brynjolfsson and Hitt 2002). Furthermore, the choice between a laptop and a desktop is based on the
need for flexibility, which is at the heart of flexible WO investment.
My PDC measure is able to capture many features of WO investment. First, I find that PDC strongly
predicts future intangible asset investment, including R&D and computer software, while it can only
predict short-term PP&E investment. This finding is also documented in Bresnahan, Brynjolfsson and Hitt
(2002), where the authors find WO is a strong predictor of future IT investment but not f uture physical
capital investment. Second, I find that PDC is positively correlated with explorative innovation but not
9

with exploitative innovation. Exploration and exploitation are two major innovation patterns and are
strongly associated with certain organizational structures. A large body of management literature
documents an association between exploration, in which innovations are independent from a firm’s
previous innovations, and flexible and loosely coupled organization systems. By contrast, exploitation, in
which innovations are largely based on previous intellectual property, is associated with mechanistic
structures, path dependence, control and bureaucracy (He and Wong 2004, Katila and Ahuja 2002).
Next, I investigate the asset pricing implications of a firm’s WO investment using PDC as a proxy. I find
evidence that supports my hypothesis that investors demand higher risk premi ums to invest in high PDC
firms. Investors that invest in firms with positive PDC earn higher future equity returns. The average
annualized equal-weighted return differential between top- and bottom-tier PDC firms (here called the
PDC portfolio) from 1997 to 2014 is 3.4% and is statistically significant. The spread becomes 3.7% and
retains its significance after adjusting for industry effects. This result is not dominated by small firms, as
the value-weighted annualized return differential is 3.2% and significant. Panel regressions show that
PDC in firms exhibit both time-series and cross-sectional variations and that the correlation between
PDC and equity returns is strong and robust to many regression specifications. The positive correlation
between expected return and PDC is not driven by a few specific industries, as the portfolios based on
industry-level PDC do not yield similar return spreads. I consider several classic empirical asset pricing
models, and all of them fail to explain the positive PDC return spreads. The average risk-adjusted excess
annualized return (alpha) is 2.9% and significant after controlling for the Fama and French 3 factors. The
PDC portfolio negatively correlates with the HML factor, which indicates that high PDC stocks are
positively correlated with growth stocks. However, this comovement does not explain the positive
spread of the PDC portfolio, as growth minus value spread is negative. The results are similar after
controlling for other firm characteristics, such as profitability, investment rate, and organizational
capital, as proposed in Eisfeldt and Papanikolaou (2013).
10

In the dynamic model, I propose that positive firm-level idiosyncratic technology shocks create
investment opportunities by increasing the efficiency of asset accumulation. Flexible WO and capital
investment are modeled as complements; hence, the demand for capital investment, as well as for
future capital investment, increases due the investment in WO. However, firms that invest are exposed
to aggregate investment-specific technology (IST) shocks, which carry a positive price of risk. Therefore,
firms with higher WO investments have higher exposure (higher beta) to these IST shocks, and they have
higher expected returns. I provide empirical evidence that supports this core mechanism. Empirically, I
show that high PDC firms experience high conditional IST betas in the future.
More broadly, to understand how PDC spread relates to systematic risks, I study the interaction
between WO investment and two fundamental drivers of economic growth: total factor productivity
(TFP) shocks, which affect the assets-in-place type of capital, and IST shocks, which affect newly installed
capital. Consistent with my model’s implications, I find that the PDC return spread has a significant
positive exposure to measures of aggregate IST shocks, while it is not associated with TFP shocks. In
contrast, HML (value-growth) portfolio return is negatively correlated with IST shocks, which suggests
that the negative correlation between HML and PDC portfolios is driven by their simultaneous opposing
exposures to fundamental shocks. In addition, allowing firms’ exposure to these fundamental shocks to
vary over time, I show that firms with high PDC subsequently experience higher IST shock exposure, as
their future conditional IST betas positively correlate with current PDC. I do not find a similar pattern for
TFP shock exposure. Finally, I provide direct evidence from firm investment data that firms with high
current PDC are making more future investments in both physical and intangible assets. I find the
complementarity between investment in WO and intangible assets to be persistent, whereas the effects
for the physical asset are short lived.
11

This paper belongs to two growing strands of literature. First, the relationship between various types of
investments and future asset prices has been a recent focus of the production-based asset pricing
literature. Tuzel (2010) finds that firms’ real estate holdings are positively related to their future equity
returns because slower depreciation of structures makes real estate harder to adjust, and increases
firms’ exposure to aggregate shocks. Belo and Lin (2012) and Jones and Tuzel (2013) study the
relationship between inventory investment and equity return. They find that the relationship is negative
and is strongly driven by the complementarity between inventories and other capital inputs, and time
variation in the cost of capital. Belo et al. (2014) focus on brand capital investment, which specifically
affects people’s willingness to buy and hedges against firms’ aggregate productivity shock risks. Eisfeldt
and Papanikolaou (2013) show that organizational capital is shared by both the company and key
employees and can affect equity returns by altering firms’ frontier technology shock exposure. My paper
explores a novel type of investment, flexible WO. In the empirical section, I document a robust and
positive relationship between WO investment and future stock returns. In addition, I find evidence that
flexible WO investment leads future capital investment, which exposes firms to higher aggregate
investment shocks.
Second, starting with Greenwood, Hercowitz, and Krusell (1997), investment-specific technological
progress has become an important topic due to the distinct economic activities in investment and
production sectors. This approach has been adopted in recent finance literature. Papanikolaou (2011),
using a two-sector general equilibrium model, theorizes that IST shocks derived from investment-
specific technological progress carry unique aggregate risk premiums. Subsequently, Kogan and
Papanikolaou (2013 and 2014) show that equity returns exhibit systematic cross-sectional differences
due to the exposures to aggregate IST risk. Li (2018) explores an investment based asset pricing model
with both aggregate productivity shocks and investment-specific technology shocks. With both
Macroeconomics shocks, his model is able to generate momentum profit and value premiums
12

simultaneously. In this paper, I present direct evidence of the interaction between flexible WO
investment and IST shocks. In the model section, firms with more investment opportunities optimally
choose the higher level of flexible WO investment, which creates persistent exposure to aggregate IST
shocks and higher future equity returns.
The paper is organized as follows: the Data and Measurement section presents the data and variable
construction method. The Empirical Findings and Portfolio Sorting section presents an analysis of
investment data, individual stock returns, portfolio returns and their interaction with fundament shocks.
The Model section presents a production-based asset pricing model with closed-form solutions. The
Asset pricing mechanism section derives model-implied propositions with corresponding proofs and
presents the underlying mechanism to generate the empirical findings. Finally, the Conclusion section
summarizes the main findings.

2. Data and Measurement
In this section, I describe the construction of the key variables used in the empirical tests as well as the
data sources used in the paper.
Flexible WO investment One challenge in this paper is obtaining a continuous proxy for flexible WO
investment for a wide range of firms. Previous literature measures WO by conducting surveys that only
cover a handful of firms with limited time-series continuation. I construct my key variable to measure
the firm’s investment in flexible WO by calculating a firm’s change in its composition of laptop and
desktop computers (portable to desktop change, or PDC in short). The laptop and computer investment
data are from the Computer Intelligence Technology Database (CiTDB). CiTDB is a proprietary database
owned by Harte-Hanks that provides detailed survey data regarding firms’ information and technology
13

investments at the establishment level. The survey is conducted annually via telephone and includes
approximately 500,000 establishments before 2010 and approximately 3.2 million establishments after
2010. CiTDB has been widely used by researchers in the management and finance literatureto measure
firms’ different technological investments (e.g., Bloom et al. 2014 and Tuzel and Zhang 2018). CiTDB
started collecting the number of laptop from 1996, and I have access to CiTDB data ending in year 2013.
As a result, the main empirical analysis is conducted using data from 1996 to 2013.
I first calculate the proportion of laptops as the number of laptops divided by the sum of laptops and
desktops at the establishment level and then calculate the ratio change (PDC) by taking the ratio
difference between two periods. I then obtain firm level PDC by averaging establishment-level PDCs,
weighted by last period’s share of total computers in that specific establishment. Specifically:
𝑃𝐷𝐶 𝑗 ,𝑡 = ∑ 𝑤 𝑖 ,𝑗 ,𝑡 −1
𝑃𝐷𝐶 𝑖 ,𝑗 ,𝑡 𝑖
where 𝑤 𝑖 ,𝑗 ,𝑡 −1
=
𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑐𝑜𝑚𝑝𝑢𝑡𝑒𝑟𝑠 𝑓𝑟𝑜𝑚 𝑒𝑠𝑡𝑎𝑏𝑙𝑖𝑠𝑚𝑒𝑛𝑡 𝑖 ( 𝑡𝑜𝑡𝑎𝑙 𝑐𝑜𝑚𝑝𝑢𝑡𝑒𝑟𝑠 𝑓𝑟𝑜𝑚 𝑡 ℎ𝑎𝑡 𝑓𝑖𝑟𝑚 𝑗 )
in year t-1.
𝑃𝐷𝐶 𝑖 ,𝑗 ,𝑡 −1
=
𝑛𝑢𝑚𝑏 𝑒 𝑟 𝑜𝑓 𝑙𝑎𝑝𝑡𝑜𝑝𝑠 𝑡 ( 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑙𝑎𝑝𝑡𝑜𝑝𝑠 + 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑑𝑒𝑠𝑘𝑡𝑜𝑝𝑠 )
𝑡 −
𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑙𝑎𝑝𝑡𝑜𝑝𝑠 𝑡 −1
( 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑙𝑎𝑝𝑡𝑜𝑝𝑠 + 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑑𝑒𝑠𝑘𝑡𝑜𝑝𝑠 )
𝑡 −1

2.1 Financial and accounting data Data on firm characteristics are from Compustat. I apply standard
filters to the Compustat data and exclude firms without positive total asset (AT) data. I calculate various
ratios, i.e., book-to-market ratio following standard literature Fama-French (1993); profitability using
gross profit divided by total current asset
𝐺𝑃
𝐴𝑇
(Novy-Marx 2013); investment rate using the change in
total assets divided by last period’s total assets
∆𝐴𝑇
𝐴𝑇
𝑡 −1
(Novy-Marx 2013); and firm size using the
logarithm of total assets log ( 𝐴𝑇 ) . Because the focus of the paper is WO investment and its relationship
14

with other types of investments, I consider two additional investments: 1. intangible assets (Intan),
which includes a firm’s investment in R&D capital, intellectual property, and software and 2. PP&E
(PPEGT), which includes a firm’s investment in building and equipment.
2.2 Monthly Stock Return Data are from the Center for Research in Security Prices (CRSP). Similar to
Fama and French (1993), my sample includes firms with ordinary common equity as classified by CRSP,
excluding ADRs, REITs, and units of beneficial interest. I match CRSP stock return data from July of year t
to June of year t + 1 with accounting information (Compustat) for the fiscal year ending in year t – 1, as
in Fama and French (1992, 1993), to allow for a minimum gap of six months between fiscal year end and
the return tests.
2.3 Macro Factor Price Data are from the U. S. Bureau of Economic Analysis (BEA). I mainly investigate
IST shocks, and I construct two versions of capital-embodied technical change proxies for IST. The first
measure is based on the quality-adjusted price of new capital goods (Greenwood, Hercowitz and Krusell
1997, 2000). The BEA provides macro data on consumption as well as investment good prices at a
quarterly frequency. The second measure is based on a long–short portfolio of stock returns between
investment and consumption good producers (IMC portfolio). To construct the IMC portfolio, I closely
follow the methodology introduced in Gomes, Kogan and Yogo (2009). I first match firms to
corresponding industries based on their NAICS codes and then take industry average monthly return
differences between the investment industry and the consumption industry. The IMC portfolio is at a
monthly frequency, which allows us to conduct empirical tests with more detailed variations.
2.4 Patent citation data are from the KPSS patent database and are downloaded directly from Professor
Noah Stoffman’s Indiana University Kelley School of Business website. The data series is available from
1926 to 2010.
15

Hand Matching CiTDB and Compustat is required because CiTDB provides establishment-level data that
are identified by the variable siteid and does not provide a link table between siteid and gvkey in
Compustat. Therefore, I have to construct a link table by matching company names. First, I filter out
perfectly matched names in both databases and focus on the names with close matched counterparts. I
then calculate the Levenshtein distance for each firm name in Compustat in each year and select the top
5% firm names in CiTDB that have the shortest Levenshtein distances as candidates for the hand match.
Finally, I hand match each firm, produce a link table between gvkey and siteid, and use other variables
such as the headquarters’ location zip code to cross validate the matching results. Table 1 summarizes
the matching result:
[Insert Table 1 here]
As reported in Table 1, CiTDB expanded the number of surveyed sites over time. The surveyed sites
jump from 25,323 to 130,699 in 1996 and continued to increase over the years. In 2010, the database
surveys 3,629,435 sites and then scales back to 528,505. For the next two years, the linked ratio is
consistently high from 1987 to 1995 and drops to 0.04 in 1996 although the number of firms matched
increases from 508 to 1528, which indicates that the CiTDB includes many private firms in its survey. The
average number of sites in each matched firm is 6.2 (excluding year 2010), which indicates that the
CiTDB mainly focuses on large firms. In 2010, Harte-Hanks acquired Information Arts and announced a
major restructure of CiTDB. At that point, CiTDB stopped surveying most existing sites and included a
large number of new sites for the same company for 2010; as a result, I do not have data continuity and
must discard all observations in year 2010. Overall, I have approximately 1200 firm observations each
year from 1997 to 2013.
3. Empirical Findings and Portfolio Sorting
16

3.1 Panel regression and Fama-MacBeth regression I first investigate the empirical relationship
between a firm’s PDC and future stock returns. According to the core mechanism in my model, if firms
are investing in flexible WO, they would be investing more capitals in the future therefore be more
exposed to future aggregate IST shocks. Because IST shocks carry a positive risk premium in the sample
period, I expect a positive relationship between current WO investments and future equity returns. I
examine this empirical relationship by running a Fama-MacBeth regression as well as panel regressions
with various controls and fixed effects:
𝑟 𝑖 ,𝑡 +1
𝑒 = 𝑏 0
+ 𝑏 1
𝑃𝐷𝐶 𝑖 ,𝑡 + 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠 + 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝜖 𝑖 ,𝑡
where 𝑟 𝑖 ,𝑡 +1
𝑒 is the excess annual return from July of year t+1 to June of year t+2. I include all major
control variables commonly found in the literature, such as B/M, profitability, investment rate and size,
with both year and firm fixed effects. The regression results are summarized in Table 2:
[Insert Table 2 here]
I find that all PDC coefficients are significantly positive at the 1% level (standard errors are clustered at
firm level) with or without firm fixed effect or control variables. The Fama-MacBeth regression shows
similar results, and the PDC coefficient is significantly positive at the 5% level. Many control variable
coefficients are statistically significant, with the correct sign expected in the literature. Fama and French
(1993) finds value firms on average outperform growth firms. In my regressions, the BM coefficient is
significantly positive at the 1% level, which indicates that firms with higher book-to-market ratios have
higher expected future returns. Cochrane (1991), Lamont (2000), Xing (2008) and Hou Xue and Zhang
(2015) finds that the expected return is negatively correlated with the investment rate and higher
investment rate firms earn lower subsequent average stock returns than low physical investment rate
firms do. My investment rate coefficients are significantly negative, which validates Xing’s finding. The
size coefficient is negative for all regressions, which validates the size premium found in Fama and
17

