Technological Change and Stock Return Volatility:
Evidence from eCommerce Adoptions
Deepak Agrawal∗
Sreedhar T. Bharath†
Siva Viswanathan‡
March 2003§
∗
K.M.V Corporation, 1620 Montgomery Street Suite 140 San Francisco, CA 94111
Assistant Professor, Finance Department, University of Michigan Business School, Room D6209, Davidson
Hall, 701 Tappan Street, Ann Arbor, MI 48109-1234. E-mail : sbharath@umich.edu.
‡
Assistant Professor, 4313, Van Munching Hall, Robert H. Smith School of Business, University of Maryland, College Park, MD-20742. E-mail : sviswana@rhsmith.umd.edu [Corresponding Address]
§
We thank M.P. Narayanan and Nagpurnanand Prabhala for helpful comments. All errors are our own.
Viswanathan acknowledges financial support from the Center for Electronic Markets and Enterprises, UMCP
†
Technological Change and Stock Return Volatility: Evidence
from eCommerce Adoptions
Abstract
This paper is among the first to use a unique controlled empirical setting - traditional
firms’ adoption of the Internet for commerce - to investigate the impact of changes in firms’
technological environment on their stock return volatility. Using three distinct empirical
methodologies we detect a significant and corresponding increase in the idiosyncratic and
total stock return volatility when a firm initiates eCommerce. Interestingly, this increase in
volatility is observed only for firms that moved online post-June 1998, a period when Internet growth reached critical mass. An increase in the implied volatility of at-the-money call
options of our sample firms reinforces our findings. We find that this increase in volatility
is attributable to changes in the firms’ product markets, specifically increased demand uncertainty, resulting from the adoption of a new technology-driven channel. Relevant controls
rule out firm-specific characteristics as well as market microstructural factors as possible explanatory variables. We also find that this surge in volatility is accompanied by a positive
abnormal return of stock prices. Overall, our results provide strong evidence of the impact
of real activity within a firm on its stock return volatility and highlight the importance of
understanding changes in firms’ technological environment.
Keywords: Technological Change, Stock Return Volatility, Product Markets, Structural
Break Analysis, Event Studies.
JEL Classification: D80, G12, G14, O33
1
1
Introduction
The last few decades have witnessed significant technological developments, the most recent
and the most revolutionary being the advent of the Internet. Financial markets have also
been affected by these developments, particularly with technology firms increasingly dominating these markets. For instance, it is widely believed that the dominance of technology
firms has made stock market returns more volatile; but the evidence in this regard has been
largely circumstantial. Two notable exceptions in this context are the recent work by Schwert
(2002) and Campbell et al (2001). Schwert (2002) identifies ’technological factors’ common
to computer, biotechnology as well as telecommunication portfolios, rather than firm size
or age, as the primary driver of the observed increase in Nasdaq volatility in the post-June
1998 period. Campbell et al (2001) also attribute the significant increase in the idiosyncratic volatility of several industries that they observe, to the emergence of new technologies.
However, what is it about technology that affects firms’ volatility, is a question that remains
unanswered. This study is an attempt in that direction.
This study seeks to examine the relationship between technological change and changes
in a firm’s stock return volatility, and also investigate the drivers of these changes. Rather
than examining firms in the technology sector, we look at a sample of traditional ’brickand-mortar’ firms that have embraced a new technological environment, the Internet, for
commerce. The adoption of eCommerce by a traditional firm provides a ’clean’ natural
experimental setting to examine the impact of changes in a firm’s technological environment
on its stock return volatility. As we report in detail later, we do find evidence of a definitive
and substantial increase in both idiosyncratic as well as total firm volatility when a firm
adopts the Internet for commerce. To the best of our knowledge, this is among the first
papers to study the relationship between stock return volatility and real activity within the
firm. Studying this linkage is important because changes in stock return volatility due to
firm-level decisions may affect among other things, the stock valuation, cost of capital, capital
structure, the firm’s ability to use its stock in acquisitions and pay-for performance policies.
Also, increased volatility could increase risk of default and consequently exacerbate agency
problems between stockholders and bondholders. The effects of stock volatility on these
2
issues are clearly of first order importance to the firm’s management and stake holders.
How does the adoption of the Internet by traditional firms change their technological
environment and where exactly lies the source of increased volatility? To take an extreme
example, while a company like Amazon.com (a pure-Internet retailer) is certainly a posterchild for the ’new economy’, a traditional book retailer like Barnes and Noble (a good example
of the type of firms we study here) is by no means considered a technology-oriented firm.
However, adopting a technological channel like the Internet for commerce, catapults a traditional firm like Barnes and Noble into a new environment, with characteristics shared by
several ’technology firms’. While the Internet enables traditional firms to complement their
existing marketing channels, its impact on firms has been more profound. First and foremost,
a traditional firm moving online is faced with a new competitive landscape dominated by
online pure-plays1 like EBay, Amazon.com and E*Trade, to name a few. The Internet also
enables traditional firms to offer customized products and services2 and alters their value
creation activities. In addition, lower costs of entry to many businesses and the erosion of
the traditional barriers of space and time differentiate the online channel from traditional
ones. The intense competition fuelled further by lower search costs for consumers, ease of
price comparison, and reduced switching costs, have forced traditional firms moving online
to realign their pricing and product positioning strategies (Viswanthan, 2002). Several traditional firms including books, CDs, grocery, and consumer electronics retailers, as well as
brokerages, real-estate and insurance services have reduced their prices in the online channel,
faced with competition from Internet firms3 . Thus, the initiation of eCommerce alters the
competencies of conventional firms and portends a transition from a ’low-tech’ to a ’high-tech’
environment.
As noted by Klein (1977), changes in technological regime result in greater demand uncer1
It is pertinent to note here that while online startups (’Internet pure-plays’) dominated the early phases
of eCommerce, traditional firms have established a significant presence online in several sectors including
apparel, books, music, computers, consumer electronics, brokerages(Bakos et al, 2003), and travel services.
2
P&G, Dell, Mattel, McGraw-Hill, Wells Fargo, and Nike are just a few of the diverse group of firms that
have used the Internet to provide highly customized offerings, not available through their traditional channels.
3
Brynjolfsson and Smith (2000) find that prices for books and CDs are 9-16% lower online compared to
traditional retailers. Brown and Goolsbee (2002) find that term life insurance prices fell by 8-15 percent from
1995 to 1997 due to the growth of online intermediaries.
3
tainty in the product markets due to a turbulent industry structure and changing consumer
tastes. In addition, new technologies trigger rampant experimentation, by both companies
and customers and more so in the initial stages of adoption and growth. Rapid technological
changes in the automobile industry in 1930’s as well as the PC industry in the 1980’s have
led to periods of industrial turbulence characterized by high entry and exit rates, as well
as rapidly falling prices (Mazzucato, 2002). Just as the ’competence-altering’ technological
developments in these industries led to market share instability, the adoption of the Internet,
a new technology-driven channel, increases turbulence in the product markets of traditional
firms. The reconfiguration of existing industry structure by the addition of a new marketing
channel to existing ones is thus expected to have a direct impact on the risk-return profile
of firms. These effects are more prominent in retailing sectors characterized by thin profit
margins. The adoption of eCommerce by traditional firms thus, provides us an excellent
setting in which to examine the link between real activity within the firm and stock return
volatility, a issue that has received little attention (Schwert, 2002).
In this paper, we study the effect of initiation of eCommerce by existing ’brick-and-mortar’
retailers on their total as well as idiosyncratic volatilities, after controlling for changes in
market volatility. We find a significant and long term increase in both volatility measures
associated with firms moving online. More interestingly, we find that this increase in volatility
occurs only for firms that moved online after the growth spurt in Internet related activities,
post-June 1998 - a phenomenon attributable to financial markets’ recognition of the potential
of the Internet to impact a traditional firm’s business. This post-June 1998 surge in firm-level
volatility parallels Schwert’s (2002) findings of an increase in market volatility of technology
intensive firms after mid-1998.
We use three distinct methodologies to examine the impact of the event on both the total
(systematic and idiosyncratic) as well as idiosyncratic volatility of a firm. The first method
involves comparing the post-event average volatility ratio to the pre-event average volatility
ratio4 . Using this methodology, we find that firm-level volatility (after controlling for market
volatility) increases significantly immediately following the event. The second method applies
4
Volatility ration refers to the ratio of firm volatility to market volatility and is a simple device to control
for any changes in firm volatility due to changes in market volatility rather than the event.
4
the structural break analysis technique of Bai, Lumsdaine and Stock, (1998)5 to detect a break
in the total volatility series around the event. We find that the break occurs around the event
date, confirming our earlier results. The third method is a formal volatility event study and
closely follows the methodology of Hilliard and Savickas (2000). This analysis shows that the
event of announcing an eCommerce initiative leads to an unambiguous and sharp increase in
the idiosyncratic volatility of firms in our sample.
We then examine the source of this increased volatility and also perform a series of
robustness checks. We postulate that increased demand uncertainty (stemming from changes
to the firm’s technological environment) results in uncertain profit opportunities leading to
increases in stock return volatility. However, it is possible that that the increase in volatility
around the event date that we uncover is related to other firm specific characteristics such as
firm size, historical stock price performance and historical volatility, rather than to product
market initiatives. In order to control for these effects we construct two matched samples.
The first sample consists of firms matched on size and historical stock price performance
while, the second sample has firms matched on size and historical volatility performance.
We examine the volatility behavior of the two matched samples and find that firms in the
matched samples do not experience the increase in volatility that we document for our test
sample. This suggests that firm specific factors are not likely to be the driving influences of
our results.
A large body of empirical literature documents a positive relation between trade size or
number of transactions and stock return volatility (See Karpoff, 1987 for a survey). In order
to ensure that our results are not simply an artifact of increased trading volume and number
of transactions, we compare the post-event to pre-event trading volume and number of transactions, for all the three samples. We find that while all three samples experience similar
increases in their trading activity measures over time, their volatility increases are substantially different, as noted earlier. Thus, we conclude that the observed volatility increases in
our test sample cannot be attributed to market microstructure variables.
A second robustness check is to use a completely different metric of volatility and see if
our results survive. Following Schwert(1990), we use the implied volatility of at-the-money
5
This method is also used by Bekaert et al. (2002) to date the integration of world equity markets
5
call options as an alternative volatility measure. We track the monthly series of these implied
volatilities for firms in our test sample over a period of 4 years around our the event dates.
Many studies have shown that close-to-the-money option prices convey the most information
about the expectations of the options market concerning future volatility (Day and Lewis,
1988). We find that implied volatility increases after the event for firms in our test sample,
confirming our earlier results. We substantiate our earlier claim that these volatility increases
are driven by increased demand uncertainty in the product markets by an empirical analysis
of this relationship. We construct two alternative measures of demand uncertainty (based on
quarterly sales data of sample firms in the 1990-2002 period) and find that these measures
are strongly associated with increases in stock return volatility. Based on this cumulative
evidence, we conclude that increased demand uncertainty in firms’ product markets stemming
from changes in their technological environment due to eCommerce initiatives is the driving
force behind the observed volatility increase.
