University of Southern California Dissertations and Theses

Private disclosure contracts within supply chains and managerial learning from stock prices

(USC Thesis Other)

Private disclosure contracts within supply chains and managerial learning from stock prices

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content PRIVATE DISCLOSURE CONTRACTS WITHIN SUPPLY CHAINS
AND MANAGERIAL LEARNING FROM STOCK PRICES
by
Vivek Pandey
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(BUSINESS ADMINISTRATION)
May 2022
Copyright 2022 Vivek Pandey
Dedication
To Mansi, Aashray, and my parents. This would have been impossible without you.
ii
Acknowledgements
I sincerely thank my dissertation chair, K.R. Subramanyam, for his invaluable guidance, support,
and suggestions. For their many useful comments and suggestions, I am grateful to my disserta-
tion committee members: Clive Lennox, Regina Wittenberg-Moerman, and Rodney Ramcharan.
I also thank Randy Beatty, AJ Chen, Jonathan Craske, Patty Dechow, Mark DeFond, Ryan Erhard,
Jesse Gardner, Shane Heitzman, Jungkoo Kang, Shelley Li, Carmen Mann, Maria Ogneva, Fangfei
Shu, Matt Shaffer, Richard Sloan, David Tsui, Holly Yang, Jun Yoon, and workshop participants
at Boston College, Duke University, Harvard Business School, University of Michigan, Univer-
sity of Rochester, University of Southern California, and University of Texas at Dallas. I thank
my PhD colleagues and the Leventhal staff for constant support and encouragement. I grate-
fully acknowledge generous financial support from the University of Southern California and the
Deloitte Foundation.
iii
TableofContents
Dedication ii
Acknowledgements iii
ListofTables vi
ListofFigures vii
Abstract viii
Chapter1: Introduction 1
Chapter2: BackgroundandHypothesisDevelopment 9
2.1 Demand forecast disclosure contracts . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Hypothesis development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Chapter3: Data 17
3.1 Data sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Controlling for determinants of DF contracts . . . . . . . . . . . . . . . . . . . . . 18
3.3 Sample construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.4 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Chapter4: Results 27
4.1 DF contracts and investment-price sensitivity . . . . . . . . . . . . . . . . . . . . 27
4.1.1 Measuring managerial learning from stock prices . . . . . . . . . . . . . . 27
4.1.2 Primary results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.2 Parallel trends assumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.3 Robustness tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.3.1 Pseudo-experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.3.2 Coefficient stability and correlated omitted variables . . . . . . . . . . . . 38
4.3.3 Alternative design choices . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.3.4 Investors’ private information in stock prices . . . . . . . . . . . . . . . . 43
4.4 Credibility of demand forecast disclosures:
Binding forecasts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
iv
Chapter5: AdditionalAnalysis 48
5.1 Demand volatility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.2 Customer accounting quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.3 Informativeness of stock prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.4 Product differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.5 Lower learning from prices or less efficient investments . . . . . . . . . . . . . . . 57
5.5.1 Sensitivity to non-price measures of investment opportunities . . . . . . . 58
5.5.2 Future firm performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Chapter6: Conclusions 62
References 64
AppendixA 69
Excerpts from Contracts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
AppendixB 70
Variable Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
v
ListofTables
3.1 Determinants of Demand Forecast (DF) Contracts . . . . . . . . . . . . . . . . . . 20
3.2 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.1 DF Contracts and Investment-Q Sensitivity (Continued) . . . . . . . . . . . . . . 31
4.2 Parallel Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.3 Robustness Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.5 Credibility of Demand Forecast Disclosures: Binding Forecasts . . . . . . . . . . 47
5.1 Demand Volatility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.2 Customer Accounting Quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.3 Price Informativeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.4 Differentiated Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.5 Sensitivity of Investment to Non-Price Measures of Investment Opportunities . . 59
5.6 Future Firm Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
vi
ListofFigures
4.1 Dynamic DiD coefficients in event time. . . . . . . . . . . . . . . . . . . . . . . . . 35
4.2 Pseudo ‘Diff-in-Diff’ coefficient estimates . . . . . . . . . . . . . . . . . . . . . . . 38
vii
Abstract
I examine whether private disclosure contracts that obligate the customer to periodically disclose
forecasts of the customer’s future demand for the supplier’s products (“DF contracts”) influence
how the supplier makes investment decisions. I expect that if suppliers find private disclosures
under DF contracts useful in making corporate investment decisions, they will reduce their de-
pendence on other forward-looking sources of investment-relevant information. Motivated by
the extensive literature that shows managers learn about investments from their own stock prices,
I hypothesize that if suppliers use demand forecasts disclosed under DF contracts when making
corporate investment decisions, they will reduce their reliance on stock prices after entering into
such contracts. Using novel hand-collected data on DF contracts, I find that suppliers’ corporate
investments become significantly less sensitive to stock prices after suppliers enter into a DF
contract for the first time, compared to that of control firms over the same time period. I sub-
stantiate my inference more directly by showing that the result is more pronounced for contracts
with partially binding disclosures of demand forecasts. In additional cross-sectional analysis, I
find that the primary result is largely non-existent for suppliers with less volatile demand, higher
customer accounting quality, more informative stock prices, and highly differentiated products.
Inconsistent with decrease in investment-price sensitivity indicating more inefficient investments
viii
post-DF, sensitivity of investment to non-price measures of investment opportunities do not de-
crease and future firm performance improves. Overall, my results suggest that suppliers use
demand forecast disclosures when making corporate investment decisions and thereby reduce
their reliance on learning about investments from stock prices.
ix
Chapter1
Introduction
Over the past three decades, a large literature has studied voluntary public disclosures made
by managers to the capital markets (see, e.g., Healy and Palepu, 2001; Beyer, Cohen, Lys, and
Walther, 2010 for a review of this literature). In contrast, only recently have a few studies exam-
ined contractual mechanisms for privately disclosing information. For example, while Carrizosa
and Ryan (2017) examine contractual private disclosures between firms and lenders, Nagar and
Schoenfeld (2021) and Bushee, Kim-Gina, and Leung (2020) examine such contractual disclosures
between firms and shareholders, and suppliers and customers, respectively. In this paper, I exam-
ine the real effects of private disclosure contracts between suppliers and customers. Specifically,
I examine how contracts that obligate customers to provide suppliers with private forecasts of
their future demand for the supplier’s products (“DF contracts” henceforth) influence the manner
in which suppliers make corporate investment decisions.
While customers often voluntarily share future demand information with their suppliers, such
private disclosures often suffer from credibility and usefulness issues (e.g., Cohen, Ho, Ren, and
Terwiesch, 2003). In particular, customers are motivated to make strategic disclosures to their
suppliers, with incentives to disclose both overoptimistic forecasts (e.g., Armony and Plambeck,
1
2005; Oh and Özer, 2013) and overpessimistic forecasts (e.g., Mussa and Rosen, 1978; Chu, Shamir,
and Shin, 2017).
∗
In addition to strategic disclosure incentives, customers, much like all other
agents in the economy, find it difficult to accurately and precisely forecast future demand. This
severely limits the usefulness and credibility of such forward looking private disclosures for sup-
pliers. DF contracts differ from voluntary private disclosures as the former is intended to improve
on the latter by explicitly enhancing the usefulness and credibility of the disclosures themselves.
At the outset, DF contracts implement a highly structured private disclosure policy, wherein the
customer periodically discloses to the supplier its forecasts of future demand for the supplier’s
products. Such demand forecasts are typically rolling forecasts, with a prespecified frequency
and horizon, and can be quite granular, broken down into individual months and products.
Importantly, DF contracts also address credibility issues in private disclosures of demand fore-
casts by making inaccurate demand forecasts costly for customers. For example, DF contracts
may include partially binding forecasts that make it mandatory for the customer to buy at least
a prespecified portion of the disclosed forecast. This ensures that customers also bear the cost of
disclosing overoptimistic forecasts. Other ways in which DF contracts enhance credibility are by
imposing explicit restrictions on customers’ ability to revise previously disclosed forecasts, and
by including performance targets for the accuracy of customers’ demand forecast disclosures.
To examine the real effects of DF contracts, I focus on corporate investments because they
are one of the primary ways in which firms create value (e.g., Roychowdhury, Shroff, and Verdi,
2019). Because future demand is one of the primary drivers of corporate investments, suppli-
ers may use demand forecasts disclosed under DF contracts when making corporate investment
∗
Customer’s overoptimistic disclosure incentives stem from its need to avoid costly supply shortages. Cus-
tomer’s overpessimistic disclosure incentives stem from its efforts to prevent suppliers from raising prices.
2
decisions. However, theory on investment under uncertainty suggests that suppliers may ignore
the preliminary information in demand forecasts and rather wait for uncertainty about payoffs to
investment to resolve (e.g., Dixit and Pindyck, 1994). While I can observe when a supplier enters a
DF contract with a customer, I do not observe actual demand forecasts privately disclosed by the
customer under the DF contract. This limitation precludes me from directly examining whether
suppliers use the information in demand forecasts disclosed through DF contracts when making
investment decisions. Accordingly, I adopt an indirect approach that relies on the notion that if
suppliers use demand forecasts disclosed under DF contracts when making investment decisions,
they may reduce their reliance on other investment-relevant information sources. Motivated by
the substantive literature in accounting and finance that shows that managers use information
from their own stock price when making investment decisions (e.g., Dye and Sridhar, 2002; Luo,
2005; Chen, Goldstein, and Jiang, 2007; Jayaraman and Wu, 2020; Goldstein, Liu, and Yang, 2021),
I choose supplier’s stock price as the alternative information source to examine whether DF con-
tracts affect the way suppliers make investment decisions. I hypothesize that after entering into
a DF contract for the first time, suppliers will reduce their dependence on stock prices when
making investment decisions.
To test my hypothesis, I hand-collect a novel data set on DF contracts included in 10-K, 10-Q,
and 8-K filings (for details, see Section 3.1). Following prior literature, I measure managerial learn-
ing from stock price through the investment-price sensitivity (i.e., the coefficient when future
corporate investment is regressed on the current Tobin’s Q). I then incorporate the investment-
price sensitivity in a difference-in-differences (DiD) research design. My DiD model estimates
the change in suppliers’ investment-price sensitivity after they enter into a DF contract for the
first time (“DF firms"), compared to the change in investment-price sensitivity of control firms
3
that never enter into a DF contract (“non-DF firms"), over the same time period. To alleviate con-
cerns that common factors simultaneously affect suppliers’ decision to enter into a DF contract
and subsequent learning from stock prices, I construct the primary sample by matching DF firms
with non-DF firms on the probability of entering into a DF contract, based on a determinants
model of DF contracts. The primary sample consists of 13,369 observations between 1996 and
2017.
Consistent with my hypothesis, I find that the average investment-price sensitivity of DF
firms significantly decreases after firms enter into a DF contract for the first time, compared to
that of non-DF firms over the same time period. In addition to firm and year fixed effects, these
results are robust to the inclusion of controls, interactions of price (Tobin’s Q) with controls, and
interactions of price with industry and year fixed effects. Economically, the investment-price
sensitivity of DF firms in the post-DF period is around 52% lower than that in the pre-DF period.
These results are robust to several sensitivity checks, including parallel trends and alternative
design choices. These findings suggest that suppliers use the demand forecasts disclosed in DF
contracts to make investment decisions and, consequently, reduce their reliance on stock prices
when making investment decisions.
To more directly substantiate my primary result, I examine whether the result is stronger
when demand forecasts are partially binding, that is when the DF contract mandates the customer
to purchase at least a portion of the disclosed forecast. By legally requiring that the customer buy
a portion of the disclosed forecast, binding forecasts make inaccurate disclosures costly for the
customer, thereby increasing credibility of forecast disclosures. I predict that the higher credibil-
ity of binding demand forecasts should result in greater substitution between learning from DF
contracts and learning from stock prices. Consistent with my prediction, I find that the reduction
4
in investment-price sensitivity post-DF is stronger for DF contracts with partially binding fore-
casts compared to that for DF contracts without binding forecasts. Economically, the reduction
in investment-price sensitivity post-DF for partially binding forecasts is around 1.5 times to 2.2
times as large as that for nonbinding forecasts.
To shed more light on which suppliers use (or do not use) demand forecast disclosures in
investment decisions, I perform several cross-sectional tests. Because the disclosure is about
demand, I begin with examining whether DF contracts have a greater effect on investment-price
sensitivity for firms that face more volatile demand. Demand volatility can make predicting future
demand difficult for both investors and customers. I expect my primary results to be stronger
for suppliers with high demand volatility if suppliers rely more on customer demand forecasts
compared to stock prices to guide investment decisions when facing volatile demand. This can be
the case if suppliers find demand forecasts more useful when there is more uncertainty to resolve.
I expect the opposite if suppliers believe that markets may be more informative than customers
during volatile demand conditions. Using the volatility of sales and the volatility of earnings
as proxies for demand volatility, I find that my primary results are stronger (insignificant) for
suppliers with high (low) demand volatility. This is consistent with the notion that for investment
decisions, demand forecasts are more (less) useful for suppliers with more (less) volatile demand.
Motivated by the prior literature that argues that accounting information plays an impor-
tant role in supply chains (e.g., Bowen, DuCharme, and Shores, 1995; Raman and Shahrur, 2008;
Pandey and Subramanyam, 2020), next, I explore whether customer accounting quality (AQ) mod-
erates my primary results. Suppliers may find other sources of information, including stock prices
and demand forecasts, incrementally less useful if they have access to high-quality customer ac-
counting information. These suppliers are less likely to change their learning behavior after the
5
onset of DF contracts. Using modified Dechow and Dichev (2002) accruals quality and earnings
persistence as measures of accounting quality, I find that the change in investment-price sensitiv-
ity post-DF is significantly weaker for suppliers with high customer AQ. This finding is consistent
with high customer AQ reducing the incremental usefulness of alternative information sources
and reinforces the important role played by accounting information in supply chains.
Next, I explore whether managers are less likely to reduce learning from more informative
stock prices. While demand forecasts provide supplier with signals of future demand from the
customer, stock prices may reveal various other signals relevant for investment decisions - say,
total demand, industry trends, competition, input costs, or risk. If more informative stock prices
also contain more information not found in demand forecast disclosures, the onset of DF contract
may not change supplier’s learning from stock prices significantly. Using probability of informed
trading and price non-synchronicity as measures of informativeness of stock prices for managers
(e.g., Chen et al., 2007; Jayaraman and Wu, 2019), I find evidence consistent with this notion − the change in investment-price sensitivity post-DF is largely insignificant for suppliers with more
informative stock prices.
Finally, I explore whether suppliers are less likely to use demand forecast disclosures when
making investment decisions if they sell highly differentiated products. Basing irreversible invest-
ment decisions on preliminary information can be costly, especially for firms that make differen-
tiated products that need specific investments with few alternative uses (e.g., Dixit and Pindyck,
1994). If that is the case, suppliers with highly differentiated products may not base their invest-
ment decisions as much on demand forecast disclosures. Consistent with this notion, using a
measure of product differentiation based on Hoberg and Phillips (2016), I find that the change in
6
investment-price sensitivity post-DF is insignificant for suppliers with more differentiated prod-
ucts. This suggests that the use of demand forecast disclosures when making investment decisions
could be influenced by the type of products a supplier sells.
My study contributes to three distinct streams of research. First, I contribute to the nascent
but growing literature examining contracted private disclosures. While much of the large litera-
ture on voluntary disclosure focuses on public voluntary disclosures made by firms to the capital
markets, recently researchers have started examining private disclosure contracts between var-
ious stakeholders, such as between firms and lenders (e.g., Carrizosa and Ryan, 2017), between
firms and shareholders (e.g., Nagar and Schoenfeld, 2021), and between suppliers and customers
(Bushee et al., 2020). In particular, Bushee et al. (2020) examine how the presence of private con-
tracted disclosures by customers to suppliers (DF contracts) affects the nature of voluntary public
disclosures made by such suppliers. I contribute to this literature by examining the real effects
of such private disclosure contracts and showing that DF contracts affect the manner in which
suppliers make investment decisions.
Next, I contribute to the literature on managerial learning from stock prices. A large liter-
ature has shown that stock prices can inform managers’ strategic decisions, such as decisions
about capital investments, R&D, and acquisitions (e.g., Luo, 2005; Chen et al., 2007; Bakke and
Whited, 2010; Foucault and Frésard, 2012; Jayaraman and Wu, 2020; Goldstein et al., 2021). The
intuition behind this hypothesis is that stock prices aggregate private information from a myriad
of traders and thus may reflect information that managers do not possess (Dye and Sridhar, 2002;
Bond, Edmans, and Goldstein, 2012; Gao and Liang, 2013). Existing studies have more broadly
focused on the notion that managers learn from stock prices, with less empirical evidence on the
specific types of information managers glean from stock prices that help them make investment
7
decisions. I contribute to this literature by highlighting one important type of information that
managers likely learn from stock prices: future demand for their products and services. While
theory suggests that managers can learn about future demand from stock prices (e.g., Subrah-
manyam and Titman, 1999; Edmans, Goldstein, and Jiang, 2015), there is hitherto no direct em-
pirical evidence on this issue. I show that managerial learning from stock prices is reduced in
the presence of a rich alternative information source about future demand, which suggests that
information about future demand indeed is an important type of managerial learning from stock
prices.
Finally, I contribute to the large literature that studies private information flows within the
supply chain. Several studies have examined information sharing within supply chains, mainly
employing theoretical models (e.g., Lee, Padmanabhan, and Whang, 1997; Özer and Wei, 2006; Oh
and Özer, 2013; Shamir and Shin, 2016). While this literature has primarily focused on voluntary
sharing of demand information, the effects of contracted demand forecast disclosures have not
been examined in a systematic manner.
†
I contribute to this literature by hand-collecting a large,
novel sample of DF contracts and using it to systematically examine their real effects.
†
Most of the studies in this domain employ theoretical models, but a few studies have empirically examined
private demand forecasts using proprietary data obtained from one or a few companies (e.g., Cattani and Hausman,
2000; Cohen et al., 2003; Terwiesch, Ren, Ho, and Cohen, 2005).
8
Chapter2
BackgroundandHypothesisDevelopment
2.1 Demandforecastdisclosurecontracts
Private voluntary disclosures are a common way to exchange information in supply chains (see,