French 1993. The coefficient for size is only significant with a firm fixed effect, which suggests that the
sample we used does not have sufficient size variation. This is not surprising given that CiTDB mainly
focuses on surveying large firms.
3.2 Organization capital and workplace organization investment Eisfeldt and Papanikolaou 2013
recognize that organization capital is embedded in key members within the firm and carries unique risk
that deserves a risk premium. In this section, I will show that WO investment has a different risk profile
than organizational capital by controlling for firm organization capital in my PDC regression. Following
Eisfeldt and Rampini 2006, I construct organizational capital for each firm using the perpetual inventory
method and recursively construct organizational capital stock by accumulating the deflated value of
selling, general and administrative expenses (SG&A):
𝑂 𝑖𝑡
= ( 1 − 𝛿 𝑜 ) 𝑂 𝑖𝑡 −1
+
𝑆𝐺𝐴 𝑖𝑡
𝐶𝑃𝐼 𝑡
with the initial 𝑂 𝑖 0
=
𝑆𝐺𝐴 𝑖 1
𝑔 +𝛿 𝑜 , CPI denotes consumer price index, growth rate 𝑔 = 10%, and depreciation
rate 𝛿 𝑜 = 15%: the PDC regression result is presented in Table 2:
From the first regression in Panel B, I validate the empirical finding in Eisfeldt and Papanikolaou (2013)
that organizational capital is positively related to future stock returns. The coefficient of OC/K is 0.052,
with a t-stat of 5.56. After controlling for organizational capital, I find that PDC coefficients are still
positive and statistically significant. Moreover, I find that after controlling for other firm characteristics
such as book-to-market ratio, size, investment rate and profitability, the organizational capital rate
coefficient becomes insignificant, which suggests that in my sample the risk premium in OC is absorbed
by other firm characteristics.
3.3 PDC source of variation in PDC With the statistically significant PDC coefficient shown above, one
natural question is where the variation in PDC originates. PDC is calculated as the change in the ratio
18

between laptops and the sum of laptops and desktops. An increased ratio means firms are investing
more in laptops relative to desktops, or it alternatively means firms are decreasing investments in
desktops. Here, I explore the major source of PDC variation following the method in Tuzel and Jones
(2013). I consider three related panel regressions:
1. 𝑃𝐷𝐶 𝑖 ,𝑡 𝑠𝑖𝑡𝑒 = 𝑏 0
+ 𝑏 1
∆𝐷𝑒𝑠𝑘𝑡𝑜𝑝 𝑖 ,𝑡 + 𝜖 𝑖 ,𝑡
2. 𝑃𝐷𝐶 𝑖 ,𝑡 𝑠𝑖𝑡𝑒 = 𝑏 0
+ 𝑏 1
∆𝐿𝑎𝑝𝑡𝑜𝑝 𝑖 ,𝑡 + 𝜖 𝑖 ,𝑡
3. 𝑃𝐷𝐶 𝑖 ,𝑡 𝑠𝑖𝑡𝑒 = 𝑏 0
+ 𝑏 1
∆𝐷𝑒𝑠𝑘𝑡𝑜𝑝 𝑖 ,𝑡 + 𝑏 1
∆𝐷𝑒𝑠𝑘𝑡𝑜𝑝 𝑖 ,𝑡 + 𝜖 𝑖 ,𝑡
The panel regression results are presented in Table 3:
[Insert Table 3 here]
According to Table 2 column one, despite the low R squared, the coefficient on desktop growth is
significantly positive. Therefore, establishments with higher PDC are not scaling back on the number of
desktops they purchase and, quite the opposite, desktop growth is positively associated with PDC.
Column 2 shows that the coefficient on laptops is significantly positive with a high R squared (0.54). This
shows that laptop growth drives the variation in PDC. Column 3 combines both independent variables.
The sign of desktop growth switches to significantly negative due to controlling the level of laptop
growth.
3.4 Portfolio sorting on PDC To explore the economic significance of the investment in flexible WO and
the link between PDC and expected return, I perform portfolio sorting on the PDC variable and create a
zero-cost portfolio based on the strategy of long–high PDC firms and short–low PDC firms. For each year,
I sort firms into 5 bins based on their PDC and match them with monthly returns from July of year t+2 to
June of year t+3.
1
This procedure allows for a gap between the realization of the sorting variables and
19

1. PDC in year t is measured from the change in the laptop to desktop ratio from year t to t+1. To prevent a potential look-ahead bias, I start
looking at returns in year t+2.
returns, as in Fama and French (1992). The 5
th
bin contains firms with the highest PDC in any given year;
in other words, these firms invest heavily in flexible WO and show the greatest switch in their computer
composition toward laptops. The summary characteristics of those 5 portfolios are presented in Table 4:
[Insert Table 4 here]
Table 4 shows that PDC spreads out evenly, with the highest ratio change of 0.066 in the top portfolio
and the lowest of -0.086 in the bottom. Desktop growth does not change monotonically. The top
portfolio has the largest growth for both desktops and laptops, which indicates that those firms are
expanding, with a high focus on flexibility. By contrast, the bottom portfolio has the lowest growth for
both desktops and laptops, which indicates that those firms are disinvesting and becoming less adaptive.
The monotonic pattern in laptop growth combined with the flat pattern in desktop growth validates that
the main source of variation in PDC is, indeed, laptop growth.
The book-to-market ratio decreases with PDC, with a mean value of 0.69 in the bottom portfolio and a
mean of 0.64 in the top, which indicates that the top PDC portfolio behaves like growth stocks.
Profitability decreases with PDC, with 0.96 in the bottom portfolio and 0.92 in the top. Both size and
investment rate do not show a trend from the 1
st
to 4
th
portfolio, although the top portfolio has the
smallest firm size and highest investment. Therefore, there is no significant association between PDC
and firm size or physical investment rate. However, later in the discussion, I will further explore the
relationship between PDC and different types of future investments. The average firm age in all five
portfolios is 20 years, which suggests that the sample is mainly composed of large, older firms and that
there is no relationship between WO investment and firm age.
20

3.5 Risk factor model regressions To explore the implications of PDC for the cross-section of stock
returns, I compute both equal-weighted and value-weighted monthly returns of five PDC portfolios from
July 1998 to June 2015. Moreover, PDC may systematically differ across industries; therefore, I also
explore whether the industry-level PDC effect exists. Table 5 summarizes the portfolio sorting results.
[Insert Table 5 here]
Panel A reports both equal- and value-weighted annualized average monthly excess returns for each
PDC portfolio. The excess returns are monotonically increasing. The bottom portfolio has an average
annualized excess return of 0.072 with a 1.55 t-stat, and the top portfolio has a 0.106 annualized
average excess return with a 2.24 t-stat. The PDC portfolio spread is 0.034 and is statistically significant
(1.98 t-stat). Panel B reports equal-weighted average excess returns. Industry-adjusted PMD is 0.037,
which is slightly larger than the 0.034 in Panel A, and the t-stat is 2.16, which is also slightly larger.
Finally, Panel C reports the industry-level PDC portfolio sorting results. In this panel, I sort industries
based on their average PDC level, and take the excess average return of the industries. Although the top
portfolio still has the highest return, the monotonic increasing pattern is not presented. Most
importantly, the PMD spread is only 0.011 annually and is not statistically significant. Combining findings
from all three panels, I conclude that the PDC effect is not an industry phenomenon, and exists between
firms within and across industries.
To explore the risk profiles of PDC portfolios, we regress the PDC-sorted portfolio returns on various risk
factors. Table 6 and Table 7 summarizes results from CAPM, Fama and French (1993) three-factor and
Fama and French (2015) five-factor model regressions of equal weighted and value weighted portfolio
returns.
[Insert Table 6 here]
21

[Insert Table 7 here]

Table 6 reports regression results. All three models fail to explain excess return pattern and risk -
adjusted excess returns a.k.a. alphas from all three models are statistically signifi cantly positive and
generally are monotonically increasing for all 5 portfolios (except middle one). Panel A shows the results
from the CAPM model. CAPM provides little explanatory power for PDC portfolio return spread. The
annualized CAPM alpha is 3%, R squared is only 0.8% and the market coefficient is not statistically
significant.
Panel B reports the classic Fama-French three-factor regression results. Similar to the CAPM model, the
Fama-French three-factor model fails to capture the return spread of the PDC portfolios. R squared
(approximately 4%) is higher than the CAPM R squared, but the annualized alpha is unchanged at 3%.
One interesting observation is that the only statistically significant coefficient is on HML, and it is
negative (-0.08) with a -2.13 t-stat. This finding suggests that the PMD portfolio behaves like a growth
company. The loading pattern almost monotonically decreases, with the bottom portfolio having the
highest HML loading and the top portfolio having the lowest loading. The empirical results of Fama-
French (1993) find that growth firms typically earn lower returns compared with value firms, and my
PMD portfolios document the opposite effect, with portfolios behaving like growth firms earning higher
excess stock returns.
Hou Xue and Zhang (2015) find that empirically, there is a negative relation between investment rate
and expected return and a positive relation between profitability and expected return. Therefore, Fama-
French (2015) develop an extended factor model with two additional factors: RMW (profitability) and
CMA (investment rate). Panel C reports results from the Fama-French 5 factor model. The annualized
five-factor alpha is still 3% and statistically significant. All five portfolios, except the middle portfolio,
22

have negative investment rate loadings with insignificant t-statistics. The positive relation between
profitability and expected return holds, as all five portfolios have positive profitability loadings with
significant t-statistics. However, the profitability loading pattern remains flat; therefore, profitability also
fails to explain the PDC portfolio return pattern. Moreover, the negative loading on the HML factor
drops to -0.07 and becomes insignificant, with a t- stat of -1.26.
3.6 Relation to macroeconomic shocks and investment I now study how PDC relates to fundamental
macroeconomic shocks. I focus on two major aggregate shocks that drive cross-sectional expected
return differences: TFP shocks and IST shocks. TFP shocks are well studied in most production-based
asset pricing literature, such as Gomes, Kogan and Zhang (2003) and Zhang (2005). TFP shocks affect the
productivity of all assets in place. A recent strand of literature emphasizes the role of IST shocks, which
are embodied in new capital goods Papanikolaou (2011).
I use quarterly data on TFP shocks from Fernald (2015), which are directly available from the Federal
Reserve Bank of San Francisco website. I consider three proxies for IST shocks. The first two proxies are
based on innovation in the quality-adjusted price of new capital goods, as in Greenwood, Hercowitz and
Krusell (1997, 2000). I use both the quality-adjusted price series for new equipment constructed by
Gordon (1990) and the quarterly quality-adjusted price series for intellectual property to calculate
proxies. Followoing Kogan and Papanikolaou (2010), I normalize the price series of new equipment and
intellectual property by the NIPA consumption deflator and then detrend the log of the quality-adjusted
price series by regressing the series on a piecewise linear time trend:
𝑝 𝑡 𝐼 = 𝑎 0
+ 𝑏 0
𝐼 1982
+ ( 𝑎 1
+ 𝑏 1
𝐼 1982
)∗ 𝑡 − 𝑍 𝑡 𝐼
𝑝 𝑡 𝐼 is the log of the price of new equipment deflated by the price of consumption at time t; 𝐼 1982
is a
dummy variable that equals zero prior to 1982 and one after 1982. I adopt this piecewise regression
method because, as Fisher (2006) note, real price series experience an abrupt increase in their average
23

rate of decline in 1982, which could be due to the effect of more accurate quality adjustments in more
recent data. Finally, we measure IST shocks as the difference in residuals:
𝐼𝑆𝑇 𝑠 ℎ𝑜𝑐𝑘 = ∆𝑍 𝑡 𝐼 = 𝑍 𝑡 𝐼 − 𝑍 𝑡 −1
𝐼
While IST shocks based on macro price shocks are at a quarterly frequency, my third IST shock proxy is
based on the monthly stock return spread between investment and consumption good producers (IMC
portfolio). Kogan and Papanikolaou (2010) show that the IMC portfolio serves as a natural candidate as a
proxy because it is spanned by IST shocks. Because the NIPA macroeconomic price series are at quarterly
frequencies, having a monthly return series proxy can potentially provide me more time-series
observations to boost statistical power. To calculate the IMC spread, I first map firms into investment
and consumption categories based on their NAICS codes. The category classification is available on
Professor Yogo’s website, and Gomes, Kogan and Yogo (2009) describe the details of this classification
procedure.
I regress PDC portfolio returns on all three proxy series. I use monthly PDC returns for the IMC
regression and aggregate monthly PDC returns to quarterly returns for TFP and the quality-adjusted
price shock regressions. The results are reported in Table 8.
[Insert Table 8 Here]
From the first row result, I conclude that the PDC return does not covary with TFP shock at all. Both
coefficients and t-statistics are close to zero under all specifications. From rows 1 to 4, PDC returns
covary with macroeconomic shocks related to new equipment prices. Although the t-statistics are
relatively low with one exception in row 2, I conclude that firms that invest in flexible WO are more
exposed to future new equipment price shocks than their low PDC counterparts are. One striking result
comes from the last row, where I regress the monthly PDC return onto the IMC portfolio. The R
imc

24

coefficient is 0.12 and is highly statistically significant, with a t-stat of 3.1. This validates my finding from
columns 1 to 4 that firms with high PDC are more exposed to future investment shocks and risks.
3.7 IST shocks and firm-level investment decisions What I have shown previously is under the
assumption that a firm’s return exposure to IST shocks does not vary over time and exists in a portfolio
setting. However, the nature of the investment goods needed by firms is different and varies over time,
especially when firms are making individual adjustments. Therefore, given what I have found in Table 8,
I study the association between future investments and firms’ current PDC investment decisions to the
individual firm level while allowing time variations in IST exposure. To investigate the implications of
PDC for future firm risk exposure and how it changes over time, I first estimate firms’ conditional IST
betas as well as the conditional equity beta, following Lewellen and Nagel (2006). Both conditional IST
and equity betas are estimated using short-window regressions and monthly returns and do not require
that the conditioning information be specified. I correct for nonsynchronous trading following the
methodology described in Lewellen and Nagel (2006). The detailed regression I run is as follows:
𝑅 𝑖 ,𝑡 = 𝛼 𝑖 + 𝛽 𝑖 ,0
𝑖𝑚𝑐 𝐼𝑀𝐶 𝑡 + 𝛽 𝑖 ,1
𝑖𝑚𝑐 𝐼𝑀𝐶 𝑡 −1
+ 𝘀 𝑖 ,𝑡
𝑅 𝑖 ,𝑡 = 𝛼 𝑖 + 𝛽 𝑖 ,0
𝑚 𝑅 𝑀 ,𝑡 + 𝛽 𝑖 ,1
𝑚 𝑅 𝑀 ,𝑡 −1
+ 𝘀 𝑖 ,𝑡
𝑅 𝑖 ,𝑡 is the firm’s monthly excess return, 𝐼𝑀𝐶 𝑡 is the investment minus the consumption portfolio
monthly return, and 𝐼𝑀𝐶 𝑡 −1
is the one-month lag of the IMC return. Similarly, 𝑅 𝑀 ,𝑡 is the monthly
market excess return, and 𝑅 𝑀 ,𝑡 −1
is the one-month-lag market excess return. Finally, the conditional IM
beta is the sum of the coefficients for the contemporaneous and lagged IMC returns. The conditional
equity beta is the sum of the coefficients for the contemporaneous and lagged market returns:
𝛽 𝑖 ,𝑡 𝑖𝑚𝑐 = 𝛽 𝑖 ,0
𝑖𝑚𝑐 + 𝛽 𝑖 ,1
𝑖𝑚𝑐
𝛽 𝑖 ,𝑡 𝑚𝑘𝑡 = 𝛽 𝑖 ,0
𝑚 + 𝛽 𝑖 ,1
𝑚
25