We also perform a traditional event-study analysis controlling for event-induced variance,
and find that the firms under consideration also experience significant positive abnormal
returns attributable to their adoption of eCommerce. More interestingly, we find that these
positive abnormal returns only for firms that moved online after June 1998, a finding that is
similar to the results of our earlier volatility analysis.
The rest of the paper is organized as follows. In Section 2, we review the literature directly
related to this study. In section 3, we briefly note the growing importance of eCommerce
in today’s economy. We state our null and alternative hypotheses and advance economic
arguments to support them. Section 4 describes our test data sample. Section 5 describes
the empirical methodologies used. Section 6 describes the main results. Section 7 concludes.
2
Literature Review
Our study is primarily concerned with firm-level (total and idiosyncratic) volatility6 . Total stock return volatility can basically be decomposed into two components: systematic
volatility - the component of volatility that can be explained by common market or industry6
A related stream of research (for instance, see French, Schwert and Stambaugh (1987) and Schwert (1990))
examines aggregate stock market volatility and finds that aggregate volatility varies over time.
6
specific factors; and unsystematic or firm-level idiosyncratic volatility. Firm-level idiosyncratic volatility, in particular, is important for large holders of individual stocks (who are
restricted from diversifying), and arbitrageurs who trade to exploit the mis-pricing of an
individual stock. Firm-level volatility is also important for pricing options and in event studies, as the significance of abnormal event-related returns is determined by the volatility of
individual stock returns relative to the market or industry (Campbell et al., 2001).
As noted by Campbell et al.(2001), there is surprisingly little research on firm-level volatility, in contrast to research on market volatility. Their study documents a dramatic increase
in firm-level idiosyncratic volatility over the last three decades (compared to industry and
market volatility), phenomena they believe is attributable to technological changes. However, they highlight the need for a more detailed investigation of the drivers of increased
volatility. They also note that most volatility studies present a statistical description rather
than a structural economic model. In keeping with these observations, our paper focuses primarily on the impact of firms’ adoption of eCommerce, an economically significant event, on
total and firm-specific stock volatility. Also, as noted earlier our study complements Schwert (2002), who finds that the value-weighted portfolio of Nasdaq stocks shows unusually
high volatility since mid-19987 . Of particular interest, is Schwert’s (2002) conclusion that
the unusual volatility of large Nasdaq firms since mid-1998 is attributable to the fact that
these firms belong to the technology sector, rather than to firm size. Our paper adds to
these findings, by identifying changes in firms’ product markets (that stem from changes in
the firms’ technological environment) as being the primary source of the increased volatility.
The results of our study are also in line with the recent findings of Mazzucato (2002), that
stock prices are the most volatile during periods of high market-share instability and radical
technological change using data from automobile and Computer industries. Likewise, the
adoption of eCommerce with its potential to radically change the technological environment
of well-established traditional firms and consequently their product markets, also significantly
affects their financial volatility.
Our paper is also closely related to a number of studies that examine the impact of various
7
This finding contradicts earlier research, where smaller NYSE stocks were found to be more volatile than
larger NYSE stocks.
7
corporate events on firms’ stock return volatility. Clayton et al (2000) analyze the impact
of CEO turnover on firm’s return volatility and find that forced turnovers lead to greater
volatility increases compared to voluntary ones. In contrast to the traditional volatility
studies’ focus on capital structure or management changes within the firm, our primary
concern here is the effect of changes in a firm’s product market resulting from changes to its
technological environment, on its risk-return profile. Mazzucato and Semmler (1999) consider
changes in industry life-cycle in the US auto industry and relate it to stock price volatility.
In particular they relate changes in market shares of firms aggregated at the industry level
to the excess volatility implied by a dividend discount model and find them to be correlated.
Another related stream of research is on the interaction between financial structure and
product market (real activity) decisions, an issue that has received little attention. Recent
theoretical models by Brander and Lewis (1986), Maksimovic (1990,1995), Bolton and Scharfstein (1990) have formalized ways in which industry product markets may be influenced by
corporate financial decisions. Phillips(1995), and Chevalier (1994) have also attempted to
establish the relationship between financial structure choices of firms and their corresponding product market outcomes at the industry level. Demers and Lewellen (2003) show the
marketing benefits of IPO underpricing for a sample of Internet firms. These papers are
concerned with the effect of financial decisions on product markets. This paper complements
this growing body of literature by providing evidence of the reverse interaction - viz. the
influence of product market decisions of firms (the adoption of eCommerce) on their financial
structure, more specifically on their stock return volatility . In particular, we use the advent
of eCommerce as an unique exogenous event that impacts traditional firms in multiple ways.
The primary issue in the context of asset price volatility relating to the Internet and traditional brick-and-mortar firms, is “excess volatility”. Excess volatility is generally defined
as the volatility in prices that cannot be explained by the fundamentals alone. It is traditionally considered synonymous with the market irrationality. Shiller (1981) and Black (1986)
are two notable papers in this area. A large empirical literature followed and it broadly
concludes that volatility is primarily driven by trading. In particular, changes in trading
patterns, increasing institutionalization of equity ownership, increase in day-trading, among
other factors affecting investors’ discount rates are also believed to influence idiosyncratic
8
volatility (Campbell et al, 2001). In contrast, this paper argues that fundamental factors
such as adoption of a new technology-driven marketing channel, that increase uncertainty
in product markets, rather than market microstructure variables, can also lead to increased
volatility. This rational explanation for increased volatility has not been empirically analyzed
before.
Finally, another related branch of literature is one that studies the impact of information
events in the Internet sector on asset prices. Most of the existing studies in this context
are returns event-studies i.e., they examine abnormal returns attributable to specific events.
While we find that firms in our sample experience significant positive abnormal returns when
they move online, ours is the first study to investigate the impact of moving online on the
firms’ stock price volatility. Cooper, Dimitrov and Rau (2001) investigate the stock price
response to an event in which a firm just changes its name to an Internet related name such
as a “dotcom”. They find that relative to an Internet matched portfolio, these firms earn, on
average, 53% excess returns around the event. A few papers, e.g. Conell and Liu (2000) and
Lamont and Thaler (2000) look at the stock price response to equity carveouts where the
subsidiary is a public company in the “new economy” (i.e. the Internet sector). They find
that in over 10 such cases, the marked to market value of the shares of the subsidiary held by
the parent exceeded the entire parent’s market value! Ofek and Richardson (2001) provide
a comprehensive survey of how the fundamentals and events in the Internet sector impact
asset prices. In addition to the common context of eCommerce, our study is related to Ofek
and Richardson (2001) to the extent that they investigate, among other areas, (1) volatility
of asset prices and (2) response of stock price to information-based events. Our study is also
related to Perotti and Rossetto (2000) who theoretically investigate the impact of demand
uncertainty in product markets on firm’s profits for the purpose of valuation of Internet
portal firms as a portfolio of entry options. However, a significant point of departure is that
most of the above Internet-related studies focus primarily on ’pure Internet’ firms. While
the dominance of pure Internet firms in the early phases of eCommerce generated a lot of
interest, we believe that the Internet’s impact on traditional brick-and-mortar firms is much
more profound. Consequently, we study traditional brick-and-mortar retailing firms that
have chosen to embrace the Internet as an additional marketing channel for their products
9
and services and the impact of such choices on their risk-return profiles.
3
The Hypotheses
While several online start-ups failed to live up to the hype, online sales have nevertheless
been growing steadily. According to the government Department of Census, online sales in
the US accounted for $5.5 billion, $9.4 billion and $11.2 billion in revenues in the fourth
quarter of 1999, 2000 and 2001 respectively, a 43% cumulative annual growth rate. Recent
surveys by Ernst and Young8 indicate that by the year 2005, online retailing would account
for more that 10% to 12% of sales in categories such as apparels, accessories and toys and as
much as 20% to 25% of total sales in categories such as books, music, software and consumer
electronics. Also, nearly two-thirds of consumers surveyed had purchased products online in
the last 12 months and more importantly, more than half of these purchases would normally
have been made in retail outlets.
Although online sales have been growing steadily, the launch of Internet retailing operations may or may not be a significant event for an existing traditional retailer. It can be
argued that the revenues from Internet retailing are likely to be just a fraction of the total
revenues from retail sales (more so in the early phases of eCommerce). Consequently, shareholders’ expected returns and risks may not be significantly affected by a traditional firm’s
adoption of eCommerce, and the event under consideration may be inconsequential to the
stock volatility, at least in the short run. Thus, our Null hypothesis in this study is as follows:
The commencement of eCommerce operations by a traditional firm has no significant impact
on its stock return volatility.
While Internet retailing is still in its infancy, it is expected to grow quickly keeping pace
with the rapid growth of the Internet and consequently affording new profit opportunities for
firms. However, as highlighted earlier online retailing in an emerging technology-driven channel is also fraught with risks. Although the volume of online retailing is small compared to
traditional retailing, even a small shift of sales to the online channel in the future is expected
to have a very significant impact on the revenues and profitability of traditional firms (Ernst
8
Source: Global Online Retailing: An Ernst and Young Special Report, 2001.
10
and Young Global Online Retailing, 2001). As noted by Lusch (1995), retailers usually face
a break-even point of 85 to 92 percent of their sales which suggests that even a modest drop
in sales volume due to increased competition online, a retailer would incur significant losses.
Thus the adoption of eCommerce places traditional firms in a new and uncertain environment, exposing them to fresh competitive forces9 . This leads to our alternative hypothesis
viz. launching of eCommerce operations is a significant event in a firm’s life because (i) it
places the firm in a new environment that offers a lot of potential for growth and (ii) it
increases risks stemming from increased competition and price-wars. This implies that the
adoption of eCommerce should have an immediate impact on stocks’ expected returns and
risks, as shareholders re-evaluate the firm’s risk-return tradeoff.
We thus postulate that adopting a new technology-driven channel and operating in new,
unexplored markets is associated with higher uncertainty of product demand. Higher demand
uncertainty leads to higher variance of profits and higher perceived risk by the stockholders,
resulting in higher volatility of stock returns. A short formal derivation (Refer Appendix
A) illustrates how increased demand uncertainty in the product markets leads to a higher
volatility of profits.
To sum up, there are two competing hypotheses about the impact of Internet-retailing
on the firm’s stock price volatility and their resolution is an empirical matter. We use a
combination of event study methodology, structural break methodology and the study of
event effects on unsystematic volatility to examine this question.
4
Data
An event in the context of this study is the announcement of an online retailing initiative by a
traditional brick-and-mortar firm. The online retailing initiative is defined as the launch of a
Web-site that enables consumers to conduct online retail transactions. Thus, we disregard the
firms that launched just an informational Web-site without any transaction capabilities. The
9
As noted by Porter(2001), the paradox of the Internet is that its very benefits - making information widely
available; reducing the difficulty of purchasing, marketing and distribution; allowing buyers and sellers to find
one another and transact business with one another more easily - also make it more difficult for existing
companies to capture those benefits as profits, due to heightened competition.
11
announcement dates were collected from leading news sources viz. PR Newswire, Business
Wire, Hoover’s Online and the Lexis/Nexis database. All the firms included in the sample
are those whose stocks are publicly traded and listed on NYSE or NASDAQ. The sample
of firms so obtained was further restricted to satisfy several criteria. We excluded all firms
which did not have a two years history of publicly listed stock price prior to the event date.