e.g., Ha and Tang, 2017 for a review). Information about future demand is of particular interest for
suppliers, who use it to make real decisions.
∗
However, private disclosures of demand suffer from
credibility issues due to customer’s strategic disclosure incentives, severely limit the usefulness
of the disclosure to the supplier (Özer and Wei, 2006; Özer, Zheng, and Chen, 2011). An impor-
tant asymmetric incentive at the core of the credibility issue: the cost of supply shortage for the
customer is much larger than the cost of sharing overoptimistic information with the supplier.
†
Such an asymmetric incentive inspire the customer to often engage in making strategically op-
timistic private disclosures (e.g., Armony and Plambeck, 2005). Customer’s also have incentives
to be strategically pessimistic in private disclosures of demand because they don’t want to give
∗
For example, key questions in the CFO Survey conducted annually by Duke University and the Federal Reserve
Banks of Richmond and Atlanta ask about CFOs’ expectations of future revenue (https://www.richmondfed.
org/research/national_economy/cfo_survey/data_and_results).
†
A case in point is the pervasive discussion in the media on how supply shortages are forcing many companies
to cut output and halt production (e.g., McLean, 2021; Ludwikowski and Sjoberg, 2021).
9
suppliers additional bargaining power to raise prices (e.g., Mussa and Rosen, 1978; Chu et al.,
2017). Some studies have gone so far as to argue that customers’ strategic disclosure incentives
are so high that no voluntary private disclosures can be made credibly (Oh and Özer, 2013; Chu
et al., 2017). Even without strategic incentives, predicting future demand is often very difficult
for any agent in the economy. Using proprietary data on demand forecasts made by a buyer to a
supplier in the semiconductor equipment supply chain, Cohen et al. (2003) conclude that fearing
order cancellations, suppliers largely ignore demand forecasts.
Prior work has shown that contractual mechanisms are one way to address incentive prob-
lems in supply chains (see, e.g., Baiman and Rajan, 2002; Costello, 2013).
‡
Demand forecast dis-
closure contracts (“DF contracts") are an important way in which firms enforce the usefulness
and credibility of private disclosures of future demand from customers. A DF contract obligates
the customer to privately and periodically disclose to the supplier, the forecasts of the customer’s
future demand for the supplier’s products. Demand forecast disclosures under DF contracts are
typically rolling-window forecasts, with a prespecified frequency and horizon, say, a 12-month
rolling window forecast updated monthly. These forecasts can be very granular, such as breaking
down each forecast into individual months and product codes. Essentially, DF contracts enforce a
highly structured private disclosure policy with a level of granularity, periodicity, and timeliness
in the private disclosures between suppliers and customers that may not be possible through ad
hoc private disclosures. This can significantly raise the usefulness of private disclosures about
future demand for the supplier.
‡
Costello (2013) shows that supply agreements may include accounting and nonaccounting covenants that limit
suppliers’ and customers’ opportunism and address information asymmetry.
10
Importantly, DF contracts help overcome the aforementioned credibility issues in private dis-
closures. First, DF contracts can include binding forecasts wherein it is mandatory for customer
to buy at least a prespecified portion of the demand forecast. Such risk-sharing provisions ensure
that the customer also has to bear the cost of disclosing overoptimistic forecasts to supplier (e.g.,
Crawford and Sobel, 1982; Kartik, 2009). Second, DF contracts can impose limits on how much
customer is allowed to revise previously disclosed forecasts, essentially making it more costly
for customer to disclose inaccurate forecasts. For instance, under restricted forecast revisions,
overoptimistic forecasts can result in customers buying more than needed, and overpessimistic
forecasts can lead to customers facing supply shortages. Finally, such contracts can include per-
formance targets for the accuracy of customers’ demand forecasts. Explicitly measuring forecast
accuracy can make forecast errors more tractable and salient, thereby inspiring customers’ com-
mitment to reduce biased or erroneous forecasts.
§
Collectively, these contractual features make
it costly for customers to disclosure inaccurate forecasts, raising customers’ commitment to dis-
closing more accurate forecasts, and in turn making demand forecasts more credible and useful
for the supplier. The contractual mechanisms inherent in DF contracts that increase their useful-
ness and credibility also distinguish them from noncontractualadhoc private disclosures between
customers and suppliers.
While DF contracts can help suppliers learn about future demand, not every firm enters in
a DF contract, which suggests that DF contracts also entail costs. First, customers may have
to bear setup and ongoing costs toward systems that can generate the granular, periodic, and
forward-looking information required under DF contracts. Second, prior literature finds that
§
Based on my own anecdotal reading of DF contracts, binding forecasts and restrictions on forecast revisions
appear to be much more common than explicit performance targets for customers’ forecast accuracy.
11
proprietary information spillover concerns can shape firm behavior (e.g., Verrecchia and Weber,
2006; Glaeser, 2018; Kang, Lennox, and Pandey, 2022). Contractual commitment to periodic and
detailed forward-looking disclosures about future demand can heighten customers’ (perceived)
risk of leakage of proprietary information through the supplier.
¶
Finally, because customers can-
not perfectly know their future purchase requirements, forecast errors are unavoidable. Con-
tractual provisions, such as binding forecasts in DF contracts, can impose costs on customer for
disclosing information that turns out to be inaccurate in the future, even when customer is not
being strategic.
Extant theoretical models, anecdotes, or small-sample studies that examine one or a few com-
panies (e.g., Cattani and Hausman, 2000; Cohen et al., 2003; Terwiesch et al., 2005) inform our
understanding of DF contracts. One exception, however, is Bushee et al. (2020), who, in a recent
working paper, examine how customer’s contractual private disclosures of future demand affect
supplier’s public disclosures. Specifically, they find that management guidance bias spillover from
customer to supplier is attenuated in the presence of DF contracts, suggesting that managers find
DF contracts useful. However, to the best of my knowledge, I am the first to empirically and
systematically examine the real effects of DF contracts.
¶
As the prior literature has noted, the competitive harm in the case of leakage can be so high that even the
perceived risk of proprietary information leakage can shape a firm’s decisions. For example, firms avoid sharing
their auditors or investment banks with competitors if they have concerns about proprietary information spillovers
(Asker and Ljungqvist, 2010; Kang et al., 2022).
12
2.2 Hypothesisdevelopment
To shed light on the real effects of DF contracts, I examine how DF contracts affect suppliers’
corporate investments. My focus on corporate investments is motivated by the notion that in-
vestments, which are often irreversible, are one of the primary means through which firms create
value (e.g., Roychowdhury et al., 2019). Because future demand is one of the main drivers of cor-
porate investments, DF contracts can help suppliers in making corporate investment decisions
by helping them learn about future demand.
A data limitation hinders my empirical examination of this issue. Specifically, I can observe
when a supplier enters a DF contract with a customer, but not the actual demand forecasts dis-
closed by the customer under the DF contract. For this reason, I cannot directly examine whether
suppliers use the demand forecasts disclosed by the customer through DF contracts when making
investment decisions. Accordingly, I adopt an indirect approach that relies on the notion that if
managers find private disclosures in DF contracts useful in making investment decisions, they
will reduce their reliance on other investment-relevant information sources after entering into
such contracts.
I focus on stock price as the alternative (observable) investment-relevant information source
for managers. The choice of stock price as the alternative information source is motivated by a
substantive literature in accounting and finance that shows that managers can gain valuable in-
formation from stock prices when making investment decisions. The intuition behind this claim
in the literature is that investors have private information that managers may not have, and by
trading on this information investors impound their private information into stock prices (e.g.,
Dye and Sridhar, 2002; Chen et al., 2007; Gao and Liang, 2013; Jayaraman and Wu, 2019). For
13
instance, investors may be better informed about how external factors, as opposed to firms’ in-
ternal affairs, may affect future demand for a firm’s products or services. By aggregating useful
private signals from various investors, prices revealnew information to managers and guide their
corporate investment decisions (see, e.g., Bond et al. (2012) for a review). This notion is often la-
beled in the literature as the “information feedback effect" of stock prices or “managerial learning"
from stock prices (Goldstein et al., 2021). In particular, because both demand forecasts and stock
prices contain forward-looking information relevant for current investments, I can make intu-
itive economic predictions about how managers’ use of one forward-looking information source
(DF contracts) affects managers’ use of another forward-looking information source (stock price),
when making corporate investment decisions.
Prior literature has argued that prices may contain information that managers already pos-
sess, and, therefore, prices are useful for guiding investment decisions only due to the portion of
the information in prices that is new to managers (e.g., Bond et al., 2012; Gao and Liang, 2013;
Edmans, Jayaraman, and Schneemeier, 2017; Jayaraman and Wu, 2020; Pinto, 2020).
∥
Consistent
with managers learning information they do not already possess from stock price, the prior lit-
erature has found that the sensitivity of corporate investments to stock price is increasing in
investors’ private information in price that is unknown to managers (Chen et al., 2007; Bakke
and Whited, 2010). In a similar vein, research shows that firms use market reactions to M&A
deal announcements to decide whether to proceed with an announced deal (Luo, 2005); firms
adjust actual capex in response to market reactions to managers’ capex forecasts (Jayaraman and
Wu, 2020); the sensitivity of management’s guidance revision to stock returns is increasing in
∥
To describe this idea, Bond et al. (2012) use the term “revelatory price efficiency” (RPE), which is the ability of
price to reveal to managers new information that is useful for making value-maximizing real decisions. In contrast,
the ability of price to accurately reflect a security’s future value is termed “forecasting price efficiency” (FPE).
14
investors’ private information in stock price (Zuo, 2016); and insider trading restrictions increase
the sensitivity of investment to stock price (Edmans et al., 2017).
Survey evidence in Goldstein et al. (2021) strongly confirms managerial learning from the
price channel: in a survey of more than 3,600 public firms in China, 68% of firms report paying
attention to stock price to acquire new information relevant for investment decisions. The sur-
vey also finds that such “learning from price” is the top-most reason managers pay attention to
stock price. Theory also posits that an important reason prices can inform corporate investment
decisions is because investors may be better at collecting information about factors external to
the firm, such as how competition or other market forces may affect future demand for a firm’s
products or services (e.g., Dow and Gorton, 1997; Subrahmanyam and Titman, 1999; Chen et al.,
2007; Edmans et al., 2015; Zuo, 2016; Jayaraman and Wu, 2019).
∗∗
The previous discussion suggests that stock prices guide corporate investment decisions, pos-
sibly by informing managers about future demand. When a supplier firm enters into a DF con-
tract, they gain access to a rich, periodic, and timely source of information about future de-
mand. Accordingly, information in demand forecasts may (partially or fully) substitute for the
investment-relevant information in stock prices. Such crowding out of information in price that
isnew to managers can reduce managerial learning about investment-relevant information from
stock prices after firms enter into DF contracts. This leads to my hypothesis, stated in an alter-
native form:
∗∗
Subrahmanyam and Titman (1999) argue that managers find stock price useful because it can reveal investors’
information about demand for a firm’s products. Chen et al. (2007) and Edmans et al. (2015) point out that information
in price that is useful for managers may be about future demand for firms’ products or other strategic issues, such
as competition. Foucault and Fresard (2014) suggest that an important reason firms can learn about investment
decisions from peers’ stock price is that peers’ demand is correlated with firms’ demand. Similarly, Zuo (2016),
Jayaraman and Wu (2019), and Goldstein and Yang (2019) also suggest that an important reason prices can inform
managers is that compared to making predictions about firms’ internal affairs, investors may be better at predicting
how factors external to the firm, such as future demand or competition, may affect the firm in the future.
15
Hypothesis: Suppliers’ corporate investments becomes less sensitive to stock prices after sup-
pliersenterintoDFcontracts,comparedtotheinvestmentsofcontrolfirmsoverthesametimeperiod.
Suppliers may not reduce their reliance on stock prices when making investment decisions af-
ter entering into a DF contract for several reasons. First, theory on investment under uncertainty
suggests that firms want to wait for the uncertainty about payoffs to irreversible investments to
resolve before making the investment decision (Dixit and Pindyck, 1994). Therefore, suppliers
may not engage in irreversible investments based on preliminary information in demand fore-
casts. Second, if customers’ strategic incentive to distort forecasts completely garble demand
information or if all useful information is already shared informally without an underlying DF
contract, then suppliers may not attach any weight to the demand forecasts in the DF contracts.
Third, if firms do not learn about future demand from stock price or if there is little overlap in
the information in stock price and demand forecasts (say, demand forecasts inform managers
about future demand and stock prices inform managers about future margins or risk), then de-
mand forecast disclosures may not significantly substitute for information in stock price. This
may prevent one from observing a change in investment-price sensitivity post-DF. Therefore,
whether DF contracts are associated with reduced sensitivity of corporate investment to stock
prices remains an empirical question.
16
Chapter3
Data
3.1 Datasources
I collect data on demand forecast disclosure contracts (DF contracts) from all material contracts
from 10-K, 10-Q, and 8-K SEC filings between January 1, 1996, and December 31, 2018. I begin
with a broad textual search for all material contracts that meet the following two conditions: First,
it should have combinations of the terms “supply," “sale," “purchase," or “procurement" collocated
within a three-word distance of “agreement" or “contract." Second, it should have combinations
of the terms “demand" or “rolling" within a five-word distance of “forecast."
∗
This results in a
broad shortlist of contracts that may have a demand forecast disclosure arrangement. From this
broad shortlist of contracts, I identify DF contracts by manually reading through each contract
and removing the ones that do not have a demand forecasting arrangement. This step results in
an initial sample of 2,101 DF contracts. Appendix A provides examples of DF contracts.
∗
The choice of keywords is based on my extensive reading of several contracts before performing exhaustive
data collection. I also allow synonyms of the search words and words that share root words contained within the
search words. For example, the search term “procurement" also searches for synonyms of “procurement" and other
terms based on the word “procure."
17
Next, I manually read through each of the 2,101 DF contracts to collect additional details,
such as the name of the supplier, the name of the customer, the filing date of the contract, the
effective date of the contract, and whether a portion of the forecasted demand is mandatory
for the customer to purchase. Finally, I merge DF contracts with Compustat data by matching
supplier firm names to Compustat names through a combination of programming and manual
matching.
†
This results in a sample of 1,124 DF contracts for 502 supplier firms.
Other than the hand-collected DF contracts data, I obtain firm fundamentals from Compustat,
stock returns data from CRSP, supplier-major customer links data from Compustat Segments on
WRDS, and PIN data from Stephen Brown’s website. The sample period is between 1996 to 2017.
‡
3.2 ControllingfordeterminantsofDFcontracts
The focus of this study is to examine whether investment-to-price sensitivity changes after sup-
plier firms enter into DF contracts. However, entry into a DF contract is not a random event.
It is possible that correlated omitted variables that affect firms’ investment-price sensitivity also
influence a firm’s decision to enter into DF contracts. Failure to account for this possibility could
lead to spurious inferences. One of the first steps I take to address the nonrandom nature of firm-
entry in DF contract is that I match supplier firms that enter into DF contracts for the first time
(DF firms or treatment firms) with supplier firms who never enter into a DF contract but have a
similar probability of entering into a DF contract (non-DF firms or control firms).
To determine the probability of a supplier firm entering into a DF contract, I estimate a firm-
level model for the determinants of DF contracts. Specifically, I estimate a probit model, where
†
I also manually verify each name that is matched through programming.
‡
The start of sample period is limited by DF contracts data and the end of the sample period is limited by
supplier-customer links database.
18
the dependent variable is EnterDFContract, an indicator that equals one in the year a firm first
enters a DF contract and zero otherwise. I draw independent variables from the prior literature
that suggests that firms may share demand forecasts to alleviate uncertainty about their operat-
ing environment (e.g., Cohen et al., 2003; Özer and Wei, 2006; Özer et al., 2011). Accordingly, to
capture the uncertainty of the operating environment, I have included in my determinants model
volatility of sales (VOL(SALES)) and volatility of cash flow from operations ( VOL(CFO)). I include
sales growth (SALEGROWTH) and cash flow from operations ( CFO) because firms experiencing
poor performance may want to be better informed about future demand. In addition, because
this study examines the real effects of DF contracts, I also include variables that may influence
corporate investment decisions. Accordingly, I include market-to-book ratio (Q) to proxy for
investment opportunities and leverage (LEV ) and market capitalization (SIZE) to proxy for any
financing constraints. Industry distribution of the DF contracts and prior literature suggest that
DF contracts may be more common in some industries than others (e.g., Terwiesch et al., 2005).
Industry membership may capture a variety of characteristics that cumulatively define the firm’s
operating environment and therefore influence a firm’s entry into DF contracting. Therefore,
I include Fama-French 30 industry fixed effects to control for any time-invariant industrywide
attributes. Finally, I include year fixed effects to control for any time-varying economywide at-
tributes. Appendix B defines all variables.
The sample for estimation of the determinants model consists of DF firms and all the non-DF
supplier firms. I identify non-DF supplier firms using all the firms that appear as a supplier in the
supplier-major customer links database, but never enter a DF contract during my sample period.
I limit non-DF firms to supplier firms because firms that are not suppliers in a supply chain may
never enter a DF contract (e.g., consumer-facing firms). This restriction allows comparison of DF
19
firms with non-DF firms that could have potentially entered into DF contracts. I drop observations
without available data for all variables, and firms in financial and utilities industries (SIC codes
between 6000–6999 and 9000–9999). Because the objective of this model is to estimate first-time
entry in DF contract, I drop observations for DF firms after they have entered into their first DF
contract. The final estimation sample for the determinants model consists of 27,515 observations
for 356 DF firms (with 2,158 observations) and 4,133 non-DF firms (with 25,357 observations).
The sample period is 1996–2017.
Table 3.1: Determinants of Demand Forecast (DF) Contracts
Table 3.1 presents the model for determinants of supplier firms entering in demand forecast receiving contracts for
the first time. t-stats in parentheses are based on standard errors clustered at both firm and industry-year level. All
continuous variables are winsorized at 1% and 99% of the yearly distributions to reduce the influence of outliers. ***,
**, and * indicate statistical significance (two-sided) at the 0.01, 0.05, and 0.10 levels, respectively. All variables are
defined in Appendix B.
(1) (2)
EnterDFContract EnterDFContract
Lag Q 0.029
∗∗∗ 0.021
∗∗∗ (4.53) (3.21)
Lag CFO − 0.472
∗∗∗ − 0.412
∗∗∗ (− 7.36) (− 5.96)
Lag SIZE 0.075
∗∗∗ 0.088
∗∗∗ (5.93) (6.15)
Lag LEV − 0.098 0.079
(− 0.65) (0.53)
Lag SALEGROWTH − 0.184
∗∗∗ − 0.138
∗∗ (− 4.03) (− 2.57)
Lag VOL(SALES) 0.331
∗∗∗ 0.189
∗∗ (4.70) (2.42)
Lag VOL(CFO) − 0.090 − 0.056
(− 1.45) (− 0.88)
Ind FE Y
Year FE Y
Obs. 27,515 27,515
PseudoR
2
0.052 0.106
AUC 0.705 0.794
20
Table 3.1 reports the coefficients for the determinants model. I find that higher sales volatil-