We use both conditional IMC and equity betas as proxies for firm time-varying risks, and we examine the
effect of PDC on both time-varying risks by running the following panel regressions with fixed effects
and control variables:
𝛽 𝑖 ,𝑡 𝑐𝑜𝑛𝑑𝑖𝑡𝑖𝑜𝑛𝑎𝑙 = 𝛼 𝑖 + 𝛽 0
𝑃𝐷𝐶 𝑡 −1
+ 𝐹𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠 + 𝘀 𝑖 ,𝑡
Here, the regression is predictive. I regress the conditional beta at time t onto PDC one year before. The
regression coefficients and their standard errors are reported in Tables 10 and 11:
[Insert Table 9 here]
[Insert Table 10 here]
The dependent variables are the annual equity and IMC betas. I present results with and without
controlling for well-known firm-level predictors of returns, such as size, book-to-market ratio,
profitability and investment rate. In Table 8, I show that the PDC-based long–short portfolio is market
neutral and has high exposure to IST shocks. Therefore, I expect the regression coefficient 𝛽 0
𝑚𝑘𝑡 to be
zero and 𝛽 0
𝐼𝑀𝐶 > 0. Namely, firms with high PDC will have positive exposure to future IST risk but not to
future market risk. The standard errors reported in both tables are clustered at the firm level. Table 9
reports coefficients for the conditional equity betas. Overall, the coefficients are stable and consistently
negative, and most of them are statistically insignificant. This result validates my hypothesis and serves
as a double check on my previous findings. Panel A does not include firm fixed effects, and Panel B
includes firm fix effects. I do not include any control variables in row one for either Panel A or B, and the
conditional beta coefficients are negative and statistically significant just below the 10% threshold.
However, the statistical significance of this relation does not hold for all other rows, as I include various
control variables. R squared is generally very low (in the 2% range), which demonstrates that PDC
variation does not explain future market risk exposure variation very well. Table 10 reports the
26

coefficients of conditional IMC beta regressions, which are all positive, with high t-statistics. The relation
is statistically significant and holds up for all different specifications, with various control variables and
firm year fixed effects. Moreover, all regressions in both panels have relatively high R squared values
(above 10%), which indicates that PDC is a major source of the variation in future IMC risk exposure.
Combining results from both tables, I show that at the firm level, PDC decisions dynamically affect firms’
risk profiles, as firms increasing their PDC will increase their exposure to future IST shocks but not to
aggregate TFP shocks.
3.8 Firm future investment and PDC Previous findings indicate that firms that increase PDC will expose
themselves to higher future IST shocks. The mechanism behind this result is that firms increase future
investments following an improvement of their current WO. In this section, I provide direct evidence
using firm investment data and explore how firm investment behavior changes after PDC changes. I
obtain three types of investment data from Compustat: property plants and equipment, total assets and
intangible assets. I calculate the percentage changes in those data as three different investment rate
measurements, then I regress the investment rate on the flexible WO investment measurement PDC. To
consider PDC’s transitory and long-lasting effects on future investments, I include the one-period
forward investment rate and a cumulative three-period-forward investment rate. I also conduct placebo
tests using the current period investment rate and the previous period investment rate. The specific
regressions I run are presented here:
𝐼𝑛𝑣𝑒𝑠𝑡 _𝑟𝑎𝑡𝑒 𝑖 ,𝑡 ,𝑡 +3
= 𝛼 𝑖 + 𝛽 0
𝑃𝐷𝐶 𝑡 −1
+ 𝐹𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝘀 𝑖 ,𝑡
𝐼𝑛𝑣𝑒𝑠𝑡 _𝑟𝑎𝑡𝑒 𝑖 ,𝑡 = 𝛼 𝑖 + 𝛽 0
𝑃𝐷𝐶 𝑡 −1
+ 𝐹𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝘀 𝑖 ,𝑡
𝐼𝑛𝑣𝑒𝑠𝑡 _𝑟𝑎𝑡𝑒 𝑖 ,𝑡 = 𝛼 𝑖 + 𝛽 0
𝑃𝐷𝐶 𝑡 + 𝐹𝑖𝑥 𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝘀 𝑖 ,𝑡
𝐼𝑛𝑣𝑒𝑠𝑡 _𝑟𝑎𝑡𝑒 𝑖 ,𝑡 −1
= 𝛼 𝑖 + 𝛽 0
𝑃𝐷𝐶𝐶 𝑡 + 𝐹𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + +𝘀 𝑖 ,𝑡
27

I use the first regression to measure the persistent investment effect. The dependent variable
𝐼𝑛𝑣𝑒𝑠𝑡 _𝑟𝑎𝑡𝑒 𝑖 ,𝑡 ,𝑡 +3
represents the cumulative investment return, and the independent variable is the last
period’s PDC. I use the one-period forward investment rate as the dependent variable in the second
regression to measure the short-term investment effect. The third regression is contemporaneous, and
the last regression uses the one-period back investment rate for the placebo test. The regression results
are presented in Table 11:
[Insert Table 11 here]
Table 11 shows that among all three types of investments in the table, all forward investment rate
coefficients are positive, with various statistical significances. All current period investment rates have
insignificant positive relations. Therefore, an increase in PDC is associated with a higher future
investment rate. In Panel A, without firm fixed effects, both the one- and the three-period-forward total
asset change rates are significantly positive, with t-statistics equal to 2.16 and 1.86. However, the
coefficients become insignificant when I include firm fixed effects. Row three shows that there is a
negative correlation between past investment rate and PDC. Panel B looks at one specific type of
investment: intangible assets. All forward intangible asset investment rates are positive and statistically
significant with or without firm fixed effects. Finally, Panel C reports investment rates in property, plants
and equipment. We see a significant short-term positive relation regardless of the firm fixed effect.
However, unlike intangible assets, there is no persistent relationship between PDC and the PP&E
investment rate. Most importantly, Table 11 shows differential PDC effects to different types of
investments. For intangible assets, the three-period cumulative investment coefficients are three times
the one-period forward coefficients, whereas the three-period coefficients for PP&E investment rates
are almost identical to the one-period coefficients. Overall, I provide clear evidence that firms that
28

increase PDC expect to invest more in the future, and the result is stronger and more persistent for
types of capital that embody new technologies.
A figure version of the investment rate regression is presented in Figure 1. In Figure 1, I compare the
average de-trended accumulated investment rates for both top PDC portfolio and bottom PDC portfolio.
The portfolio is formed at year zero (0 on the x-axis). I track the investment rates from t-3 to t+3 and use
them to calculate the accumulated investment rates for all three types of investments covered in Table
11. The parallel trends exist in all three types of investments from t-3 to time 0. The differential gaps
then grow after time 0 as the blue and red lines never cross again for all three types of investments.
Among all three investments, the intangible asset investment rate gap has the largest magnitude and
grows consistently larger.
[Insert Figure 1 here]
4. Model
4.1 Firm production and capital accumulation
In this section, I present an equilibrium asset pricing model motived by Lin (2012). The economy
comprises a series of ex ante identical firms producing homogeneous products. Firms follow a constant
return to scale Cobb-Douglas production function with two types of capital inputs: regular capital K and
flexible WO capital PD. The production function is presented here:
𝑌 𝑡 = 𝐾 𝑡 𝛼 𝑃𝐷
𝑡 1−𝛼
Both capital inputs require corresponding investments and follow different laws of motion, which are
specified below:
𝐾 𝑡 +1
= ( 1 − 𝛿 𝑘 ) 𝐾 𝑡 + 𝑍 𝑡 𝑖 ( 𝑃𝐷
𝑡 +1
𝜌 + 𝑖 𝑘 ,𝑡 𝜌 )
1/𝜌
29

𝑃𝐷
𝑡 +1
= ( 1 − 𝛿 𝑃𝐷
) 𝑃 𝐷 𝑡 + 𝑖 𝑝𝑑 ,𝑡
The regular capital accumulation is subjected to 𝑍 𝑡 𝑖 , which is idiosyncratic IST shock, and firms with
positive idiosyncratic IST shocks are more effective at accumulating regular capital. This feature models
a firm’s idiosyncratic growth opportunity because firms with higher growth opportunities accumulate
capital more efficiently. Moreover, regular capital accumulation also requires WO capital as input. This is
grounded by the fact that WO can affect the efficiency of capital investment. To model the
complementary nature of WO investment and regular capital investment, the parameter 𝜌 is negative.
Let ∅ = ( 𝑃𝐷
𝑡 +1
𝜌 + 𝑖 𝑘 ,𝑡 𝜌 )
1/𝜌 and because ∅ is a homogenous degree of one (HD1), the regular capital law
of motion becomes:
𝐾 𝑡 +1
= ( 1 − 𝛿 𝑘 ) 𝐾 𝑡 + 𝑍 𝑡 𝑖 ∅ = ( 1 − 𝛿 𝑘 ) 𝐾 𝑡 + 𝑍 𝑡 𝑖 ∅
𝑖 𝑖 𝑘 ,𝑡 + 𝑍 𝑡 𝑖 ∅
𝑃 𝐷 𝑃𝐷
𝑡 +1

where ∅
𝑖 and ∅
𝑃𝐷
are first order derivatives of ∅ with respect to i and PD.
Firms also experience a standard quadratic adjustment cost:
𝐶 𝑡 =
1
2
ƞ
𝑃𝐷
(
𝑖 𝑝𝑑 ,𝑡 𝑃𝐷
𝑡 )
2
𝑃𝐷
𝑡 +
1
2𝑍 𝑡 𝑎 ƞ
𝐾 (
𝑖 𝑘 ,𝑡 𝐾 𝑡 )
2
𝐾 𝑡
Here, 𝑍 𝑡 𝑎 is an aggregate IST shock that is an investment price shock that affects all firms and their newly
installed capital. This feature is natural because a positive aggregate investment shock allows firms to
invest at a lower cost.
4.2 SDF
The stochastic discount factor is exogenous, and firms are price takers. I adopt exogenous SDF as
introduced in Kilic (2018), which features both aggregate TFP shock and aggregate IST shock:
30

𝑀 𝑡 ,𝑡 +1
= exp ( −𝑟 𝑓 )
exp ( −𝛾 𝑎 𝜎 𝑎 𝜖 𝑡 +1
𝑎 − 𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 )
𝐸 𝑡 [exp ( −𝛾 𝑎 𝜎 𝑎 𝜖 𝑡 +1
𝑎 − 𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) ]

where 𝛾 𝑎 and 𝛾 𝑧 are the market prices associated with TFP and IST shocks, respectively. The logarithm of
both TFP and IST follow a similar random walk with drift:
log (
𝑎 𝑡 +1
𝑎 𝑡 ) = 𝜇 𝑎 + 𝜎 𝑎 𝜖 𝑡 +1
𝑎

log (
𝑍 𝑡 +1
𝑎 𝑍 𝑡 𝑎 ) = 𝜇 𝑧 + 𝜎 𝑧 𝜖 𝑡 +1
𝑧
where 𝜇 𝑎 and 𝜇 𝑧 are drifts, 𝜎 𝑎 and 𝜎 𝑧 are conditional volatility, and 𝜖 𝑡 +1
𝑎 and 𝜖 𝑡 +1
𝑧 are random shocks
that follow an iid standard normal distribution.
The idiosyncratic IST shock follows an AR(1) process:
log( 𝑍 𝑡 +1
𝑖 )= ( 1 − 𝜌 𝑧𝑖
) 𝑍 ̅ + 𝜌 𝑧𝑖
log( 𝑍 𝑡 𝑖 )+ 𝜎 𝑧𝑖
𝜖 𝑡 +1
𝑧𝑖

where 𝜌 𝑧𝑖
governs the persistence, 𝑍 ̅ and 𝜎 𝑧𝑖
are the unconditional means of log IST and the conditional
volatility of the idiosyncratic IST process, respectively, and 𝜖 𝑡 +1
𝑧𝑖
is iid and has a standard normal
distribution.
4.3 Firm optimization problem
The output price is normalized to one, and firm value is defined as the expected discounted ex-cost
future dividend streams:
𝑉 ( 𝑆 𝑡 )
= 𝐸 𝑡 ∑ 𝑀 𝑡 ,𝑡 +𝑗 ( 𝑌 𝑡 +𝑗 − 𝑖 𝑘 ,𝑡 +𝑗 − 𝑖 𝑝𝑑 ,𝑡 +𝑗 − 𝐶 𝑡 +𝑗 )
∞
𝑗 =0

Therefore, firms maximize their value subject to the laws of motion for both capital and WO investment:
31

max ( 𝑉 ( 𝑆 𝑡 )
)
𝑠 . 𝑡
𝐾 𝑡 +1
= ( 1 − 𝛿 𝑘 ) 𝐾 𝑡 + 𝑍 𝑡 𝑖 ( 𝑃𝐷
𝑡 +1
𝜌 + 𝑖 𝑘 ,𝑡 𝜌 )
1/𝜌
𝑃𝐷
𝑡 +1
= ( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 + 𝑖 𝑝𝑑 ,𝑡

The Lagrange is as follows:
𝐿 = 𝐸 𝑡 ∑ 𝑀 𝑡 ,𝑡 +𝑗 {( 𝑌 𝑡 +𝑗 − 𝑖 𝑘 ,𝑡 +𝑗 − 𝑖 𝑝𝑑 ,𝑡 +𝑗 − 𝐶 𝑡 +𝑗 )
∞
𝑗 =0
− 𝑞 𝑡 +𝑗 𝑘 [𝐾 𝑡 +𝑗 +1
− ( 1 − 𝛿 𝑘 ) 𝐾 𝑡 +𝑗 − 𝑍 𝑡 +𝑗 𝑖 ∅
𝑖 𝑖 𝑘 ,𝑡 +𝑗 − 𝑍 𝑡 𝑖 ∅
𝑃𝐷
𝑃𝐷
𝑡 +𝑗 +1
]− 𝑞 𝑡 +𝑗 𝑃𝐷
[𝑃𝐷
𝑡 +𝑗 +1
− ( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 +𝑗 − 𝑖 𝑝𝑑 ,𝑡 +𝑗 ]}
The F.O.C. are presented below in the following order:
𝜕𝐿
𝜕 𝑖 𝑘 ,𝑡 ,
𝜕𝐿
𝜕 𝑖 𝑝𝑑 ,𝑡 ,
𝜕𝐿
𝜕 𝐾 𝑡 +1
and
𝜕𝐿
𝜕 𝑃𝐷
𝑡 +1

𝑞 𝑡 𝑘 =
1
𝑍 𝑡 𝑖 ∅
𝑖 ( 1 +
𝜕 𝐶 𝑡 𝜕 𝑖 𝑘 ,𝑡 )
𝑞 𝑡 𝑃𝐷
= ( 1 +
𝜕 𝐶 𝑡 𝜕 𝑖 𝑝𝑑 ,𝑡 )
𝑞 𝑡 𝑘 = 𝐸 𝑡 𝑀 𝑡 ,𝑡 +1
[
𝑑 𝑌 𝑡 +1
𝑑 𝐾 𝑡 +1
−
𝑑 𝐶 𝑡 +1
𝑑 𝐾 𝑡 +1
+ 𝑞 𝑡 +1
𝑘 ( 1− 𝛿 𝑘 ) ]
𝑞 𝑡 𝑃𝐷
= 𝑞 𝑡 𝑘 𝑍 𝑡 𝑖 ∅
𝑃𝐷
+ 𝐸 𝑡 𝑀 𝑡 ,𝑡 +1
[
𝑑 𝑌 𝑡 +1
𝑑 𝑃𝐷
𝑡 +1
−
𝑑 𝐶 𝑡 +1
𝑑 𝑃𝐷
𝑡 +1
+ 𝑞 𝑡 +1
𝑃𝐷
( 1 − 𝛿 𝑃𝐷
) ]
The first two equations explain elements that affect the shadow prices of regular capital investment as
well as the WO capital. The last two equations are expressions of optimal conditions where marginal
32

cost is equal to the marginal benefits of additional capital. The left-hand side is the cost of additional
units of capital and WO investment, and the right-hand side is the marginal gain from investing.
With the transversality condition holding:
lim
𝑗 →∞
𝐸 𝑡 𝑀 𝑡 ,𝑡 +𝑗 𝑞 𝑡 +𝑗 𝑘 𝐾 𝑡 +1+𝑗 = 0
lim
𝑗 →∞
𝐸 𝑡 𝑀 𝑡 ,𝑡 +𝑗 𝑞 𝑡 +𝑗 𝑃𝐷
𝑃𝐷
𝑡 +1+𝑗 = 0
The firm value can be expressed as follows (see proof 1 in the appendix):
𝑉 ( 𝑆 𝑡 )
= 𝑌 𝑡 −
𝜕 𝐶 𝑡 𝜕 𝐾 𝑡 𝐾 𝑡 −
𝜕 𝐶 𝑡 𝜕 𝑃𝐷
𝑡 𝑃𝐷
𝑡 + 𝑞 𝑡 𝑘 ( 1 − 𝛿 𝑘 ) 𝐾 𝑡 + 𝑞 𝑡 𝑃𝐷
( 1− 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡
The firm value can also be decomposed into the current period’s dividend and the expected future
value, which is the ex-dividend firm price:
𝑉 ( 𝑆 𝑡 )
= 𝐶𝐹
𝑡 + 𝐸 𝑡 𝑀 𝑡 ,𝑡 +1
𝑉 ( 𝑆 𝑡 +1
)
= 𝐶𝐹
𝑡 + 𝑃 𝑡
The price has the following expression (see proof 2 in the appendix):
𝑃 𝑡 = 𝑉 ( 𝑆 𝑡 )
− 𝐶𝐹
𝑡 = 𝑉 ( 𝑆 𝑡 )
− 𝑌 𝑡 + 𝑖 𝑘 ,𝑡 + 𝑖 𝑝𝑑 ,𝑡 + 𝐶 𝑡 = 𝑞 𝑡 𝑘 𝐾 𝑡 +1
+ ( 𝑞 𝑡 𝑃𝐷
− 𝑞 𝑡 𝑘 𝑍 𝑡 𝑖 ∅
𝑃𝐷
) 𝑃𝐷
𝑡 +1