We also omitted firms whose announcements coincided with other events with potentially
confounding effect on stock prices, e.g. earnings announcements or announcements about
alliances and mergers. Further, since the focus of the study is on business-to-consumer
segment of eCommerce activity as the end consumer of the product is clearly defined, we
omitted announcements regarding business-to-business eCommerce. Our final sample consists
of 166 firms with event dates spread over the years 1995 to 2000. The year-wise breakup of
these dates is given in Table I, Panel A. This table shows that event dates are sufficiently
spread out over time and that clustering of events is unlikely to be a significant issue in the
study. However, two thirds of the firms announced their online initiatives post June 1998.
Table I, Panel B shows the industry wide distribution of the firms in the sample. The largest
group of firms belong to the Computer Hardware and Consumer Electronics sector, followed
by Speciality Retailers. The daily stock price data for each firm in the sample is taken from
Center for Research in Security Prices (CRSP) daily database. Our proxy for the market
return is the equally weighted return on S&P500 index and the data for this index also comes
from CRSP files.10 In order to analyze the implied volatilities of at the money call options
for our sample firms, we use the Option Metrics database.
5
Empirical Methodology
In this paper, we use three distinct methodologies to detect if the event under consideration
has an impact on firm-specific volatility. The first methodology involves comparing the
pre-event and post-event volatilities of stock returns, after controlling for the overall market
volatility in the corresponding periods. Next, we use the methodology of Bai, Lumisdaine and
Stock (1998) to detect a structural break in the average volatility series. The average volatility
10
Using value weighted return on the index, produced qualitatively similar results
12
series is constructed by taking the cross-sectional average of stock volatilities in event time.
The formal test for structural break uses a Wald statistic to search for the break in the series
around event date zero, and is used to confirm the above event-study results. The third
methodology is designed to measure the impact of an event on the unsystematic volatility
of a firm. Similar to a traditional returns event-study, it involves estimating the parameters
of a market model of security returns over an estimation window. The market model is
augmented by assuming a parametric model for the evolution of volatility. The estimated
parameters are then used over an event window to determine the impact of the event on
unsystematic volatility. The following paragraphs describes these different methodologies.
5.1
Comparison of pre-event and post-event volatilities
Following the standard event study methodology, we realign the stock returns of the sample
firms in event time. We estimate the firm’s stock return volatility σi2 over two different
windows viz. (1) the pre-event window [−L, 0] and the (2) post-event window [0, L], where L
is the length of each window and 0 represents the event date. Four different window lengths
2 is
are considered (three months, six months, one year and two years). Market volatility σm
also estimated over the same two windows. The simplest way to estimate the volatilities is
by the sample standard deviation of daily returns over the relevant windows. This method of
estimating volatility of returns has been used in a number of studies, e.g. Schwert (1989).11 .
Next, we compute the volatility ratio λ, defined as,
s
λ=
σi2
2
σm
(1)
The impact of the event on firm volatility, after controlling for any changes in the market
volatility, can be studied by comparing the volatility ratio λ over the pre-event and post-event
windows. Similar to standard event studies, the firms in the sample are stacked together in
11
French, Schwert and Stambaugh (1987) consider an alternative estimator for volatility, which adjusts
for autocorrelation in returns that includes sum of the products of adjacent returns, and apply it on stock
portfolios. As they note, this modification has little effect on their results. However in our case if the
autocorrelation of individual stocks is less than -0.5, it causes the volatility estimate to become negative and
hence we use the simpler estimator. This effect never happens with portfolio volatility.
13
event time and the tests are then performed on cross-sectional average λs. This methodology
has been used in earlier volatility event studies. For instance, Clayton et al. (2000) use this
methodology to examine the impact of CEO turnover on equity volatility12 .
Any differences in the average λ between the pre-event and post-event windows can be
formally examined with a simple t-test or a non parametric Wilcoxon test. In addition, we
also confirm our results using a more sophisticated econometric technique viz. the test for
the location of structural break in a time series. The methodology for detecting a structural
break in a time-series was pioneered by Banerjee, Lumsdaine, and Stock (1992) (BLS1) and
Bai, Lumsdaine, and Stock (1998) (BLS2). These papers contain two key observations viz.
(i) that tests can be constructed to determine whether or not a structural break occurred
in a given time series, (ii) that confidence intervals can be computed enabling inference
about the break date. They demonstrate this for both stationary vector autoregressions and
cointegrated systems. The details of this methodology are provided in Appendix B.
Under our null hypothesis, there should be no structural break in the total volatility series
around the event date, while such a break will be expected under the alternative hypothesis.
In order to conduct this test, we calculate a time series of average monthly volatility,(defined
as the sample standard deviation of daily returns over the relevant month) for two years prior
to and after the event date. The firm volatilities are then averaged cross-sectionally for each
month in event-time. The BLS methodology is then used to identify a structural break in
the average volatility series.
5.2
Event Induced Unsystematic Volatility
We also study the event’s effect on the firm’s unsystematic volatility using the methodology
proposed in Hilliard and Savickas (2002) (hereafter HS). HS specify a market model for
security returns, to separate the systematic and unsystematic components of volatility. They
postulate the following diffusion processes for the instantaneous market return m and its
12
The volatilities themselves may be estimated in alternative ways. One possible approach is to use option
implied volatilities. Mayhew (1995) reviews event studies which use implied volatility. We present results using
implied volatility later. Another possible approach to volatility estimation is to use a parametric technique
like GARCH. We also repeated our analysis using GARCH(1,1) volatilities with very similar results.
14
volatility Vm ,
dm = µm dt +
p
Vm dZm ; dVm = (ωm − Θm Vm )dt + bm Vm dZvm
(2)
where dZm ∼ N (0, dt), dZvm ∼ N (0, dt) and Corr(dZm , dZvm ) = 0. The instantaneous
security return p of a firm is given by the market model,
dp = αdt + βdm +
p
Vε dZε , dVε = (ωε − Θε Vε )dt + bε Vε dZvε
where dZε ∼ N (0, dt), dZvε ∼ N (0, dt) and Corr(dZε , dZvε ) = 0.
(3)
√
Vε dZε is the unsystematic
volatility of the firm.
The estimation of model parameters follow traditional event studies. Unlike the usual
return models, the estimation of above volatility models is complicated by two factors (a)
the volatilities are unobserved and (b) the models are in continuous time, while the data
are observed in discrete time. HS overcome these problems by using a discrete stochastic
volatility model (a filter) that converges to the above continuous model as time interval
shrinks to zero. Their filter is based on the general result in Nelson and Foster (1994) who
show that the optimal filter for the diffusion,
dx = µdt + σdW1 ,
dσ 2 = (ω − Θσ 2 )dt +
√
2aσ 2 Vm dW2 ,
(4)
where
dW1 ∼ N (0, dt), dW2 ∼ N (0, dt)
and
Corr(dW1 , dW2 ) = 0
is,
xt+∆ − xt = µ∆ +
√
∆ξt+∆ ; ξt+∆ ∼ N (0, yt ),
yt+∆ = ω∆ + (1 − Θ∆ −
15
√
√
2
∆a)yt + ∆aξt+∆
.
(5)
Nelson and Foster (1994) prove that the discrete time model in (5) converges in distribution to the continuous time model in (4) as ∆ shrinks to zero. The discrete equations above
can be written as the following GARCH(1,1) model,
Xt+1 − Xt = c + ηt+1 ; ηt+1 |Ωt ∼ N (0, ht+1 );
2
ht+1 = a0 + a1 ηt+1−i
+ b1 ht+1−i
with c = µ∆, a0 = ω∆2 , a1 =
(6)
√
√
∆a, b1 = (1 − Θ∆ − ∆a) and ht+∆ = ∆yt .
The parameters of the GARCH model over a window can be easily estimated using the
method of maximum likelihood. Based on these estimates, we can then derive the estimates
of the parameters of the original diffusion process.
Following the above methodology, we first estimate the parameters of the diffusion process
(2) for market returns over the estimation window. We then estimate the parameters of the
continuous time market model for each firm (3). The betas in (3) are estimated separately by
running OLS regression of firm’s returns on the market returns over the estimation window.
The second part of this methodology involves hypotheses testing over the event window
using parameter estimates from the estimation window. HS introduce a parameter, the
multiplicative abnormal volatility to measure the impact of a given event on unsystematic
volatility Vε .(To avoid notational confusion we use ψ instead of the λ used by HS). The
parameter ψ measures the multiple by which unsystematic volatility increases from its no
event level, due to the event. Thus, if ψ = 1, the event has no effect on Vε ; ψ > 1 implies
a volatility increase due to the event and ψ < 1 implies a volatility decrease due to the
event. A particular event may have a different value of ψ for each day in the event window
which is denoted by a time subscript t on ψ. HS show that the estimate of ψt for each
day t in the event window can be obtained by computing the cross-sectional variance of the
standardized GARCH(1,1) residuals, ηt in equation (6). The cumulative abnormal volatility
Cψk,m between event days k and m is the sum of daily abnormal volatilities over these days.
Cψk,m =
m
X
t=k
16
ψt
(7)
As HS illustrate, the null hypothesis regarding the effects of an event on volatility can be
expressed in terms of either ψt or Cψk,m . In terms of ψt , it is,
H0 : ψt = 1
(8)
H0 : Cψk,m = m − k + 1
(9)
and, in terms of Cψk,m , it is,
HS show that this hypothesis (8) can be tested using a test statistic st ≡ (N − 1)ψbt which
is distributed χ2N −1 under the null. Similarly, hypothesis (9) can be tested using the test
2
\
statistic Csk,m ≡ (N − 1)Cψ
k,m which is distributed χ(N −1)(m−k+1) under the null. If the
observed value of the test statistic exceeds the critical value, the null hypothesis is rejected
and we can conclude that the event has a statistically significant (daily or cumulative) impact
on the (daily or cumulative) unsystematic volatility of stock returns.
6
6.1
Results
Comparison of pre-event and post-event volatilities
First we present the results of our analysis using
r the simple methodology of comparing the
pre-event and post-event volatility ratios λ ≡
σi2
2 .
σm
We use the pre-event and post-event
windows of 3 months, six months, one year and two years. The resulting pre-event and
post-event average λś are reported in Table II, Panel (A). It is interesting to note that the
average λ jumps significantly from pre-event to post-event period. The ratio of post-event
to pre-event average λ’s varies between 1.28 to 1.07. The jump is statistically significant at
the 1% level using a standard t-test and a wilcoxon test. This indicates that moving online
is associated with a significant volatility increase. We then split the sample into two parts,
one with event dates prior to June 1998 and the other with event dates post June 1998. The
analysis of the two split samples is presented in Table II, Panels (B) and (C). Results show
that firms with event dates prior to June 1998 have no increase in the volatility associated
with the event and the volatility increase for the overall sample can be attributed to the firms
which moved online post June 1998.