ity, higher market-to-book, lower cash flow from operations, bigger size, and lower sales growth
predict the likelihood of firms entering into DF contracts. On the other hand, leverage and cash
flow volatility do not appear to predict firm entry into DF contract. The prediction model per-
forms well in correctly classifying observations as EnterDFContract = 0 or EnterDFContract = 1,
with an area under the ROC curve (AUC) of 0.794 in column 2.
§
Of all the variables in the model,
industry membership appears to provide one of the largest improvements in the discrimination
ability of the model (i.e. improvement in AUC). This is consistent with the notion that industry
characteristics may drive the need for learning about future demand. Next, I use the specification
in column 2 to predict the probability of a firm entering into a DF contract and construct the
matched sample, which I will describe in the next section.
3.3 Sampleconstruction
To construct the primary sample for this study, I perform the following matching procedure. I
start with 223 DF firms with available data on investment and Tobin’s Q, and at least one obser-
vation in the pre- and the post-DF periods. In relative year = 1 (one year prior to the first entry
into a DF contract), I match each DF firm with replacement to non-DF supplier firms (supplier
firms that never enter a DF contract) within the same Fama-French 30 industry, in the same year,
§
The AUC is used to assess a prediction model’s discrimination ability, and its value can theoretically vary
between 0.50 (same as a random guess) to 1.00 (perfect discrimination). A general rule of thumb is that an AUC
between 0.70 and 0.80 is considered “good," and 0.80–0.90 is considered “excellent” (Hosmer and Lemeshow, 2000;
Kim and Skinner, 2012).
21
with an absolute difference in the predicted probability of less than 0.002, and at least one ob-
servation in both pre- and post-DF periods.
¶
This step results in 199 unique DF firms. I retain
observations for DF and non-DF firms for seven years each in the pre- and the post-DF periods
and drop the year that the firm entered into the DF contract (i.e., I drop the relative year = 0).
∥
I
drop observations with missing values for the control variables. The final sample consists of 199
DF firms with 1,918 observations and 823 non-DF firms with 11,451 observations, resulting in a
total of 13,369 observations.
3.4 Descriptivestatistics
Table 3.2 presents the descriptive statistics. Panels A and B show firm-year-level descriptive
statistics for DF contracts.
∗∗
Panel A shows the industry distribution for the 1,918 firm-year obser-
vations for DF firms. Some of the most common industries with DF contracts are Pharmaceutical
products, Electronic equipment, Medical equipment, Business services, and Computers. Panel B
presents the characteristics of the DF contracts. Despite the variation inforecasthorizon (from as
low as 3 months to as long as 60 months), over 76% of the forecasts have a horizon of 12 months
or longer, 59% of the forecasts have a 12-month horizon, and around 17% of the forecasts have a
horizon of longer than 12 months. Note that the rolling nature of the forecast means that even
¶
I use the probability in the relative year = 2 if the probability in the relative year = 1 is missing. Additionally, the
primary result of the study is robust to an alternative cutoff for the difference in probability of 0.001 (untabulated).
∥
Two empirical patterns that I observed during data collection drove my choice of event window, [7, 7]: for
supply contracts with demand forecasting arrangements, (1) the most common term is 5 years or more, and (2)
automatic renewals are very common (e.g., contracts commonly include a provision for autorenewal of the contract
unless the contract is explicitly breached or terminated). Combining these two patterns, I choose an event window
of [7, 7] years. Table 4.3, panel B, shows that the primary result is robust to the alternative choice of event windows
[5, 5] or [3, 3] years.
∗∗
For descriptive purposes, for each DF firm I keep characteristics of their first DF contract constant throughout
the event window. For firms that enter into more than one DF contracts for the first time in the same year, I pick the
longest forecast horizon and the most-frequent forecast revision among the multiple DF contracts for the same firm.
22
for a 12-month forecast horizon, during one full year, the supplier firm can receive cumulative
private disclosures of two years of product demand from the customer.
††
Forecast revision fre-
quency can vary from weekly to every 12 months. The most common revision frequencies are
every month (around 47%) and every 3 months (around 42%), respectively. The mean (median)
revision frequency is 2 months (1 month).
‡‡
Panel C shows descriptive statistics for the primary variables used in this study. 14.3% of the
observations in my sample belong to DF firms ( DF). The mean investment (INV ) in the sample is
17.8% of the total assets, while the average Q (i.e., Tobin’s Q, which is the ratio of market value
to the book value of assets) is 2.1. The average firm in the sample has a market capitalization
(SIZE) of $361 million (natural logarithm = 5.890), sales growth (SALEGROWTH) of 12.5%, and
cash flow from operations ( CFO) of 5.9% of the assets. Around 46% (i.e., 0.067/0.143) of the DF
contracts include binding requirements for the customer to purchase at least a portion of the
forecasted demand (BINDING_DF). Finally, panel D shows the correlation matrix. Appendix B
defines all variables.
††
For example, consider a DF contract with a 12-month forecast horizon and monthly revisions. Say, the firm
receives its first forecast in January 2010 for demand for the period February 2010 to January 2011. By December
2010, the firm receives forecasts covering the period up to December 2011. Thus, during the year 2010, the firm
receives cumulative demand forecasts of two years of product demand.
‡‡
The number of observations is fewer than 1,918 for the forecast disclosure arrangement statistics because some-
times the forecast horizon and/or frequency are unavailable or have been redacted in SEC filings.
23
Table 3.2: Descriptive Statistics
Table 3.2 presents the descriptive statistics for the primary sample. For panels C and D: all variables are defined
in Appendix B, and all continuous variables are winsorized at 1% and 99% of the yearly distributions to reduce the
influence of outliers. In Panel D, † and * indicate statistical significance at the 0.01 and 0.05 levels, respectively.
Panel A: Industry distribution of DF firms
Industry Obs. Percent
Food Products 23 1.20
Entertainment 28 1.46
Consumer Goods 8 0.42
Apparel 13 0.68
Healthcare 12 0.63
Medical Equipment 207 10.79
Pharmaceutical Products 572 29.82
Chemicals 36 1.88
Rubber and Plastic Products 28 1.46
Construction Materials 39 2.03
Construction 10 0.52
Steel Works Etc 36 1.88
Fabricated Products 13 0.68
Machinery 62 3.23
Electrical Equipment 14 0.73
Automobiles and Trucks 31 1.62
Defense 9 0.47
Precious Metals 10 0.52
Petroleum and Natural Gas 19 0.99
Utilities 8 0.42
Communication 30 1.56
Business Services 143 7.46
Computers 123 6.41
Electronic Equipment 230 11.99
Measuring and Control Equipment 69 3.60
Business Supplies 14 0.73
Transportation 19 0.99
Wholesale 70 3.65
Retail 22 1.15
Restaraunts, Hotels, Motels 7 0.36
Others 13 0.68
Total 1,918 100.00
24
Panel B: Characteristics of demand forecast disclosure arrangement
Forecast Horizon (Months)
Months Obs. Percent
3 150 11.16
4 9 0.67
6 151 11.24
9 9 0.67
12 800 59.52
15 11 0.82
18 103 7.66
24 63 4.69
36 39 2.90
60 9 0.67
Total 1,344 100.00
Mean Forecast Horizon = 12.31
Median Forecast Horizon = 12.00
Forecast Revision Frequency (Months)
Months Obs. Percent
.25 93 6.15
.5 14 0.93
1 718 47.46
2 6 0.40
3 641 42.37
4 11 0.73
6 11 0.73
12 19 1.26
Total 1,513 100.00
Mean Forecast Freq. = 2.00
Median Forecast Freq. = 1.00
Panel C: Descriptive statistics for the variables
N Mean SD P25 P50 P75
DF 13,369 0.143 0.351 0.000 0.000 0.000
POST 13,369 0.524 0.499 0.000 1.000 1.000
INV 13,369 0.178 0.172 0.070 0.134 0.224
Q 13,369 2.100 1.716 1.156 1.590 2.408
CFO 13,369 0.059 0.169 0.006 0.079 0.147
SIZE 13,369 5.890 2.002 4.524 5.906 7.201
LEV 13,369 0.165 0.180 0.001 0.105 0.282
SALEGROWTH 13,369 0.125 0.358 − 0.027 0.067 0.214
VOL(SALES) 13,369 0.326 0.247 0.157 0.254 0.423
VOL(CFO) 13,369 0.118 0.173 0.041 0.072 0.124
NON_BINDING_DF 13,369 0.077 0.266 0.000 0.000 0.000
BINDING_DF 13,369 0.067 0.250 0.000 0.000 0.000
CUST_DDAQ 9,638 0.065 0.037 0.039 0.060 0.082
CUST_EARNPERSIST 9,853 0.355 0.303 0.176 0.360 0.539
VOL(EARN) 13,369 0.152 0.278 0.035 0.076 0.155
ROA 13,369 − 0.012 0.200 − 0.056 0.031 0.086
1− R
2
12,843 0.831 0.153 0.734 0.871 0.965
PIN 10,227 0.182 0.098 0.111 0.161 0.232
INVPRC 13,340 0.214 0.430 0.039 0.080 0.198
CAPEX 13,183 0.352 0.397 0.131 0.239 0.428
R&D 13,369 0.088 0.115 0.000 0.053 0.130
25
Panel D: Correlation Matrix
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (21)
(1) DF 1.00
(2) POST 0.08
†
1.00
(3) INV 0.08
†
-0.05
†
1.00
(4) Q 0.13
†
-0.07
†
0.26
†
1.00
(5) CFO -0.10
†
-0.03
∗ -0.06
†
0.04
†
1.00
(6) SIZE 0.11
†
-0.03
†
0.02 0.29
†
0.34
†
1.00
(7) LEV 0.00 0.03
∗ -0.08
†
-0.20
†
0.07
†
0.06
†
1.00
(8) SALEGROWTH 0.04
†
-0.14
†
0.04
†
0.12
†
0.09
†
0.08
†
-0.02
∗ 1.00
(9) VOL(SALES) 0.10
†
-0.10
†
0.16
†
0.19
†
-0.21
†
0.01 -0.06
†
0.33
†
1.00
(10) VOL(CFO) 0.04
†
-0.10
†
0.10
†
0.17
†
-0.21
†
-0.13
†
-0.17
†
0.16
†
0.43
†
1.00
(11) NON_BINDING_DF 0.74
†
0.03
†
0.04
†
0.06
†
-0.05
†
0.06
†
-0.03
∗ 0.03
∗ 0.07
†
0.01 1.00
(12) BINDING_DF 0.63
†
0.08
†
0.08
†
0.12
†
-0.09
†
0.10
†
0.04
†
0.02 0.07
†
0.05
†
-0.05
†
1.00
(13) CUST_DDAQ 0.05
†
0.15
†
0.05
†
0.01 -0.07
†
-0.01 -0.04
†
-0.07
†
0.09
†
0.06
†
0.04
†
0.04
†
1.00
(14) CUST_EARNPERSIST -0.04
†
-0.06
†
-0.08
†
-0.08
†
0.01 -0.05
†
0.01 0.04
†
-0.01 0.03
∗ 0.02 -0.07
†
-0.10
†
1.00
(15) VOL(EARN) 0.05
†
-0.06
†
0.10
†
0.15
†
-0.22
†
-0.09
†
-0.16
†
0.09
†
0.38
†
0.69
†
0.01 0.07
†
0.10
†
0.00 1.00
(16) ROA -0.08
†
-0.05
†
-0.11
†
0.07
†
0.74
†
0.31
†
0.05
†
0.13
†
-0.23
†
-0.22
†
-0.04
†
-0.08
†
-0.08
†
0.03
∗ -0.24
†
1.00
(17)1− R
2
-0.04
†
-0.08
†
0.06
†
-0.04
†
-0.16
†
-0.59
†
0.02
∗ 0.07
†
0.08
†
0.13
†
-0.02
∗ -0.03
†
-0.09
†
0.01 0.07
†
-0.11
†
1.00
(18) PIN -0.08
†
0.03
∗ -0.05
†
-0.22
†
-0.17
†
-0.71
†
0.02 -0.10
†
-0.13
†
0.01 -0.05
†
-0.07
†
-0.05
†
0.04
†
-0.01 -0.15
†
0.58
†
1.00
(19) INVPRC -0.01 0.09
†
0.05
†
-0.03
†
-0.27
†
-0.42
†
-0.00 -0.18
†
0.04
†
0.10
†
-0.02 0.00 0.12
†
-0.03
†
0.14
†
-0.27
†
0.26
†
0.34
†
1.00
(20) CAPEX 0.00 -0.07
†
0.35
†
0.34
†
0.02 0.04
†
-0.19
†
0.13
†
0.15
†
0.14
†
0.01 -0.01 0.03
∗ -0.03
†
0.13
†
0.09
†
0.07
†
-0.08
†
0.03
†
1.00
(21) R&D 0.14
†
-0.00 0.56
†
0.32
†
-0.36
†
-0.09
†
-0.26
†
-0.03
†
0.21
†
0.24
†
0.07
†
0.14
†
0.14
†
-0.07
†
0.24
†
-0.36
†
0.03
∗ -0.03
∗ 0.19
†
0.19
†
1.00
26
Chapter4
Results
4.1 DFcontractsandinvestment-pricesensitivity
4.1.1 Measuringmanageriallearningfromstockprices
To measure managerial learning from stock price, I follow the extensive prior literature and
regress future corporate investment (INV ) on current Tobin’s Q (Q) (e.g., Chen et al., 2007; Fou-
cault and Frésard, 2012; Edmans et al., 2017; Jayaraman and Wu, 2019). INV is corporate in-
vestment, defined as the sum of capital expenditures, research and development expenses, and
acquisition costs, scaled by lagged total assets (e.g., Shroff, 2017). Q represents Tobin’s Q, a nor-
malized measure of stock price, defined as total assets plus the market value of equity minus the
book value of equity, scaled by total assets. Q is a price-based signal of firm’s investment op-
portunities, widely considered one of the primary determinants of corporate investments. The
coefficient on Q in the investment-price regression is known as investment-price sensitivity and
is used in the literature as the measure of managerial learning from stock price. The extensive
27
literature on managerial learning from stock price shows that investment-price sensitivity is in-
creasing in precisely those situations in which one would expect managers to learn more about
their investment opportunities from stock price; this gives investment-price sensitivity the inter-
pretation of managers’ learning from stock prices (e.g., Bond et al., 2012). See Section 2.2 for a
detailed discussion of the investment-price sensitivity framework.
4.1.2 Primaryresults
I begin the analyses by examining whether supplier firms’ corporate investments become less
sensitive to stock prices after they enter into demand forecast contracts, compared to investments
of control firms over the same time period. Specifically, I augment the investment-price sensitivity
regression to implement variations of the following difference-in-differences research design:
INV
it+1
=β 1
Q
it
+β 2
POST
it
+β 3
Q
it
× DF
i
+β 4
Q
it
× POST
it
+β 5
POST
it
× DF
i
+β 6
Q
it
× POST
it
× DF
i
+ω
i
Firm
i
+γ t
Year
t
+θControls
it
+β k
Q
it
× (Controls
it
+Ind
j
+Year
t
)+ϵ it+1
(4.1)
In eq. (4.1), i, j, and t index supplier, industry, and year, respectively. INV and Q represent
corporate investment and Tobin’s Q, as defined in Section 4.1.1. DF is an indicator variable that
equals one for suppliers that enter into a DF contract with a customer at some point in time dur-
ing the sample period and zero otherwise (i.e., firms with DF = 0 never enter into a DF contract).
POST is an indicator variable that equals one for DF firms in the years after they have entered into
28
a DF contract for the first time during the sample period.
∗
POST for control firms (i.e., the non-DF
firms) takes the value of POST for the corresponding DF firms to which they are matched.
†
The
coefficient of interest is the difference-in-differences coefficient, β 6
, which compares the change
in investment-price sensitivity of DF firms after entering into DF contract for the first time, com-
pared to the change in investment-price sensitivity of control firms over the same time period. If
corporate investments become less sensitive to stock price post-DF, I expect a significant negative
coefficient on Q× POST × DF ; that is, I expectβ 6
< 0 in eq. (4.1).
Controls include a vector of control variables drawn from the prior literature and the deter-
minants analysis in Section 3.2, which may influence either firms’ investment decisions or firms’
decision to enter into a DF contract (e.g., Chen et al., 2007; Foucault and Frésard, 2012; Dessaint,
Foucault, Frésard, and Matray, 2019; Jayaraman and Wu, 2019). Based on investment-price sensi-
tivity literature, I control for cash flow from operations ( CFO), market capitalization (SIZE), sales
growth (SALEGROWTH), and leverage (LEV ). In addition, based on the determinants analysis in
Section 3.2, I control for variables that may influence firms’ decision to enter into DF contract:
volatility of sales (VOL(SALES)) and volatility of cash flow from operations ( VOL(CFO)). I include
firm fixed effects ( Firm
i
) to control for the time-invariant characteristics of each firm that may
affect its investment decisions or decision to enter into DF contracts.
‡
I include year fixed effects
(Year
t
) to control for any time-varying economywide factors. To control for the heterogeneity
in investment-price sensitivity due to various time-varying firm characteristics, time-invariant
industry characteristics, or time-varying economywide factors, I also include interactions of Q
∗
I can only observe contracts filed since the year 1996. This limitation could potentially induce measurement
error in the definitions of both DF andPOST variables. However, I do not believe this situation will create a bias that
can overturn the main results, but I highlight the caveat that the results are subject to this potential measurement
error.
†
This essentially creates a post-DF period for non-DF firms.
‡
Firm fixed effects absorb the main term DF in eq. (4.1).
29
with control variables and interactions ofQ with Fama-French 30 industry fixed effects and year
effects ( Q
it
× (Controls
it
+Ind
j
+Year
t
)).
§
To account for any time-series or cross-sectional
correlation in error terms, I two-way cluster standard errors at the firm and industry-year level
(Petersen, 2009; Gow, Ormazabal, and Taylor, 2010).
¶
Table 4.1: DF Contracts and Investment-Q Sensitivity
Table 4.1 examines whether corporate investment becomes less sensitive to stock price after firms enter in demand
forecast disclosure contracts. Columns (1)-(4) present variants of Eq.(4.1). INV is corporate investment, defined
as sum of capital expenditures, research and development expenses, and acquisitions costs, scaled by lagged total
assets. Q is Tobin’sQ, defined as total assets plus market value of equity minus book value of equity, divided by total
assets. DF is an indicator variable that equals one for supplier firms that enter in DF contracts anytime in the sample
period, zero otherwise. POST is an indicator variable that equals one (zero) after (before)DF firms enter in their first
DF contract. For non-DF firms, it takes the value of POST corresponding to the DF firm to which they are matched.
CFO is cash flow from operations scaled by lagged total assets. SIZE is the natural logarithm of market value of
equity. LEV is leverage defined as long term debt plus short term debt divided by total assets. SALEGROWTH is
the change in sales scaled by lagged total assets. VOL(SALES) is volatility of sales, defined as within-firm standard
deviation of sales in the last five years, scaled by average sales in the last five years. VOL(CFO) is volatility of
cash flow from operations, defined as within-firm standard deviation of CFO in the last five years. t-stats in
parentheses are based on standard errors clustered at both firm and industry-year level. All continuous variables
are winsorized at 1% and 99% of the yearly distributions to reduce the influence of outliers. ***, **, and * indicate sta-
tistical significance (two-sided) at the 0.01, 0.05, and 0.10 levels, respectively. All variables are defined in Appendix B.
§
Q is not commonly interacted with variables such as CFO in the “managerial learning from price” literature.
However, to be conservative, in specifications where I interact Q with any control variable, I interact Q with all
the control variables. Nevertheless, before I present results for specifications that include interactions of Q, I also
tabulate specifications that do not include such interactions and find consistent results.
¶
To avoid problems with too-few clusters (Cameron and Miller, 2015), I do not cluster standard errors by year
because my sample has only 22 years of data. Instead, I cluster standard errors at the firm and Fama-French 30
industry times year level (resulting in 383 clusters). Table 4.3 shows that the primary result is robust to alternative
clustering schemes: SIC-2 industry level (47 clusters) or firm level (1,022 clusters).
30
Table 4.1: DF Contracts and Investment-Q Sensitivity (Continued)
(1) (2) (3) (4)
INV
t+1
INV
t+1
INV
t+1
INV
t+1
Q 0.019
∗∗∗ 0.023
∗∗∗ 0.056
∗∗∗ (7.47) (7.91) (8.18)
POST − 0.022
∗∗∗ − 0.027
∗∗∗ − 0.026
∗∗∗ − 0.024
∗∗∗ (− 3.53) (− 4.70) (− 4.60) (− 4.17)
Q× DF 0.012
∗∗∗ 0.011
∗∗∗ 0.014
∗∗∗ 0.019
∗∗∗ (3.03) (2.89) (3.03) (3.78)
Q× POST 0.010
∗∗∗ 0.013
∗∗∗ 0.012
∗∗∗ 0.011
∗∗∗ (3.72) (4.98) (4.75) (4.29)
POST × DF 0.010 0.019 0.021 0.021
(0.60) (1.18) (1.24) (1.17)
Q× POST × DF − 0.026
∗∗∗ − 0.026
∗∗∗ − 0.026
∗∗∗ − 0.027
∗∗∗ (− 4.47) (− 4.46) (− 4.24) (− 3.98)
CFO 0.030 0.001 0.005
(1.61) (0.05) (0.18)
SIZE − 0.024
∗∗∗ − 0.016
∗∗∗ − 0.019
∗∗∗ (− 5.98) (− 4.37) (− 5.06)
Lag SALEGROWTH 0.003 0.006 0.010
(0.46) (0.55) (0.88)
Lag LEV − 0.158
∗∗∗ − 0.252
∗∗∗ − 0.256
∗∗∗ (− 8.62) (− 7.29) (− 9.32)
VOL(SALES) − 0.028
∗∗ − 0.043
∗∗ − 0.047
∗∗∗ (− 2.09) (− 2.49) (− 2.64)
VOL(CFO) 0.023 − 0.006 − 0.008
(1.05) (− 0.20) (− 0.26)
Q× CFO 0.017
∗∗ 0.016
∗∗ (2.02) (2.02)
Q× SIZE − 0.006
∗∗∗ − 0.006
∗∗∗ (− 8.56) (− 7.63)
Q× Lag SALEGROWTH − 0.001 − 0.003
(− 0.30) (− 0.70)
Q× Lag LEV 0.037
∗∗∗ 0.039
∗∗∗ (2.64) (3.78)
Q× VOL(SALES) 0.007 0.009
∗ (1.39) (1.69)
Q× VOL(CFO) 0.007 0.009
(0.76) (0.94)
Firm FE Y Y Y Y
Year FE Y Y Y Y
Q× Ind FE Y
Q× Year FE Y
Obs. 13,369 13,369 13,369 13,369
Adj. R
2
0.406 0.421 0.432 0.436
31
Table 4.1 presents the results from estimating eq. (4.1). In column (1), which includes firm and
year fixed effects, I find a negative and statistically significant coefficient on Q× POST × DF
(coefficient = − 0.026 and t-stat = − 4.47). Columns (2)–(4) progressively build on column (1).
Column (2), in addition to column (1), includes control variables and finds very similar results
(coefficient = − 0.026,t-stat =− 4.46). Column (3), in addition to column (2), includes interactions
of Q with all the control variables (coefficient = − 0.026, t-stat = − 4.24). Finally, column (4),
in addition to column (3), includes interactions of Q with industry fixed effects and year fixed
effects. I continue to find similar results (coeff. = − 0.027, t-stat =− 3.98). The strikingly similar
and stable results across all specifications are consistent with my primary hypothesis and suggest
that investment-price sensitivity declines after supplier firms enter into DF contracts.
These results are economically meaningful. Based on the estimates in column (1), the investment-
price sensitivity of DF firms in post-DF period is around 52% lower than the investment-price sen-
sitivity of DF firms in the pre-DF period.
∥
The economic magnitude of my results is comparable
to those reported by previous or concurrent studies. For example, Edmans et al. (2017) find that
insider trading restrictions increase investment-price sensitivity by 38%; Foucault and Frésard
(2012) find that cross-listing in the US almost doubles the investment-price sensitivity; and Mc-
Clure, Shi, and Watts (2020) and Goldstein, Yang, and Zuo (2020) find that the implementation
of electronic disclosure systems, such as EDGAR (in the US) and CEDS (international) reduced
investment-price sensitivity by 20% and 54%, respectively. My results suggest that when mak-
ing investment decisions, managers significantly reduce learning from stock prices if they can
learn about future demand through private disclosure contracts with customers. Put differently,
∥
Based on column (1), investment-price sensitivity for DF firms in pre-DF period = Q (0.019) +
Q× DF (0.012), and investment-price sensitivity for DF firms in post-DF period = Q(0.019)+Q× DF (0.012)+
Q× POST (0.010)+Q× POST × DF (− 0.026).
32
investment-relevant information revealed through DF contracts crowds out the ability of prices
to reveal new investment-relevant information to managers. Accordingly, managers appear to
putlessweight onprice-based signals for making investment decisions (i.e.,Q) when private dis-
closure contracts reveal arguably more direct non-price signals of investment opportunities (i.e.,