The excess return is (see proof 3 in the appendix):
33

𝑅 𝑡 +1
𝑒 =
𝑃 𝑡 +1
+ 𝐶𝐹
𝑡 +1
𝑃 𝑡 =
1
𝑃 𝑡 [𝑞 𝑡 +1
𝑘 𝐾 𝑡 +2
+ ( 𝑞 𝑡 +1
𝑃𝐷
− 𝑞 𝑡 +1
𝑘 𝑍 𝑡 +1
𝑖 ∅
𝑃𝐷
) 𝑃𝐷
𝑡 +2
+ 𝑌 𝑡 +1
− 𝑖 𝑘 ,𝑡 +1
− 𝑖 𝑝𝑑 ,𝑡 +1
− 𝐶 𝑡 +1
]
=
1
𝑃 𝑡 [( 1 − 𝛿 𝑘 ) 𝑞 𝑡 +1
𝑘 +
𝜕 𝑌 𝑡 +1
𝜕 𝐾 𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
]𝐾 𝑡 +1
+
1
𝑃 𝑡 [𝑞 𝑡 +1
𝑃𝐷
( 1 − 𝛿 𝑃𝐷
)+
𝜕 𝑌 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
] 𝑃𝐷
𝑡 +1
=
𝐾 𝑡 +1
𝑞 𝑡 𝑘 𝑃 𝑡 [
( 1 − 𝛿 𝑘 ) 𝑞 𝑡 +1
𝑘 +
𝜕 𝑌 𝑡 +1
𝜕 𝐾 𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
𝑞 𝑡 𝑘 ]
+
𝑃𝐷
𝑡 +1
𝑞 𝑡 𝑃𝐷
𝑃 𝑡 [
𝑞 𝑡 +1
𝑃𝐷
( 1− 𝛿 𝑃𝐷
)+
𝜕 𝑌 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
𝑞 𝑡 𝑃𝐷
]
From the expression above, the firm return is a weighted average of capital return and WO investment
marginal return. Let 𝑅 𝑡 +1
𝑘 = [
( 1−𝛿 𝑘 ) 𝑞 𝑡 +1
𝑘 +
𝜕 𝑌 𝑡 +1
𝜕 𝐾 𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
𝑞 𝑡 𝑘 ] and 𝑅 𝑡 +1
𝑃𝐷
= [
𝑞 𝑡 +1
𝑃𝐷
( 1−𝛿 𝑃𝐷
) +
𝜕 𝑌 𝑡 +1
𝜕 𝑃 𝐷 𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
𝑞 𝑡 𝑃𝐷
]; I
summarize the excess return expression as follows:
𝑅 𝑡 +1
𝑒 =
𝐾 𝑡 +1
𝑞 𝑡 𝑘 𝑃 𝑡 𝑅 𝑡 +1
𝑘 +
𝑃𝐷
𝑡 +1
𝑞 𝑡 𝑃𝐷
𝑃 𝑡 𝑅 𝑡 +1
𝑃𝐷

Define the asset return as follows:
𝑟 𝑡 +1
𝑠 =
𝑉 ( 𝑆 𝑡 +1
)
𝑉 ( 𝑆 𝑡 )
− 𝑑 𝑡
Then, a conditional risk premium expression can be written as follows:
𝐸 𝑡 ( 𝑟 𝑡 +1
𝑠 )= 𝑟 𝑓 −
𝑐𝑜𝑣 𝑡 ( 𝑀 𝑡 ,𝑡 +1
,𝑟 𝑡 +1
𝑠 )
𝑣𝑎𝑟 𝑡 ( 𝑀 𝑡 ,𝑡 +1
)

𝑣𝑎𝑟 𝑡 ( 𝑀 𝑡 ,𝑡 +1
)
𝐸 𝑡 ( 𝑀 𝑡 ,𝑡 +1
)
= 𝑟 𝑓 + 𝛽 𝑡 𝜎𝛾
where 𝛽 𝑡 = −
𝑐𝑜𝑣 𝑡 ( 𝑀 𝑡 ,𝑡 +1
,𝑟 𝑡 +1
𝑠 )
𝑣𝑎𝑟 𝑡 ( 𝑀 𝑡 ,𝑡 +1
)
is the conditional beta, which measures the risk exposure.
34

5. Asset Pricing Mechanism
In this section, I present five propositions and provide corresponding proofs derived from the model.
These propositions capture all empirical findings in this paper. Finally, I summarize the main asset
pricing mechanism.
5.1 Proposition
Proposition I (see proof in the appendix): WO investment is positively related to the marginal return on
physical capital.
In the proof, I show
𝜕 𝑅 𝑡 +1
𝑘 𝜕 𝑖 𝑝𝑑 ,𝑡 > 0. Intuitively, a positive WO investment improves the marginal productivity
of physical capital and simultaneously decreases the cost of additional physical capital investment by
making the physical accumulation more efficient.

Proposition II (see proof in the appendix): Physical capital investment is negatively related to the
marginal return on physical capital.
I show that
𝜕 𝑅 𝑡 +1
𝑘 𝜕 𝑖 𝑘 ,𝑡 < 0. Intuitively, additional physical capital investment will reduce the marginal
productivity of physical capital due to diminishing returns and will increase the marginal adjustment cost
of physical capital and the marginal cost of physical capital investment.

Proposition III (see proof in the appendix): WO investment is negatively related to the marginal return
on WO investment, 𝑅 𝑡 +1
𝑃𝐷
. Physical capital investment is positively related to 𝑅 𝑡 +1
𝑃𝐷
.
35

Physical capital investment will increase the efficiency of WO investment’s marginal productivity and, at
the same time, lower the cost of investing in WO. On the other hand, additional investment in WO will
decrease the marginal productivity and raise the marginal investment cost.

Proposition IV (see proof in the appendix): The asset return is positively correlated with WO investment
and negatively related to physical capital investment when the market value of physical capital is larger
than the market value of WO investment.
Proposition V (see proof in the appendix): The conditional IST shock beta 𝛽 𝑡 𝐼𝑆𝑇 is increasing w.r.t
additional WO investment.
Intuitively, WO investment increase future capital investment which is subjected to higher quadratic
adjustment cost. Moreover, the IST shock which affects investment price is embedded in the capi tal
adjustment cost. Therefore a higher WO investment will lead to a greater IST shock exposure through
adjustment cost function.

5.2 Mechanism summary The model and propositions above can help us understand all the empirical
evidence presented in the Empirical Findings and Portfolio Sorting. When the firm experiences a
positive idiosyncratic IST shock, its physical capital accumulation becomes more efficient, and the firm
begins to invest to take advantage of the positive efficiency shock. WO investment and physical capital
are complements; therefore, the firm increases investment in both of them together. The market value
of physical capital is greater than the market value of WO, and the marginal physical capital return
dominates the asset return. Therefore, we observe that future stock return is positively correlated with
WO investment while negatively correlated with physical capital investment, ceteris paribus. Moreover,
36

given that the marginal adjustment cost dominates, additional workplace investment increases the
aggregate IST shock exposure, which provides us with a risk-based explanation for the higher future
returns I observe.

6. Conclusion
The paper shows that a firm’s investment in flexible WO is informative about the firm’s future growth
plan and its future risk profile related to fundamental risk factors. I study the interaction between a
firm’s current PDC decision and future expected returns following a production-based asset pricing
framework. Empirically, I find strong evidence that higher PDC investment relates to higher future
expected returns. The higher current PDC also directly relates to higher future firm investments. Firms
that invest in flexible WO are those that expect to expand beyond their typical operations. Therefore,
those firms experience higher fluctuations in technological progress embodied in new capital and have
risk profiles similar to growth firms but with higher expected returns due to the positive IST risk
premium. I also document that the PDC effect is not dominated by any particular industry but rather is a
firm-level effect. Finally, I explain the mechanism behind this novel relation between PDC and expected
return using a partial-equilibrium asset pricing model. The model successfully explains the new
interaction as well as a few established patterns in the cross-section of firms’ returns.

37

Reference
Belo, Frederico, Xiaoji Lin. 2012. “The Inventory Growth Spread.” The Review of Financial Studies,
25:278–313.
Belo, Frederico, Xiaoji Lin, and Maria Ana Vitorino. 2014. “Brand Capital and Firm Value.” Review of
Economic Dynamics, Vol. 17, pp. 150–169

Black, Sandra, and Lisa Lynch. 2001. “How to Compete: The Impact of Workplace Practices and
Information Technology on Productivity.” Review of Economics and Statistics, LXXXIII, 434–445.
Bloom, Nicholas, Luis Garicano, Raffaella Sadun, and John Van Reenen. 2014. “The Distinct Effects of
Information Technology and Communication Technology on Firm Organization.” Management
Science, 60 (12): 2859–85.
Bresnahan, Timothy F., Erik Brynjolfsson, Lorin M. Hitt. 2002. “Information Technology, Workplace
Organization, and the Demand for Skilled Labor: Firm-Level Evidence.” The Quarterly Journal of
Economics, Volume 117, Issue 1, 1 February 2002, Pages 339–376.
Cochrane, John H., 1991. “Production-Based Asset Pricing and the Link between Stock Returns and
Economic Fluctuations.” Journal of Finance, 46, 209–237.
Dessein, Wouter, and Tano Santos. 2006. “Adaptive Organizations.” Journal of Political Economy, 114(5):
956–95.
Eisfeldt, Andrea, and Dimitris Papanikolaou. 2013. “Organization Capital and the Cross-Section of
Expected Returns.” Journal of Finance 68, 1365–1406.

Eisfeldt, Andrea L., and Adriano A. Rampini. 2006. “Capital reallocation and liquidity.” Journal of
Monetary Economics 53, 369–399.

Fama. Eugene F. and Kenneth R. French. 1992. “The Cross-Section of Expected Stock Returns.” Journal of
Finance 47. 427-465.

Fama, Eugene F., and Kenneth R. French. 1993. “Common Risk Factors in the Returns on Bonds and
Stocks.” Journal of Financial Economics 33, 3-53.

Fama, Eugene F., and Kenneth R. French. 2015. “A Five-Factor Asset Pricing Model.” Journal of Financial
Economics 116: 1–22.

Fernald, John G. 2015. “Productivity and Potential Output before, during, and after the Great
Recession.” NBER Macroeconomics Annual 2014, Volume 29, 1–51. Chicago: University of
Chicago Press.
38

Fisher, Jonas D. M. 2006. “The Dynamic Effects of Neutral and Investment-Specific Technology Shocks.”
Journal of Political Economy, 114(3): 413–451.

Gomes, Joao F., Leonid Kogan, and Motohiro Yogo. 2009. “Durability of Output and Expected Stock
Returns.” Journal of Political Economy, 117 (October): 941–86.
Gomes, Joao, Leonid Kogan, and Lu Zhang. 2003. “Equilibrium Cross Section of Returns.” Journal of
Political Economy 111, 693-732.

Gordon, Robert. 1990. “Measurement of Durable Goods Prices.” Chicago: University of Chicago Press.
Greenwood, Jeremy, Zvi Hercowitz, and Per Krusell. 1997. "Long-Run Implications of Investment-
Specific Technological Change." American Economic Review, 87(3): 342-62.

Greenwood, Jeremy, Zvi Hercowitz, and Per Krusell. 2000. “The Role of Investment-Specific
Technological Change in the Business Cycle.” European Economic Review, 44(1): 91–115.

Gulek, James C., & Hakan Demirtas. 2005. “Learning with Technology: The Impact of Laptop Use on
Student Achievement.” Journal of Technology, Learning, and Assessment, 3(2).
He, Zi-Lin, Poh-Kam Wong. 2004. “Exploration and Exploitation: An Empirical Test of the Ambidexterity
Hypothesis.” Organization Science. 15(4) 481–494.
Hou, Kewei, Chen Xue, and Lu Zhang. 2015. “Digesting Anomalies: An Investment Approach.” The Review
of Financial Studies 28:650–705.
Jones, Christopher S., and Selale Tuzel. 2013a. “New Orders and Asset Prices.” The Review of Financial
Studies 26, 115–157.

Jones, Christopher S., and Selale Tuzel. 2013b. “Inventory Investment and the Cost of Capital.” Journal of
Financial Economics, 107(3), 557–579.

Katila, Riitta, and Gautam Ahuja. 2002. “Something Old, Something New: A Longitudinal Study of Search
Behavior and New Product Introduction.” Academy of Management Journal. 45 1183–1194.
Kilic, Mete. 2018. “Asset Pricing Implications of Hiring Demographics.” Working paper
Kogan, Leonid, and Dimitris Papanikolaou. 2010. “Growth Opportunities and Technology Shocks.”
American Economic Review 100, 532–536.

Kogan, Leonid, and Dimitris Papanikolaou. 2013. “Firm Characteristics and Stock Returns: The Role of
Investment- Specific Shocks.” The Review of Financial Studies 26, 2718-2759.
39

Kogan, L., and D. Papanikolaou. 2014. “Growth Opportunities, Technology Shocks, and Asset Prices.”
Journal of Finance 69:675–718.
Lamont, Owen A. 2000. “Investment Plans and Stock Returns.” Journal of Finance 55, 2719–2745.
Lewellen, Jonathan, and Stefan Nagel. 2006. “The Conditional CAPM Does not Explain Asset Pricing
Anomalies.” Journal of Financial Economics 82 (2), 289–314.
Li, Jun. 2018. “Explaining Momentum and Value Simultaneously.” Management Science 64, 4239-
4260.
Lin, Xiaoji. 2012. “Endogenous Technological Progress and the Cross Section of Stock Returns.” Journal of
Financial Economics. 103:411–428.
Nicol, David J., and Iain A. MacLeod. 2005. “Using a Shared Workspace and Wireless Laptops to Improve
Collaborative Project Learning in an Engineering Design Class.” Computers & Education, 44(4),
459-475.
Novy-Marx, Robert. 2013. “The Other Side of Value: The Gross Profitability Premium.” Journal of
Financial Economics 108:1–28.
Osterman, Paul. 1994. “How Common is Workplace Transformation and Who Adopts It?” Industrial and
Labor Relations Review, XLVII, 173–188.
Papanikolaou, Dimitris. 2011. “Investment Shocks and Asset Prices.” Journal of Political Economy 119,
639–685.

Smith, Vicki. 1997. “New Forms of Work Organization.” Annual Review of Sociology. 23:315-39
Tuzel, Selale. 2010. “Corporate Real Estate Holdings and the Cross-Section of Stock Returns.” The Review
of Financial Studies 23 (6): 2268–2302.
Tuzel, Selale, and Miao Zhang 2018. “Economic Stimulus at the Expense of Routine-Task Jobs.” Working
paper.
Zhang, Lu. 2005. “The Value Premium.” The Journal of Finance 60, 67-103.

40

Table 1: Hand-matched link table summary
This table reports the statistics of the hand-matched link table. CiTDB data are available from 1987 to 2013. The first column is
year, and the second column is the total number of siteid included in each year. The third column records the number of siteid
that have valid company names matched with Compustat firm names. Column 4 contains firms that are linked between the two
databases. Column 5 records how many sites are linked in both databases and is calculated using column 3 divided by column 2.
Column 6 measures how many siteid each matched firm has and is calculated using column 3 divided by column 4.

Year Total siteid
Siteid linked to
Compustat
Firm linked to
Compustat
Linked ratio
Number of
siteid per
firm
1987 21261 2077 393 0.098 5.3
1988 25537 2389 404 0.094 5.9
1989 27054 2840 415 0.105 6.8
1990 28204 2961 407 0.105 7.3
1991 30128 3807 464 0.126 8.2
1992 30287 4192 493 0.138 8.5
1993 28761 3522 472 0.122 7.5
1994 27041 3871 469 0.143 8.3
1995 25323 4447 508 0.176 8.8
1996 130699 5609 1528 0.043 3.7
1997 180556 7938 1999 0.044 4.0
1998 212090 14735 2916 0.069 5.1
1999 237109 14478 2707 0.061 5.3
2000 217141 10958 2419 0.050 4.5
2001 179914 8206 1967 0.046 4.2
2002 161655 6956 1770 0.043 3.9
2003 389461 8425 1905 0.022 4.4
2004 395238 7466 1819 0.019 4.1
2005 341022 9980 2056 0.029 4.9
2006 207176 5089 1358 0.025 3.7
2007 140733 3806 1115 0.027 3.4
2008 450747 8522 1821 0.019 4.7
2009 208963 7019 1577 0.036 4.5
2010 3629435 79277 2545 0.022 31.2
2011 528505 8215 1191 0.016 6.9
2012 646442 22183 1902 0.034 11.7
2013 3094251 36964 2435 0.012 15.2

41

Table 2: Panel regression and Fama-MacBeth results
Table 2 reports panel regression as well as Fama-MacBeth regression results. In Panel A, we regress annualized future monthly
returns on PDC and a set of control variables for each firm with no firm fixed effects. In Panel B, we repeat this exercise with
firm fixed effects. All coefficients are presented in the table, and the t-statistic is included in the parentheses. All standard
errors are clustered at the firm level. Panel C is the Fama-Macbeth regression result with all 6 control variables (separate and
together). I only include the PDC coefficients. PDC is the relative ratio change defined in the data section; BM is the book-to-
market ratio constructed following Fama and French (1992); Inv_rate is the investment rate and is defined as the change in the
firm’s total assets divided by last period’s total assets; Profitability is measured as gross profit divided by total assets as in Novy-
Marx (2013); Size is calculated as the log of the firm’s total asset. OC/K is ratio of organizational capital over total assets.