17
The results highlight the changing perceptions of the financial markets about the potential
impact of the Internet on a firm’s business. This shift in market perceptions coincides with
the spurt in the growth of the Internet. The growth of the Internet is significantly higher in
the post-June 1998 period compared to the pre-June 1998 period as measured by the growth
in the number of Websites in existence. Figure 1A illustrates the growth of Web-sites over
the past several years13 . We observe an exponential growth in the number of Web-sites in the
post-June 1998 period. The difference between the two periods is also seen in Figure 1B which
shows the evolution of AMEX Internet index over the two sample periods. Schwert(2002)
shows the unusual increase in Nasdaq volatility in the post June 1998 period. Based on all
this evidence, we analyze our sample separately for the pre and post June 98 sample periods.
Figure 2 graphically shows the changes in abnormal volatility around the event date. We
compute the monthly λ of each stock (using daily returns) for 24 months before and after the
event. Individual λs in each month were then averaged cross-sectionally in the event time.
The time-series of average λ are plotted in Figure 2. The notable result is the spike in λ at
the event date, which can be seen in Figure 2(A). This again confirms our results in Table
II that the event is associated with a surge in volatility. The corresponding graphs for two
sub-samples and the results are in Figure 2(B) and 2(C). These figures confirm that firms
which moved online prior to June 1998 did not experience a volatility increase, in marked
contrast to firms that moved online post June 1998.
Next, we use the methodology of Bai, Lumisdaine and Stock (1998) to detect a structural
break in the average firm volatility series. The volatility of a stock is computed as the 12month rolling sample standard deviation of monthly returns following Officer (1973) and
Merton (1980). The average volatility series is constructed by taking the cross-sectional
average of stock volatilities in event time. The average volatility series for the full, pre-June
1998 and post-June 1998 samples are shown in Figures 3(A),(B) and (C) respectively. The
pattern here is similar to the one seen in Figure 2 with the λ series. In the overall sample,
we observe a surge in volatility at event date, which is caused solely by the events after June
1998. The formal test for structural break using a Wald statistic confirms this result (Table
III). Figure 3 (D),(E) and (F) show the time series of the Wald statistic for the full sample,
13
c
°Robert
H. Zakon, adapted with permission from http://www.zakon.org/robert/internet/timeline/
18
pre-June 1998 sample and post-June 1998 sample respectively. In the overall sample, as in
the post-June 1998 sample, we see a sharp rise in the Wald statistic around the event date.
The 1% confidence band of the break date is around the event date and is quite narrow,
covering a period of only about 3 months and includes the event date (month 0) for the full
sample and the post-98 sample. This is very clear evidence of a break around the event date
for the average volatility series. It can be seen that the pre-1998 sample shows a structural
break in average volatility 3 months before the event and thus, does not have any change in
volatility associated with the event. This analysis confirms the above event study results.
It is possible that our sample of firm experience increases in volatility for reasons unrelated
to their launch of eCommerce operations and our tests detect just these increases. In other
words, the volatility increase and the event of moving online could both be jointly caused by
other unobservable factors. To rule out this possibility, we repeat the volatility analysis in a
matched sample of firms.
We construct a return-matched sample of firms as follows. For each firm in the test
sample, we find a matched firm satisfying two criteria viz., (a) it is in the same NYSE
market capitalization decile as the firm in the test sample and (b) in this size decile, it is
the firm which has a compounded return in the two years prior to the event date which is
the closest to that of the test firm. This methodology closely follows that recommended by
Barber and Lyon (1996). If the matching firm thus selected is found in our sample or if it is a
’pure Internet’ firm or has operations in an Internet-related domain, we select the next-best
match. Size and past returns are well known to be among the most important influences on
the volatility of individual stocks (see, Duffee, 1995). Thus, a control by size and past returns
should be able to account for volatility changes not related to the event under consideration.
We also construct a volatility-matched sample of firms as follows. For each firm in the
test sample, we find a matched firm which satisfies two criteria viz., (a) it is in the same
NYSE market capitalization decile as the firm in the test sample and (b) in this size decile,
it is the firm which has a natural logarithm of the ratio of the sample standard deviation for
year -1 in event time to the sample standard deviation for year -2 (where 1 year consists of
250 trading days) that is the closest to that of the test firm. If the matching firm is found
to be in our sample or if it is a ’pure Internet’ firm or has operations in an Internet-related
19
domain, we select the next-best match as earlier.
The results of the above analysis for the matched sample of firms appear in figures 2 and
3. Figure 2 depicts the monthly average λs for the return and volatility matched samples. No
spike in average λ is seen at the event date for these two samples either in the pre-98 or the
post-98 period. Figure 3 confirms this pattern by constructing a rolling volatility series, even
though the general level of volatility is higher in the post-event period than in the pre-event
period for all the 3 samples.
14 .
The contrasting results found with our test sample of firms and the matched sample
confirm that the volatility surge around the event cannot be attributed to other well known
determinants of stock return volatility viz. past returns, past volatility run up and firm size.
Since we have already omitted other firm specific announcements and events which could
have a confounding effect on the stock volatility, we conclude that the observed surge in
volatility can be attributed solely to the commencement of eCommerce operations. We now
proceed using the HS methodology to confirm our above findings.
6.2
Event Induced Unsystematic Volatility-Results
The results for this analysis are summarized in Figures IV (A),(B) and (C), which show
the estimated cumulative abnormal volatility Cλt over the event window for the test sample,
return matched sample and the volatility matched sample. They also illustrate the cumulative
abnormal volatility that would be expected under the null hypothesis of no event induced
volatility, given the values of the parameter estimates. The same results are summarized
in a tabular form in Table IV, which shows the values of abnormal volatility λ cumulative
abnormal volatility Cλ for the event window. The event window covers the period [-26,+25]
days around the event date.
Figure IV(A) shows the results for the entire sample. First we note that the cumulative
abnormal volatility drifts up (compared to its expected value under the null) for the test
sample as well as for the two matched samples. This is also reflected clearly in Table IV(A).
Note that under the null hypotheses (8) and (9) above, the value of λ should be 1 on each day
14
The individual stock volatility has increased much more in recent years compared to the market volatility.
See Campbell, Lettau, Malkiel and Xu (2001) for a detailed analysis.
20
and the value of Cλ should start at 1 on day -26 and increase by 1 for each day over the event
window. The p-values of the test statistic for testing the null hypothesis (9) of no event effect
on cumulative abnormal volatility is also reported. The null is resoundingly rejected for all
the three samples on all days except day -26. This reflects that, in general, the idiosyncratic
volatility of stocks has risen over time in the past few years. This is also consistent with the
findings in Campbell et al.(2001) that the unsystematic volatility of individual stocks have
risen over the last few decades.
Although all three samples show an increase in unsystematic volatility over time, the
impact of the event is evident from figure IV(A). The cumulative abnormal volatility surges
significantly at the event date for the test sample only, but not for the two matched samples.
The deviation of the cumulative abnormal volatility from its value under null is also the
maximum for the test sample. These results indicate that while the statistical tests proposed
by HS show a significant increase in unsytematic volatility for all three samples on the event
date, the event clearly has the strongest effect on unsystematic volatility for the test sample.
The test sample is then split into two - sample 1, consisting of firms with event dates
prior to June 1998 and sample 2 consisting of firms with event dates after June 1998. The
results for these 2 samples are shown in Figures IV(B) and IV(C)respectively15 and also in
Panels (B) and (C) of Table IV. We find in sample 1, that the event had no effect on the
unsystematic volatility a result that confirms our earlier analysis. The entire effect that we
observe for the full sample is due to sample 2.
6.3
Additional Tests for Robustness
A large body of empirical literature documents a positive relation between trade size or the
number of transactions and stock return volatility (See Karpoff(1987) for a survey). In order
to ensure that our results are not simply an artifact of increased trading volume and number
of transactions for the test firms, we compare the post-event to pre-event trading volume and
number of transactions for all three samples. The results in Table V suggest that all three
samples in our study had statistically significant and the same order of magnitude increases
15
In the estimation, we dropped the firms whose volatility process parameters showed the presence of
non-stationarities. Hence, the number of firms are unequal across the samples.
21
in their trading activity measures over time, whether measured as volume or number of
transactions, while their volatility experience as noted above, was very different. We therefore
conclude that increases in trading activity cannot be an explanation for increases in volatility
in our eCommerce sample.
Finally, as corroborative evidence, following Schwert(1990) we track the monthly series
of implied volatility of at-the-money call options for firms in our eCommerce sample over
a period of 4 years around our event date. We obtain the data from the Optionmetrics
database. Many studies have shown that close-to-the-money option prices convey the most
information about the expectations of the options market concerning future volatility (Day
and Lewis, 1988). The results are presented in Table VI and figure 5. We find that implied
volatility increased for our eCommerce sample after the event of moving online, confirming
our earlier results. This increase is statistically significant at the 5% level or higher.
We also provide additional evidence for higher demand uncertainty relating to the Internet
and more specifically to eCommerce. In addition to earlier discussions, the evidence that
eCommerce activity is associated with a higher uncertainty in product demand comes from
several sources. First, the US Census Bureau data of Internet-retail sales versus overall retail
sales are reproduced in Table VII, Panel A . These data show that Internet sales have been
far more volatile compared to traditional sales. The coefficient of variation was 22.7% for
Web-sales and only 5.6% for the total retail sales16 . Table VII, Panel B shows the changes in
annual sales (both total retail sales as well as eCommerce sales) across different sectors. On
an average, Internet-driven sales rose by 89% between 1999 and 2000, compared to a 7.1%
increase in overall sales for the same time period. In addition, we also find that the agencies
which forecast eCommerce activity (such as online advertising spending by US firms) have
widely different views about eCommerce, reflecting the inherent uncertainties. Table VII,
Panel C provides estimates made in June 2000 by various consulting firms for the period
2000-2002 of Internet advertising spending by US firms to promote their online operations.
While this is not direct evidence on product demand, advertising is certainly an important
component of a firm’s marketing plan that stimulates product demand. As can be seen, there
is wide variation in their estimates reflecting the diversity of expectations about the potential
16
Coefficient of variation is the standard deviation normalized by the mean.
22
of eCommerce operations. The cross-sectional standard deviation of these projections is 35%
of the mean projection for the year 2000, 52% for the year 2001 and 66% for the year 2002.
17 .
Taken together, panels A-C of table VII suggest that eCommerce activity is associated
with a higher demand uncertainty compared to traditional retail channels.
Finally we attempt to relate demand uncertainty in the product markets to stock return
volatility. Ideally we would like to construct a measure of demand uncertainty using quarterly sales data before and after the introduction of eCommerce operations (using eCommerce
sales in the latter period) and relate it to the increase in stock return volatility. Due to data
limitations we are unable to obtain a long enough time series of sales after the adoption of
eCommerce operations at the firm level to undertake this test. Thus we adopt an alternative
approach and show that the measures of demand uncertainty we construct are positively
related to stock return volatility. This evidence taken together with the evidence from Table VII suggests that the increased demand uncertainty due to introduction of eCommerce
operations is associated with increases in stock return volatility.