forecasts of future demand).
∗∗
4.2 Paralleltrendsassumption
A key identifying assumption of the difference-in-differences research design is the parallel trends
assumption: DF and non-DF firms would have followed similar trends in investment-price sen-
sitivity if DF firms had not entered into a DF contract. Stated differently, the assumption is that
the trend in investment-price sensitivity for non-DF firms provides an appropriate counterfactual
for the trend in investment-price sensitivity of DF firms. While this assumption is not directly
testable, a credible approach would be an examination of the trend in investment-price sensi-
tivity during the pre-DF period (Angrist and Pischke, 2008). To examine the parallel trends, I
modify eq. (4.1) and replace the termPOST with separate event-time indicators for each year rel-
ative to DF firms’ entry into DF contracts for the first time: REL
<=− 4
,REL
− 3
,REL
− 2
,REL
0
,
REL
1
, REL
2
, REL
3
, and REL
>=4
. I omit REL
− 1
and, thus, the relative year =− 1 becomes
the benchmark year. I estimate the following model:
∗∗
A related stream of research uses observable characteristics such as speed of earnings news, management
guidance incidence or accuracy, to measure internal information quality, and examine its effect on real decisions (e.g.,
Goodman, Neamtiu, Shroff, and White, 2014; Gallemore and Labro, 2015; Heitzman and Huang, 2019). My results
complement that literature by showing how investment decision making can be influenced by access to contractual
private disclosures about future demand which in turn can shape the quality of manager’s internal information.
33
INV
it+1
=β 1
Q
it
+β 2k
X
k̸=− 1
REL
ik
+β 3
Q
it
× DF
i
+β 4k
Q
it
× X
k̸=− 1
REL
ik
+β 5k
DF
i
× X
k̸=− 1
REL
ik
+β 6k
Q
it
× DF
i
× X
k̸=− 1
REL
ik
+ω
i
Firm
i
+γ t
Year
t
+θControls
it
+β k
Q
it
× (Controls
it
+Ind
j
+Year
t
)+ϵ it+1
(4.2)
where k indexes the relative years <= − 4,− 3,− 2, 0, 1, 2, 3, and >= 4. Fig. 4.1 plots the
dynamic diff-in-diff coefficients ( Q× DF × REL
k
) and the confidence intervals, in event-time.
The coefficient on Q× DF × REL
k
is the difference between two differences: the difference
in investment-price sensitivity between DF and non-DF firms when the relative year = k, and
the difference in investment-price sensitivity between DF and non-DF firms when the relative
year =− 1. Top panel in Fig. 4.1, which includes firm and year fixed effects, shows no change
in the difference in investment-price sensitivity between DF and non-DF firms in the pre-period.
Only in the post-period is a sharp decline in investment-price sensitivity for DF firms compared
to that for non-DF firms obvious. Put differently, the counterfactual treatment effect in the pre-
period is statistically indistinguishable from zero. Next, I continue to find similar evidence in the
bottom panel in Fig. (4.1), which additionally includes control and interactions ofQ with controls,
industry fixed effects, and year fixed effects.
34
.06 .04 .02 0 .02
DiD coefficient and 90% CI
<=4 3 2 1 0 1 2 3 >=4
Relative Year
.06 .04 .02 0 .02
DiD coefficient and 90% CI
<=4 3 2 1 0 1 2 3 >=4
Relative Year
Figure 4.1: Dynamic DiD coefficients in event time.
This figure plots the dynamic difference-in-differences coefficients in event time. The x-axis denotes the years relative
to the first-time entry in DF contract (Relative Year). The y-axis plots the difference-in-differences coefficients and
90% confidence intervals, estimated by replacing the term POST (and all its interactions) with the separate indicators
for each relative year (and their respective interactions with all other terms). Relative Year =− 1 is omitted and is
the comparison group. Top Panel includes firm and year fixed effects. Bottom Panel includes firm and year fixed
effects, controls, and the interactions of Q with controls, with industry fixed effects, and with year fixed effects. For
the purpose of plotting the dynamic DiD coefficients, Relative Year = 0 is included in the sample.
35
Table 4.2: Parallel Trends
Table 4.2 examines parallel trends. The sample is restricted to pre-DF period (i.e. all observations with POST = 0).
TREND is a trend variable that takes discrete values between− 7 and− 1, corresponding to year relative to first-time
entry in DF contract. t-stats in parentheses are based on standard errors clustered at both firm and industry-year
level. All continuous variables are winsorized at 1% and 99% of the yearly distributions to reduce the influence of
outliers. ***, **, and * indicate statistical significance (two-sided) at the 0.01, 0.05, and 0.10 levels, respectively. All
other variables are defined in Appendix B.
(1) (2) (3) (4)
INV
t+1
INV
t+1
INV
t+1
INV
t+1
Q 0.016
∗∗∗ 0.025
∗∗∗ 0.064
∗∗∗ (4.04) (5.32) (6.69)
TREND 0.002 0.000 − 0.001 − 0.000
(0.83) (0.02) (− 0.57) (− 0.18)
Q× DF 0.017 0.015 0.016 0.020
(1.49) (1.15) (1.17) (1.56)
Q× TREND − 0.000 0.001 0.001 0.001
(− 0.20) (0.67) (1.26) (0.87)
TREND× DF − 0.009 − 0.005 − 0.003 − 0.005
(− 1.21) (− 0.64) (− 0.38) (− 0.59)
Q× TREND× DF 0.002 0.001 0.001 0.001
(0.62) (0.40) (0.22) (0.31)
Firm FE Y Y Y Y
Year FE Y Y Y Y
Controls Y Y Y
Q× Controls Y Y
Q× Ind FE Y
Q× Year FE Y
Obs. 6,359 6,359 6,359 6,359
Adj.R
2
0.438 0.455 0.467 0.470
In an alternative approach to examining parallel trends, I modify eq. (4.1) to replace POST
with a trend variable, TREND, that takes discrete values between− 7 and− 1, corresponding to
relative years. I restrict the sample for this analysis to the pre-DF period. The coefficient on
Q× TREND× DF is the difference between the trends in the investment-price sensitivity
of DF and non-DF firms in the pre-DF period. Table 4.2 presents the results for this analysis.
Consistent with Fig. 4.1, the coefficients on Q× TREND× DF are statistically and economically
36
insignificant ( t-statistic = 0.22 to 0.62). Overall, this evidence does not support violation of the
parallel trends assumption in my sample.
4.3 Robustnesstests
In this section, I perform several tests to check the robustness of my primary results.
4.3.1 Pseudo-experiments
I begin by performing pseudo-experiments after inserting pseudo-randomness in the data along
two dimensions: the cross-section (which firm is a DF firm) and the time-series (when does a DF
firm enter a DF contract for the first time). Specifically, in each pseudo-experiment, I randomly
assign a fraction of the firms to become pseudo-DF firms and randomly assign a pseudo-POST
period to each firm.
††
I then estimate eq. (4.1) using the pseudo-DF and pseudo-POST vari-
ables. I repeat the whole process 1,000 times and plot the resultant pseudo-DiD coefficients on
Q× POST × DF in Fig. 4.2. Panel (a) is estimated using the specification in Table 4.1, column
(1), and panel (b) is estimated using the specification in Table 4.1, column (4). The means and
medians of the resultant distributions are zero. In panel a (b), 4 (1) of 1,000 pseudo-experiments
result in a pseudo-DiD coefficient of at least as large as the actual DiD coefficient. This suggests
that any spurious cross-sectional or time-series variations in DF contracts do not seem to explain
the primary result.
††
I match the fraction of pseudo-DF firms to the fraction of actual DF firms in my sample.
37
0 .02 .04 .06 .08 .1
Fraction
Actual DiD Coefficient
.04 .03 .02 .01 0 .01 .02 .03 .04
Pseudo-DiD coefficient
0 .02 .04 .06 .08 .1
Fraction
Actual DiD Coefficient
.04 .03 .02 .01 0 .01 .02 .03 .04
Pseudo-DiD coefficient
Figure 4.2: Pseudo ‘Diff-in-Diff’ coefficient estimates
This figure plots the distribution of pseudo diff-in-diff coefficients based on 1,000 pseudo “experiments”. In each such
pseudo “experiment”, I assign demand forecast contracts to randomly chosen firms with randomly chosen effective
years. Using the resulting pseudo-DF firms and pseudo-POST periods, I re-estimate the DiD regression in eq.(4.1).
I repeat the whole process 1,000 times. I plot the resulting distribution of 1,000 pseudo DiD coefficients, using a
bin-width of 0.002. Top Panel is estimated using specification in Table 4.1, column (1), and bottom Panel is estimated
using specification in Table 4.1, column (4). As a benchmark for the pseudo estimates, the solid maroon vertical line
represents the DiD coefficient on Q× POST × DF from actual data. The resulting distributions have a mean and
median of zero. In top (bottom) panel, four (one) out of 1,000 pseudo-experiments result in a DiD coefficient equal
or to the left of actual DiD coefficient.
4.3.2 Coefficientstabilityandcorrelatedomittedvariables
The nonrandom nature of DF contracts makes it difficult to draw causal inferences from my re-
sults. So far, I have taken several steps to address this issue: (a) match firms on the probability of
entering into a DF contract, (b) include firm and time fixed effects, (c) include control variables
drawn from the prior literature and determinants analysis, (d) interact Q with control variables
and fixed effects, (e) examine parallel trends, and (f) perform pseudo-experiments. Still, one could
38
argue that a correlated omitted variable drives the primary result. This interpretation seems less
likely because such a correlated omitted variable would need to provide an explanation orthog-
onal to the plethora of measures I have taken so far.
Nevertheless, to further alleviate concerns about correlated omitted variable bias, I now more
formally assess the potential impact of correlated omitted variables on my primary result using
the technique from Oster (2019) and implemented in Mian and Sufi (2014); Call, Martin, Sharp,
and Wilde (2018), among others. The objective is to find a factor called the “coefficient of pro-
portionality” (δ ) using the changes in the coefficient of interest and R
2
when moving from a
specification without controls to a specification with all the controls. δ conveys the following
intuition: compared to observables, how important unobservables would have to be to suppress
the coefficient of interest to zero? For example, a δ = 5 means that if the unobservables were five
times as important as the observables, they would suppress the coefficient of interest to zero. An
absolute value ofδ greater than one is considered a robust result (|δ | = 1 means that unobserv-
ables as important as observables should suppress the coefficient of interest to zero.) (Oster, 2019;
Call et al., 2018).
I implement the Oster (2019) technique in Table 4.3, panel A.
‡‡
The estimation ofδ requires
the following inputs: the coefficient of interest and R
2
in regressions with no controls and all
controls, andR
2
max
. R
2
max
is the hypotheticalR
2
from a regression that includes all controls and
all unobservables. While the theoretical maximum value ofR
2
max
is 1, it is not the recommended
value because Oster (2019) argues that in most settings, finding meaningful variables that will
fully explain the outcome of interest appears impossible to do. Following prior studies, I use the
Oster (2019) recommendedR
2
max
equal to 1.3 times theR
2
from the regression with all observable
‡‡
I use the STATA command, psacalc, made available by Emily Oster.
39
controls.
§§
Panel A reports all the inputs and the resultantδ . Rows (1)-(4) correspond to columns
(1)–(4) of Table 4.1. Theδ estimate reported in first row (corresponding to model (1) from Table
4.1) suggests that any unobservables will have to be 49.5 times as important as the observables to
suppress the DiD coefficient on Q× POST × DF to zero. Similarly, for specifications (2)–(4),
the magnitude of δ is comfortably above the recommended threshold of|δ | = 1. Lastly, I also
find (untabulated) that my inferences are unchanged if I use the theoretical maximum of R
2
max
=
1 (which means I assume unobserved variables that when included in the specification will make
R
2
=1). Overall, these tests further alleviate concerns about correlated omitted variable bias in
my sample.
§§
Note that theR
2
’s reported in Table 4.3 are slightly different than the corresponding adjusted R
2
’s reported
in Table 4.1. For example, model (1) in Table 4.3 reports anR
2
equal to 0.452 instead of 0.406, which is reported in
Table 4.1. This is because the Oster technique uses R
2
, whereas Table 4.1 reports the adjusted R
2
. However, the
untabulated (unadjusted)R
2
’s from Table 4.1 look identical to those reported in Table 4.3.
40
Table 4.3: Robustness Tests
Table 4.3 presents various robustness tests for the primary result. Panel A examines how important correlated
ommitted variables would need to be to overturn the primary result, using the methodology of Oster (2019).
Panels B-D examine alternative event windows, alternative measures of investment, and alternative clustering of
standard errors, respectively. Main terms and all lower order interactions are included but only DiD coefficients
are shown for brevity. PIN is the probability of informed trading measure obtained from Stephen Brown’s website.
1− R
2
is the price non-synchronicity measured as 1 minusR
2
from the regression of daily stock return on CRSP
value-weighted market return during the year. t-statistics in parentheses are based on standard errors clustered at
both firm and industry-year level. All continuous variables are winsorized at 1% and 99% of the yearly distributions
to reduce the influence of outliers. ***, **, and * indicate statistical significance (two-sided) at the 0.01, 0.05, and 0.10
levels, respectively. All variables are defined in Appendix B.
Panel A: How Important are Correlated Omitted Variables
Coefficient on Q× POST × DF R
2
Model # Without With Without With Π R
2
max
δ (|δ |>1
(col. # from Controls Controls Controls Controls (R
2
× Π ) suggests coef
Table 4.1) stability)
(1) − 0.031 − 0.026 0.108 0.452 1.3 0.590 − 49.57
(2) − 0.031 − 0.026 0.108 0.467 1.3 0.609 − 62.49
(3) − 0.031 − 0.026 0.108 0.477 1.3 0.620 − 391.86
(4) − 0.031 − 0.027 0.108 0.483 1.3 0.626 − 61.32
Panel B: Alternative Event Windows
Event Window = [− 5,5] [− 3,3]
(1) (2) (3) (4) (5) (6) (7) (8)
INV
t+1
INV
t+1
INV
t+1
INV
t+1
INV
t+1
INV
t+1
INV
t+1
INV
t+1
Q× POST × DF − 0.027
∗∗∗ − 0.027
∗∗∗ − 0.026
∗∗∗ − 0.027
∗∗∗ − 0.031
∗∗∗ − 0.030
∗∗∗ − 0.031
∗∗∗ − 0.029
∗∗∗ (− 4.14) (− 4.14) (− 3.81) (− 3.59) (− 3.71) (− 3.73) (− 3.92) (− 3.32)
Firm FE Y Y Y Y Y Y Y Y
Year FE Y Y Y Y Y Y Y Y
Controls Y Y Y Y Y Y
Q× Controls Y Y Y Y
Q× Ind FE Y Y
Q× Year FE Y Y
Obs. 10,817 10,817 10,817 10,817 7,366 7,366 7,366 7,366
Adj.R
2
0.402 0.417 0.427 0.432 0.420 0.432 0.441 0.447
41
Table 4.3: Robustness Tests (Continued)
Panel C: Alternative Definitions of Investment
(1) (2) (3) (4) (5) (6) (7) (8)
CAPEX
t+1
CAPEX
t+1
CAPEX
t+1
CAPEX
t+1
R&D
t+1
R&D
t+1
R&D
t+1
R&D
t+1
Q× POST × DF − 0.057
∗∗ − 0.058
∗∗ − 0.049
∗∗ − 0.049
∗∗ − 0.022
∗∗∗ − 0.022
∗∗∗ − 0.018
∗∗∗ − 0.018
∗∗∗ (− 2.27) (− 2.40) (− 2.32) (− 2.52) (− 3.88) (− 4.05) (− 3.52) (− 3.42)
Firm FE Y Y Y Y Y Y Y Y
Year FE Y Y Y Y Y Y Y Y
Controls Y Y Y Y Y Y
Q× Controls Y Y Y Y
Q× Ind FE Y Y
Q× Year FE Y Y
Obs. 13,183 13,183 13,183 13,183 13,369 13,369 13,369 13,369
Adj.R
2
0.252 0.259 0.266 0.272 0.730 0.752 0.763 0.767
Panel D: Alternative Clustering of Standard Errors
Cluster s.e. by = SIC-2 Firm
(1) (2) (3) (4) (5) (6) (7) (8)
INV
t+1
INV
t+1
INV
t+1
INV
t+1
INV
t+1
INV
t+1
INV
t+1
INV
t+1
Q× POST × DF − 0.026
∗∗∗ − 0.026
∗∗∗ − 0.026
∗∗∗ − 0.027
∗∗∗ − 0.026
∗∗∗ − 0.026
∗∗∗ − 0.026
∗∗∗ − 0.027
∗∗∗ (− 7.29) (− 6.39) (− 10.31) (− 8.47) (− 4.25) (− 4.29) (− 4.17) (− 4.24)
Firm FE Y Y Y Y Y Y Y Y
Year FE Y Y Y Y Y Y Y Y
Controls Y Y Y Y Y Y
Q× Controls Y Y Y Y
Q× Ind FE Y Y
Q× Year FE Y Y
Obs. 13,369 13,369 13,369 13,369 13,369 13,369 13,369 13,369
Adj.R
2
0.406 0.421 0.432 0.436 0.406 0.421 0.432 0.436
Panel E: Investors’ Private Information in Prices
(1) (2) (3) (4) (5) (6)
PIN PIN PIN 1− R
2
1− R
2
1− R
2
POST 0.001 0.001 0.002 0.007
∗∗ 0.007
∗∗ 0.007
∗∗ (0.47) (0.43) (0.70) (2.40) (2.30) (2.37)
POST × DF 0.001 0.001 − 0.004 0.009 0.008 0.008
(0.12) (0.19) (− 0.78) (0.94) (0.88) (0.94)
Lag SIZE − 0.035
∗∗∗ − 0.033
∗∗∗ − 0.032
∗∗∗ − 0.043
∗∗∗ − 0.043
∗∗∗ − 0.041
∗∗∗ (− 17.58) (− 16.17) (− 14.84) (− 14.58) (− 14.70) (− 14.09)
Lag INVPRC − 0.006 − 0.005 − 0.004 − 0.001 − 0.001 − 0.000
(− 0.79) (− 0.62) (− 0.60) (− 0.15) (− 0.18) (− 0.06)
Other Controls from eq. (4.1) N Y Y N Y Y
Firm FE Y Y Y Y Y Y
Year FE Y Y Y Y
Ind× Year FE Y Y
Obs. 10,203 10,198 10,198 12,816 12,811 12,811
Adj.R
2
0.676 0.678 0.685 0.708 0.709 0.727
42
4.3.3 Alternativedesignchoices
Table 4.3, panels B–D, show the robustness of the primary result to additional alternative re-
search design choices. Panel B estimates eq. (1) using alternative event window lengths of [− 5,
5] and [− 3, 3] and finds similar results. For example, the coefficient on Q× POST × DF in
column (4) is identical to that of Table 4.1. I find similar results for an even shorter window of
[− 3, 3]. Panel C estimates eq. (1) using alternative definitions of investment. Columns (1)–(4)
define investment as capital expenditures scaled by lagged PP&E, whereas columns (5)–(8) define
investment as research and development expenditures scaled by lagged total assets. These alter-
native definitions of investment do not change my inferences. Finally, panel D estimates eq. (1)
after clustering standard errors in alternative ways. Columns (1)–(4) cluster standard errors by
two-digit SIC industry, and columns (5)–(8) cluster standard errors by firm.
¶¶
These alternative
schemes for clustering standard errors do not change the inferences. Overall, the primary result
is robust to alternative design choices.
4.3.4 Investors’privateinformationinstockprices
Finally, recall the discussion in Section 2.2 that managers learn about investment decisions from
stock price because investors impound their private information into prices, thereby revealing
new decision-relevant information to managers. Therefore, if for some reason, investors’ pri-
vate information in stock prices decreases post-DF, this decrease can also result in decreased
investment-price sensitivity. To examine this possibility, I directly examine whether investors’
private information in stock price decreases post-DF. Investors’ private information in stock price
¶¶
When clustering by industry, I do not cluster by Fama-French 30 (FF30) because doing so can lead to too few
clusters. However, the inference remains unchanged if I cluster by FF30 industry, FF30× year, or SIC-2× year.
43
is defined using two common measures used in the prior literature: (a) probability of Informed
trading (PIN), and (b ) nonsynchronicity of stock price (1− R
2
) (Durnev, Morck, and Yeung, 2004;
Chen et al., 2007; Bakke and Whited, 2010; Foucault and Fresard, 2014; Jayaraman and Wu, 2019).
PIN data come from Stephen Brown’s website (Brown and Hillegeist, 2007). Price nonsynchronic-
ity is one minusR
2
, whereR
2
for each firm-year is calculated from the regressions of daily stock
returns on CRSP value-weighted market returns. I regressPIN and price nonsynchronicity mea-
sures on POST, DF × POST , Controls, and fixed effects. Controls include the lagged values of
size (SIZE), the inverse of price (INVPRC), cash flow from operations ( CFO), sales growth (SALE-
GROWTH), leverage (LEV ), the volatility of sales (VOL(SALES)), and the volatility of cash flow
from operations (VOL(CFO)). I include firm and year (or industry × year) fixed effects. My interest
is in the coefficient on DF× POST . I present the results of this analysis in Table 4.3, panel E. I do
not find any significant decrease in investors’ private information in stock price post-DF across
any specifications ( t-stat =− 0.78 to 0.94). This suggests that a reduction in investors’ private
information in stock price does not seem to be a plausible explanation for the primary result.
4.4 Credibilityofdemandforecastdisclosures:
Bindingforecasts
The results so far provide robust evidence that managers reduce their reliance on stock prices
in guiding their investment decisions after entering into a DF contract. I now provide more
direct evidence to substantiate my primary inferences. At the heart of the main argument is the
supplier’s use of demand forecasts in making investment decisions. However, as discussed in
Section 2.1, cheap talk concerns can severely hurt the credibility and thus usefulness of demand
44
forecasts. In particular, for customers, the cost of a supply shortage is usually much larger than
the cost of disclosing inflated forecasts and canceling the order later (Armony and Plambeck,
2005; Özer and Wei, 2006; Oh and Özer, 2013). This customer incentive is costly for suppliers
because suppliers can make real decisions based on inflated demand forecasts that don’t result in
any actual sales ex post, making demand forecasts less credible to suppliers ex ante. I expect that
contractual provisions that make inflated demand forecasts costly for customers reduce cheap
talk concerns and improve the credibility of the demand signals shared through DF contracts
(e.g., Kartik, 2009), resulting in greater learning for the supplier. Greater learning should result in
more pronounced substitution between stock prices and demand forecast disclosures in guiding
suppliers’ corporate investment decisions. Specifically, I expect that the effect of DF contracts
on investment-price sensitivity should be stronger for DF contracts where its more costly for
customers to share inflated forecasts.
To test this idea, I exploit the fact that DF contracts can include risk-sharing provisions that
make it mandatory for customer to purchase a portion of the forecasted demand disclosed to
the supplier (see Section 2.1). I call such demand forecasts binding forecasts.
∗∗∗
By contractually
requiring a customer to buy a portion of the forecasted demand, binding forecasts increase costs
for customers who disclose inflated forecasts. This aspect of DF contracts can reduce concerns of
cheap talk and increase the credibility of demand forecast disclosures. I collect data on whether
or not forecasts in a DF contract are (partially) binding by manually reading through the portions
of the contract that describe the terms of the demand forecast. I split the DF variable into two
types: BINDING_DF and NON_BINDING_DF. BINDING_DF is an indicator variable that equals
∗∗∗
For simplicity, I call such forecasts “binding forecasts,” but they are technically “partially binding forecasts,”
because usually a fraction of the forecast is binding.
45
one for DF firms if the customer in the DF contract is required to purchase at least a portion of the
forecasted demand (and zero otherwise). NON_BINDING_DF is an indicator variable that equals
one for DF firms without any binding provisions for demand forecasts (and zero otherwise).
†††
I modify eq. (4.1) to replace DF (and its interactions) with the terms BINDING_DF and NON_-