Panel A: Panel regression controlling for firm characteristics (No firm fixed effects)
1 2 3 4 5 6 7
PDC 0.18
(2.68)
0.20
(2.67)
0.25
(3.35)
0.26
(3.37)
0.24
(3.22)
0.254
(3.36)
0.271
(3.3)
BM

0.07
(5.09)

0.082
(5.24)
Inv_rate

-0.05
(-3.32)

-0.039
(-2.26)
Profit

0.02
(2.34)

0.01
(1.09)
Size

-0.004
(-1.41)
0
(-0.01)
OC/K

0.005
(1.9)
0.0039
(1.24)
Year fixed Y Y Y Y Y Y Y
Firm fixed N N N N N N N
Obs 16268 16268 16268 16268 16268 16268 16268
R2 0.08 0.09 0.08 0.08 0.08 0.09 0.12

42

Table 2: Panel regression and Fama-MacBeth results (Continue)

Panel B: Panel regression controlling for firm characteristics (With firm fixed effects)
1 2 3 4 5 6 7
PDC 0.25
(3.05)
0.22
(2.59)
0.32
(3.56)
0.32
(3.58)
0.32
(3.55)
0.324
(3.6)
0.286
(3.11)
BM

0.2
(8.8)

0.21
(8.03)
Inv_rate

-0.12
(-6.78)

-0.0414
(-2.17)
Profitability

0.01
(0.49)

-0.071
(-3.11)
Size

-0.2
(-9.01)
-0.17
(-7.98)
OC/K

0.052
(5.56)
0.0066
(0.74)
Year fixed Y Y Y Y Y Y Y
Firm fixed Y Y Y Y Y Y Y
Obs 16268 16268 16268 16268 16268 16268 16268
R2 0.08 0.07 0.08 0.08 0.07 0.1 0.14
Panel C: Fama-MacBeth regression
1 2 3 4 5 6 7
PDC 0.23
(2.23)
0.26
(2.24)
0.22
(2.09)
0.23
(2.23)
0.23
(2.13)
0.22
(2.1)
0.266
(2.28)

43

Table 3: Site level panel regression of PDC on desktop and laptop growth
Table 3 reports panel regression coefficients and their corresponding t-statistics. The standard error is double clusters on both
site and year. The dependent variable is site level PDC and is calculated as the laptop to total computer ratio difference
between two years. Column 1 represents the first equation in the PDC source of the variation section, and desktop growth is
measured as the change in the number of desktops divided by the last period’s total number of desktops. Column 2 represents
the second equation in the PDC source of the variation section, and laptop growth is measured as the change in the number of
laptops divided by the last period’s total number of laptops. Column 3 represents the third equation in the PDC source of the
variation section and includes both desktop growth and laptop growth.
1 2 3
Desktop growth 0.025
(8.6)
-0.07
(-9.67)
Laptop growth

0.08
(43.89)
0.1
(22.64)
Obs 47364 47364 47364
R2 0.02 0.544 0.63

44

Table 4: Firm characteristics summary
Table 4 summarizes average firm characteristics in each of the five portfolios. The portfolios are sorted based on their PDC from
L to H in ascending order. The mean values of PDC, desktop growth and laptop growth are presented in the top three rows. The
average and median value of book-to-market ratio, firm size, profitability and investment rate are winsorized at the 0.005 level
and are presented in rows 4 to 11.

Low PDC 2 3 4 High PDC
PDC -0.086 -0.017 0.002 0.017 0.066
Desktop growth -0.03 0.026 0.022 -0.002 0.06
Laptop growth -0.56 -0.08 0.07 0.2 0.58
BM
Mean 0.69 0.68 0.64 0.65 0.64
Median 0.52 0.5 0.5 0.49 0.5
Size
Mean 5.8 5.8 5.8 5.8 5.6
Median 5.7 5.7 5.8 5.7 5.5
Profitability
Mean 0.96 0.99 0.95 0.94 0.92
Median 0.81 0.83 0.81 0.8 0.79
Investment rate
Mean 0.17 0.17 0.16 0.17 0.19
Median 0.07 0.08 0.07 0.08 0.08
OC/K
Mean 2.83 2.94 2.71 2.79 2.71
Median 2.14 2.23 2.02 2.06 2.08

45

Table 5: Portfolio sorting results
Table 5 reports various portfolio sorting results. Portfolios are sorted based on PDC, with L portfolios having the lowest PDC and
H having the highest PDC. The PDC portfolio is the zero-cost top (H) minus the bottom (L) portfolio. Panel A reports both equal-
weighted and value-weighted excess annualized returns for each portfolio and the annualized returns of PDC portfolio. The T-
score is included in parentheses. Panel B reports equal-weighted excess annualized returns for industry-adjusted portfolios. In
each portfolio formation year, instead of sorting directly on PDC, we adjust the industry effect by taking out the mean value of
PDC in each industry, which is categorized by the three digits SIC code. Panel C reports the returns of portfolios sorted on
industry PDC. For each formation year, we sort industries based on their average PDC level and then take the excess average
return of the industries within each portfolio.

Low PDC 2 3 4 High PDC PDC Portfolio
Panel A: excess returns
r-rf (equal
weighted)
0.072
(1.55)
0.084
(1.8)
0.087
(1.82)
0.095
(1.99)
0.106
(2.24)
0.034
(1.98)
r-rf (value
weighted)
0.079
(2.04)
0.092
(2.44)
0.103
(2.46)
0.096
(2.37)
0.11
(2.62)
0.032
(1.87)
Panel B: industry adjusted excess return
r-rf (equal
weighted)
0.069
(1.44)
0.095
(2.07)
0.095
(1.81)
0.081
(1.89)
0.105
(2.26)
0.037
(2.16)
r-rf (value
weighted)
0.072
(1.43)
0.104
(2.19)
0.103
(1.92)
0.08
(1.74)
0.113
(2.15)
0.04
(2.65)
Panel C: industry sorting excess return
r-rf (equal
weighted)
0.092
(1.87)
0.069
(1.52)
0.062
(1.37)
0.109
(2.12)
0.114
(2.35)
0.01
(0.58)
r-rf (value
weighted)
0.101
(2.0)
0.067
(1.29)
0.071
(1.54)
0.106
(1.99)
0.126
(2.44)
0.025
(1.47)

46

Table 6: Factor model regressions (equal weighted)
Table 6 reports regression results for three risk factor models. The dependent variable is equal weighted monthly excess
returns for each portfolio, and independent variables are risk factors from the Fam– French data library. Panel A reports the
regression results of the CAPM model, and t-statistics are in parentheses. Panel B reports regression results of the Fama-French
3 factor model, and t-statistics are in parentheses. Panel C reports regression results of the Fama-French 5 factor model, and t-
statistics are in parentheses.
Low PDC 2 3 4 High PDC PDC Portfolio

Panel A: CAPM

Alpha 0.0036
(2.65)
0.0047
(3.2)
0.0043
(2.96)
0.0058
(3.84)
0.006
(3.79)
0.0025
(1.88)

MKT 1.0845
(25.11)
1.0977
(25.52)
1.141
(26.1)
1.0576
(25.44)
1.1186
(25.54)
0.0341
(1.25)
R
2

0.77 0.77 0.78 0.77 0.77 0.01

Panel B: Fama-French 3 factors

Alpha 0.0017
(2.28)
0.0029
(3.0)
0.0022
(2.87)
0.0039
(3.86)
0.0041
(3.68)
0.0024
(1.84)

MKT 1.047
(31.04)
1.0594
(31.63)
1.1016
(36.18)
1.0175
(31.7)
1.06379
(30.46)
0.0167
(0.6)
SMB 0.4438
(9.27)
0.4473
(9.41)
0.4913
(11.37)
0.4459
(9.79)
0.4976
(10.04)
0.0538
(1.36)
HML 0.4067
(8.74)
0.403
(8.73)
0.4747
(11.31)
0.3831
(8.66)
0.32435
(6.73)
-0.0823
(-2.13)
R
2

0.87 0.87 0.9 0.88 0.87 0.04

Panel C: Fama-French 5 factors

Alpha 0.0009
(1.63)
0.0023
(2.52)
0.0012
(2.08)
0.0029
(3.04)
0.0034
(3.04)
0.0025
(1.82)

MKT 1.0869
(26.17)
1.083
(26.2)
1.146
(30.8)
1.062
(26.81)
1.0979
(25.43)
0.01098
(0.31)
SMB 0.5029
(9.26)
0.499
(9.25)
0.5575
(11.46)
0.493
(9.52)
0.5489
(9.73)
0.046
(1.01)
HML 0.3513
(5.03)
0.384
(5.53)
0.413
(6.6)
0.306
(4.6)
0.2774
(3.82)
-0.0738
(-1.26)
INV -0.02
(-0.22)
-0.0757
(-0.82)
-0.02376
(-0.29)
0.047
(0.53)
-0.0197
(-0.2)
0.00059
(0.01)
PROF 0.159
(2.19)
0.1264
(1.75)
0.178
(2.73)
0.142
(2.04)
0.138
(1.82)
-0.0216
(-0.35)
R
2

0.87 0.88 0.91 0.88 0.87 0.05

47

Table 7: Factor model regressions (value weighted)
Table 6 reports regression results for three risk factor models. The dependent variable is value weighted monthly excess
returns for each portfolio, and independent variables are risk factors from the Fam– French data library. Panel A reports the
regression results of the CAPM model, and t-statistics are in parentheses. Panel B reports regression results of the Fama-French
3 factor model, and t-statistics are in parentheses. Panel C reports regression results of the Fama-French 5 factor model, and t-
statistics are in parentheses.

Low PDC 2 3 4 High PDC PDC Portfolio
Panel A: CAPM
Alpha 0.004
(2.99)
0.004
(2.92)
0.004
(3.11)
0.005
(3.59)
0.0056
(3.65)
0.0014
(1.8)
MKT 1.1
(25.45)
1.14
(25.82)
1.18
(27.75)
1.06
(25.64)
1.15
(26.2)
0.0426
(1.57)
R2 0.77 0.78 0.8 0.76 0.78 0.01
Panel B: Fama-French 3 factors
Alpha 0.0025
(2.68)
0.0023
(2.61)
0.0025
(2.93)
0.0035
(3.41)
0.0038
(3.44)
0.0013
(1.73)
MKT 1.07
(29.64)
1.11
(30.78)
1.15
(34.72)
1.03
(29.6)
1.1
(30.1)
0.03
(1.08)
SMB 0.39
(7.66)
0.42
(8.24)
0.409
(8.69)
0.37
(7.47)
0.437
(8.4)
0.044
(1.1)
HML 0.379
(7.62)
0.4
(2.61)
0.43
(9.38)
0.36
(7.41)
0.332
(6.58)
-0.047
(-1.22)
R2 0.85 0.87 0.89 0.85 0.86 0.03
Panel C: Fama-French 5 factors
Alpha 0.0005
(1.47)
0.001
(1.68)
0.001
(1.96)
0.002
(2.53)
0.003
(2.56)
0.002
(1.88)
MKT 1.16
(26.77)
1.17
(26.72)
1.22
(30.09)
1.09
(25.32)
1.16
(25.77)
0
(-0.01)
SMB 0.49
(8.71)
0.5
(8.73)
0.49
(9.25)
0.42
(7.54)
0.5
(8.45)
0.003
(0.942)
HML 0.23
(3.16)
0.3
(4.05)
0.33
(4.9)
0.26
(3.56)
0.23
(3.04)
0
(-0.01)
INV 0.069
(0.71)
0.024
(0.25)
0.018
(0.2)
0.076
(0.79)
0.066
(0.65)
-0.003
(-0.04)
PROF
0.297
(3.92)
0.228
(2.98)
0.223
(3.15)
0.166
(2.21)
0.184
(2.56)
-0.113
(-1.8)
R2 0.87 0.087 0.9 0.86 0.86 0.05

48

Table 8: Macroeconomic shock regressions
Table 8 reports the regression results for the PDC portfolio on various macroeconomic shocks. From columns 1 to 4, we
aggregate PDC monthly returns to quarterly returns because price shocks come in a quarterly frequency. Column 1 displays
results for the TFP shock from Fernald (2014) and a price shock for communication and software equipment as independent
variables. The second column has the price shock of communication equipment and software as the only variable. Column 3 has
the TFP shock and the price shock of new equipment as independent variables. Column 4 has the price shock of new equipment
alone. Column 5 is the CAMP model. Column 6 has both market excess return and IMC portfolio return as independent
variables. Columns 5 and 6 are performed with monthly returns.

PDC
1 2 3 4 5 6
TFP 0
(0.03)

0
(0.19)

INT 0.24
(1.66)
0.24
(1.69)

EQP

0.35
(1.64)
0.34
(1.66)

R
m

0.034
(1.25)
-0.011
(-0.37)
R
imc

0.118
(3.1)
R squared 0.04 0.04 0.04 0.04 0.01 0.06

49

Table 9: Conditional equity beta regressions
Table 9 reports conditional equity beta regression coefficients from various regression specifications. Dependent variables are
the conditional equity betas at annual frequency. All independent variables are defined exactly as in Table 2. All standard errors
are clustered at the firm level.

Panel A
1 2 3 4 5 6
PDC -0.417
(-1.81)
-0.419
(-1.68)
-0.399
(-1.53)
-0.416
(-1.6)
-0.422
(-1.62)
-0.428
(-1.53)
BM

0.1
(2.2)

0.067
(1.28)
Inv_rate

0.138
(2.84)

0.127
(2.33)
Profitability

-0.167
(-4.72)
-0.16
(-4.03)
Size

-0.031
(-2.15)
-0.029
(-1.75)
year fixed Y Y Y Y Y Y
firm fixed N N N N N N
Obs 16268 16268 1626268 16268 16268 16268
R2 0.024 0.025 0.028 0.031 0.027 0.034
Panel B
1 2 3 4 5 6
PDC -0.41
(-1.71)
-0.387
(-1.5)
-0.324
(-1.2)
-0.33
(-1.23)
-0.33
(-1.23)
-0.333
(-1.16)
BM

0.12
(1.97)

0.086
(1.28)
Inv_rate

0.11
(2.16)

0.11
(1.83)
Profitability

-0.04
(-0.68)
0.002
(0.03)
Size

0.057
(1.13)
0.067
(1.08)
year fixed Y Y Y Y Y Y
firm fixed Y Y Y Y Y Y
Obs 16268 16268 1626268 16268 16268 16268
R2 0.03 0.03 0.03 0.04 0.03 0.05

50

Table 10: Conditional IMC beta regressions
Table 10 reports conditional IMC beta regression coefficients from various regression specifications. Dependent variables are
the conditional IMC betas at annual frequency. All independent variables are defined e xactly as in Table 2. All standard errors
are clustered at the firm level.

Panel A
1 2 3 4 5 PDC
PDC 0.78
(1.93)
0.88
(2.02)
0.89
(1.97)
0.88
(1.94)
0.86
(1.91)
1.03
(2.16)
BM

-0.06
(-0.97)

-0.14
(-1.88)
Inv_rate

0.057
(0.86)

0.03
(0.4)
Profitability

-0.171
(-3.51)
-0.22
(-4.17)
Size

-0.05
(-2.5)
-0.05
(-2.29)
year fixed Y Y Y Y Y Y
firm fixed N N N N N N
Obs 16268 16268 1626268 16268 16268 16268
R2 0.137 0.128 0.127 0.131 0.129 0.129
Panel B
1 2 3 4 5 PDC
PDC 0.7
(1.62)
0.81
(1.75)
0.93
(1.96)
0.92
(1.94)
0.93
(1.96)
1.13
(2.24)
BM

-0.169
(-1.84)

-0.245
(-2.49)
Inv_rate

-0.001
(-0.02)

-0.05
(-0.64)
Profitability

0.097
(1.14)
0.15
(1.59)
Size

0.289
(4.22)
0.356
(4.51)
year fixed Y Y Y Y Y Y
firm fixed Y Y Y Y Y Y
Obs 16268 16268 1626268 16268 16268 16268
R2 0.14 0.13 0.13 0.12 0.09 0.158

51

Table 11: Investment rate and PDC regression
Table 11 reports regression coefficients for the investment rate and PDC regressions. Standard errors are clustered at the firm
level. In Panel A, dependent variables are physical investment rates calculated as the change in total assets divided by the last
period’s total assets. Inv_rate 1 is the one-period forward investment rate. Inv_rate 1_3 is the cumulative investment rate for
the next three periods. Inv_rate b1 is the last period investment rate, and inve_rate is contemporaneous. Panel B reflects the
intangible asset investment rate. Panel C is the property plant and equipment investment rate. All calculations and variable
naming formats in Panels B and C are the same as in Panel A.