We use quarterly sales data for firms in our sample for the period 1990-2002 from COMPUSTAT (data item 2) to measure demand uncertainty. We posit an AR(1) process to
estimate quarterly sales for each firm in our sample. For each firm the following AR(1)
process for quarterly sales is estimated.
ln(Sales(i, t)) = αi + βi ∗ ln(Sales(i, t − 1)) + ²it
where Sales (i,t) refers to quarterly sales for firm i in quarter t. The measure of demand
uncertainty for firm i, U N C1i is the time-series standard deviation of residuals ²it from the
above specification. In the second stage a cross sectional regression is run between λi and
U N C1i as follows18 .
λi = a + b ∗ U N C1i + ei
We also calculate another measure of demand uncertainty U N C2i . The calculation of U N C2i
follows exactly the same procedure as before except that the AR(1) specification in the first
17
18
Source : www.eMarketer.com
λi is calculated for each firm for eight years around the event date, corresponding roughly to the same
period for which AR(1) process is estimated using daily returns. Note that this definition of lambda is different
from the definition used in the rest of the tables.
23
stage is estimated without a constant term for each firm. For the regression White (1980)
corrected standard errors are reported.
Table VIII, Panel A provides the summary statistics of the 2 demand uncertainty measures
U N C1 , U N C2 and λ respectively. As can be seen, demand uncertainty measures are positively
correlated with stock return volatility. Panel B provides the results of the regression. The
results are shown to be statistically significant. Taken together, the results of tables VII and
VIII indicate a linkage between demand uncertainty stemming from eCommerce initiatives
and stock return volatility.
6.4
Volatility and Abnormal Returns
While the focus of this paper is on examining the impact of firms’ adoption of eCommerce on
their stock price volatility, a natural question that arises is the impact of the event on firms’
stock returns. To this end, we perform a traditional return-event study using a market-model
and a value-weighted index, ensuring to correct for non-synchronous trading using Scholes
William beta as well as for increases in event-induced variance using the method of Boehmer
et al. (1991). The results of the event-study are provided in Table IX. Panel A presents the
event study results for the full sample of 164 firms, for four different event windows, while
Panel B and C present the results for the pre-1998 sample and post-1998 sample of firms,
respectively. As illustrated in Table IX, it can be seen that while firms experience a positive
abnormal return on moving online, these abnormal returns are more pronounced for firms that
moved online post-June 1998, compared to firms that moved online prior to June 1998. These
findings parallel those of our volatility study and suggest a positive relationship between
firm-specific idiosyncratic volatility and abnormal returns. This is particularly interesting in
light of the recent findings by Duffee (2002) who finds a significant positive contemporaneous
relation between stock returns and firm-level idiosyncratic volatility; this positive relationship
being stronger for firms with higher betas and book-to-market ratios. Duffee (2002) theorizes
that shocks to firm value are caused by shocks to asset values, with riskier assets having larger
absolute shocks - a positive (negative) shock to firm value being accompanied by an increase
(decrease) in the value of the firm’s risky assets leading to higher idiosyncratic volatility. In
our study, the onset of eCommerce can be considered to be an exogenous event that is shown
24
to increase both the value of the firm (based on the results of the return event study) and
its idiosyncratic volatility - findings that reinforce Duffee’s (2002)arguments.
7
Conclusions
In this paper, we study the effect of real activity with in the firm on its stock return volatility.
Specifically, we focus on the impact of traditional firms’ adoption of eCommerce on the
volatility of their stock returns. We find that stock return volatility increases when the
traditional firms announce eCommerce initiatives. This increase in firm-level volatility is
detected using three different methodologies and similar abnormal volatility is absent in
samples of firms matched with the test sample. We corroborate this evidence by studying
the implied volatility of at-the-money call options of our test sample which are also found to
increase around the event date. More importantly, we find that the market’s perception of
the significance of the Internet for traditional firms plays a very important role in determining
the impact of the event on risk-return profile of firms. Only firms that moved online postJune 1998, the period when Internet-related activity reached critical mass, experience a
significant surge in volatility, an effect that is absent for firms that moved online prior to June
1998. This coincides with the findings by Schwert (2002), of an increase in market volatility
of ’technology-intensive’ firms after mid-1998. The adoption of the Internet for commerce
constitutes a significant change in a traditional firm’s technological environment. We find
that the consequence of these developments is increased turbulence in firms’ product markets
as reflected in a higher demand uncertainty, leading to higher volatility of firm’s stock returns.
We also find that the behavior of event-induced positive abnormal returns closely parallels
that of the event-induced volatility increase and suggests a positive relationship between
the two. Further research would be required to examine the contemporaneous relationship
between abnormal returns and volatility increases and to identify factors that moderate the
strength of this relationship. Overall, our results show strong evidence of the impact of real
activity within the firm on its stock return volatility and highlight the need for a better
understanding of the financial impacts of firms’ changes in technological environment and
product market initiatives.
25
8
References
Bakos, Y., Lucas, H. C., Oh, W., Simon, G., Viswanathan, S., and Weber, B., 2002. An experimental evaluation of the impact of electronic commerce on the retail brokerage industry.
Working Paper, University of Maryland, College Park.
Banerjee, A., Lumisdaine, R., Stock, J., 1992. Recursive and sequential tests of the unit root
and trend break hypothesis. Journal of Business and Economic Statistics 10, 271-288.
Bai, J., Lumsdaine, R., Stock, J., 1998. Testing for and dating common breaks in multivariate time series. Review of Economic Studies 65, 395-432.
Barber, B. M., Lyon, J. D., 1996. Detecting abnormal operating performance: The empirical
power and specification of test statistics. Journal of Financial Economics 41, 359-399.
Bekaert, G., Harvey, C. R., Lumsdaine, R., 2002. Dating the integration of world equity
markets. Forthcoming, Journal of Financial Economics.
Black, F., 1986. Noise, Journal of Finance 31, 529-543.
Boehmer E. and Musumeci J., Poulsen A.B., 1991. ”Event study methodology under conditions of event-induced variance,” Journal of Financial Economics 30, 253-272.
Bolton, P., Scharfstein, D., 1990. A theory of predation based on agency problems in financial contracting. American Economic Review 80, 93-106.
Brander, J. A., Lewis, T. R., 1986. Oligopoly and financial structure: The limited liability
effect. American Economic Review 76, 956-970.
Brown, J. R., and Goolsbee, A., 2000. Does the Internet make markets more competitive?
Evidence from the Life Insurance Industry. NBER Working Paper 7996.
Brynjolfsson, E., Smith, M. D., 2000. Frictionless commerce? A comparison of Internet and
conventional retailers. Management Science 46, 563-585.
Campbell, J. Y., Lettau, M., Malkiel, B., Xu, Y., 2001. Have individual stocks become more
26
volatile? An empirical exploration of idiosyncratic risk. Journal of Finance 56, 1-43.
Chevalier, J. A., 1994. Capital structure and product market competition: An examination
of supermarket LBOs. Harvard University Working Paper.
Clayton, M., Hartzell, J., Rosenberg, J., 2000. The impact of CEO turnover on equity volatility. Working Paper, New York University.
Cooper, M. J., Dimitrov, O., Rau, P.R., 2001. A Rose.com by any other name. Journal of
Finance 56, 2371-2388.
Cornell, B., Liu, Q., 2000. The parent company puzzle : When is the whole worth less than
one of its parts. Working Paper, UCLA.
Day, T. E., Lewis, C. M., 1988. The Behavior of the volatility implicit in the prices of stock
index options. Journal of Financial Economics 22, 103-122.
Demers, E., Lewellen, K., 2003. The marketing role of IPOs : Evidence from Internet stocks.
forthcoming, Journal of Financial Economics.
Duffee, G. R., 1995. Stock returns and volatility: A firm level analysis. Journal of Financial
Economics 37, 399-420.
Duffee, G. R., 2002. Balance sheet explantions for asymmetric volatility. Working Paper,
Haas School of Business, University of California, Berkeley.
French, K. R., Schwert, G. W., Stambaugh, R., 1987. Expected stock returns and volatility.
Journal of Financial Economics 19, 3-29.
Hilliard, J. E., Savickas, R. E., 2002. On the statistical significance of event effects on unsystematic volatility. Journal of Financial Research 25, 1-16.
Karpoff, J. M., 1987. The relation between price changes and trading volume: A survey.
Journal of Financial and Quantitative Analysis 22, 109-126.
27
Klein B. H., 1977. Dynamic economics. Cambridge,MA: Harvard University Press.
Lamont, O. A., Thaler, R., 2000. Can the market add and subtract? Mispricing in Tech
stock carve-outs. Working Paper, University of Chicago.
Lusch, R. F., 1995. Two critical determinants of retail profitability and productivity. Retailing Issues Letter 2, 1-3.
Maksimovic, V., 1990. Product market imperfections and loan commitments. Journal of
Finance 45, 1641-1653.
Maksimovic, V., 1995. Financial structure and product market competition, in Handbooks
in OR and MS, ed., R. Jarrow et al., North-Holland, Amsterdam 9, 887-920.
Mazzucato M., 2002. The PC Industry: New economy or early life-cycle? Review of Economic Dynamics, 5, 318-345.
Mazzucato M., Semmler, W., 1999. Market share instability and stock price volatility during
the US Auto industry life cycle. Journal of Evolutionary Economics, 9, 67-96.
Merton, R. C., 1980. On estimating the expected return on the market : An exploratory
investigation. Journal of Financial Economics 8, 323-361.
Mayhew, S., 1995. Implied volatility. Financial Analyst Journal 51, 8-20.
Nelson, D. B., Foster, D. P., 1994. Asymptotic filtering theory for univariate ARCH models
1995. . Econometrica 62, 1-41.
Ofek, E., Richardson, M., 2001. DotCom Mania : A Survey of market efficiency in the Internet sector. Forthcoming Journal of Finance.
Officer, R., 1973. The variability of the market factor of New York Stock Exchange. Journal
of Business 46, 434-453.
Perotti, E., Rossetto, S., 2000. Internet portals as portfolios of entry options. Working Pa28
per, University of Amsterdam.
Phillips, G. M., 1995. Increased debt and product markets: An empirical analysis. Journal
of Financial Economics 37, 189-238.
Porter, M. E., 2001. Strategy and the Internet. Harvard Business Review, 63-78.
Schwert, G. W., 1989. Why does stock market volatility change over time? Journal of Finance 54, 1115-1153.
Schwert, G. W., 1990. Stock volatility and the crash of ’87, Review of Financial studies 3,
77-102.
Schwert, G. W., 2002. Stock market volatility in the new millennium: How wacky is Nasdaq?
Journal of Monetary Economics 49, 3-26.
Shiller, R., 1981. Do stock prices move too much to be justified by subsequent changes in
dividends? American Economic Review 71, 421-436.
Viswanathan, S., 2002. Dotcoms versus Notcoms: The impact of Internet commerce on traditional firms. Unpublished Dissertation, New York University.
29
9
Appendix A : A Simple Model of Demand Uncertainty
We begin with the premise that a firm faces a more uncertain product market demand online
than in traditional retailing. The objective of this simple model is to show that such an
uncertainty in the product market can lead to a higher volatility of profits.