BINDING_DF (and all the respective interactions). I expect the coefficient on Q× POST × BINDING_DF to be more negative than the coefficient on Q× POST × NON_BINDING_DF.
Table 4.5 presents the results for the binding forecasts analysis. As expected, across all the
specifications in columns (1)–(4), I find that the coefficients on Q× POST× BINDING_DF are
negative and statistically significant ( t-stat =− 3.70 to− 4.35). Interestingly, across all the specifi-
cations, I also find that the coefficients on Q× POST× NON_BINDING_DF are negative and
statistically significant ( t-stat =− 2.83 to− 3.71). This suggests that even when the demand fore-
cast disclosures are nonbinding on customers, the supplier learns investment-relevant informa-
tion from such forecasts, which results in reduced learning about investments from stock price.
‡‡‡
Importantly, consistent with my expectations, the coefficients on Q× POST× BINDING_DF
are more negative than the coefficients on Q× POST × NON_BINDING_DF . Economi-
cally, the effect of DF contracts on investment-price sensitivity for binding forecasts is around
1.5 times to 2.2 times as large as that for nonbinding forecasts (i.e., the ratio of the coefficients
onQ× POST× BINDING_DF andQ× POST× NON_BINDING_DF ). This suggests
that the effect of DF contracts on investment-price sensitivity is stronger when demand forecast
†††
Contracts may explicitly state that a portion of the forecasted demand is binding to the customer; I classify
such forecasts asbinding. Some contracts explicitly state that the complete forecast is shared on a “good-faith" basis,
with no obligation on the customer to purchase. I classify such forecasts as nonbinding. I also classify forecasts as
nonbinding when I cannot find any binding provisions in the forecasting arrangement.
‡‡‡
Forces other than binding forecasts may bring credibility to the forecasts. Examples of such potential forces
could include other contractual provisions that can improve credibility, such as limits on forecast revisions, forecast
accuracy targets, or reputation.
46
disclosures are binding on customers. Overall, this evidence is consistent with greater manage-
rial learning from more credible demand forecasts, further reducing managers’ reliance on stock
price in guiding their investment decisions. This result not only substantiates my primary infer-
ences but also directly speaks to the unique ability of DF contracts to address the credibility of
private disclosures in supply chains.
Table 4.5: Credibility of Demand Forecast Disclosures: Binding Forecasts
Table 4.5 examines whether (non-)binding nature of the demand forecast influences the effect of DF contracts on the
investment-Q sensitivity. BINDING_DF =1 if customer is required to purchase at least a portion of the forecasted
demand, zero otherwise. NON_BINDING_DF =1 if customer has no obligation to purchase a portion of the forecasted
demand, zero otherwise. Main terms and all lower order interactions are included but only DiD coefficients are shown
for brevity. t-statistics in parentheses are based on standard errors clustered at both firm and industry-year level.
All continuous variables are winsorized at 1% and 99% of the yearly distributions to reduce the influence of outliers.
***, **, and * indicate statistical significance (two-sided) at the 0.01, 0.05, and 0.10 levels, respectively. All variables
are defined in Appendix B.
(1) (2) (3) (4)
INV
t+1
INV
t+1
INV
t+1
INV
t+1
Q× POST × BINDING_DF − 0.032
∗∗∗ − 0.034
∗∗∗ − 0.036
∗∗∗ − 0.037
∗∗∗ (− 3.91) (− 4.35) (− 3.79) (− 3.70)
Q× POST × NON_BINDING_DF − 0.021
∗∗∗ − 0.020
∗∗∗ − 0.017
∗∗∗ − 0.017
∗∗∗ (− 3.71) (− 3.13) (− 2.99) (− 2.83)
p-value
(BINDING_DF > NON_BINDING_DF) 0.08 0.04 0.02 0.02
Firm FE Y Y Y Y
Year FE Y Y Y Y
Controls Y Y Y
Q× Controls Y Y
Q× Ind FE Y
Q× Year FE Y
Obs. 13,369 13,369 13,369 13,369
Adj.R
2
0.406 0.422 0.433 0.437
47
Chapter5
AdditionalAnalysis
In this section, I develop and test additional hypotheses that help improve our understanding of
the effect of DF contracts on how supplier’s make investment decisions.
5.1 Demandvolatility
I begin by examining how the heterogeneity in the demand volatility can influence my primary
results. Demand volatility can potentially make forecasting future demand difficult for both in-
vestors and customers. Therefore, it is not clear whether suppliers, when faced with volatile
demand, would rely more on information in the stock prices or the customer-provided demand
forecasts. I expect the primary results to be stronger for suppliers with high demand volatility
if such suppliers rely more on customer demand forecasts compared to stock prices, to guide
investment decisions. On the other hand, if suppliers with high demand volatility rely on stock
prices more than on customer demand forecasts, I expect the primary results to be weaker. This
48
is possible if suppliers with high demand volatility believe that the aggregate wisdom of the in-
vestor base might be more helpful than the managers at the customer firm, in making predictions
about future demand.
I measure demand volatility using two proxies: volatile of sales (VOL(SALES)) and volatility of
earnings (VOL(EARN)). While sales volatility is a relatively more direct proxy for demand volatil-
ity, the intuition behind using earnings volatility as a proxy is that volatility in top-line sales can
eventually result in volatility in bottom-line earnings. I classify demand volatility above (below)
cross-sectional median as high (low). I estimate eq. (4.1) separately for high and low demand
volatility subsamples.
∗
Table 5.1, panel A, presents the results of theVOL(SALES) analysis. In the high demand volatil-
ity subsample in column (1), I find that the coefficient on Q× POST × DF is insignificant ( t-stat
=− 0.07). However, the coefficient on Q× POST × DF in the low demand volatility subsample
in column (2) is negative and significant ( t-stat =− 4.14). The difference in the coefficients for
Q× POST × DF between the high and low volatility subsamples is also significantly different
from zero. I find similar evidence across all specifications in columns (3)-—(8), where I incremen-
tally add controls, and interactions ofQ with controls, industry fixed effects, and year fixed effects
(low subsamplet-stat =− 0.07 to 0.33; high subsamplet-statistics =− 3.89 to− 4.14). In panel B, I
find similar results using the VOL(EARN) measure of demand volatility (low subsample, t-stat =
0.15 to 0.58; high subsample, t-stat =− 3.39 to− 4.15). These findings suggest that suppliers with
high demand volatility rely more on the customer demand forecasts compared to stock prices
when making investment decisions.
∗
Because this analysis exploits the heterogeneity in the demand volatility, I do not include VOL(SALES) and
VOL(CFO) as controls variables. However, inferences (untabulated) do not change even if I control for the two
variables.
49
Table 5.1: Demand Volatility
Table 5.1 examines whether demand volatility influences the effect of DF contracts on investment- Q sensitivity.
Panel A and B measure demand volatility using sales volatility (VOL(SALES)), and earnings volatility (VOL(EARN)),
respectively. High (Low) subsample splits are based on the demand volatility measures above (below) their
respective median. p-values for the difference in the DiD coefficients between high and low subsamples are shown
between high and low columns. Main terms and lower order interactions are included but only DiD coefficients are
shown for brevity. t-statistics in parentheses are based on standard errors clustered at both firm and industry-year
level. All continuous variables are winsorized at 1% and 99% of the yearly distributions to reduce the influence of
outliers. ***, **, and * indicate statistical significance (two-sided) at the 0.01, 0.05, and 0.10 levels, respectively. All
variables are defined in Appendix B.
Panel A: Demand Volatility Measure: Sales Volatility (VOL(SALES))
Dep Var =INV
t+1
(1) (2) (3) (4) (5) (6) (7) (8)
VOL(SALES) = Low High Low High Low High Low High
Q× POST × DF − 0.001 − 0.029
∗∗∗ − 0.001 − 0.029
∗∗∗ 0.003 − 0.029
∗∗∗ 0.005 − 0.031
∗∗∗ (− 0.07) (− 4.14) (− 0.13) (− 4.11) (0.33) (− 3.98) (0.32) (− 3.89)
p-value (High= Low) 0.02 0.03 <0.01 0.02
Firm FE Y Y Y Y Y Y Y Y
Year FE Y Y Y Y Y Y Y Y
Controls Y Y Y Y Y Y
Q× Controls Y Y Y Y
Q× Ind FE Y Y
Q× Year FE Y Y
Obs. 6,706 6,663 6,706 6,663 6,706 6,663 6,706 6,663
Adj.R
2
0.427 0.421 0.439 0.437 0.447 0.449 0.454 0.454
Panel B: Demand Volatility Measure: Earnings Volatility (VOL(EARN))
Dep Var =INV
t+1
(1) (2) (3) (4) (5) (6) (7) (8)
VOL(EARN)= Low High Low High Low High Low High
Q× POST × DF 0.006 − 0.029
∗∗∗ 0.005 − 0.029
∗∗∗ 0.002 − 0.025
∗∗∗ 0.006 − 0.025
∗∗∗ (0.58) (− 4.06) (0.45) (− 4.15) (0.15) (− 3.44) (0.34) (− 3.39)
p-value (High= Low) <0.01 <0.01 0.03 0.09
Firm FE Y Y Y Y Y Y Y Y
Year FE Y Y Y Y Y Y Y Y
Controls Y Y Y Y Y Y
Q× Controls Y Y Y Y
Q× Ind FE Y Y
Q× Year FE Y Y
Obs. 6,697 6,672 6,697 6,672 6,697 6,672 6,697 6,672
Adj.R
2
0.375 0.412 0.387 0.431 0.399 0.442 0.407 0.448
50
5.2 Customeraccountingquality
Prior literature has highlighted the important role played by accounting information in supply
chains (Bowen et al., 1995; Raman and Shahrur, 2008; Hui, Klasa, and Yeung, 2012; Bauer, Hen-
derson, and Lynch, 2018; Pandey and Subramanyam, 2020). By providing a summary measure
of customers’ fundamental performance, customers’ accounting information can serve as an al-
ternative signal of suppliers’ future demand and in turn provide investment-relevant signals.
Accordingly, customer accounting quality (AQ) can play a role in my setting and influence the
primary results in two different ways. First, suppliers may find alternative information sources,
such as stock prices and demand forecasts, incrementally less useful if they have access to high-
quality customer accounting information. Accordingly, the supplier’s reliance on stock price after
entering into a DF contract may not change as much as that for suppliers with low customer AQ,
because suppliers do not find prices or demand forecasts incrementally useful. On the other hand,
if customer AQ proxies for the quality of private demand forecasts disclosed by customers un-
der DF contracts, higher customer AQ suggests more useful demand forecasts and thus stronger
primary results. I leave the direction of the moderating effect of customer AQ to data.
I measure customers’ accounting quality using two methods: the accruals quality model of
Dechow and Dichev (2002) as modified in Francis, LaFond, Olsson, and Schipper (2005) and earn-
ings persistence (Dechow, Ge, and Schrand, 2010). Using the above measures, I first calculate
the accounting quality of customers. For each supplier, I then calculate the average customer
accounting quality, CUST_DDAQ (using Dechow and Dichev (2002)), and CUST_EARNPERSIST
51
(using earnings persistence), respectively (see Appendix B for detailed definitions). I classify cus-
tomer accounting quality above (below) cross-sectional median as high (low). I estimate eq. (4.1)
separately for high and low customer AQ subsamples.
Table 5.2, panel A, presents the results of customer AQ analysis usingCUST_DDAQ. In column
(1), I find that the coefficient for Q× POST × DF is negative and significant in low customer
AQ subsample (t-stat =− 5.14). However, in column (2) for high customer AQ subsample, the
coefficient on Q× POST × DF is of a much smaller magnitude and statistically insignificant
at conventional levels (t-stat =− 1.23). The difference in the coefficient on Q× POST × DF
between the high and low customer AQ subsamples is also significantly different from zero. I find
similar evidence across all specifications in columns (3)–(8), where I incrementally add controls
and interactions ofQ with controls, industry fixed effects, and year fixed effects (low subsample,
t-stat =− 4.70 to− 5.27; high subsample,t-stat =− 1.09 to− 1.99). In panel B, I find similar results
using theCUST_EARNPERSIST measure of customer AQ (low subsamplet-stat =− 3.84 to− 3.97;
high subsample t-stat =− 0.30 to− 0.80). Overall, these results are consistent with the notion
that suppliers with high customer AQ find other sources of investment signals incrementally less
informative when they’re making investment decisions. These results reinforce those from prior
studies: accounting information plays an important role in supply chains.
52
Table 5.2: Customer Accounting Quality
Table 5.2 examines whether the quality of customer’s accounting signals influences the effect of DF contracts on
investment-Q sensitivity. Panel A measures customer’s accounting quality using customer’s Dechow-Dichev mea-
sure of accrual quality (CUST_DDAQ). Panel B measures customer’s accounting quality using customer’s earnings
persistence (CUST_EARNPERSIST). High (Low) subsample splits are based on the customer accounting quality mea-
sures above (below) their respective median. p-values for the difference in the DiD coefficients between High and
Low subsamples are shown between respective columns. Main terms and all lower order interactions are included
but only DiD coefficients are shown for brevity. t-statistics in parentheses are based on standard errors clustered at
both firm and industry-year level. All continuous variables are winsorized at 1% and 99% of the yearly distributions
to reduce the influence of outliers. ***, **, and * indicate statistical significance (two-sided) at the 0.01, 0.05, and 0.10
levels, respectively. All variables are defined in Appendix B.
Panel A: Customer Dechow-Dichev Accruals Quality (CUST_DDAQ)
Dep Var =INV
t+1
(1) (2) (3) (4) (5) (6) (7) (8)
CUST_DDAQ = Low High Low High Low High Low High
Q× POST × DF − 0.049
∗∗∗ − 0.016 − 0.048
∗∗∗ − 0.014 − 0.056
∗∗∗ − 0.016 − 0.060
∗∗∗ − 0.022
∗∗ (− 5.14) (− 1.23) (− 5.27) (− 1.09) (− 5.11) (− 1.32) (− 4.70) (− 1.99)
p-value (High = Low) 0.02 0.02 <0.01 0.02
Firm FE Y Y Y Y Y Y Y Y
Year FE Y Y Y Y Y Y Y Y
Controls Y Y Y Y Y Y
Q× Controls Y Y Y Y
Q× Ind FE Y Y
Q× Year FE Y Y
Obs. 4,852 4,786 4,852 4,786 4,852 4,786 4,852 4,786
Adj.R
2
0.456 0.465 0.467 0.474 0.476 0.486 0.486 0.493
Panel B: Customer Earnings Persistence (CUST_EARNPERSIST)
Dep Var =INV
t+1
(1) (2) (3) (4) (5) (6) (7) (8)
CUST_EARNPERSIST = Low High Low High Low High Low High
Q× POST × DF − 0.036
∗∗∗ − 0.010 − 0.034
∗∗∗ − 0.004 − 0.036
∗∗∗ − 0.008 − 0.046
∗∗∗ − 0.007
(− 3.94) (− 0.80) (− 3.97) (− 0.30) (− 3.87) (− 0.48) (− 3.84) (− 0.42)
p-value (High = Low) 0.06 0.02 0.10 0.06
Firm FE Y Y Y Y Y Y Y Y
Year FE Y Y Y Y Y Y Y Y
Controls Y Y Y Y Y Y
Q× Controls Y Y Y Y
Q× Ind FE Y Y
Q× Year FE Y Y
Obs. 4,957 4,896 4,957 4,896 4,957 4,896 4,957 4,896
Adj.R
2
0.438 0.424 0.453 0.435 0.466 0.451 0.473 0.457
53
5.3 Informativenessofstockprices
Next, I examine whether informativeness of stock prices influence my primary results. While de-
mand forecasts contain signals of future demand from a specific customer, stock prices can con-
tain investor’s private signals on various other dimensions that are relevant for corporate invest-
ments: supplier’s total demand, industry dynamics, risk, etc. Therefore, higher informativeness
of stock prices can be due to the portion of information that is common or that is disjoint between
demand forecasts and prices. If prices are more informative due to more information that is not
found in demand forecast disclosures, I don’t expect supplier’s learning behavior from prices to
change significantly post-DF. Following prior literature, I measure informativeness of stock prices
for managers using probability of informed trading (PIN) and price non-synchronicity, as defined
in Appendix B (Chen et al., 2007; Jayaraman and Wu, 2019). Similar to previous cross-sectional
subsample tests, I estimate eq. (4.1) separately for high and low price informativeness subsamples
cut based on median.
Table 5.3, panel A, presents the results of price informativeness subsample analysis using
PIN. In the high PIN subsample in column (1), I find that the coefficient on Q× POST × DF is
negative and significant ( t-stat =− 2.81). However, the corresponding coefficient in the high PIN
subsample in column (2) is insignificant ( t-stat =− 0.20). The difference in the coefficients for
Q× POST × DF between the high and lowPIN subsamples is also significantly different from
zero. I find largely similar evidence across all specifications in columns (3)–(8) (low subsample
t-stat =− 2.58 to− 3.00; high subsample t-statistics =− 0.20 to 1.70). In panel B, I find similar
results using 1-R
2
measure of price informativeness (low subsample, t-stat =− 4.38 to− 5.07;
high subsample, t-stat = − 0.76 to − 1.94). These findings are consistent with the notion that
54
Table 5.3: Price Informativeness
Table 5.3 examines whether informativeness of stock price for managers influences the effect of DF contracts on
investment-Q sensitivity. Panel A and B measure price informativeness using probability of informed trading (PIN)
and price non-synchornicity (1-R
2
), respectively. High (Low) subsample splits are based on the informativeness
measures above (below) their respective median. p-values for the difference in the DiD coefficients between high
and low subsamples are shown between high and low columns. Main terms and lower order interactions are
included but only DiD coefficients are shown for brevity. t-statistics in parentheses are based on standard errors
clustered at both firm and industry-year level. All continuous variables are winsorized at 1% and 99% of the yearly
distributions to reduce the influence of outliers. ***, **, and * indicate statistical significance (two-sided) at the 0.01,
0.05, and 0.10 levels, respectively. All variables are defined in Appendix B.
Panel A: Probability of Informed Trading (PIN)
Dep Var =INV
t+1
(1) (2) (3) (4) (5) (6) (7) (8)
PIN= Low High Low High Low High Low High
Q× POST × DF − 0.025
∗∗∗ − 0.002 − 0.023
∗∗ − 0.005 − 0.029
∗∗∗ 0.008 − 0.029
∗∗∗ 0.014
∗ (− 2.81) (− 0.20) (− 2.58) (− 0.43) (− 3.00) (0.88) (− 2.64) (1.70)
p-value (High = Low) 0.09 0.21 <0.01 <0.01
Firm FE Y Y Y Y Y Y Y Y
Year FE Y Y Y Y Y Y Y Y
Controls Y Y Y Y Y Y
Q× Controls Y Y Y Y
Q× Ind FE Y Y
Q× Year FE Y Y
Obs. 5,123 5,104 5,123 5,104 5,123 5,104 5,123 5,104
Adj.R
2
0.413 0.499 0.436 0.511 0.450 0.528 0.457 0.539
Panel B: Price Non-Synchronicity (1-R
2
)
Dep Var =INV
t+1
(1) (2) (3) (4) (5) (6) (7) (8)
1-R
2
= Low High Low High Low High Low High
Q× POST × DF − 0.030
∗∗∗ − 0.018
∗ − 0.030
∗∗∗ − 0.018
∗ − 0.034
∗∗∗ − 0.009 − 0.036
∗∗∗ − 0.007
(− 4.38) (− 1.83) (− 4.64) (− 1.94) (− 4.63) (− 1.00) (− 5.07) (− 0.76)
p-value (High = Low) 0.29 0.27 0.03 <0.01
Firm FE Y Y Y Y Y Y Y Y
Year FE Y Y Y Y Y Y Y Y
Controls Y Y Y Y Y Y
Q× Controls Y Y Y Y
Q× Ind FE Y Y
Q× Year FE Y Y
Obs. 6,433 6,410 6,433 6,410 6,433 6,410 6,433 6,410
Adj.R
2
0.393 0.479 0.414 0.493 0.425 0.504 0.434 0.513
55
suppliers with more informative stock prices find lot of information in prices they cannot find in
demand forecast disclosures, and thus do not meaningfully reduce learning from prices post-DF.
5.4 Productdifferentiation
Next, I explore the role of product differentiation in my setting. Firms that make differentiated
or specialized products need to make investments specifically for their specialized products (e.g.,
Boccaletti and Cerasi, 2021). This can make such assets less useful for alternative uses. Irre-
versible investment decisions based on preliminary information in demand forecasts could be
more costly for such suppliers (e.g., Dixit and Pindyck (1994)). If this is the case, I expect suppli-
ers with more differentiated products to make less use of demand forecast disclosures. I measure
product differentiation using the total product similarity measure of Hoberg and Phillips (2016),
and multiply it with negative one so that the measure is increasing in product differentiation. I
estimate eq. (4.1) separately for high and low product differentiation subsamples cut based on
median.
Table 5.4 presents the results of product differentiation subsample analysis. I find that the
coefficient on Q× POST × DF in the low product differentiation subsample in column (1) is
negative and significant ( t-stat = − 4.27). However, the corresponding coefficient in the high
product differentiation subsample in column (2) is insignificant ( t-stat =− 0.08). The difference
in the coefficients for Q× POST × DF between the high and low product differentiation sub-
samples is also significantly different from zero. I find similar evidence across all specifications
in columns (3)–(8) (low subsample t-stat =− 4.09 to− 4.38; high subsample t-statistics =− 0.14
to− 0.34). These findings are consistent with the notion that suppliers with highly differentiated
56
products find making investments based on preliminary information in demand forecasts costly,
and thus their learning behavior does not change post-DF in an economically meaningful way.
Table 5.4: Differentiated Products
Table 5.4 examines whether the extent of supplier’s product differentiation influences the effect of DF contracts
on investment-Q sensitivity. Product differentiation is measured as total product similarity (Hoberg and Phillips,
2016) times negative one, so that the measure is increasing in product differentiation ( PROD_DIFFERENT). High
(Low) subsample splits are based on the product differentiation measure above (below) median. p-values for the
difference in the DiD coefficients between high and low subsamples are shown between high and low columns.
Main terms and lower order interactions are included but only DiD coefficients are shown for brevity. t-statistics
in parentheses are based on standard errors clustered at both firm and industry-year level. All continuous variables
are winsorized at 1% and 99% of the yearly distributions to reduce the influence of outliers. ***, **, and * indicate sta-
tistical significance (two-sided) at the 0.01, 0.05, and 0.10 levels, respectively. All variables are defined in Appendix B.
Dep Var =INV
t+1
(1) (2) (3) (4) (5) (6) (7) (8)
PROD_DIFFERENT = Low High Low High Low High Low High
Q× POST × DF − 0.032
∗∗∗ − 0.001 − 0.032
∗∗∗ − 0.002 − 0.033
∗∗∗ − 0.003 − 0.033
∗∗∗ − 0.004
(− 4.27) (− 0.08) (− 4.38) (− 0.14) (− 4.19) (− 0.21) (− 4.09) (− 0.34)
p-value (High = Low) 0.04 0.05 0.04 0.06
Firm FE Y Y Y Y Y Y Y Y
Year FE Y Y Y Y Y Y Y Y
Controls Y Y Y Y Y Y
Q× Controls Y Y Y Y
Q× Ind FE Y Y
Q× Year FE Y Y
Obs. 6,041 6,040 6,041 6,040 6,041 6,040 6,041 6,040
Adj.R
2
0.413 0.329 0.433 0.345 0.445 0.356 0.465 0.359
5.5 Lowerlearningfrompricesorlessefficientinvestments
Price is a market based signal of firm’s investment opportunities. Therefore, an alternative in-
terpretation of a decrease in investment-price sensitivity post-DF is that suppliers are ignoring
investment opportunities. I perform two tests to rule out this alternative explanation. First, I
argue that if suppliers are ignoring investment opportunities post-DF, I should find a decrease in