Panel A

Investment
rate t+1 to t+2
Investment
rate t+1 to t+4
Investment
rate t-1 to t
Investment
rate t to t+1
No firm fixed effects 0.167
(2.16)
0.313
(1.86)
-0.139
(-1.55)
0.127
(1.4)
With firm fixed effects 0.133
(1.57)
0.178
(0.99)
-0.19
(-1.93)
0.087
(0.89)
Panel B

Intangible
rate t+1 to t+2
Intangible
rate t+1 to t+4
Intangible
rate t-1 to t
Intangible
rate t to t+1
No firm fixed effects 0.977
(1.97)
3.287
(1.89)
-0.235
(-0.33)
0.804
(1.42)
With firm fixed effects 1.134
(1.99)
3.438
(1.82)
-0.795
(-0.99)
0.4
(0.62)
Panel C

PP&E
rate t+1 to t+2
PP&E
rate t+1 to t+4
PP&E
rate t-1 to t
PP&E
rate t to t+1
No firm fixed effects 0.109
(2.54)
0.135
(1.08)
-0.021
(-0.4)
-0.065
(-1.33)
With firm fixed effects 0.135
(2.93)
0.108
(0.83)
-0.047
(-0.82)
-0.097
(-1.06)

52

Table 12: Depth and Scope regression
Table 12 reports regression results from three regression specifications. I run two separate regressions in each of all three
regressions. The dependent variables are always scope and depth (see appendix for the detailed construction method). In Panel
A, the key independent variable is PDC, and it is laptop growth in Panel B. In Panel C, I have both laptop growth and desktop
growth. Firm fixed effects are included in all regressions, and the standard errors are clustered at the firm level.

PDC
Laptop
growth
Desktop
growth

Panel A

Scope 5.98
(1.82)
Depth 4.11
(1.3)

Panel B
Scope

0.56
(2.26)
Depth

0.39
(1.56)
Panel C
Scope

0.65
(1.73)
-0.18
(-0.49)
Depth

0.7
(1.17)
-0.58
(-0.77)
53

Figure 1: Investment
Figure 1 reports the average accumulated investment rates for both highest and lowest PDC portfolios. The accumulated
investment rates are de-trended. The Portfolios are formed at year 0 and the investment rates are recorded from year t-3 to
year t+3 and then the accumulated rates are calculated. Figure 1 considers three types of capital investments included in table
11.

54

7. Appendix
7.1 A simple model explaining the relationship between IT and flexible WO I provide a simple model to
illustrate why improving a firm’s quality of communication will improve its flexibility. The model is
inspired by Dessin and Santos (2006).
For simplicity, we assume that the firm’s profit is as follows:
𝐵 𝑓 ,𝑡 = ∑ {∅ [𝐶 − ( 𝑎 𝑡 𝑖𝑖
− 𝜃 𝑡 𝑖 )
2
] + 𝛽 (𝐶 − ∑ ( 𝑎 𝑡 𝑗𝑖
− 𝑎 𝑡 𝑖𝑖
)
2
𝑗 ≠𝑖 )}
𝑁 𝑖 =1

Firm f’s profit at time t 𝐵 𝑓 ,𝑡 is composed of the productivities of N tasks, and each task’s productivity
depends on two separate sets of actions: primary, which determine how adaptive the firm can be, and
secondary, which coordinate between other primary actions. Regarding task i, the firm takes primary
action 𝑎 𝑡 𝑖𝑖
based on local information 𝜃 𝑡 𝑖 . The importance of the primary action is governed by an
adaptation parameter ∅. The firm then coordinates secondary actions 𝑎 𝑡 𝑗𝑖
based on task i’s primary
action, and the importance of coordination is governed by 𝛽 . Adaptation is then defined as the reaction
of the i
th
primary action 𝑎 𝑡 𝑖𝑖
to the i
th
task condition 𝜃 𝑡 𝑖 , which is a random variable realized and observed
only by the primary action, and coordination is defined as the corresponding actions 𝑎 𝑡 𝑗𝑖
to the primary
action 𝑎 𝑡 𝑖𝑖
.
Prefect adaptation and coordination occur when the primary action is 𝑎 𝑡 𝑖𝑖
= 𝜃 𝑡 𝑖 and 𝑎 𝑡 𝑗𝑖
= 𝑎 𝑡 𝑖𝑖
. However,
the communication between primary and corresponding actions is not perfectly efficient, as the local
information is passed to the secondary action correctly with probability p. For example, if the worker
decides the primary actions to be exactly 𝜃 𝑡 𝑖 , he would have a harder time communicating and setting all
corresponding actions 𝑎 𝑡 𝑗𝑖
to 𝑎 𝑡 𝑖𝑖
. On the other hand, if the worker chooses not to adapt to local
information 𝜃 𝑡 𝑖 and sets primary actions for all N tasks to be the same, the risk of miscommunication
would be eliminated, but the firm would be rigid and experience loss on the adaptation portion of
productivity. The firm can also choose to improve the communication quality to minimize coordination
55

failure. The improvement in communication will increase the firm’s adaptive ability because the firm is
able to take a more customized primary action with a higher probability of correctly passing this
information to the secondary actions. The whole setup unfolds following the sequence below:
i. The local conditions 𝜃 𝑡 𝑖 are realized and observed by the primary action.
ii. Workers communicate the realizations of local information, and with an independent
probability p, those communications are successful.
iii. Everybody costlessly chooses his primary action and corresponding actions to maximize
productivity 𝐵 𝑓 ,𝑡
We solve the optimal primary and second actions backward. First, we determine the corresponding
action 𝑎 𝑡 𝑗𝑖
:
𝑎 𝑡 𝑗𝑖
= {
𝑎 𝑡 𝑖𝑖
𝑊 ℎ𝑒𝑛 𝑐𝑜𝑚𝑚𝑢𝑛𝑖𝑐𝑎𝑡𝑖𝑜𝑛 𝑖𝑠 𝑠𝑢𝑐𝑐𝑒𝑠𝑠𝑓𝑢𝑙 𝜃 𝑡 𝑖 ̂
𝑊 ℎ𝑒𝑛 𝑐𝑜𝑚𝑚𝑢𝑛𝑖𝑐𝑎𝑡𝑖𝑜𝑛 𝑖𝑠 𝑢𝑛𝑠𝑢𝑐𝑐𝑒𝑠𝑠𝑓𝑢𝑙
𝜃 𝑡 𝑖 ̂
is the mean of local information 𝜃 𝑡 𝑖 . When communication is successful, agents know 𝑎 𝑡 𝑖𝑖
perfectly and
when communication fails, agents maximize expected profit given that they have learned nothing about
primary action: 𝑚𝑎𝑥𝐸 {𝛽 (𝐶 − ∑ ( 𝑎 𝑡 𝑗𝑖
− 𝑎 𝑡 𝑖𝑖
)
2
𝑗 ≠𝑖 )}. Therefore, 𝑎 𝑡 𝑗𝑖
= 𝐸 ( 𝑎 𝑡 𝑖𝑖
) = 𝜃 𝑡 𝑖 ̂

Next, we determine the primary action 𝑎 𝑡 𝑖𝑖
given the response of the corresponding actions solved
previously: 𝑀𝑎𝑥 𝐸 {∅ [𝐶 − ( 𝑎 𝑡 𝑖𝑖
− 𝜃 𝑡 𝑖 )
2
+ 𝛽 (𝐶 − ∑ ( 𝑎 𝑡 𝑗𝑖
− 𝑎 𝑡 𝑖𝑖
)
2
𝑗 ≠𝑖 )]} = 𝑀𝑖𝑛 [∅( 𝑎 𝑡 𝑖𝑖
− 𝜃 𝑡 𝑖 )
2
+ 𝑝 ∗
𝛽 ( 𝑛 − 1) ( 𝑎 𝑡 𝑖𝑖
− 𝑎 𝑡 𝑖𝑖
)
2
+ 𝛽 ( 𝑛 − 1) ( 1 − 𝑝 )( 𝜃 𝑡 𝑖 ̂
− 𝑎 𝑡 𝑖𝑖
)
2
], and the result is as follows:
𝑎 𝑡 𝑖𝑖
= 𝜃 𝑡 𝑖 ̂
+
∅
∅ + 𝛽 ( 𝑛 − 1) ( 1 − 𝑝 )
( 𝜃 𝑡 𝑖 − 𝜃 𝑡 𝑖 ̂
)

We want to measure a firm’s adaptive ability as how accurately the primary action varies with local
information. Therefore, we define a firm’s adaptive flexibility as the covariance between 𝑎 𝑡 𝑖𝑖
and 𝜃 𝑡 𝑖 :
56

𝑐𝑜𝑣 ( 𝑎 𝑡 𝑖𝑖
, 𝜃 𝑡 𝑖 ) = [
∅
∅ + 𝛽 ( 𝑛 − 1) ( 1 − 𝑝 )
]𝜎 𝜃 2

From the expression above, we clearly see that having a high probability of succe ssful communication
will increase the comovement between a firm’s primary action and local information that measures a
firm’s flexibility.

7.2 Proofs
Proof 1:
Firm value closed-form solution
𝐿 = 𝑌 𝑡 − 𝑖 𝑘 ,𝑡 − 𝑖 𝑝𝑑 ,𝑡 − 𝐶 𝑡 − 𝑞 𝑡 𝑘 [𝐾 𝑡 +1
− ( 1 − 𝛿 𝑘 ) 𝐾 𝑡 − 𝑍 𝑡 𝑖 ∅
𝑖 𝑖 𝑘 ,𝑡 − 𝑍 𝑡 𝑖 ∅
𝑃𝐷
𝑃𝐷
𝑡 +1
]− 𝑞 𝑡 𝑃𝐷
[𝑃𝐷
𝑡 +1
− ( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 − 𝑖 𝑝𝑑 ,𝑡 + 𝐸 𝑡 ∑ 𝑀 𝑡 ,𝑡 +𝑗 {… }
∞
𝑗 =1

Plug in the first two F.O.C.
𝐿 = 𝑌 𝑡 − 𝑖 𝑘 ,𝑡 − 𝑖 𝑝𝑑 ,𝑡 −
𝜕 𝐶 𝑡 𝜕 𝑖 𝑘 ,𝑡 𝑖 𝑘 ,𝑡 −
𝜕 𝐶 𝑡 𝜕 𝑖 𝑝𝑑 ,𝑡 𝑖 𝑝𝑑 ,𝑡 −
𝜕 𝐶 𝑡 𝜕 𝐾 𝑡 𝐾 𝑡 −
𝜕 𝐶 𝑡 𝜕 𝑃𝐷
𝑡 𝑃𝐷
𝑡 −
1
𝑍 𝑡 𝑖 ∅
𝑖 ( 1
+
𝜕 𝐶 𝑡 𝜕 𝑖 𝑘 ,𝑡 ) [𝐾 𝑡 +1
− ( 1 − 𝛿 𝑘 ) 𝐾 𝑡 − 𝑍 𝑡 𝑖 ∅
𝑖 𝑖 𝑘 ,𝑡 − 𝑍 𝑡 𝑖 ∅
𝑃𝐷
𝑃𝐷
𝑡 +1
] − ( 1 +
𝜕 𝐶 𝑡 𝜕 𝐼𝑇
𝑡 ) [𝑃𝐷
𝑡 +1
− ( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 − 𝑖 𝑝𝑑 ,𝑡 ]+ 𝐸 𝑡 ∑ 𝑀 𝑡 ,𝑡 +𝑗 {… }
∞
𝑗 =1

𝐿 = 𝑌 𝑡 −
𝜕 𝐶 𝑡 𝜕 𝐾 𝑡 𝐾 𝑡 −
𝜕 𝐶 𝑡 𝜕 𝑃𝐷
𝑡 𝑃𝐷
𝑡 − 𝑞 𝑡 𝑘 [𝐾 𝑡 +1
− ( 1 − 𝛿 𝑘 ) 𝐾 𝑡 − 𝑍 𝑡 𝑖 ∅
𝑃𝐷
𝑃𝐷
𝑡 +1
] − 𝑞 𝑡 𝑃𝐷
[𝑃𝐷
𝑡 +1
− ( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 ]
+ 𝐸 𝑡 ∑ 𝑀 𝑡 ,𝑡 +𝑗 {… }
∞
𝑗 =1

57

𝐸 𝑡 ∑ 𝑀 𝑡 ,𝑡 +𝑗 {… }
∞
𝑗 =1
= 𝐸 𝑡 𝑀 𝑡 ,𝑡 +1
{𝑌 𝑡 +1
− 𝑖 𝑘 ,𝑡 +1
− 𝑖 𝑝𝑑 ,𝑡 +1
− 𝐶 𝑡 +1
− 𝑞 𝑡 +1
𝑘 [𝐾 𝑡 +2
− ( 1 − 𝛿 𝑘 ) 𝐾 𝑡 +1
− 𝑍 𝑡 +1
𝑖 ∅
𝑖 𝑖 𝑘 ,𝑡 +1
− 𝑍 𝑡 +1
𝑖 ∅
𝑃𝐷
𝑃𝐷
𝑡 +2
]
− 𝑞 𝑡 +1
𝑃𝐷
[𝑃𝐷
𝑡 +2
− ( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 +1
− 𝑖 𝑝𝑑 ,𝑡 +1
}+ 𝐸 𝑡 ∑ 𝑀 𝑡 ,𝑡 +𝑗 { … }
∞
𝑗 =2
= 𝐸 𝑡 𝑀 𝑡 ,𝑡 +1
𝜕 𝑌 𝑡 +1
𝜕 𝐾 𝑡 +1
𝐾 𝑡 +1
+
𝜕 𝑌 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
𝑃𝐷
𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
𝐾 𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
𝑃𝐷
𝑡 +1
− 𝑞 𝑡 +1
𝑘 [𝐾 𝑡 +2
− ( 1 − 𝛿 𝑘 ) 𝐾 𝑡 +1
− 𝑍 𝑡 +1
𝑖 ∅
𝑃𝐷
𝑃𝐷
𝑡 +2
] − 𝑞 𝑡 +1
𝑃𝐷
[𝑃𝐷
𝑡 +2
− ( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 +1
]
+ 𝐸 𝑡 ∑ 𝑀 𝑡 ,𝑡 +𝑗 { … }
∞
𝑗 =2

Collect terms for both 𝐾 𝑡 +1
and 𝑃𝐷
𝑡 +1
:
𝐸 𝑡 ∑ 𝑀 𝑡 ,𝑡 +𝑗 {…}
∞
𝑗 =1
= 𝐸 𝑡 𝑀 𝑡 ,𝑡 +1
{ [
𝜕 𝑌 𝑡 +1
𝜕 𝐾 𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
+ 𝑞 𝑡 +1
𝑘 ( 1− 𝛿 𝑘 ) ]𝐾 𝑡 +1
+ [
𝜕 𝑌 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
+ 𝑞 𝑡 +1
𝑃𝐷
( 1− 𝛿 𝑃𝐷
) ]𝑃𝐷
𝑡 +1
− 𝑞 𝑡 +1
𝑘 [𝐾 𝑡 +2
− 𝑍 𝑡 +1
𝑖 ∅
𝑃𝐷
𝑃𝐷
𝑡 +2
]
− 𝑞 𝑡 +1
𝑃𝐷
𝑃𝐷
𝑡 +2
}+ 𝐸 𝑡 ∑ 𝑀 𝑡 ,𝑡 +𝑗 { … }
∞
𝑗 =2
= 𝑞 𝑡 𝑘 𝐾 𝑡 +1
+ ( 𝑞 𝑡 𝑃𝐷
− 𝑞 𝑡 𝑘 𝑍 𝑡 𝑖 ∅
𝑃𝐷
) 𝑃𝐷
𝑡 +1
− 𝐸 𝑡 𝑀 𝑡 ,𝑡 +1
{ 𝑞 𝑡 +1
𝑘 [𝐾 𝑡 +2
− 𝑍 𝑡 +1
𝑖 ∅
𝑃𝐷
𝑃𝐷
𝑡 +2
] − 𝑞 𝑡 +1
𝑃𝐷
𝑃𝐷
𝑡 +2
}+ 𝐸 𝑡 ∑ 𝑀 𝑡 ,𝑡 +𝑗 { … }
∞
𝑗 =2