The firm faces a stochastically varying inverse demand process θt in the product market,
which is modeled as a diffusion process in continuous time. It maximizes the profit function
π = (θ − Q)Q, where Q is the quantity sold. The optimal profit for a monopoly firm are
π1 =
θ2
4 .
Let xt =
θt2
4 .
We can express its evolution as a Geometric Brownian Motion, i.e.
so that,
dxt = xt [µ1 dt + σ1 dWt ]
(10)
1
xt = x0 exp((µ1 − σ12 )t + σ1 Wt )
2
(11)
Thus, expected profit is,
Z
π¯1 = E[
∞
0
xt exp(−rt)dt] =
x0
r − µ1
(12)
and variance of the profit can be shown to be,
V ar(π1 ) = E[π12 ] − [E[π1 ]]2 = x0
1 2
2 σ1
(r − µ1 )2 (r − µ1 − 12 σ12 )
(13)
Now consider the same firm with eCommerce operations. We postulate two changes to
occur when the firm moves online viz. (1) there is a fixed setup cost I and (2) the demand
function becomes more uncertain, which is captured by a stochastically evolving multiplier
γ. The firm therefore maximizes the objective function γ(θ − Q)Q − I and the maximized
profit process is π2 = γx − I. The evolution of γ is assumed to be given by the process,
dγ = γ[µ2 dt + σ2 dBt ]
(14)
where corr(Bt , Wt ) = 0.
This implies that,
1
γt = γ0 exp((µ2 − σ22 )t + σ2 Bt )
2
The expected profit and the variance of profits can be calculated as follows:
Z
π¯2 = E[
0
∞
γt xt exp(−rt)dt] − I =
V ar(π2 ) = E[π22 ] − [E[π2 ]]2 = γ0 x0
γ0 x0
−I
r − µ1 − µ2
1
2
2 (σ1
(15)
(16)
+ σ22 )
+ 2I 2 (17)
(r − µ1 − −µ2 )2 (r − µ1 − 12 σ12 − µ2 − 21 σ22 )
30
The ratio of the variances in the two cases is,
µ2 + 21 σ22
µ2
2I 2
V ar(π2 )
σ2
]
+
= γ0 [1 + 22 ][1 +
]2 [1 +
(18)
V ar(π1
(r − µ1 − µ2 )
V ar(π1 )
σ1 )
(r − µ1 − µ2 − 21 σ12 − 21 σ22 )
This ratio is always greater than or equal to 1. Thus, the twin effects of higher demand
uncertainty and lower fixed cost of entry implies that the introduction of online operations
by a firm would always lead to an increase in the volatility of firm’s profit or net cash flow
each period. This will translate into a higher equity price volatility as well.
31
10
Appendix B : Structural Break Analysis
Under our null hypothesis, the event has no effect on the volatility σ, i.e. the process generating the time-series of σ remains structurally unchanged. Under the alternative hypothesis
of event-induced abnormal volatility, the process generating σt time series will have a structural break at the event date. Banerjee, Lumsdaine, and Stock (1992) (BLS1) and Bai,
Lumsdaine, and Stock (1998) (BLS2) provide an approach for detecting such a break. They
show (i) that tests can be constructed to determine the occurrence of structural break in a
set of time-series data, and (ii) that confidence intervals can be computed enabling inference
about the break date. They demonstrate this for both stationary vector autoregressions and
cointegrated systems. This paper focuses on the stationary case.
BLS2 postulate the following general form of regression relationship among dependent
variable y and vector of independent variables X (equation 2.2 from BLS2):
yt = (G0t In )θ + dt (k)(G0t In )S 0 Sδ + εt
(19)
where yt is n by 1, G0t is a row vector containing a constant, lags of yt , and row t of
the matrix of exogenous regressors, X, In is an n by n identity matrix, dt (k) = 0 for t < k
and dt (k) = 1 for t >= k. θ and δ are parameter vectors with dimension r. For example,
for a first-order vector autoregression with a vector of constants µ and parameter matrix A,
(yt = µ+Ayt−1 +²t ), θ = vec[(µ, A)] and r = n(n+1). S is a selection matrix containing zeros
and ones, with column dimension r and full row rank (equal to the number of coefficients
which are allowed to change). Note that S 0 S is idempotent with non-zero elements only
on the diagonal. If S = Ir , then the above model is a full structural change model. For
N
S = s In where s = (1, 0, ..., 0) a row vector, the model allows for a mean shift only (See
empirical examples in BLS2). The model (19) allows any or all of the coefficients to change.
More compactly,
yt = Zt0 (k)β + εt ,
(20)
where Zt0 (k) = ((G0t In ), dt (k)(G0t In )S 0 ) and β = (θ0 , (Sδ)0 )0 . If we let R = (0, I) be the
selection matrix associated with β, so that Rβ = Sδ and Σ is the covariance matrix of the
errors ²t , then the F-statistic testing Sδ = 0 is
0
b
FbT (k) = T {Rβ(k)}
{R(T −1
T
X
b
b −1 Z 0 )−1 R0 }−1 {Rβ(k)},
Zt Σ
t
(21)
t=1
b
where β(k) and Σ(k)
denote the estimators of β and Σ, respectively, evaluated at b
k, obtained
as described above. BLS2 show that FbT (k) converges in distribution via the functional central
limit theorem to F ∗, where F ∗ = {τ (1 − τ )}−1 kW (τ ) − τ W (1)k2 , where k · k represents
32
the Euclidean norm and now W (.) is a vector of independent standard Brownian motion
processes whose dimension is equal to q, the rank of S. The corresponding distribution can
be approximated by partial sums of normal random variables for each dimension. We use the
table with corresponding critical values for the rank up to q = 50 as given in the appendix
of Bekaert, Harvey and Lumsdaine(2002). By the continuous mapping theorem, the limiting
distribution of maxk FT (k) converges to max F ∗. As noted above, we focus on this test
statistic in the empirical work.
The dimension of the test statistic increases with both the dimensionality of the system
and with the number of regressors in the model whose coefficients are allowed to break. As
an example, consider an n by 1 VAR. If the order of the VAR is p and we allow for a break
in all of the coefficients, the relevant dimension of the F-statistic will be n(np + 1).
To conduct inference about the break date, theorem 4 of BLS2 shows that
[δT0 S 0 S(QΣ−1 )S 0 SδT ](b
k − k0 ) ⇒ V ∗,
where V ∗ has limiting density given by equation below (to be filled) and Q = plim T1
(22)
PT
0
t=1 Gt Gt .
Similarly, we can invert the limiting distribution to construct confidence intervals for the estimated break date, based on allowing any or all of the coefficients to experience a break.
The confidence interval is
b
bΣ
b −1 )S 0 (S δbT )]−1 ,
k ± α 1 π [(S δbT )0 S(Q
k
2
b=
where Q
1
T
PT
0
t=1 Gt Gt
(23)
ck are estimated values.
and b
k, Σ
Finite sample properties of these test statistics are investigated in BLS2. The tests
are shown to have good size and power properties under the null hypothesis of no break
and the alternative of a breaking mean, respectively. In addition, simulations regarding
confidence intervals confirm that, for fixed parameters, increasing the sample will not affect
the precision of the MLE of k0 but increasing the number of series that experience the same
break does improve precision. In addition, this precision depends on the true value of the
break magnitude.
33
Table I, Panel A : Distribution of Event Dates
This table shows the temporal distribution of the event dates in the sample of 166 firms. The sample consists
of the firms which announced an online trading initiative in business to consumer segment of the market.
Firms with other events in addition to the announcement, which might have potentially confounding effect
on stock prices were omitted.
Year
Quarter
Number of Firms
Year
Quarter
Number of Firms
1995
Q1
1
1997
Q4
3
1995
Q2
1
1998
Q1
17
1995
Q3
1
1998
Q2
0
1995
Q4
1
1998
Q3
15
1996
Q1
3
1998
Q4
11
1996
Q2
1
1999
Q1
29
1996
Q3
4
1999
Q2
15
1996
Q4
8
1999
Q3
28
1997
Q1
3
1999
Q4
17
1997
Q2
2
2000
Q1
3
1997
Q3
3
Total
166
Table I, Panel B : Distribution of firms by Industry
This table shows the industry wide distribution of the 166 firms in the sample. The sample consists of the
firms which announced an online trading initiative in business to consumer segment of the market. Firms
with other events in addition to the announcement, which might have potentially confounding effect on stock
prices were omitted.
Industry
No. of Firms
Software
20
Computer Hardware, Consumer Electronics
29
Banking, Brokerage
21
Media, Publishing, Entertainment, Services
20
General Merchandise Retailers
20
Speciality Retailers
27
Auto, Motorcycles, Hardware, Home Improvement
Catalog, Direct Marketing, Auctions, Travel
Pharma, Conglomerates, Real Estate
Total
7
12
10
166
34
Table II : Pre and Post Event Abnormal Volatility
This table shows the Pre event and Post event abnormal volatility for the firms in the sample. Abnormal
volatility, Average λ is computed as the square root of the ratio of the firm’s stock return volatility σi2 to the
2
market volatility (S and P 500 equally weighted return) σm
and averaged over all firms in the sample. This
is computed for (1) the pre-event window [−L, 0] and the (2) post-event window [0, L], where L is the length
of each window and 0 represents the event date. Four different window lengths are considered (three months,
six months,one year and two years). Post/Pre λ is the ratio of Post λ to Pre λ. t-stat and Wilcoxon are the
t-test and Wilcoxon test statistic for the 1-tail test of Post/Pre λ = 1.0 vs. Post/Pre λ > 1.0 respectively
Panel (A) : Full Sample Results (166 firms)
Firms that moved online
Window
Pre-event
Post-event
Post/Pre λ
t-stat
Wilcoxon
Avg λ
Avg λ
Mean
Median
Max
Min
Std.Dev
3 months
6.43
6.66
1.23
1.09
5.65
0.11
0.71
4.14∗∗∗
3807∗∗∗
6 months
5.74
6.54
1.25
1.14
3.83
0.14
0.62
5.21∗∗∗
5215∗∗∗
∗∗∗
7003∗∗∗
1 year
4.99
6.16
1.28
1.21
4.05
0.26
0.58
6.16
2 years
5.24
5.48
1.07
1.00
3.79
0.29
0.39
2.19∗∗
1247
Panel (B) : First half sample (before June 1998) results (56 firms)
Firms that moved online before June 1998
Window
Pre-event
Post-event
Avg λ
Avg λ
Mean
Median
Post/Pre λ
Max
Min
Std.Dev
t-stat
3 months
6.01
6.26
1.27
1.18
3.03
0.30
0.76
2.70∗∗∗
6 months
5.75
5.62
1.08
0.92
2.96
0.42
0.54
1.08
1 year
5.61
5.36
0.99
0.89
2.38
0.44
0.42
−0.21
2 years
5.74
5.25
0.93
0.90
1.65
0.49
0.22
− 2.26∗∗
Wilcoxon
502∗∗
44
-222
− 582∗∗∗
Panel (C) : Second half sample (after June 1998) results (110 firms)
Firms that moved online after June 1998
Window
Pre-event
Post-event
Post/Pre λ
t-stat
Wilcoxon
Avg λ
Avg λ
Mean
Median
Max
Min
Std.Dev
3 months
6.64
6.87
1.20
1.07
5.65
0.11
0.69
3.13∗∗∗
1667∗∗∗
6 months
5.74
7.00
1.34
1.20
3.83
0.14
0.64
5.54∗∗∗
3379∗∗∗
1 year
4.68
6.56
1.42
1.33
4.05
0.26
0.59
7.47∗∗∗
4749∗∗∗
0.43
∗∗∗
1871∗∗∗
2 years
4.99
5.60
1.13
1.05
∗ ∗ ∗, ∗∗, ∗ - Significant at 1%,5%,10% level respectively
35
3.79
0.29
3.24
Table III: Structural Break Analysis of Volatility
This table reports the estimated break dates k̂ for the structural relation of equity return volatility
σt = µ + Aσt−1 + dt (k)(λ + βσt−1 ) + ²t
Equity return volatility is computed as the 12 month rolling volatility following Officer(1973)and Merton
(1980). The median column in the table shows the estimated break date k̂ in event time for the time series of
average monthly volatility for the corresponding sample. Break dates are estimated using the Wald Statistic
F described in Appendix B. We test the null hypothesis that the post-break coefficient changes are not
significantly different from zero, i.e., that no break occurred in the sample period by comparing the maximum
value in the estimated time series F(k̂) to the 5% quantile of its limiting distribution. The null hypothesis is
rejected when the maximum value for F(k̂), reported in the Max-Wald column of the table is higher than the
critical value for the selected significance level. We use the critical values from Bekaert et.al (2002), Table
10 in the Appendix who approximate the limiting distribution of the F process with partial sums of normal
random variables for each possible dimension of the test statistic which is the dimension of S. From that table
we use the asymptotic 1% critical value of 11.81 corresponding to a rank of 1 for vector S. the 2.5th and
the 97.5th percentile columns display the estimated lower and upper bands respectively for the confidence
intervals for the ”true break dates” as per equation (23) with quantiles of the Picard (1985) distribution.
Sample
Firms
Full Sample
166
2.5th
Median
Percentile
-1
Pre 98 Sample
56
-4
Post 98 Sample
110
0
97.5th
Max-Wald
p-value
1
54.86
< 0.01
-2
29.24
< 0.01
2
32.78
< 0.01
Percentile
0∗∗∗
∗∗∗
−3
1∗∗∗
∗ ∗ ∗ - Significant at 1% level
36
Table IV : Unsystematic Volatility Induced by the Event
This table shows the estimates of the unsystematic volatility induced by the event and their statistical significance following the method of Hilliard and Savickas (2002).
Panel (A) : Full Sample Results
Test Sample
DAY
ψ
-26
0.861
-2
1.350
-1
1.311
0
1
Cψ
Matched Sample (Returns)
p-value
ψ
0.861
0.891
0.889
30.508
0.000
1.136
31.819
0.000
1.948
21.251
53.071
0.000
3.305
56.376
0.000
2
0.889
57.265
25
1.049
83.290
Cψ
Matched Sample (Volatility)
p value
ψ
Cψ
p value
0.889
0.829
1.076
1.076
0.249
38.475
0.000
1.112
33.276
0.000
40.423
0.000
1.173
34.449
0.000
1.164
41.587
0.000
1.455
35.903
0.000
1.437
43.023
0.000
1.145
37.048
0.000
0.000
1.225
44.249
0.000
1.300
38.348
0.000
0.000
1.217
75.682
0.000
1.247
67.031
0.000
Panel (B) : Pre-98 Sample Results
Test Sample
DAY
ψ
-26
1.109
-2
1.065
-1
0.841
0
1.997
1
2
25
Cψ
Matched Sample (Returns)
p value
ψ
1.109
0.277
0.820
25.594
0.274
1.177
26.434
0.331
0.710
28.431
0.086
1.489
1.114
29.545
0.074
1.147
30.691
0.060
0.875
55.888
0.004
Cψ
Matched Sample (Volatility)
p value
ψ
Cψ
p value
0.820
0.815
1.139
1.139
0.229
29.942
0.000
1.271
31.134
0.000
30.651
0.000
0.729
31.863
0.000
32.141
0.000
1.143
33.006
0.000
1.364
33.505
0.000
1.381
34.387
0.000
1.256
34.761
0.000
1.375
35.762
0.000
0.959
58.796
0.000
1.441
63.622
0.000
Panel (C) : Post-98 Sample Results
Test Sample
Matched Sample (Returns)
Matched Sample (Volatility)
DAY
ψ
Cψ
p value
ψ
Cψ
p value
ψ
Cψ
p value
-26
0.747
0.747
0.973
1.098
1.098
0.239
1.296
1.296
0.026
-2
1.493
33.122
0.000
1.255
74.087
0.000
1.372
45.972
0.000
-1
1.546
34.668
0.000
2.618
76.705
0.000
1.734
47.706
0.000
0
30.187
64.855
0.000
1.004
77.709
0.000
4.311
52.017
0.000
1
4.489
69.343
0.000
1.723
79.433
0.000
1.587
53.603
0.000
2
0.794
70.137
0.000
1.844
81.277
0.000
1.323
54.926
0.000
25
1.135
96.447
0.000
1.578
136.129
0.000
1.487
100.279
0.000
37
Table V : Robustness Checks- Volume and Number of Transactions
This table shows the ratio of the Pre event and Post event trading volume and number of transactions
for the firms in the sample for which data is available. This is computed for (1) the pre-event window [−L, 0]
and the (2) post-event window [0, L], where L is the length of each window and 0 represents the event date.
Four different window lengths are considered (three months, six months,one year and two years) and 3 different samples are considered in Panels A, B and C: Online sample, Return-matched sample and volatility
matched sample. Firms with extreme observations and errors in volume, transactions data are dropped from
the sample. t-stat is the t-test statistic for the 1-tail test of Ratio = 1.0vs.Ratio > 1.0 respectively
Panel (A) : Online Sample Results
Firms that moved online
Window
3 months
Post/Pre Volume
t-stat
Mean
Std.Dev
1.26
0.82
4.01∗∗∗
∗∗∗
154
1.56
141
2.73
6 months
1.36
0.98
4.63
1 year
1.76
1.52
5.92∗∗∗
1.54
∗∗∗
2 years
Post/Pre Transactions
2.07
7.58
No. of firms
Mean
Std.Dev
160
1.28
0.99
118
3.24
t-stat
No. of firms
2.53∗∗
79
1.33
3.66
∗∗∗
75
3.72
3.69∗∗∗
63
3.22
∗∗∗
53
5.05
Panel (B) : Return Matched Sample
Firms matched by size and past return
Window
Post/Pre Volume
Mean
Std.Dev
t-stat
Post/Pre Transactions
No. of firms
Mean
∗∗∗
153
1.43
140
1.69
3 months
1.40
1.37
3.61
6 months
1.40
0.93
5.06∗∗∗
∗∗∗
1 year
1.58
1.11
5.80
2 years
1.98
1.51
6.06∗∗∗
Std.Dev
t-stat
No. of firms
1.75
2.00
∗∗
66
2.43
2.21∗∗
61
∗∗∗
51
31
123
1.86
1.70
3.62
86
2.76
2.55
3.84∗∗∗
Panel (C) : Volatility Matched Sample
Firms matched by size and past volatility
Window
Post/Pre Volume
Mean
Std.Dev
t-stat
Mean
Std.Dev
t-stat
No. of firms
∗∗
160
1.07
0.78
0.75
77
157
1.30
1.29
2.00∗∗
75
3 months
1.22
1.12
2.43
6 months
1.26
1.28
2.56∗∗
1 year
2 years
1.44
1.38
1.91
1.16
Post/Pre Transactions
No. of firms
2.80
∗∗∗
3.46
∗∗∗
151
112
38
1.57
2.01
1.57
2.08
3.08
∗∗∗
71
3.53
∗∗∗
53
Table VI : Robustness Checks- Implied Volatility from the options market
This table shows the implied volatility (expressed as % per month) of at the money call options for the
sample of firms that moved online to commence their eCommerce operations. The ratio of the Pre event and
Post event implied volatility is also presented. This is computed for (1) the pre-event window [−L, 0] and
the (2) post-event window [0, L], where L is the length of each window and 0 represents the event date. Four
different window lengths are considered (three months, six months,one year and two years) and the (3) The
ratio of post to pre implied volatility is also presented in Panels A, B and C respectively. t-stat is the t-test
statistic for the 1-tail test of Ratio = 1.0vs.Ratio > 1.0 respectively
Panel (A) : Pre Event, Implied Volatility Results
Implied Volatility before firms moved online
Window
Implied Volatility
Mean
Median
Std.Dev
No. of firms
3 months
15.17%
14.46%
5.46%
86
6 months
15.06%
14.53%
5.44%
86
1 year
14.81%
14.17%
4.86%
86
2 years
14.35%
13.55%
4.70%
88
Panel (B) :Post Event, Implied Volatility Results
Implied Volatility after firms moved online
Window
Implied Volatility
Mean
Median
Std.Dev
No. of firms
3 months
15.75%
14.18%
5.98%
89
6 months
15.67%
14.57%
5.52%
90
1 year
16.54%
15.66%
5.78%
95
2 years
17.23%
16.27%
6.18%
97
Panel (C) : Ratio of Post/Pre Event, Implied Volatility
Window
Ratio of Implied Volatility Post/Pre
Mean
Median
Std.Dev
No. of firms
t-stat
3 months
1.05
1.02
0.29
86
1.50
6 months
1.06
1.03
0.24
86
2.39∗∗
1 year
1.12
1.12
0.20
86
5.27∗∗∗
2 years
1.21
1.20
0.22
88
8.75∗∗∗
∗ ∗ ∗, ∗∗, ∗ - Significant at 1%,5%,10% level respectively
39
Table VII : Demand Uncertainty in retail and eCommerce markets
Panel A : Estimated Quarterly US Retail Sales (Total and eCommerce)
This table shows the estimated quarterly U.S. retail sales (Total and eCommerce) as estimated by the U.S.
Census Bureau. Data are in millions of dollars and are not adjusted for seasonal, holiday and trading-day
differences. eCommerce sales refer to sales of goods and services on the internet, an extranet, Electronic Data
Interchange or other online system. Payment may or may not be made online. Source : US Department of
Commerce News, August 22, 2002.
Period
Retail Sales
eCommerce as a
Total
eCommerce
percentage of total
Q4,1999
784,278
5481
0.70%
Q1,2000
711,600
5814
0.82%
Q2,2000
771,691
6346
0.82%
Q3,2000
765,536
7266
0.95%
Q4,2000
810,311
9459
1.17%
Q1,2001
724,224
8256
1.14%
Q2,2001
805,245
8246
1.02%
Q3,2001
782,088
8236
1.05%
Q4,2001
856,285
11178
1.31%
Q1,2002
743,810
9880
1.33%
Q2,2002
825,532
10243
1.24%
Average
780,054.5
8218.6
1.05%
Standard Deviation
43,415.1
1866.8
Coefficient of variation
5.6%
22.7%
Table VII : Panel B : Annual US Retail Sales By Sector(Total and eCommerce)
This table shows the annual U.S. retail sales (Total and eCommerce) as measured by the U.S. Census Bureau.
Data are in millions of dollars and are not adjusted for seasonal, holiday and trading-day differences. eCommerce sales refer to sales of goods and services on the internet, an extranet, Electronic Data Interchange or
other online system. Payment may or may not be made online. Source : US Department of Commerce News,
August 22, 2002.
Industry / Sector
Total Retail Sales
Year 2000
Year 1999
Speciality Retailing
182211
Computers, Electronics
120969
eCommerce Sales
%Change
Year 2000
Year 1999
% Change
173216
5.2%
110008
10.0%
2279
962
136.9%
8821
5657
55.9%
Media, Entertainment, Services
86833
81778
6.2%
3777
2654
42.3%
Retailing, Toys, Cosmetics
535293
504242
6.2%
3556
1627
118.6%
Auto, Motorcycles, Hardware
1436872
1344278
6.9%
5912
2079
184.4%
Catalog, Auctions, Travel
167080
139619
19.7%
22749
12082
88.3%
Pharma, Real Estate, Food
632926
600717
5.4%
1239
498
148.8%
Total
3162184
2953858
7.1%
48333
25559
89.1%
40
Table VII : Panel C : Comparative estimates of Internet Advertising Spending in the
US,2000-2002 (in millions)
This table shows the estimates made as of June 2000 by various consulting firms on internet advertising
spending by US firms for the period 2000-2002. Data are in millions of dollars. Source : eMarketer.com
Name of Forecaster
Estimate for Year
2000
2001
2002
IDC
3,300
n.a
n.a
Giga Info. Group
3,950
5,770
8,000
Myers Group
4,320
6,480
10,368
Veronia Suhler and Assoc
4,500
5,700
6,900
Jupiter Communications
5,000
6,700
8,800
Global Internet Project
5,000
n.a
n.a
Aberdeen Group
5,100
n.a
n.a
Forrester Research
5,400
8,700
12,600
Lazard Freres
5,493
8,028
11,057
eMarketer
6,100
9,500
13,500
Simba
6,500
7,100
n.a
Internet Advertising Bureau
7,740
12,487
18,350
Internet Stock Report
8,100
11,300
15,900
n.a
16,300
22,900
Meckler-Media
ActivMedia
11,200
23,500
43,300
Average
5,835.9
10,130.4
15,606.8
Standard Deviation
2,044.7
5,252.8
10,340.9
Coefficient of variation
35.0%
51.9%
66.3%
41
Table VIII : Demand Uncertainty and Stock Return Volatility
This table shows the estimates of demand uncertainty estimated for each firm using its quarterly sales data
(obtained from COMPUSTAT) for the period 1990-2002. For each firm the following AR(1) process for
quarterly sales is estimated
ln(Sales(i, t)) = αi + βi ∗ ln(Sales(i, t − 1)) + ²it
Where Sales (i,t) refers to quarterly sales for firm i in quarter t. The measure of demand uncertainty for firm
i, U N C1i is the time series standard deviation of residuals ²it from the above specification. In the second
stage, a cross sectional regression is run between λi and U N C1i as follows
λi = a + b ∗ U N C1i + ei
λi is calculated for each firm for the same period for which AR(1) process is estimated (i.e.) 1990-2002 using
daily returns. The following table reports the distribution of U N C1i and the results of the cross sectional
regression. The calculation of U N C2i follows exactly the same procedure except that the AR(1) specification
in the first stage is estimated without a constant term for each firm. For the regression White (1980) corrected
standard errors are reported.
Panel A : Summary Statistics for Demand Uncertainty and λ
Variable
N
Mean
Std. Deviation
Min
Max
U N C1
166
0.24906
0.20176
0.0482
1.31
U N C2
166
0.285
0.23924
0.05
1.43843
λ
166
5.35159
2.62666
2.22567
13.30
Correlation (U N C1 , λ): 0.37871 (significant at 0.01% level.)
Correlation (U N C2 , λ): 0.33256 (significant at 0.01% level.)
Panel B : Regression Results
λi = a + b ∗ U N C1i + ei
Variable
Coefficient
Std. Error
t Value (white)
Intercept
b
4.12043
0.2662
15.48
4.94333
0.99082
4.99
Adjusted R-Squared: 0.139. F-Value: 27.63
λi = a + b ∗ U N C2i + ei
Variable
Coefficient
Std. Error
t Value (white)
Intercept
4.31097
0.26576
16.22
b
3.65127
0.88257
4.14
Adjusted R-Squared: 0.105
F-Value: 20.39
42
Table IX : Event Study Results
This table reports the results of an event study on the announcement by firms to move online. The event
study results take into account non-synchronous trading by using Scholes William Beta in the market model
estimation and event-induced variance increases for assessing test Statistics. The event study uses a market
model with a value-weighted index. SCS Z is the standardized cross section Z statistic that corrects for event
induced variance along the lines of Boehmer, Musumeci and Poulsen, (1991).The symbols $,*,**, and ***
denote statistical significance at the 10%, 5%, 1% and 0.1% levels, respectively, using a 1-tail test.
Panel A :Event Study Results - Full Sample
Window
N
CAR
Wtd. CAR
SCS Z
(-8,+2)
162
4.50%
2.52%
2.204∗
(-4,+2)
162
3.98%
2.35%
2.258∗
(-2,+2)
162
2.80%
1.46%
1.474$
(0,0)
162
2.21%
1.49%
2.955∗∗
Table IX : Panel B :Event Study Results - Pre 98 Sample
Window
N
CAR
Wtd. CAR
SCS Z
(-8,+2)
56
3.33%
1.30%
0.633
(-4,+2)
56
3.55%
1.59%
0.801
(-2,+2)
56
3.23%
1.32%
0.678
(0,0)
56
1.24%
0.51%
0.928
Table IX : Panel C :Event Study Results - Post 98 Sample
Window
N
CAR
Wtd. CAR
SCS Z
(-8,+2)
106
5.12%
3.37%
2.551∗∗
(-4,+2)
106
4.21%
2.90%
2.576∗∗
(-2,+2)
106
2.58%
1.55%
1.568$
(0,0)
106
2.72%
2.19%
2.849∗∗
43
Figure 1
The Hobbes’ Internet Timeline and the Amex Internet Index
The top panel (Figure 1A) of this figure shows that the growth rate of number of Websites,
which was approximately linear till about June 1998, became exponential in the later period.
This shows that the importance and size of the Internet, as measured by this metric increased
dramatically after this date. The bottom panel (Figure 1B) of this figure shows the evolution
of AMEX internet index in periods corresponding to Pre and Post June 1998. Copyright :
BigCharts.com.
44
Figure 2
Time Series of Average Monthly Volatility Ratio
This Figure shows the time-series of average monthly volatility ratio over a four year interval around the
event date, covering the period [-2,2] years in event time. Volatility ratio, λ for each firm is computed as
the ratio of the square root of the firm’s stock volatility to the market volatility over the same monthly
period. Market volatility is computed using the return on the equally weighted index. These ratios are
then averaged across firms in the event time. The panels show the average ratios for the net sample (Firms
that moved online) and the same ratios for 2 samples in which each net sample firm is matched to a corresponding firm by size and 2 year past return (return sample) and size and 2 year past volatility (volatility
sample). Pre98 and Post98 samples are firms that moved online before and after June 1998 respectively.
Average lambda
9
8
7
6
5
4
3
12
14
16
18
20
22
24
12
14
16
18
20
22
24
12
14
16
18
20
22
24
10
8
6
4
2
0
-2
-4
-6
-8
-10
-12
-14
-16
-18
-20
-22
-24
month
Net Sample
returnsample
volatility sample
Average lambda - Pre 98 Sample
9
8
7
6
5
4
3
10
8
6
4
2
0
-2
-4
-6
-8
-10
-12
-14
-16
-18
-20
-22
-24
month
Net Sample
returnsample
volatility sample
Average lambda - Post 98 Sample
9
8
7
6
5
4
3
10
8
6
45
4
returnsample
2
0
-2
-4
-6
-8
-10
-12
-14
-16
-18
-20
-22
-24
Net Sample
volatility sample
Figure 3
Structural Break methodology
The left panels of this figure show the volatility defined and computed as the Average of the monthly %
standard deviation of returns across firms in event time. The panels show the average for the net sample
(Firms that moved online) and the same for 2 samples in which each net sample firm is matched to a
corresponding firm by size and 2 year past return (return sample) and size and 2 year past volatility (volatility
sample). Pre98 and Post98 samples are firms that moved online before and after June 1998 respectively.
The right panels show the Wald Test Statistics computed for the purpose of testing for a structural break
in the volatility series of the actual net sample in the corresponding left panel. The date of structural
break is the date at which the Wald Statistics peaks. The 1% confidence intervals are shown in the boxes.
!
"
!
#
!
46
Figure 4
Cumulative Abnormal Volatility
This figure shows the behavior of the Cumulative Abnormal Volatility (CAV) over the event
window of [-26,25] days. Each panel shows four graphs which correspond to the CAV plots for
(a) the test sample of firms (b) the size-returns matched sample of firms (c) the size-volatility
matched sample of firms and (d) under the null hypothesis of no event effect on volatility.
Panel (A) shows the results for the full sample period, Panel (B) for pre-June 1998 sample
and Panel (C) for post-June 1998 sample.
) * + , - . / 0 1 + 2 - 354 . 1
K < . - + ? . ? + ;5J- + * = 1 82 - 354 . 1
I 1 + , 9 7J- + * = 1 82 - 354 . 1
6 7 8 1 9 0 , .. : ; 4 < + = 1 > ?>
"# ! "$ ! "" ! "% ! &' ! &# ! &$ ! &" ! &% ! ' ! # ! $ ! " ! %
$(#
"
'
&%
&$
&"
&#
"%
&'
""
"$
5 @ A BC AD EF G
) * + , - . / 0 1 + 2 - 354 . 1
K < . - + ? . ? + ;5J- + * = 1 82 - 354 . 1
I 1 + , 9 7J- + * = 1 82 - 354 . 1
6 7 8 1 9 0 , .. : ; 4 < + = 1 > ?>
"# ! "$ ! "" ! "% ! &' ! &# ! &$ ! &" ! &% ! ' ! # ! $ ! " ! %
$H#
"
'
&%
&$
&"
&#
"%(""
&'
ABC G D EF AG
I 1 + , 9 7J- + * = 1 82 - 354 . 1
) * + , - . / 0 1 + 2 - 354 . 1
K < . - + ? . ? + ;5J- + * = 1 82 - 354 . 1
6 7 8 1 9 0 , .. : ; 4 < + = 1 > ?>
"# ! "$ ! "" ! "% ! &' ! &# ! &$ ! &" ! &% ! ' ! # ! $ ! " ! %
47
" $(#
'
&%
&"
&$
&#
&'
"%
"" "$
"$
Figure 5
Implied Volatility from the options market
This figure shows the behavior of the average monthly implied volatility of at the money call
options for the firms that moved online, and for which data is available. This is plotted for
4 years around the event date. The graph also shows the number of firms in each month for
which options data was available.
48