57
sensitivity of investment to other non-price measures of investment opportunities. I employ two
common non-price measures of investment opportunities - cash flow from operations and sales
growth (Badertscher, Shroff, and White, 2013; Jayaraman and Wu, 2019; Edmans et al., 2017).
5.5.1 Sensitivitytonon-pricemeasuresofinvestmentopportunities
Table 5.5, panel A, presents the results of investment sensitivity to cash flow from operations
(CFO). I augment the eq. (4.1) by exhaustively including interactions of CFO with DF and POST.
The coefficient for CFO× POST × DF indicates the diff-in-diff coefficient for the investment-
cfo sensitivity. Across columns (1)-(4), consistent with the primary result in Table 4.1, I con-
tinue to find negative and significant coefficient for Q× POST × DF . Importantly, the co-
efficient for CFO× POST × DF is positive and largely statistically significant at 10% level
or better (t-stat = 1.73 to 2.59). In panel B, I find qualitatively similar results when using sales
growth as the non-price measure of investment opportunities. In particular, the coefficient for
SALEGROWTH× POST × DF is positive in all specifications and statistically signigicant in
three out of four specifications ( t-stat = 1.49 to 2.08). This evidence is inconsistent with suppliers
ignoring investment opportunities post-DF, and in fact suggests that suppliers respond more to
investment opportunities post-DF.
5.5.2 Futurefirmperformance
In the second test, I argue that if suppliers are making less efficient investments post-DF, we
should observe poorer firm performance in the future. I measure firm performance using cash
58
Table 5.5: Sensitivity of Investment to Non-Price Measures of Investment Opportunities
Table 5.5 examines whether supplier’s corporate investment becomes less sensitive to non-price measures of
investment opportunities after they enter in demand forecast disclosure contracts. Investment opportunities are
measured using cash from operations (CFO) in panel A and sales growth (SALEGROWTH) in panel B. All lower
order interactions are included but only DiD coefficients are shown for brevity. Controls does not include CFO
(SALEGROWTH) in panel A (B). t-statistics in parentheses are based on standard errors clustered at both firm and
industry-year level. All continuous variables are winsorized at 1% and 99% of the yearly distributions to reduce
the influence of outliers. ***, **, and * indicate statistical significance (two-sided) at the 0.01, 0.05, and 0.10 levels,
respectively. All variables are defined in Appendix B.
Panel A: Investment-CFO sensitivity
(1) (2) (3) (4)
INV
t+1
INV
t+1
INV
t+1
INV
t+1
Q× POST × DF − 0.024*** − 0.024*** − 0.024*** − 0.022***
(− 4.12) (− 4.06) (− 3.98) (− 3.29)
CFO× POST × DF 0.163** 0.133** 0.098* 0.104*
(2.59) (2.19) (1.73) (1.83)
Firm FE Y Y Y Y
Year FE Y Y Y Y
Controls Y Y Y
Q× Controls, CFO× Controls Y Y
Q× Ind FE, CFO× Ind FE Y
Q× Year FE, CFO× Year FE Y
Obs. 13,369 13,369 13,369 13,369
Adj.R
2
0.407 0.422 0.435 0.443
Panel B: Investment-Sales growth sensitivity
(1) (2) (3) (4)
INV
t+1
INV
t+1
INV
t+1
INV
t+1
Q× POST × DF − 0.029
∗∗∗ − 0.028
∗∗∗ − 0.028
∗∗∗ − 0.028
∗∗∗ (− 4.74) (− 4.67) (− 4.30) (− 4.01)
SALEGROWTH× POST × DF 0.071
∗∗ 0.064
∗ 0.050 0.057
∗ (2.08) (1.84) (1.49) (1.66)
Firm FE Y Y Y Y
Year FE Y Y Y Y
Controls Y Y Y
Q× Controls, SALEGROWTH × Controls Y Y
Q× Ind FE, SALEGROWTH × Ind FE Y
Q× Year FE, SALEGROWTH × Year FE Y
Obs. 13,369 13,369 13,369 13,369
Adj.R
2
0.407 0.422 0.434 0.440
59
flow from operations and return on assets. Table 5.6 presents the difference-in-differences esti-
mates of the future firm performance. Specification in Column (1) measures firm performance
usingCFO and includes firm and year fixed effects. I find that diff-in-diff coefficient, POST× DF ,
is positive and statistically significant. I find similar evidence in Columns (2)-(3) that addition-
ally include the vector of control variables from eq. (4.1), and more granular industry-year fixed
effects instead of year fixed effects ( t-stat = 2.16 to 2.76). In columns (4)-(6), I continue to find sim-
ilar evidence using ROA as a measure of firm performance ( t-stat = 2.38 to 3.13). Economically,
the estimate in Column (3) equates to 0.142 standard deviations improvement in future cash flow
from operations post-DF (= 0.024/0.169). Similarly, the estimate in Column (6) equates to 0.195
standard deviations improvement in future return on assets post-DF (= 0.039/0.200). The signif-
icant improvement in future firm performance seems inconsistent with suppliers making more
inefficient investments post-DF. Overall, evidence in Tables 5.5 and 5.6 fails to support the alter-
native interpretation that lower investment-price sensitivity post-DF indicates suppliers making
more inefficient investment decisions.
60
Table 5.6: Future Firm Performance
Table 5.6 examines future firm performance. t-statistics in parentheses are based on standard errors clustered at
both firm and industry-year level. All continuous variables are winsorized at 1% and 99% of the yearly distributions
to reduce the influence of outliers. ***, **, and * indicate statistical significance (two-sided) at the 0.01, 0.05, and 0.10
levels, respectively. All variables are defined in Appendix B.
(1) (2) (3) (4) (5) (6)
CFO
t+3
CFO
t+3
CFO
t+3
ROA
t+3
ROA
t+3
ROA
t+3
POST × DF 0.032
∗∗∗ 0.032
∗∗∗ 0.024
∗∗ 0.051
∗∗∗ 0.053
∗∗∗ 0.039
∗∗ (2.76) (2.81) (2.16) (2.96) (3.13) (2.38)
POST − 0.007
∗∗ − 0.008
∗∗ − 0.008
∗∗ − 0.007 − 0.007 − 0.008
(− 1.97) (− 2.12) (− 2.11) (− 1.36) (− 1.44) (− 1.62)
Q 0.002 0.002 0.002 0.002
(0.93) (0.84) (0.58) (0.48)
SIZE − 0.002 − 0.002 − 0.011
∗∗ − 0.011
∗∗ (− 0.71) (− 0.57) (− 2.58) (− 2.51)
SALEGROWTH 0.012
∗ 0.012
∗ 0.012 0.013
(1.83) (1.75) (1.32) (1.33)
LEV 0.025 0.025 0.018 0.021
(1.40) (1.35) (0.84) (0.98)
VOL(SALES) − 0.007 − 0.007 0.010 0.012
(− 0.62) (− 0.57) (0.63) (0.74)
VOL(CFO) − 0.022 − 0.023 0.010 0.006
(− 1.13) (− 1.19) (0.39) (0.22)
Firm FE Y Y Y Y Y Y
Year FE Y Y Y Y
Ind× Year FE Y Y
Obs. 12,583 12,564 12,564 12,597 12,578 12,578
Adj.R
2
0.612 0.612 0.611 0.546 0.547 0.546
61
Chapter6
Conclusions
While the majority of the voluntary disclosure literature focuses on disclosures made by firms to
capital markets, I focus on private disclosure contracts between suppliers and customers. I exam-
ine how DF contracts influence suppliers’ investment decisions by examining how DF contracts
affect suppliers’ reliance on stock prices, another important forward-looking investment-relevant
information source. Using a hand-collected novel data on DF contracts, I hypothesize and find
strong evidence that suppliers reduce their reliance on stock prices in guiding their investment
decisions after entering into a DF contract. Consistent with the unique role of DF contracts in
addressing the credibility of private disclosures, these results are stronger for DF contracts with
binding forecasts.
I contribute to the nascent literature on private disclosure contracts by examining the real
effects of private disclosure contracts between suppliers and customers. My results suggest that
private disclosure contracts significantly affect suppliers’ investment decisions. I also contribute
to the literature on managerial learning about real decisions from stock prices. By showing that
managerial learning from stock prices is reduced in the presence of a rich alternative information
source about future demand, my results highlight a specific type of information managers can
62
learn from stock prices, namely, future demand, a notion suggested in theory but hitherto lacking
direct empirical evidence. Finally, I contribute to the large literature on private information flows
in supply chains by collecting a large, novel sample of DF contracts and providing systematic
empirical evidence on their real effects.
63
References
Angrist, J. D., Pischke, J.-S., 2008. Mostly harmless econometrics. Princeton university press.
Armony, M., Plambeck, E. L., 2005. The impact of duplicate orders on demand estimation and
capacity investment. Management Science 51, 1505–1518.
Asker, J., Ljungqvist, A., 2010. Competition and the structure of vertical relationships in capital
markets. Journal of Political Economy 118, 599–647.
Badertscher, B., Shroff, N., White, H. D., 2013. Externalities of public firm presence: Evidence
from private firms’ investment decisions. Journal of Financial Economics 109, 682–706.
Baiman, S., Rajan, M. V., 2002. The role of information and opportunism in the choice of buyer-
supplier relationships. Journal of Accounting Research 40, 247–278.
Bakke, T. E., Whited, T. M., 2010. Which firms follow the market? An analysis of corporate
investment decisions. Review of Financial Studies 23, 1941–1980.
Bauer, A. M., Henderson, D., Lynch, D. P., 2018. Supplier Internal Control Quality and the Dura-
tion of Customer-Supplier Relationships. The Accounting Review 93, 59–82.
Beyer, A., Cohen, D. A., Lys, T. Z., Walther, B. R., 2010. The financial reporting environment:
Review of the recent literature. Journal of Accounting and Economics 50, 296–343.
Boccaletti, S., Cerasi, V., 2021. Liquidation value of productive assets and product differentiation.
University of Milan Bicocca Department of Economics, Management and Statistics Working
Paper .
Bond, P., Edmans, A., Goldstein, I., 2012. The real effects of financial markets. Annual Review of
Financial Economics 4, 339–360.
Bowen, R. M., DuCharme, L., Shores, D., 1995. Stakeholders’ implicit claims and accounting
method choice. Journal of accounting and economics 20, 255–295.
Brown, S., Hillegeist, S. A., 2007. How disclosure quality affects the level of information asymme-
try. Review of Accounting Studies 2007 12:2 12, 443–477.
Bushee, B. J., Kim-Gina, J., Leung, E., 2020. Public and Private Information Channels along Supply
Chains: Evidence from Contractual Private Forecasts. Working Paper. .
Call, A. C., Martin, G. S., Sharp, N. Y., Wilde, J. H., 2018. Whistleblowers and outcomes of financial
misrepresentation enforcement actions. Journal of Accounting Research 56, 123–171.
64
Cameron, A. C., Miller, D. L., 2015. A practitioner’s guide to cluster-robust inference. Journal of
human resources 50, 317–372.
Carrizosa, R., Ryan, S. G., 2017. Borrower private information covenants and loan contract mon-
itoring. Journal of Accounting and Economics 64, 313–339.
Cattani, K., Hausman, W., 2000. Why Are Forecast Updates Often Disappointing? Manufacturing
& Service Operations Management 2, 119.
Chen, Q., Goldstein, I., Jiang, W., 2007. Price informativeness and investment sensitivity to stock
price. Review of Financial Studies 20, 619–650.
Chu, L. Y., Shamir, N., Shin, H., 2017. Strategic Communication for Capacity Alignment with
Pricing in a Supply Chain. Management Science 63, 4366–4388.
Cohen, M. A., Ho, T. H., Ren, Z., Terwiesch, C., 2003. Measuring Imputed Cost in the Semicon-
ductor Equipment Supply Chain. Management Science 49, 1653–1670.
Costello, A. M., 2013. Mitigating incentive conflicts in inter-firm relationships: Evidence from
long-term supply contracts. Journal of Accounting and Economics 56, 19–39.
Crawford, V. P., Sobel, J., 1982. Strategic Information Transmission. Econometrica 50, 1431–1451.
Dechow, P., Ge, W., Schrand, C., 2010. Understanding earnings quality: A review of the proxies,
their determinants and their consequences. Journal of accounting and economics 50, 344–401.
Dechow, P. M., Dichev, I. D., 2002. The Quality of Accruals and Earnings: The Role of Accrual
Estimation Errors. The Accounting Review 77, 35–59.
Dessaint, O., Foucault, T., Frésard, L., Matray, A., 2019. Noisy stock prices and corporate invest-
ment. Review of Financial Studies 32, 2625–2672.
Dixit, A., Pindyck, R., 1994. Investment under uncertainty. Princeton University Press .
Dow, J., Gorton, G., 1997. Stock Market Efficiency and Economic Efficiency: Is There a Connec-
tion? The Journal of Finance 52, 1087–1129.
Durnev, A., Morck, R., Yeung, B., 2004. Value-enhancing capital budgeting and firm-specific stock
return variation. The Journal of Finance 59, 65–105.
Dye, R. A., Sridhar, S., 2002. Resource Allocation Effects of Price Reactions to Disclosures. Con-
temporary Accounting Research 19, 385–410.
Edmans, A., Goldstein, I., Jiang, W., 2015. Feedback Effects, Asymmetric Trading, and the Limits
to Arbitrage. American Economic Review 105, 3766–97.
Edmans, A., Jayaraman, S., Schneemeier, J., 2017. The source of information in prices and
investment-price sensitivity. Journal of Financial Economics 126, 74–96.
Foucault, T., Frésard, L., 2012. Cross-listing, investment sensitivity to stock price, and the learning
hypothesis. Review of Financial Studies 25, 3305–3350.
65
Foucault, T., Fresard, L., 2014. Learning from peers’ stock prices and corporate investment. Journal
of Financial Economics 111, 554–577.
Francis, J., LaFond, R., Olsson, P., Schipper, K., 2005. The market pricing of accruals quality.
Journal of accounting and economics 39, 295–327.
Gallemore, J., Labro, E., 2015. The importance of the internal information environment for tax
avoidance. Journal of Accounting and Economics 60, 149–167.
Gao, P., Liang, P. J., 2013. Informational feedback, adverse selection, and optimal disclosure policy.
Journal of Accounting Research 51, 1133–1158.
Glaeser, S., 2018. The effects of proprietary information on corporate disclosure and transparency:
Evidence from trade secrets. Journal of Accounting and Economics 66, 163–193.
Goldstein, I., Liu, B., Yang, L., 2021. Market Feedback: Evidence from the Horse’s Mouth. Available
at SSRN 3874756 .
Goldstein, I., Yang, L., 2019. Good disclosure, bad disclosure. Journal of Financial Economics 131,
118–138.
Goldstein, I., Yang, S., Zuo, L., 2020. The real effects of modern information technologies. Working
Paper 27529, National Bureau of Economic Research.
Goodman, T. H., Neamtiu, M., Shroff, N., White, H. D., 2014. Management forecast quality and
capital investment decisions. The Accounting Review 89, 331–365.
Gow, I. D., Ormazabal, G., Taylor, D. J., 2010. Correcting for cross-sectional and time-series de-
pendence in accounting research. The accounting review 85, 483–512.
Ha, A. Y., Tang, C. S. (eds.), 2017. Handbook of Information Exchange in Supply Chain Manage-
ment, vol. 5 ofSpringerSeriesinSupplyChainManagement. Springer International Publishing.
Healy, P. M., Palepu, K. G., 2001. Information asymmetry, corporate disclosure, and the capital
markets: A review of the empirical disclosure literature. Journal of Accounting and Economics
31, 405–440.
Heitzman, S., Huang, M., 2019. Internal Information Quality and the Sensitivity of Investment to
Market Prices and Accounting Profits. Contemporary Accounting Research 36, 1699–1723.
Hoberg, G., Phillips, G., 2016. Text-based network industries and endogenous product differenti-
ation. Journal of Political Economy 124, 1423–1465.
Hosmer, D. W., Lemeshow, S., 2000. Applied logistic regression. John Wiley & Sons.
Hui, K. W., Klasa, S., Yeung, P. E., 2012. Corporate suppliers and customers and accounting con-
servatism. Journal of Accounting and Economics 53, 115–135.
Jayaraman, S., Wu, J., 2020. Should I stay or should I grow? Using voluntary disclosure to elicit
market feedback. The Review of Financial Studies 33, 3854–3888.
66
Jayaraman, S., Wu, J. S., 2019. Is silence golden? Real effects of mandatory disclosure. Review of
Financial Studies 32, 2225–2259.
Kang, J. K., Lennox, C., Pandey, V., 2022. Client concerns about information spillovers from shar-
ing audit partners. Journal of Accounting and Economics 73, 101434.
Kartik, N., 2009. Strategic communication with lying costs. The Review of Economic Studies 76,
1359–1395.
Kim, I., Skinner, D. J., 2012. Measuring securities litigation risk. Journal of Accounting and Eco-
nomics 53, 290–310.
Lee, H. L., Padmanabhan, V., Whang, S., 1997. Information distortion in a supply chain: The
bullwhip effect. Management science 43, 546–558.
Ludwikowski, M., Sjoberg, W., 2021. Semiconductor shortage and the U.S. auto industry. Reuters
.
Luo, Y., 2005. Do insiders learn from outsiders? Evidence from mergers and acquisitions. Journal
of Finance 60, 1951–1982.
McClure, C., Shi, S. X., Watts, E. M., 2020. Disclosure Processing Costs and Market Feedback
Around the World. Chicago Booth Research Paper .
McLean, S., 2021. Toyota to Cut Output as Chip Shortage Finally Catches Up to It. Wall Street
Journal .
Mian, A., Sufi, A., 2014. What Explains the 2007–2009 Drop in Employment? Econometrica 82,
2197–2223.
Mussa, M., Rosen, S., 1978. Monopoly and product quality. Journal of Economic theory 18, 301–
317.
Nagar, V., Schoenfeld, J., 2021. Shareholder monitoring and discretionary disclosure. Journal of
Accounting and Economics 72, 101422.
Oh, S., Özer, Ö., 2013. Mechanism Design for Capacity Planning Under Dynamic Evolutions of
Asymmetric Demand Forecasts. Management Science 59, 987–1007.
Oster, E., 2019. Unobservable Selection and Coefficient Stability: Theory and Evidence. Journal
of Business and Economic Statistics 37, 187–204.
Özer, Ö., Wei, W., 2006. Strategic commitments for an optimal capacity decision under asymmetric
forecast information. Management Science 52, 1238–1257.
Özer, Ö., Zheng, Y., Chen, K. Y., 2011. Trust in forecast information sharing. Management Science
57, 1111–1137.
Pandey, V., Subramanyam, K. R., 2020. In Customer We Trust: Supplier Investment when Cus-
tomer Misreports. Working Paper .
67
Petersen, M. A., 2009. Estimating Standard Errors in Finance Panel Data Sets: Comparing Ap-
proaches. The Review of Financial Studies 22, 435–480.
Pinto, J., 2020. Does reduced mandatory disclosure increase price efficiency. Working Paper .
Raman, K., Shahrur, H., 2008. Relationship-specific investments and earnings management: Evi-
dence on corporate suppliers and customers. Accounting Review 83, 1041–1081.
Roychowdhury, S., Shroff, N., Verdi, R. S., 2019. The effects of financial reporting and disclosure
on corporate investment: A review. Journal of Accounting and Economics 68, 101246.
Shamir, N., Shin, H., 2016. Public Forecast Information Sharing in a Market with Competing
Supply Chains. Management Science 62, 2994–3022.
Shroff, N., 2017. Corporate investment and changes in GAAP. Review of Accounting Studies 22,
1–63.
Subrahmanyam, A., Titman, S., 1999. The Going-Public Decision and the Development of Finan-
cial Markets. The Journal of Finance 54, 1045–1082.
Terwiesch, C., Ren, Z. J., Ho, T. H., Cohen, M. A., 2005. An Empirical Analysis of Forecast Sharing
in the Semiconductor Equipment Supply Chain. Management Science 51, 208–220.
Verrecchia, R. E., Weber, J., 2006. Redacted disclosure. Journal of Accounting Research 44, 791–
814.
Zuo, L., 2016. The informational feedback effect of stock prices on management forecasts. Journal
of Accounting and Economics 61, 391–413.
68
AppendixA
ExcerptsfromContracts
Demand Forecast Schedule between Astrazeneca and Ironwood Pharmaceuticals:
Demand Forecast Schedule between Abott and NeoGenomics:
69
AppendixB
VariableDefinitions
70
Variable Description
Primary Variables
INV Corporate Investment, defined as sum of capital expenditures, research
and developmen expenses, and acquisitions costs, scaled by lagged to-
tal assets.
Q Tobin’s Q, defined as total assets plus market value of equity minus
book value of equity, divided by total assets.
DF Indicator variable that equals one for supplier firms that enter in DF
contracts anytime in the sample period, zero otherwise.
POST Indicator variable that equals one (zero) after (before) DF firms enter
into their first DF contract. Non-DF firms take the value of POST cor-
responding to that of the DF firm to which they are matched. If the
effective date of the contract is not available, then I code the variable
based on the filing date of the contract.
All other variables
BINDING_DF An indicator variable that equals one for DF firms if the customer in the
DF contract is required to purchase at least a portion of the forecasted
demand, and zero otherwise.
CAPEX Capital expenditures scaled by lagged property, plant, and equipment.
CFO Cash flow from operations scaled by lagged total assets.
CUST_DDAQ Average customer accruals quality. I first estimate following accru-
als model for each FF30 industry-year with at least 20 observations in
Compustat universe (Francis et al., 2005): TCA
t
=α 0
+α 1
CFO
t− 1
+
α 2
CFO
t
+α 3
CFO
t+1
+α 4
∆ Rev
t
+α 5
PPE+ϵ t
; where TCA is cur-
rent accruals, CFO is cash flow from operations, REV is revenue, and
PPE is property, plant, and equipment. I calculate accruals quality for
each customer as the standard deviation of the residuals of the accru-
als model over years t-4 to t. For each supplier, I calculate the average
customer accruals quality as the average of accruals quality measure
for all the customers of the supplier over the last three years. I multi-
ply the measure by 1 so that customer accruals quality is increasing in
CUST_DDAQ.
71
Variable Description
CUST_EARNPERSIST Average customer earnings persistence. I calculate earnings persis-
tence for each customer as the coefficient on within-customer firm re-
gression of next period ROA on current ROA (i.e. β 1
from the regres-
sion ROA
t+1
= β 0
+β 1
ROA
t
+ϵ t+1
) using observations of the last
ten years. For each supplier, I calculate the average customer earnings
persistence as the average of the earnings persistence measure for all
the customers of the supplier over the last three years.
INVPRC Inverse price, defined as one divided by stock price at the end of the
fiscal year.
LEV Leverage, defined as long term debt plus short term debt divided by
total assets.
NON_BINDING_DF An indicator variable that equals one for DF firms if the customer in
the DF contract is not required to purchase a portion of the forecasted
demand, and zero otherwise.
PIN Probability of informed trading. (Downloaded from Stephen Brown’s
website)
PROD_DIFFERENT Product differentiation measured as total product similarity measure
of Hoberg and Phillips (2016) times negative one.
ROA Return on assets, defined as income before extraordinary items, scaled
by lagged total assets.
R&D Research and development expenses scaled by lagged total assets.
SALEGROWTH Change in sales scaled by lagged total assets.
SIZE Natural logarithm of market value of equity.
VOL(CFO) Volatility of cash flow from operations, defined as within-firm standard
deviation of CFO in the last five years.
VOL(SALES) Volatility of sales, defined as within-firm standard deviation of sales in
the last five years, scaled by average sales in the last five years.
VOL(EARN) Volatility of earnings, defined as within-firm standard deviation of ROA
in the last five years.
1R
2
Price non-synchronicity, defined as one minus R
2
, whereR
2
for each
firm-year is calculated from the regressions of daily stock returns on
CRSP value-weighted market returns.
72