𝐿 = 𝑌 𝑡 −
𝜕 𝐶 𝑡 𝜕 𝐾 𝑡 𝐾 𝑡 −
𝜕 𝐶 𝑡 𝜕 𝑃𝐷
𝑡 𝑃𝐷
𝑡 − 𝑞 𝑡 𝑘 [𝐾 𝑡 +1
− ( 1 − 𝛿 𝑘 ) 𝐾 𝑡 − 𝑍 𝑡 𝑖 ∅
𝑃𝐷
𝑃𝐷
𝑡 +1
] − 𝑞 𝑡 𝑃𝐷
[𝑃𝐷
𝑡 +1
− ( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 ]
+ 𝐸 𝑡 ∑ 𝑀 𝑡 ,𝑡 +𝑗 { … }
∞
𝑗 =1
= 𝑌 𝑡 −
𝜕 𝐶 𝑡 𝜕 𝐾 𝑡 𝐾 𝑡 −
𝜕 𝐶 𝑡 𝜕 𝑃𝐷
𝑡 𝑃𝐷
𝑡 − 𝑞 𝑡 𝑘 [𝐾 𝑡 +1
− ( 1 − 𝛿 𝑘 ) 𝐾 𝑡 − 𝑍 𝑡 𝑖 ∅
𝑃𝐷
𝑃𝐷
𝑡 +1
]
− 𝑞 𝑡 𝑃𝐷
[𝑃𝐷
𝑡 +1
− ( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 ] + 𝑞 𝑡 𝑘 𝐾 𝑡 +1
+ ( 𝑞 𝑡 𝑃𝐷
− 𝑞 𝑡 𝑘 𝑍 𝑡 𝑖 ∅
𝑃𝐷
) 𝑃𝐷
𝑡 +1
− 𝐸 𝑡 𝑀 𝑡 ,𝑡 +1
{ 𝑞 𝑡 +1
𝑘 [𝐾 𝑡 +2
− 𝑍 𝑡 +1
𝑖 ∅
𝑃𝐷
𝑃𝐷
𝑡 +2
]− 𝑞 𝑡 +1
𝑃𝐷
𝑃𝐷
𝑡 +2
}+ 𝐸 𝑡 ∑ 𝑀 𝑡 ,𝑡 +𝑗 { … }
∞
𝑗 =2
= 𝑌 𝑡 −
𝜕 𝐶 𝑡 𝜕 𝐾 𝑡 𝐾 𝑡 −
𝜕 𝐶 𝑡 𝜕 𝑃𝐷
𝑡 𝑃𝐷
𝑡 + 𝑞 𝑡 𝑘 ( 1 − 𝛿 𝑘 ) 𝐾 𝑡 + 𝑞 𝑡 𝑃𝐷
( 1− 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 − 𝐸 𝑡 𝑀 𝑡 ,𝑡 +1
{ 𝑞 𝑡 +1
𝑘 [𝐾 𝑡 +2
− 𝑍 𝑡 +1
𝑖 ∅
𝑃𝐷
𝑃𝐷
𝑡 +2
]− 𝑞 𝑡 +1
𝑃𝐷
𝑃𝐷
𝑡 +2
}+ 𝐸 𝑡 ∑ 𝑀 𝑡 ,𝑡 +𝑗 { … }
∞
𝑗 =2

Keep expanding the last term 𝐸 𝑡 ∑ 𝑀 𝑡 ,𝑡 +𝑗 { … }
∞
𝑗 =2
, and with the transversality condition, the value of the
firm becomes:
58

𝑉 = 𝐿 = 𝑌 𝑡 −
𝜕 𝐶 𝑡 𝜕 𝐾 𝑡 𝐾 𝑡 −
𝜕 𝐶 𝑡 𝜕 𝑃𝐷
𝑡 𝑃𝐷
𝑡 + 𝑞 𝑡 𝑘 ( 1 − 𝛿 𝑘 ) 𝐾 𝑡 + 𝑞 𝑡 𝑃𝐷
( 1− 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡
Proof 2:
𝑃 𝑡 = 𝑉 ( 𝑆 𝑡 )
− 𝐶𝐹
𝑡 = 𝑉 ( 𝑆 𝑡 )
− 𝑌 𝑡 + 𝑖 𝑘 ,𝑡 + 𝑖 𝑝𝑑 ,𝑡 + 𝐶 𝑡 = 𝑖 𝑘 ,𝑡 + 𝑖 𝑝𝑑 ,𝑡 + 𝐶 𝑡 −
𝜕 𝐶 𝑡 𝜕 𝐾 𝑡 𝐾 𝑡 −
𝜕 𝐶 𝑡 𝜕 𝑃𝐷
𝑡 𝑃𝐷
𝑡 + 𝑞 𝑡 𝑘 ( 1 − 𝛿 𝑘 ) 𝐾 𝑡 + 𝑞 𝑡 𝑃𝐷
( 1− 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 = 𝑞 𝑡 𝑘 ( 1− 𝛿 𝑘 ) 𝐾 𝑡 + 𝑞 𝑡 𝑃𝐷
( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 + 𝑖 𝑘 ,𝑡 + 𝑖 𝑝𝑑 ,𝑡 +
𝜕 𝐶 𝑡 𝜕 𝑖 𝑘 ,𝑡 𝑖 𝑘 ,𝑡 +
𝜕 𝐶 𝑡 𝜕 𝑖 𝑝𝑑 ,𝑡 𝑖 𝑝𝑑 ,𝑡 = 𝑞 𝑡 𝑘 ( 1− 𝛿 𝑘 ) 𝐾 𝑡 + 𝑞 𝑡 𝑃𝐷
( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 + (1 +
𝜕 𝐶 𝑡 𝜕 𝑖 𝑘 ,𝑡 ) 𝑖 𝑘 ,𝑡 + (1 +
𝜕 𝐶 𝑡 𝜕 𝑖 𝑝𝑑 ,𝑡 ) 𝑖 𝑝𝑑 ,𝑡 = 𝑞 𝑡 𝑘 ( 1− 𝛿 𝑘 ) 𝐾 𝑡 + 𝑞 𝑡 𝑃𝐷
( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 + 𝑍 𝑡 𝑖 ∅
𝑖 𝑞 𝑡 𝑘 𝑖 𝑘 ,𝑡 + 𝑞 𝑡 𝑃𝐷
𝑖 𝑝𝑑 ,𝑡 = 𝑞 𝑡 𝑘 [( 1− 𝛿 𝑘 ) 𝐾 𝑡 + 𝑍 𝑡 𝑖 ∅
𝑖 𝑖 𝑘 ,𝑡 ]+ 𝑞 𝑡 𝑃𝐷
[( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 + 𝑖 𝑝𝑑 ,𝑡 ]
= 𝑞 𝑡 𝑃𝐷
𝑃𝐷
𝑡 +1
+ 𝑞 𝑡 𝑘 ( 𝐾 𝑡 +1
− 𝑍 𝑡 𝑖 ∅
𝑃𝐷
𝑃𝐷
𝑡 +1
) = 𝑞 𝑡 𝑘 𝐾 𝑡 +1
+ (𝑞 𝑡 𝑃𝐷
− 𝑞 𝑡 𝑘 𝑍 𝑡 𝑖 ∅
𝑃𝐷
) 𝑃𝐷
𝑡 +1

Proof 3:
𝑃 𝑡 +1
+ 𝐶𝐹
𝑡 +1
= 𝑞 𝑡 +1
𝑘 𝐾 𝑡 +2
+ ( 𝑞 𝑡 +1
𝑃𝐷
− 𝑞 𝑡 +1
𝑘 𝑍 𝑡 +1
𝑖 ∅
𝑃𝐷
) 𝑃𝐷
𝑡 +2
+ 𝑌 𝑡 +1
− 𝑖 𝑘 ,𝑡 +1
− 𝑖 𝑝𝑑 ,𝑡 +1
− 𝐶 𝑡 +1
= 𝑞 𝑡 +1
𝑘 [( 1 − 𝛿 𝑘 ) 𝐾 𝑡 +1
+ 𝑍 𝑡 +1
𝑖 ∅
𝑖 𝑖 𝑘 ,𝑡 +1
+ 𝑍 𝑡 +1
𝑖 ∅
𝑃𝐷
𝑃𝐷
𝑡 +2
] + ( 𝑞 𝑡 +1
𝑃𝐷
− 𝑞 𝑡 +1
𝑘 𝑍 𝑡 +1
𝑖 ∅
𝑃𝐷
) 𝑃𝐷
𝑡 +2
+ 𝑌 𝑡 +1
− 𝑖 𝑘 ,𝑡 +1
− 𝑖 𝑝𝑑 ,𝑡 +1
− 𝐶 𝑡 +1
= 𝑞 𝑡 +1
𝑘 [( 1 − 𝛿 𝑘 ) 𝐾 𝑡 +1
+ 𝑍 𝑡 +1
𝑖 ∅
𝑖 𝑖 𝑘 ,𝑡 +1
]+ 𝑞 𝑡 +1
𝑃𝐷
[( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 +1
+ 𝑖 𝑝𝑑 ,𝑡 +1
]+ 𝑌 𝑡 +1
− 𝑖 𝑘 ,𝑡 +1
− 𝑖 𝑝𝑑 ,𝑡 +1
− 𝐶 𝑡 +1

Because 𝑌 𝑡 +1
and 𝐶 𝑡 +1
are homogenous degree 1, we can further expand the expression above into the
following:
59

𝑃 𝑡 +1
+ 𝐶𝐹
𝑡 +1
= 𝑞 𝑡 +1
𝑘 [( 1− 𝛿 𝑘 ) 𝐾 𝑡 +1
+ 𝑍 𝑡 +1
𝑖 ∅
𝑖 𝑖 𝑘 ,𝑡 +1
] + 𝑞 𝑡 +1
𝑃𝐷
[( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 +1
+ 𝑖 𝑝𝑑 ,𝑡 +1
] +
𝜕 𝑌 𝑡 +1
𝜕 𝐾 𝑡 +1
𝐾 𝑡 +1
+
𝜕 𝑌 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
𝑃𝐷
𝑡 +1
− 𝑖 𝑘 ,𝑡 +1
− 𝑖 𝑝𝑑 ,𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
𝐾 𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
𝑃𝐷
𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝑖 𝑝𝑑 ,𝑡 +1
𝑖 𝑝𝑑 ,𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝑖 𝑘 ,𝑡 +1
𝑖 𝑘 ,𝑡 +1
= 𝑞 𝑡 +1
𝑘 [( 1 − 𝛿 𝑘 ) 𝐾 𝑡 +1
+ 𝑍 𝑡 +1
𝑖 ∅
𝑖 𝑖 𝑘 ,𝑡 +1
] + 𝑞 𝑡 +1
𝑃𝐷
[( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 +1
+ 𝑖 𝑝𝑑 ,𝑡 +1
] +
𝜕 𝑌 𝑡 +1
𝜕 𝐾 𝑡 +1
𝐾 𝑡 +1
+
𝜕 𝑌 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
𝑃𝐷
𝑡 +1
− (1 +
𝜕 𝐶 𝑡 +1
𝜕 𝑖 𝑘 ,𝑡 +1
) 𝑖 𝑘 ,𝑡 +1
− (1 +
𝜕 𝐶 𝑡 +1
𝜕 𝑖 𝑝𝑑 ,𝑡 +1
) 𝑖 𝑝𝑑 ,𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
𝐾 𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
𝑃𝐷
𝑡 +1
= 𝑞 𝑡 +1
𝑘 [( 1 − 𝛿 𝑘 ) 𝐾 𝑡 +1
+ 𝑍 𝑡 +1
𝑖 ∅
𝑖 𝑖 𝑘 ,𝑡 +1
] + 𝑞 𝑡 +1
𝑃𝐷
[( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 +1
+ 𝑖 𝑝𝑑 ,𝑡 +1
] +
𝜕 𝑌 𝑡 +1
𝜕 𝐾 𝑡 +1
𝐾 𝑡 +1
+
𝜕 𝑌 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
𝑃𝐷
𝑡 +1
− 𝑞 𝑡 +1
𝑘 𝑍 𝑡 +1
𝑖 ∅
𝑖 𝑖 𝑘 ,𝑡 +1
− 𝑞 𝑡 +1
𝑃𝐷
𝑖 𝑝𝑑 ,𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
𝐾 𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
𝑃𝐷
𝑡 +1
= 𝑞 𝑡 +1
𝑘 [( 1 − 𝛿 𝑘 ) 𝐾 𝑡 +1
]+ 𝑞 𝑡 +1
𝑃𝐷
[( 1 − 𝛿 𝑃𝐷
) 𝑃𝐷
𝑡 +1
] +
𝜕 𝑌 𝑡 +1
𝜕 𝐾 𝑡 +1
𝐾 𝑡 +1
+
𝜕 𝑌 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
𝑃𝐷
𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
𝐾 𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
𝑃𝐷
𝑡 +1
= [( 1 − 𝛿 𝑘 ) 𝑞 𝑡 +1
𝑘 +
𝜕 𝑌 𝑡 +1
𝜕 𝐾 𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
] 𝐾 𝑡 +1
+ [𝑞 𝑡 +1
𝑃𝐷
( 1 − 𝛿 𝑃𝐷
)+
𝜕 𝑌 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝑃𝐷
𝑡 +1
]𝑃𝐷
𝑡 +1

Proof for proposition I:
I need to explore the comparative static of
𝜕 𝑅 𝑡 +1
𝑘 𝜕 𝑖 𝑝𝑑 ,𝑡 . The return on physical capital is defined as
( 1−𝛿 𝑘 ) 𝑞 𝑡 +1
𝑘 +
𝜕 𝑌 𝑡 +1
𝜕 𝐾 𝑡 +1
−
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
𝑞 𝑡 𝑘 , and there are three terms in the numerator and one term in the denominator.
The first term in the numerator contains variables in the t+1 period; therefore, changes in IT at time t
have a negligible effect on this term. The third term concerns marginal physical capital adjustment cost
and is not related to WO investment. The second term concerns the marginal productivity of physical
capital at t+1 and is positively related to WO investment at t:
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 𝜕 𝑌 𝑡 +1
𝜕 𝐾 𝑡 +1
=
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 𝛼 𝐾 𝑡 +1
𝛼 −1
𝑃𝐷
𝑡 +1
1−𝛼 =
𝜕 𝑃𝐷
𝑡 +1
𝜕 𝑖 𝑝𝑑 ,𝑡 𝜕 𝜕 𝑃𝐷
𝑡 +1
𝛼 𝐾 𝑡 +1
𝛼 −1
𝑃𝐷
𝑡 +1
1−𝛼 > 0
60

The denominator is the marginal cost of physical capital investment and is negatively related to WO
investment:
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 𝑞 𝑡 𝑘 =
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 1
𝑍 𝑡 𝑖 ∅
𝑖𝑡
(1 +
𝜕 𝐶 𝑡 𝜕 𝑖 𝑘 ,𝑡 )
=
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 1
𝑍 𝑡 𝑖 ∅
𝑖𝑡
~
𝜕 𝑃𝐷
𝑡 +1
𝜕 𝑖 𝑝𝑑 ,𝑡 𝜕 𝜕 𝑃𝐷
𝑡 +1
(
1
∅
𝑖𝑡
)~
𝜕 𝜕 𝑃𝐷
𝑡 +1
( 𝑃𝐷
𝑡 +1
𝜌 + 𝑖 𝑡 𝜌 )
𝜌 −1/𝜌 𝑖 𝑘 ,𝑡 1−𝜌 < 0
The numerator of the physical capital return is positively related to 𝑖 𝑝𝑑 ,𝑡 , and the denominator is
negatively related. Therefore, I conclude
𝜕 𝑅 𝑡 +1
𝑘 𝜕 𝑖 𝑝𝑑 ,𝑡 > 0. Intuitively, a positive WO investment improves the
marginal productivity of physical capital and simultaneously decreases the cost of additional physical
capital investment by making the physical accumulation more efficient.