Abstract (if available)

Abstract I examine whether private disclosure contracts that obligate the customer to periodically disclose forecasts of the customer’s future demand for the supplier's products (“DF contracts”) influence how the supplier makes investment decisions. I expect that if suppliers find private disclosures under DF contracts useful in making corporate investment decisions, they will reduce their dependence on other forward-looking sources of investment-relevant information. Motivated by the extensive literature that shows managers learn about investments from their own stock prices, I hypothesize that if suppliers use demand forecasts disclosed under DF contracts when making corporate investment decisions, they will reduce their reliance on stock prices after entering into such contracts. Using novel hand-collected data on DF contracts, I find that suppliers' corporate investments become significantly less sensitive to stock prices after suppliers enter into a DF contract for the first time, compared to that of control firms over the same time period. I substantiate my inference more directly by showing that the result is more pronounced for contracts with partially binding disclosures of demand forecasts. In additional cross-sectional analysis, I find that the primary result is largely non-existent for suppliers with less volatile demand, higher customer accounting quality, more informative stock prices, and highly differentiated products. Inconsistent with decrease in investment-price sensitivity indicating more inefficient investments post-DF, sensitivity of investment to non-price measures of investment opportunities do not decrease and future firm performance improves. Overall, my results suggest that suppliers use demand forecast disclosures when making corporate investment decisions and thereby reduce their reliance on learning about investments from stock prices.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Gone with the big data: institutional lender demand for private information

PDF

Disclosure distance and earnings announcement returns

PDF

Statistical learning in High Dimensions: Interpretability, inference and applications

PDF

Real-time controls in revenue management and service operations

PDF

Three essays on industrial organization

PDF

Efficient policies and mechanisms for online platforms

PDF

Essays on information design for online retailers and social networks

Asset Metadata

Creator Pandey, Vivek (author)

Core Title Private disclosure contracts within supply chains and managerial learning from stock prices

School Marshall School of Business

Degree Doctor of Philosophy

Degree Program Business Administration

Degree Conferral Date 2022-05

Publication Date 04/07/2022

Defense Date 03/09/2022

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

Tag Contracts,disclosure contracts,investments,learning from prices,OAI-PMH Harvest,private disclosure,real effects,voluntary disclosure

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Subramanyam, K.R. (committee chair), Lennox, Clive (committee member), Ramcharan, Rodney (committee member), Wittenberg-Moerman, Regina (committee member)

Creator Email vivekpandey1987@gmail.com,vpandey@marshall.usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC110886257

Unique identifier UC110886257

Document Type Dissertation

Format application/pdf (imt)

Rights Pandey, Vivek

Type texts

Source 20220408-usctheses-batch-920 (batch), 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 author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

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

Repository Email cisadmin@lib.usc.edu