Proof for proposition II:
I need to explore the comparative static of
𝜕 𝑅 𝑡 +1
𝐾 𝜕 𝑖 𝑘 ,𝑡 . Following the same procedure above, the marginal
productivity of physical capital at t+1 is negatively related to the physical capital investment:
𝜕 𝜕 𝑖 𝑘 ,𝑡 𝜕 𝑌 𝑡 +1
𝜕 𝐾 𝑡 +1
=
𝜕 𝐾 𝑡 +1
𝜕 𝑖 𝑘 ,𝑡 𝜕 𝜕 𝐾 𝑡 +1
𝛼 𝐾 𝑡 +1
𝛼 −1
𝑃𝐷
𝑡 +1
1−𝛼 < 0
The marginal adjustment cost of physical capital investment
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
is increasing w.r.t physical capital
investment:
𝜕 𝜕 𝑖 𝑘 ,𝑡 𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
=
𝜕 𝐾 𝑡 +1
𝜕 𝑖 𝑘 ,𝑡 𝜕 𝜕 𝐾 𝑡 +1
(−
1
2
𝑖 𝑘 ,𝑡 +1
2
𝐾 𝑡 +1
2
) > 0
The marginal cost of physical capital investment is positively related to the physical capital investment:
61

𝜕 𝜕 𝑖 𝑘 ,𝑡 𝑞 𝑡 𝑘 =
𝜕 𝜕 𝑖 𝑘 ,𝑡 1
𝑍 𝑡 𝑖 ∅
𝑖𝑡
(1+
𝜕 𝐶 𝑡 𝜕 𝑖 𝑘 ,𝑡 ) =
𝜕 𝜕 𝑖 𝑘 ,𝑡 1
𝑍 𝑡 𝑖 ∅
𝑖𝑡
( 1 +
𝑖 𝑘 ,𝑡 𝐾 𝑡 ) > 0
Collectively, the numerator of the physical capital return is decreasing w.r.t 𝑖 𝑘 ,𝑡 , and the denominator is
increasing. Therefore, I have shown that
𝜕 𝑅 𝑡 +1
𝑘 𝜕 𝑖 𝑘 ,𝑡 < 0. Intuitively, additional physical capital investment will
reduce the marginal productivity of physical capital due to diminishing returns and will increase the
marginal adjustment cost of physical capital and the marginal cost of physical capital investment.

Proof for proposition III:
The proof is similar to the proof for propositions I and II.
𝜕 𝑅 𝑡 +1
𝑃𝐷
𝜕 𝑖 𝑘 ,𝑡 > 0 and
𝜕 𝑅 𝑡 +1
𝑃𝐷
𝜕 𝑖 𝑝𝑑 ,𝑡 < 0. Physical capital
investment will increase the efficiency of WO investment’s marginal productivity and, at the same time,
lower the cost of investing in WO. On the other hand, additional investment in WO will decrease the
marginal productivity and raise the marginal investment cost.

Proof for proposition IV:
The proof is trivial. Recall that the asset return can be decomposed into the weighted average of
marginal returns on physical capital and WO investment: 𝑅 𝑡 +1
𝑒 =
𝐾 𝑡 +1
𝑞 𝑡 𝑘 𝑃 𝑡 𝑅 𝑡 +1
𝑘 +
𝑃𝐷
𝑡 +1
𝑞 𝑡 𝑃𝐷
𝑃 𝑡 𝑅 𝑡 +1
𝑃𝐷
. When
the market value of physical capital 𝐾 𝑡 +1
𝑞 𝑡 𝑘 is greater than the market value of WO investment
𝑃𝐷
𝑡 +1
𝑞 𝑡 𝑃𝐷
, the comparative static of the asset return will exhibit a pattern similar to that of marginal
return of physical investment. Combining with propositions I and II, I have shown proposition IV.

62

Proof for proposition V: I need to explore
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 𝛽 𝑡 𝐼𝑆𝑇 . Recall that the conditional beta 𝛽 𝑡 =
−
𝑐𝑜𝑣 𝑡 ( 𝑀 𝑡 ,𝑡 +1
,𝑟 𝑡 +1
𝑠 )
𝑣𝑎𝑟 𝑡 ( 𝑀 𝑡 ,𝑡 +1
)
, so the effect of 𝑖 𝑝𝑑 ,𝑡 on the conditional IST beta becomes the effect on the conditional
covariance between the aggregate IST shock part of the SDF and the asset return. Recall exogenous
𝑀 𝑡 ,𝑡 +1
= exp ( −𝑟 𝑓 )
exp ( −𝛾 𝑎 𝜎 𝑎 𝜖 𝑡 +1
𝑎 −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 )
𝐸 𝑡 [exp ( −𝛾 𝑎 𝜎 𝑎 𝜖 𝑡 +1
𝑎 −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) ]
and let ∆= exp ( −𝑟 𝑓 )
1
𝐸 𝑡 [exp ( −𝛾 𝑎 𝜎 𝑎 𝜖 𝑡 +1
𝑎 −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) ]
and 𝜎 𝑀 =
𝑣𝑎𝑟 𝑡 ( 𝑀 𝑡 ,𝑡 +1
) . Because ∆ and 𝜎 𝑀 are positive constants and do not affect the direction of comparative
statics, the task of exploring
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 𝛽 𝑡 𝐼𝑆𝑇 becomes that of exploring
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 [−𝑐𝑜𝑣 𝑡 ( exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) , 𝑟 𝑡 +1
𝑠 ) ].
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 𝛽 𝑡 𝐼𝑆𝑇 ~
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 [−𝑐𝑜𝑣 𝑡 ( exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) ,𝑟 𝑡 +1
𝑠 ) ]~
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 [−𝑐𝑜𝑣 𝑡 ( exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) , 𝑅 𝑡 +1
𝐾 )
− 𝑐𝑜𝑣 𝑡 ( exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) ,𝑅 𝑡 +1
𝑃𝐷
) ]
The aggregate IST shock is only imbedded in the physical capital adjustment cost (see the adjustment
cost expression); therefore, 𝑐𝑜𝑣 𝑡 ( exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) , 𝑅 𝑡 +1
𝑃𝐷
) = 0, and I have:
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 𝛽 𝑡 𝐼𝑆𝑇 ~
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 [−𝑐𝑜𝑣 𝑡 ( exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) , 𝑅 𝑡 +1
𝐾 ) ]
The marginal return on physical capital has three components in the numerator and one component in
the denominator, and of those four components, the aggregate IST shock only affects two: the cost of
physical capital investment and the marginal adjustment cost:
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 𝛽 𝑡 𝐼𝑆𝑇 ~
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 [−𝑐𝑜𝑣 𝑡 ( exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) , 𝑞 𝑡 +1
𝑘 )
+ 𝑐𝑜𝑣 𝑡 (exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ),
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
)] ~
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 [−𝑐𝑜𝑣 𝑡 ( exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) ,
𝜕 𝐶 𝑡 +1
𝜕 𝑖 𝑡 +1
)]
+
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 [𝑐𝑜𝑣 𝑡 ( exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) ,
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
) ]
63

Therefore, the investment in WO can affect the covariance through two channels: the physical capital
investment cost channel (first term with
𝜕 𝐶 𝑡 +1
𝜕 𝑖 𝑘 ,𝑡 +1
) and the marginal adjustment cost channel (the second
term with
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
) . Intuitively, 𝑖 𝑝𝑑 ,𝑡 directly affects the physical capital accumulation 𝐾 𝑡 +1
, which in turn
affects both the marginal cost of physical capital investment 𝑞 𝑡 +1
𝑘 through
𝜕 𝐶 𝑡 +1
𝜕 𝑖 𝑘 ,𝑡 +1
and the marginal
adjustment cost of physical capital
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
.
Physical capital investment cost channel:
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 [−𝑐𝑜𝑣 𝑡 (exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ),
𝜕 𝐶 𝑡 +1
𝜕 𝑖 𝑘 ,𝑡 +1
)]~
𝜕 𝐾 𝑡 +1
𝜕 𝑖 𝑝𝑑 ,𝑡 𝜕 𝜕 𝐾 𝑡 +1
[−𝑐𝑜𝑣 𝑡 (exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) ,
𝜕 𝐶 𝑡 +1
𝜕 𝑖 𝑘 ,𝑡 +1
)]
𝜕 𝐶 𝑡 +1
𝜕 𝑖 𝑘 ,𝑡 +1
=
∆
𝐾 𝑍 𝑡 +1
𝑎 𝑖 𝑘 ,𝑡 +1
𝐾 𝑡 +1

The WO investment effect’s direction through this channel is indeterminate because it depends on
whether we are investing 𝑖 𝑘 ,𝑡 +1
> 0 or disinvesting 𝑖 𝑘 ,𝑡 +1
< 0. When 𝑖 𝑘 ,𝑡 +1
> 0the
𝜕 𝐶 𝑡 +1
𝜕 𝑖 𝑘 ,𝑡 +1
is positive and
the direction of the WO investment’s effect through this channel is positive. . Because we know that 𝑖 𝑝𝑑

and 𝑖 𝑘 are complements, a positive 𝑖 𝑝𝑑 ,𝑡 is associated with a positive 𝑖 𝑘 ,𝑡 +1
.
Therefore, the direction of the WO investment’s effect through this channel is positive due to
complementarity assumption.
Marginal adjustment cost channel:
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 [𝑐𝑜𝑣 𝑡 ( exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) ,
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
)] ~
𝜕 𝐾 𝑡 +1
𝜕 𝑖 𝑝𝑑 ,𝑡 𝜕 𝜕 𝐾 𝑡 +1
[𝑐𝑜𝑣 𝑡 ( exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ),
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
)] > 0
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
= −
∆
𝐾 2𝑍 𝑡 +1
𝑎 (
𝑖 𝑘 ,𝑡 +1
𝐾 𝑡 +1
)
2

64

Therefore, the direction of the WO investment’s effect through this channel is strictly positive.
In conclusion, an additional 𝑖 𝑝𝑑 ,𝑡 will increase the conditional IST beta.
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 𝛽 𝑡 𝐼𝑆𝑇 ~
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 [−𝑐𝑜𝑣 𝑡 ( exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) ,𝑟 𝑡 +1
𝑠 ) ]~
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 [−𝑐𝑜𝑣 𝑡 ( exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) ,
𝜕 𝐶 𝑡 +1
𝜕 𝑖 𝑘 ,𝑡 +1
)] +
𝜕 𝜕 𝑖 𝑝𝑑 ,𝑡 [−𝑐𝑜𝑣 𝑡 (exp ( −𝛾 𝑧 𝜎 𝑧 𝜖 𝑡 +1
𝑧 ) ,
𝜕 𝐶 𝑡 +1
𝜕 𝐾 𝑡 +1
)] >0

7.3 Exploitation vs. exploration (external validation) In this section, I provide additional evidence
supporting PDC as a valid proxy for flexible WO investment. The management literature also identified
the association between flexible WO and explorative innovation (exploration), while firms with rigid
structures are more likely to innovate exploitatively (exploitation); He and Wong 2004, Katila and Ahuja
2002.
To test out the validity of PDC as a proxy of WO investment, I calculate firms’ degree of exploration and
exploitation using patent citation data and regress them onto my PDC measurement. The rationale is
that if higher PDC investment represents a more flexible WO, I should see a positive relationship
between PDC and exploration. Following Katila and Ahuja 2002, I construct a variable scope, which
represents exploration, and depth, which represents exploitation. The variables 𝑆 𝑐 𝑜𝑝𝑒 𝑖 ,𝑡 and 𝐷𝑒𝑝𝑡 ℎ
𝑖 ,𝑡
are defined as follows:
𝑆𝑐𝑜𝑝𝑒 𝑖 ,𝑡 −1
=
𝑛𝑒𝑤 𝑐𝑖𝑡𝑎𝑡𝑖𝑜𝑛 𝑖 ,𝑡 −1
𝑡𝑜𝑡𝑎𝑙 𝑐𝑖𝑡𝑎𝑡𝑖𝑜𝑛 𝑖 ,𝑡 −1

𝐷𝑒𝑝𝑡 ℎ
𝑖 ,𝑡 −1
=
∑ 𝑟𝑒𝑝𝑒𝑡𝑖𝑡𝑖𝑜𝑛 𝑐𝑜𝑢𝑛𝑡 𝑖 ,𝑦 𝑡 −2
𝑦 =𝑡 −6
𝑡𝑜𝑡𝑎𝑙 𝑐𝑖𝑡𝑎 𝑡 𝑖𝑜𝑛 𝑖 ,𝑡 −1

The variable scope is the proportion of the citations that were never cited before in year t-1 and
corresponds to the theoretical notion of the exploration of new knowledge. A high scope ratio indicates
that the firm often explores unrelated paths, which requires a high degree of flexibility and
65

interdisciplinary communication. The variable depth is the average number of times a firm repeatedly
used the citations in the patents it applied for over the past 5 years and describes the accumulation of
search experience by exploiting existing knowledge elements. Depth measures how often the firm
exploits its existing knowledge, which is often associated with routinized rigid methodology.
The regression I run is presented here:
𝑆𝑐𝑜𝑝𝑒 𝑖 ,𝑡 = 𝑏 0
+ 𝑏 1
𝑅𝑅𝐶 𝑖 ,𝑡 + 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝜖 𝑖 ,𝑡
𝐷𝑒𝑝𝑡 ℎ
𝑖 ,𝑡 = 𝑏 0
+ 𝑏 1
𝑅𝑅𝐶
𝑖 ,𝑡 + 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝜖 𝑖 ,𝑡

The regression results are presented in Table 12.
[Insert Table 12 here]
Panel A in Table 12 reports a statistically significant positive correlation between Scope and PDC, while
the correlation between Depth and PDC is positive but not significant. This result shows that firms that
are switching to a higher portion of laptops in their computer composition are exploration types.
Because exploration is associated with flexible WO, this result validates PDC as a proxy for a firm’s WO
investment. I then investigate what drives the PDC result found in Panel A. In Panels B and C, I find
evidence that the variation in laptop growth is driving the scope coefficients. In Panel B, the coefficient
between laptop growth alone and scope is positive and has a t-statistic equal to 2.26. In Panel C, the
laptop growth coefficient is positive and significant, while the desktop coefficient is negative and
insignificant

Abstract (if available)

Abstract Flexible workplace organization (WO) increases the efficiency of a firm’s future capital investment. Hence, investment in WO positively predicts future capital investment and exposure to future aggregate investment-specific technology (IST)shocks. As a result, firms that invest in WO become riskier and systematically experience higher equity returns. This paper documents a robust positive correlation between firms’ WO investments and future equity returns. Firms with higher industry-adjusted WO investments have average returns that are 3.7% higher than those of firms with lower WO investments. In addition, firms with high WO investments experience a high conditional IST beta in the future. I tie all my empirical findings together in aproduction-based equilibrium model with closed-form solutions.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Three essays on behavioral finance with rational foundations

PDF

Picky customers and expected returns

PDF

Mutual fund screening versus weighting

PDF

Asset prices and trading in complete market economies with heterogeneous agents

PDF

Essays in empirical asset pricing

PDF

Three essays in derivatives, trading and liquidity

PDF

Essays on information and financial economics

PDF

Executive compensation: the trend toward one size fits all

PDF

Behavioral approaches to industrial organization

PDF

Voluntary disclosure responses to mandated disclosure: evidence from Australian corporate tax transparency

PDF

Innovation: financial and economics considerations

PDF

Essays on firm investment, innovation and productivity

PDF

Accrual quality and expected returns: the importance of controlling for cash flow shocks

PDF

Essays in asset pricing

PDF

Essays in behavioral and financial economics

PDF

Incentives and relative-wealth concerns: theory and evidence

PDF

A general equilibrium model for exchange rates and asset prices in an economy subject to jump-diffusion uncertainty

PDF

Evolution of returns to scale and investor flows during the life cycle of active asset management

PDF

Real asset liquidity and asset impairments

PDF

Essays in financial intermediation

Asset Metadata

Creator Wan, Yuan (author)

Core Title Workplace organization and asset pricing

School Marshall School of Business

Degree Doctor of Philosophy

Degree Program Business Administration

Publication Date 07/30/2019

Defense Date 03/19/2019

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag asset pricing,macro economics,OAI-PMH Harvest,production based asset pricing

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Tuzel, Selale (committee chair), David, Joel (committee member), Jones, Christopher (committee member), Zapatero, Fernando (committee member)

Creator Email prcharming@live.com,yuanwan@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-200069

Unique identifier UC11663586

Identifier etd-WanYuan-7687.pdf (filename),usctheses-c89-200069 (legacy record id)

Legacy Identifier etd-WanYuan-7687.pdf

Dmrecord 200069

Document Type Dissertation

Format application/pdf (imt)

Rights Wan, Yuan

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA