The Quarterly Review of Economics and Finance 52 (2012) 243–255
Contents lists available at SciVerse ScienceDirect
The Quarterly Review of Economics and Finance
journal homepage: www.elsevier.com/locate/qref
Value relevance and the dot-com bubble of the 1990s夽
John J. Morris a,∗ , Pervaiz Alam b
a
b
Kansas State University, Department of Accounting, 109 Calvin Hall, Manhattan, KS 66506-0502, United States
Department of Accounting, College of Business Administration, Kent State University, P.O. Box 5190, Kent, OH 44242, United States
a r t i c l e
i n f o
Article history:
Received 30 November 2010
Received in revised form 31 January 2012
Accepted 5 April 2012
Available online 17 April 2012
Keywords:
New economy
Value relevance
Capital markets
Equity valuation
Earnings quality
Analyst forecast
a b s t r a c t
During the dot-com bubble of the 1990s, equity market valuation was a popular topic for investors,
financial analysts and academics. Some questioned whether traditional accounting and financial information had lost its value relevance, as stocks traded at multiples of earnings well in excess of historic
levels, leading Alan Greenspan to caution against “irrational exuberance.” This study examines the relation between market valuation and traditional accounting/financial information before, during and after
the bubble. We confirm previous research that documents a decline in the relation between market value and traditional accounting information leading up to the bubble period. However, we also
document that after the collapse of the bubble in 2000 this trend reverses. We also examine two
related metrics that may provide a rational explanation for this phenomenon, including the quality
of earnings, and the aggressiveness of financial analysts’ forecasts, finding some support that earnings quality may contribute to the changes in value relevance, but not the aggressiveness of analyst
forecasts.
© 2012 The Board of Trustees of the University of Illinois. Published by Elsevier B.V. All rights reserved.
1. Introduction
During the dot-com bubble of the 1990s, many questioned
the value of fundamental financial information for investment
decision-making purposes. Stocks were trading at record multiples of earnings. In fact many companies with no earnings at
all, experienced significant increases in their stock prices during
the latter half of the 1990s. A number of academic studies documented a decline in the linear relationship between earnings and
stock returns (e.g., Brown, Lo, & Lys, 1999; Ely & Waymire, 1999;
Francis & Schipper, 1999; Lev & Zarowin, 1999). Investors called
for additional information beyond the traditional financial statements based on Generally Accepted Accounting Principles (GAAP).
Some argued that earnings no longer mattered and that other
metrics such as number of clicks or page views were more appropriate in the new economy (Penman, 2003). Others argued that
bad accounting and poor accounting standards contributed to the
1990s bull market (Krugman, 2004; Stiglitz, 2003). In response to
these demands, companies began releasing so called “pro-forma”
financial information that presented what the company’s financial
statements would look like if they did not have to follow current
夽 Data availability: Data used in this study are available from public sources.
∗ Corresponding author. Tel.: +1 785 532 6185; fax: +1 785 532 5959.
E-mail addresses: jjmorris@ksu.edu (J.J. Morris), palam@kent.edu (P. Alam).
accounting guidelines. Amazon.com Inc. started the trend in the
second quarter of 1998 by excluding amortization expenses on
intangible assets, and was quickly followed by Yahoo! Inc. and others. By the middle of 2001, the majority of S&P 500 companies
excluded some GAAP expenses when reporting financial performance in their press releases (Best, 2006). The practice became
such a concern to the SEC that on December 4, 2001 it issued an
advisory statement that cautioned public companies not to mislead
investors, providing five propositions for guidance on the dissemination of pro-forma information (SEC, 2001).1
Penman (2003) describes the bubble period of the 1990s as
a pyramiding chain letter where momentum investing displaced
fundamental investing. The mood was perhaps best described by
Alan Greenspan, Chairman of the Federal Reserve Board, when he
cautioned against irrational exuberance in a speech at The American
Enterprise Institute for Public Policy Research on December 5, 1996
(Greenspan, 1996). These prophetic words have been the subject
of many discussions since, and they motivate our investigation of
this bubble and its subsequent collapse.
1
The propositions include: (1) antifraud provisions apply to pro-forma, (2) differences from GAAP and pro-forma should be clearly spelled out, (3) materiality of
omitted information is important consideration, (4) companies should follow guidelines developed by the Financial Executives International and the National Investors
Relations Institute for pro-forma style, and (5) investors are encouraged to compare
pro-forma results to GAAP-based results.
1062-9769/$ – see front matter © 2012 The Board of Trustees of the University of Illinois. Published by Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.qref.2012.04.001
244
J.J. Morris, P. Alam / The Quarterly Review of Economics and Finance 52 (2012) 243–255
Specifically, we examine the value relevance of accounting
and financial information during the period of time surrounding the dot-com bubble, from 1995 to 2000. This period of
time is referred to as the new economy period (NEP) by several researchers, who examine trends in value relevance leading
up to this time period (e.g., Core, Guay, & Van Buskirk, 2003;
Demers & Lev, 2001; Trueman, Wong, & Zhang, 2003). Our study
extends these prior studies into the post dot-com bubble period
by examining a broad cross-section of firm-year observations
from 1989 through 2006 and a sub-sample of firms in high
technology industries that are thought to be more representative of the so called “new economy firms.” We find that value
relevance as measured by regression R2 decreased during the bubble period, from 1995 to 2000, consistent with prior research,
but increased after the collapse of the bubble in 2000. We find
similar results for both high technology and low technology
sub-samples.
We also examine two related metrics as possible explanations
for this phenomenon. First, we use a proxy for perceived earnings quality and find no significant difference in perceived earnings
quality occurs during the dot-com bubble period, suggesting that
the decline in value relevance cannot be explained by a perceived
decline in the quality of financial reporting for the overall sample.
However, when we split the sample between high technology and
low technology firms, we find that the perception of earnings quality for high technology firms declined during the dot-com bubble
period, and remained low for the next four years after the bubble
collapsed. In contrast our low-tech sample reflects no change during the bubble period and improved quality perceptions following
the bust, suggesting that the investing public was still apprehensive
about the quality of financial reporting for the high-tech segment
of the economy, but not the low-tech segment. Therefore, the overall value relevance trend, which declines during the bubble period
then increases following the bursting of the bubble, can only be
explained, in part, by perceived earnings quality. We also examine
whether financial analysts may have contributed to the decline in
value relevance during the dot-com bubble period by being overly
aggressive with their forecasts. We find no evidence, of increased
aggressiveness in forecasting during the bubble period. However,
we do find a significant decline in aggressiveness following the
bubble bursting.
The results of our study are likely to be of interest to academics,
accountants, financial professionals, investors, and regulators for
a number of reasons. First, the general decline in value relevance identified by the academic community in prior research
appears to have stopped and reversed in the post bubble period,
which to our knowledge has not been previously documented. For
accounting and financial professionals, the decline in value relevance and perceived earnings quality during the dot-com bubble
period followed by an increase in value relevance after the collapse should serve to encourage continued improvement in the
quality of financial reporting. From the investor perspective, the
results indicate that reliance on traditional accounting and financial information for investment decision making still has merit.
For accounting regulators, the results provide support for resisting
any call to reduce reporting standards just because certain interest groups argue that a “new economy” requires new or different
information.
The remainder of the paper is organized as follows: Section 2 reviews prior research and develops our hypotheses,
Section 3 discusses research methods and models used in the
study, Section 4 summarizes the sample selection process, Section 5 presents the empirical results, and Section 6 summarizes
results and offers some conclusions and opportunities for future
research.
2. Prior research and hypotheses development
2.1. Prior research
The demand for additional information by investors during the
1990s motivated a number of academic studies that demonstrate
a decline in the linear relationship between earnings and stock
returns (e.g., Brown et al., 1999; Ely & Waymire, 1999; Francis &
Schipper, 1999; Lev & Zarowin, 1999). These value relevance studies typically use regressions of returns on earnings, finding that
the slope coefficients and R2 s decline over time. Ryan and Zarowin
(2003) investigate two explanations for this decline, lag and asymmetry, finding that annual earnings reflect news with a lag relative
to stock prices, and that earnings reflect good and bad news in
an asymmetric fashion. Sinha and Watts (2001) use an analytical
model to argue that an increase in alternate sources of information, either indirectly through financial analysts or directly from
companies due to increased pressure from regulators, may be contributing to the decline in relevance of financial reports. Core et al.
(2003) study a broad sample of firms from 1975 to 1999 to explore
whether, and to what extent, traditional proxies for future cash
flows are relevant for explaining equity values of firms operating
in the so called “new economy period.” They find mixed support
for their hypothesis that the new economy period is characterized
by significant changes in the relation between equity values and
traditional explanatory variables.
A growing literature also examines the role of Internet companies during the 1996–2000 stock market bubble. Keating, Lys, and
Magee (2003) investigate the decline in value of Internet firms in
the spring of 2000 and conclude that stock prices during the period
are explained more by revised investor assessments of annual
report data than by any new information. Lewellen (2003), in a
discussion of the Keating et al. (2003) paper, argues that the results
tell more about investor perceptions or misperceptions than they
tell about the underlying economics of Internet firms. He argues
that the market was irrational at that time, and offers evidence that
prices were too high. In another new economy study that focused
on anomalous stock returns around earnings announcements by
Internet firms, Trueman et al. (2003) find little evidence to suggest
that returns can be explained either by earnings news disclosed or
by risk changes, suggesting some level of irrationality in the pricing of Internet stocks. Ofek and Richardson (2003) demonstrate
that the market valuation of Internet companies was higher during
the bubble period because of the market’s limited ability to short
Internet stocks. Cooper, Khorana, Osobov, Patel, and Rau (2005)
find that internet-related firms that change their name by adding
a dot-com during the boom and removing it after the bust experience large gains in shareholder wealth. Pastor and Veronesi (2006)
argue that the Internet Nasdaq bubble during the late 1990s was
not irrational but it was a result of high volatility of stock prices due
to uncertainty about average future firm profitability. Bharath and
Viswanathan (2006) report that U.S. publicly traded Internet firms
lost nearly $428 billion during the March–December 2000 period as
a result of the bubble burst. These studies suggest the importance
and contribution of Internet firms to the stock market bubble of the
1990s.
More recent research examines other issues related to bubbles. For instance, Leger and Leone (2008) find departures from
fundamental pricing during the bubble in the UK and that consumer confidence is a strong explanatory variable in the pre-bubble
period. Louis and Eldomiaty (2010) use the Dow Jones and NASDAQ
indices to test the robustness of Binswanger’s (2004a,b,c) findings
that US stock price movements after the 1982 debt crisis are mainly
governed by non-fundamental shocks. They conclude that US stock
prices are mostly governed by speculative bubbles or irrational
J.J. Morris, P. Alam / The Quarterly Review of Economics and Finance 52 (2012) 243–255
245
exuberance. Anderson, Brooks, and Katsaris (2010) find that the
bubble of the 1990s was not confined to the information technology sector, but included financials and general industry segments
as well.
to the decline in value relevance, we would expect to see a pattern
similar to value relevance, with a decrease in quality during the
bubble followed by an increase in quality after. This leads to our
next hypothesis:
2.2. Value relevance hypothesis
H2. The quality of financial reporting as measured by perceived
earnings quality will decline during the dot-com bubble period, and
increase following the collapse of the bubble.
Although there has been a significant amount of research related
to bubbles, as discussed in the prior section, to our knowledge no
one has examined whether the value relevance findings leading
into the dot-com bubble period continue to hold in the post-bubble
period. The previous studies, in one way or another, raise the possibility that the markets are not always rational. In fact, a somewhat
controversial line of research in the finance field argues that markets may be driven more by human behavior characteristics than
by the efficient market theory. For instance, Jegadeesh and Titman
(1993, 2001) develop and test behavioral models, including the
concept of momentum strategies, finding evidence that investors
tend to overreact to momentum. In other words, when markets are
going up they tend to buy beyond the point where they should, and
likewise on the downside tend to hold on and not sell when they
should.
Assuming behavioral finance academics are at least partially
correct that the market does from time to time react irrationally,
we posit that one of those times was during the dot-com bubble
of the 1990s. Irrationality would explain why researchers would
struggle in reconciling their results with an efficient market theory.
Irrationality would also help explain why traditional accounting
measures would reflect a decline in value relevance as discovered
by Collins, Maydew, and Wiess (1997), Francis and Schipper (1999),
Brown et al. (1999), Core et al. (2003) and others. Aharon, Gavious,
and Yosef (2010) study the effects of the stock market bubble on
mergers and acquisitions, concluding that investors seem to experience a learning process in terms of the type of variables preferred
and are more cautious after the bursting of the bubble. Therefore, we further posit that if irrationality contributed to the decline
in value relevance during the bubble period, then after the bubble bursts, investors will return to a more fundamental (rational)
approach to market valuation, which leads to our first hypothesis:
2.4. Forecast accuracy hypothesis
Another rational explanation for the decline in value relevance
may be related to the influence of financial analysts. One of the
groups that Healy and Palepu (2003) single out for contributing
to the dot-com bubble is the financial analyst community. They
point out that many investors rely on financial analysts to interpret
the financial information provided by management and to assess
company strategy and management strengths and to forecast the
performance of the companies that they track. Armed with this
information, individual investors then make the final buy, sell, or
hold decisions. They note that following the collapse of the bubble,
investigators found evidence of conflicts of interest with analysts
providing “puffed up” prospects for companies that were also IPO
clients of their firm. Liu and Song (2001) examined analysts’ forecasts of earnings for Internet companies surrounding the market
crash in March 2000. They reported that analysts were more optimistic before than after the March 2000 period suggesting that
analysts’ optimism may have caused the stock market bubble. Similarly, O’Brien and Tian (2006) conclude that analysts were more
optimistic in their recommendation for Internet companies than
for non-Internet companies during the 1990s bubble period. Glaum
and Friedrich (2006) find that following the collapse of the bubble
analysts have changed their focus from revenue-oriented measures
towards an assessment of profitability and cash flow generation.
Given this situation, we expect to see a trend in analyst forecast
accuracy that is similar to the trend in value relevance of financial
information and the trend in earnings quality. In other words, if
financial analysts were overly optimistic during the bubble period,
the collapse of the bubble and the subsequent regulatory response
should mitigate that behavior, which leads to our next hypothesis:
H1. The value relevance of traditional accounting information,
which declined during the dot-com bubble period, will increase in
the period following the collapse of the bubble.
H3. Earnings forecasts by financial analysts will be more aggressive during the dot-com bubble period, and less aggressive
following the collapse of the bubble.
2.3. Earnings quality hypothesis
3. Research models
One rational explanation for the decline in value relevance may
be related to a decline in the quality of financial information provided to investors. Healy and Palepu (2003) argue that in addition
to “irrational exuberance,” other forces were at work during the
bubble that led to the collapse. They point to a number of high profile business failures including Enron, WorldCom, Tyco, and Global
Crossing who misrepresented their financial condition, and blame
auditors, financial analysts, and mutual fund managers for ignoring fundamental analysis and contributing to the “herd mentality.”
They point out that as a result of the finger pointing that followed
the scandals, and the market collapse, regulators and legislators
stepped in to change the way these professions function. An example is the oversight provided by the Sarbanes-Oxley Act (SOX),
which was quickly rushed through the US Congress. Jenkins, Kane,
and Velury (2006) find a significant increase in the magnitude of
discretionary accruals and a significant decrease in the earnings
response coefficient (ERC) during the period 1997–1999, relative to
the period 1990–1996, suggesting an overall decrease in earnings
quality. Assuming poor quality financial reporting did contribute
3.1. Value relevance model
The exact start of the new economy period is uncertain. However, Core et al. (2003, p. 54) point out that the term “new economy”
shows up as early as 1994, and that “comments from the Vice Chairman of the Federal Reserve Board suggest that the economic factors
denoting an NEP significantly accelerated around the year 1995.”
Therefore, we use 1995 as the first year of the new economy period
and the start of the related dot-com bubble period. The end of the
bubble is easier to select because the stock market collapsed in
2000. Therefore, we assume that the year 2000 is the last year of
the bubble period.
To test our value relevance hypothesis (H1), we look for a
valuation regression model that relates equity value to various
accounting and financial variables. Ohlson (1995) was one of the
first to model the market value of equity (price) as a function of
current book value plus the present value of future earnings. Subsequent researchers have modified this basic model to examine
various issues related to the association between equity values and
246
J.J. Morris, P. Alam / The Quarterly Review of Economics and Finance 52 (2012) 243–255
accounting information (Brown et al., 1999; Collins et al., 1997;
Core et al., 2003; Francis & Schipper, 1999; Lev & Thiagarijan, 1993).
We use the formulation developed by Core et al. (2003), which
has been widely cited in subsequent research (Armstrong, Davila,
Foster, & Hand, 2011; Ashton, 2005; Balachandran & Mohanram,
2011; Bonson, Cortijo, & Escobar, 2008; Hao, Jin, & Zhang, 2011;
Henderson, Kobelsky, & Richardson, 2010; Jeon & Kim, 2011;
Kothari & Shanken, 2003; Maines et al., 2003; Murgulov & Bornholt,
2009; Shah & Akbar, 2008; Simpson, 2008; Skinner, 2008; Wang &
Alam, 2007; Xu, Magnan, & Andre, 2008). Core et al. (2003) extend
their research through 1999, just prior to the bubble bursting,
which gives us a reference point for comparison purposes. Their formulation is also appropriate for our study because it includes more
detailed accounting data points than other models. Our objective,
like Core et al. (2003), is to select accounting variables that theoretically and empirically explain cross-sectional variation in stock
prices, and are expected to be robust over time. Our objective is
not to maximize the fit of any particular model therefore we do not
test it against other models.
Core et al. (2003) regress the market value of equity on the book
value of equity, current earnings, and proxies for expected earnings
growth. Following Core et al. (2003), we use income before extraordinary items as the measure of current earnings, and a variable for
loss years because prior literature documents differences in the
valuation of losses and profits (e.g., Basu, 1997; Collins et al., 1997;
Hayn, 1995). The loss-year variable is set equal to income before
extraordinary items for firm-years with losses (the negative value),
and zero for firm-years with profits. Proxies for growth include:
advertising expenditures, research and development expenditures,
capital expenditures, and sales growth over the previous year.
Brown et al. (1999) caution researchers that the use of per share
or firm level data can provide misleading results when comparing R2 values from different samples unless they control for the
scale factor effect on the estimated coefficients. In fact, they show
that in two prior studies that had concluded that R2 was increasing (Collins et al., 1997; Francis & Schipper, 1999), after controlling
for the coefficient of variation, the values decreased. Brown et al.
(1999) recommend two approaches to control for this effect. The
first is to include a proxy for the coefficient of variation of scale,
and the second is to deflate individual observations by a proxy for
scale. To be consistent with Core et al. (2003), we follow the second
recommendation and use a scaled version of the model, deflated
by the book value of equity, which they suggest is consistent with
several other valuation studies in the new economy period, such as
Rajgopal, Kothari, and Venkatachalam (2000) and Trueman, Wong,
and Zhang (2000). Following is our valuation regression model:
(MVEi,t+4m /BVEi,t ) = ˛0 + ˛1 (1/BVE)i,t + ˛2 (IBX/BVE)i,t
+ ˛3 (NEG IBX/BVE)i,t + ˛4 (RND/BVE)i,t
+ ˛5 (ADVERT/BVE)i,t + ˛6 (CAP EX/BVE)i,t
+ ˛7 (SALES GR/BVE)i,t + ε
(1)
where MVEi,t+4m = market value of equity for firm i four months
after fiscal-year end t (Compustat MKVALM)2 ; BVEi,t = Book value
of equity for firm i at fiscal-year end t (Compustat #216);
IBXi,t = net income before extraordinary items for firm i for
fiscal-year t (Compustat #18); NEG IBXi,t = net income before
extraordinary items for firm i for fiscal-year t if < 0, otherwise = 0;
RNDi,t = R&D expenditures for firm i for fiscal-year t (Compustat
#46); ADVERTi,t = advertising expenditures for firm i for fiscal-year
2
The four month lag is common practice in accounting research to allow the
market to incorporate results of the prior fiscal year financial reports.
t (Compustat #45); CAP EXi,t = capital expenditures for firm i for
fiscal-year t (Compustat #30); SALES GRi,t = 1 year change in sales
for firm i for fiscal-year t (Compustat #12 change).
Based on prior research, we expect a positive coefficient on
each of the variables except the variable for negative income
(NEG IBX), which both Core et al. (2003) and Collins et al. (1997)
found to be negative, indicating that stock prices reflect expectations of investors that large losses precede higher future cash
flows.
3.2. Earnings quality model
To test our earnings quality hypothesis (H2), we use the methodology developed by Ecker, Francis, Kim, Olsson, and Schipper (2006)
to measure perceived earnings quality by adding a variable for
accrual quality (AQfactor), to the standard CAPM asset pricing
regression model3 as follows:
Rj,t − RF,t = ˛j,t + ˇj,T (RM,t − RF,t ) + ej,T AQfactort + εj,t
(2)
where t = index for the number of trading days in year T; Rj,t = firm
j’s return on day t; RF,t = the risk free rate on day t; RM,t = the market
return on day t; AQfactort = accrual quality factor on day t.
AQfactors are daily measures of accrual quality developed by
Ecker et al. (2006) using a model first developed by Dechow and
Dichev (2002) as modified by McNichols (2002), which measures
accrual quality as the residuals from firm-specific regressions of
changes in working capital on past, present, and future operating
cash flows. It is a time specific, not firm specific, measure developed
using a dynamic portfolio technique with industry cross-sections
for each of the Fama and French (1997) industries. Ecker et al.
(2006), show that AQfactor can be correlated with the returns of any
firm to determine its exposure to poor quality because it is timespecific and not firm-specific. See Appendix A for a more detailed
explanation of how AQfactor is constructed.
The coefficient on the AQfactor variable, referred to as the eloading by Ecker et al. (2006), is a returns-based representation of
earnings quality. Similar to the way the CAPM beta captures exposure to market risk, e-loading captures investor perceptions of the
firm’s earnings quality exposure in year t. Ecker et al. (2006) show
that e-loadings exhibit predictably positive correlations with most
other proxies used to represent earnings quality. They also show
that firms with higher e-loadings have lower earnings response
coefficients and more dispersed and less accurate analysts’ forecasts. They argue that this characteristic is consistent with market
participants perceiving that higher e-loading firms have noisier
earnings signals compared to lower e-loading firms. They also find
a decline in the magnitude of e-loadings and an increase in autocorrelation over time, as firms mature. In a final test, they find
that e-loadings are highest during years containing restatement
announcements, lawsuit filings, and bankruptcies, all events that
are indicative of poor earnings quality. Based on this prior work,
we believe e-loadings represent a reasonable proxy for earnings
quality for our study. A larger e-loading coefficient implies greater
sensitivity to poor earnings quality; therefore we expect this coefficient to be higher during the dot-com bubble period and lower
before and after the bubble period.
3
In addition to the CAPM model, Ecker et al. (2006) use the 3-factor Fama and
French model which includes variables for SMB (small-minus-big) and HML (highminus-low) portfolios. We also used this model, but only report the results from the
primary CAPM model because they lead to similar conclusions.
J.J. Morris, P. Alam / The Quarterly Review of Economics and Finance 52 (2012) 243–255
247
Table 1
Three digit SIC codes used to identify Hi-Tech sub-sample.
Code
Description
283
357
360
361
362
363
364
365
366
368
481
737
873
Drugs
Computer and office equipment
Electrical machinery and equipment, excluding computers
Electrical transmission and distribution equipment
Electrical industrial apparatus
Household appliances
Electrical lighting and wiring
Household audio, video equipment, audio receiving
Communications equipment
Computer hardware
Telephone communications
Computer programming, software, data processing
Research, development, testing services
3.3. Forecast accuracy
Abarbanell and Bushee (1997, 1998) find that accounting fundamentals are associated with analysts forecast revision. Therefore,
we predict that analysts forecast accuracy will decline in the bubble period if analysts contributed to the bubble. In particular, we
expected to see that decline in analysts forecast accuracy more in
the Hi-Tech sector than in the Lo-tech sector during the bubble
period. To test our forecast accuracy hypothesis (H3), we use an
equation that measures the accuracy of the one-year-ahead analysts’ forecast of earnings per share (EPS) as a ratio of forecasted
earnings per share to actual earnings per share as follows:
Analysts forecast accuracy ratio =
forecasted EPS
actual EPS
(3)
Based on our hypothesis, we expect the analysts’ forecast accuracy
ratio to reflect more aggressively optimistic earnings during the
dot-com bubble period than during the period before and after the
bubble period. We use the median analysts’ forecast of one year
ahead EPS from Institutional Broker’s Estimate System (IBES).
3.4. Test of significance model
To measure the significance of the changes in R2 , e-loadings, and
forecast ratios between the periods, we regress each of the three
factors on dummy variables representing the periods before and
after the dot-com bubble period using the following model:
Factor = ˇ0 + ˇ1 (BEFORE) + ˇ2 (AFTER) + ε
(4)
where Factor = the adjusted R2 from the year-by-year regressions
from Eq. (1), or the e-loadings from the year-by-year regressions
from Eq. (2), or the forecast accuracy ratios for each year from Eq.
(3); BEFORE = dummy variable is equal to 1 before the dot-com bubble period, otherwise 0; AFTER = dummy variable is equal to 1 after
the dot-com bubble period, otherwise 0.
In this model, using OLS regression, the coefficient on the intercept represents the mean value during the dot-com bubble period.
The coefficients on the BEFORE (AFTER) variables represent the differences or change from the intercept (the bubble period) to the
period before (after) the bubble. We report our results using auto
regression with the Yule–Walker method and lag of 2, which adjust
the coefficients and the related t-statistics for autocorrelation.
4. Sample data selection
Fig. 1. Comparison of adjusted R2 over time.
Services (WRDS) system. We define a firm-year consistent with
Compustat data conventions. For example, the year 1999 corresponds to firms with fiscal years ending between June 1999 and
May 2000. Although Core et al. (2003) collected data for years
beginning in 1975, our data are only collected for years 1989–2006,
since our intent is not to replicate their study, but rather to use their
model for the years leading up to the dot-com bubble and extending
it into the post-bubble period. Since the bubble period lasted for six
years (1995–2000), we used six years before and six years after for
our pre- and post-bubble periods. Following Core et al. (2003), and
Morck, Shleifer, and Vishney (1998), we set R&D, advertising, and
capital expenditures equal to zero when their values are missing.
We start with an initial sample of 147,344 total firm-years from
the Compustat database. Following prior studies (Brown et al.,
1999; Collins et al., 1997; Core et al., 2003), we exclude observations
for all financial institutions,4 observations with negative or missing book values and/or market values, and observations with other
missing data. Also, consistent with prior studies, in order to mitigate the effect of extreme values, we delete observations in the top
and bottom one-half percent of market-to-book value ratios each
year, which leaves a final sample of 68,298 firm-year observations.
To address outliers in the independent variables, we winsorize all
values at the 1 percent and 99 percent levels by year.
Because many of the firms in high technology industries were
the ones that argued for using non-traditional financial measures
for market valuations, we divide the sample into two sub-samples:
(1) observations from Hi-Tech firms and (2) the remaining observations referred to as Lo-Tech firms. Following Core et al. (2003),
we select the sub-sample of Hi-Tech firms by using the same classification scheme used by Francis and Schipper (1999), which is
based on whether firms in the industry are likely to have significant unrecorded intangible assets. Examples of these industries
4.1. Value relevance data
We begin by selecting data from the Compustat database for all
active US companies (CS Active) using the Wharton Research Data
4
Financial institutions (SIC codes 6000–6999) are excluded because they have
significantly different disclosure requirements and accounting rules.
248
J.J. Morris, P. Alam / The Quarterly Review of Economics and Finance 52 (2012) 243–255
Table 2
Comparison of adjusted R2 values.
Total
Hi-Tech
Lo-Tech
Panel A: Mean values by year
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
0.446
0.403
0.347
0.442
0.329
0.318
0.277
0.271
0.285
0.244
0.245
0.245
0.312
0.391
0.430
0.376
0.383
0.402
0.544
0.465
0.352
0.399
0.220
0.239
0.257
0.250
0.276
0.217
0.206
0.173
0.313
0.405
0.451
0.434
0.429
0.529
0.435
0.393
0.342
0.485
0.373
0.350
0.272
0.273
0.301
0.279
0.257
0.319
0.332
0.428
0.422
0.336
0.347
0.304
Panel B: Mean values by period
Pre-bubble period (1989–1994)
Bubble period (1995–2000)
Post-bubble period (2001–2006)
0.381
0.261
0.382
0.370
0.230
0.427
0.397
0.284
0.361
Sign
Panel C: Test of significance
(+)
Intercept
(+)
BEFORE
AFTER
(+)
Observations
Total R2
Coef.
p-Value
Coef.
p-Value
Coef.
p-Value
0.255
0.127
0.128
18
0.723
<0.001
<0.001
<0.001
0.238
0.133
0.190
18
0.618
<0.001
0.042
0.007
0.280
0.116
0.086
18
0.610
<0.001
<0.001
0.001
Values in bold represent highest and lowest values for each column.
Adjusted R2 results are from the following regression model:
(MVEi,t+4m /BVEi,t ) = ˛0 + ˛1 (1/BVE)i,t + ˛2 (IBX/BVE)i,t + ˛3 (NEG IBX/BVE)i,t + ˛4 (RND/BVE)i,t + ˛5 (ADVERT/BVE)i,t + ˛6 (CAP EX/BVE)i,t + ˛7 (SALES GR/BVE)i,t + ε
where MVEi,t+4m = market value of equity for firm i four months after fiscal-year end t (Compustat MKVALM); BVEi,t = book value of equity for firm i at fiscal-year end t
(Compustat #216); IBXi,t = net income before extraordinary items for firm i at fiscal-year end t (Compustat #18); NEG IBXi,t = net income before extraordinary items for firm i
at fiscal-year end t if < 0, otherwise = 0; RNDi,t = R&D expenditures for firm i at fiscal-year end t (Compustat #46); ADVERTi,t = advertising expenditures for firm i at fiscal-year
end t (Compustat #45); CAP EXi,t = capital expenditures for firm i at fiscal-year end t (Compustat #30); SALES GRi,t = 1 year change in sales for firm i at fiscal-year end t
(Compustat #12 change).
Test of significance is based on auto regressions of year-by-year adjusted R2 s using SAS PROC AUTOREG with the Yule–Walker method and lag of 2 with the following model:
Adjusted R2 = ˇ0 + ˇ1 (BEFORE) + ˇ2 (AFTER) + ε
where Intercept = mean value during the bubble period (1995–2000); BEFORE = dummy variable is equal to 1 before the dot-com bubble (1989–1994), otherwise is equal to
0; AFTER = dummy variable is equal to 1 after the dot-com bubble (2001–2006).
include: computer hardware and software, pharmaceuticals, electronic equipment and telecommunications. This selection process
resulted in a sub-sample of 20,679 Hi-Tech and 47,619 Lo-Tech
firm-year observations over the 18 year time period. Table 1 provides a listing of the three-digit SIC codes used to select the Hi-Tech
sub-sample.
by eliminating the top and bottom 1 percent of daily returns each
year leaving a sample of 23,116,343 daily observations. We also follow Ecker et al. (2006) and exclude any firm-year regressions with
less than 100 daily observations, resulting in a total of 97,176 firmyear regressions. We use the same approach to segment the sample
into Hi-Tech and Lo-Tech that we used for the value relevance
data.
4.2. Earnings quality data
4.3. Analysts’ forecast accuracy data
We use 4288 observations of market index returns (RM ), risk
free rates (Rf ), and accruals quality index returns (AQfactor) for
the period January 1, 1989 through December 31, 2005.5 We then
extract from the Center for Research in Securities Prices (CRSP)
database the daily returns for all firms listed during the same period
(1/1/1989 to 12/31/2005). We eliminate observations with missing returns and for all financial institutions. We address outliers
5
We thank Jennifer Francis for making this data available to us. In this section, our
analysis is limited to the time period ending 12/31/2005, which is the time period
for the data provided on the Francis website.
For the forecast accuracy data, we begin by selecting all firms
for which one-year ahead forecast earnings per share (EPS) data
are available in the Thomson Financial I/B/E/S database from 1989
through 2006. We exclude those observations for which SIC codes
were not available and all financial institutions leaving a sample of
50,004 firm-year observations. We use the median forecast rather
than the mean forecast in order to accommodate the impact that
a small number of analysts following each firm may have on the
mean value. We then use the same approach to segment the sample into Hi-Tech and Lo-Tech that we used for the value relevance
data.
J.J. Morris, P. Alam / The Quarterly Review of Economics and Finance 52 (2012) 243–255
249
Table 3
Regression of MVE/BVE by time period.
Pre-bubble
Panel A: All firms
Intercept
1/BVE
IBX/BVE
NEG IBX/BVE
RND/BVE
ADVERT/BVE
CAP EX/B/BVE
SALES GR/BVE
Observations
Adjusted R2
Panel B: Hi-Tech
Intercept
1/BVE
IBX/BVE
NEG IBX/BVE
RND/BVE
ADVERT/BVE
CAP EX/B/BVE
SALES GR/BVE
Observations
Adjusted R2
Panel C: Lo-Tech
Intercept
1/BVE
IBX/BVE
NEG IBX/BVE
RND/BVE
ADVERT/BVE
CAP EX/B/BVE
SALES GR/BVE
Observations
Adjusted R2
Bubble period
Post-bubble
Estimate
(t-stat)
Estimate
(t-stat)
Estimate
(t-stat)
0.425
4.822
10.935
−13.295
6.058
1.819
1.112
0.514
13,360
0.348
3.82***
8.09***
15.44***
−17.96***
9.76***
2.48**
4.05***
4.61***
0.740
4.459
12.153
−14.020
9.278
0.950
2.383
0.618
23,972
0.218
5.31***
8.40***
16.01***
−16.67***
14.12***
0.82
5.50***
5.92***
0.471
5.396
9.202
−10.933
4.418
3.940
4.184
0.924
30,966
0.334
4.20***
12.27***
13.94***
−15.71***
9.29***
3.14***
7.18***
8.05***
0.393
5.565
11.078
−13.894
5.465
−1.400
1.667
0.864
3,255
.328
1.55
7.07***
9.64***
−10.27***
6.20***
−0.58
1.47
3.37***
1.680
4.970
12.321
−13.632
8.360
−0.436
5.138
1.047
7,375
.169
5.26
4.68***
7.33***
−7.74***
9.56
−0.10
3.67***
3.94***
1.038
5.685
5.718
−7.075
4.217
2.463
5.262
1.616
10,049
.346
6.62***
7.66***
6.25***
−7.07***
7.58***
0.71
3.38***
7.29***
0.448
4.661
10.892
−13.086
5.599
2.354
1.085
0.431
10,105
0.354
3.64***
6.19***
12.53***
−14.89***
5.45***
3.11***
4.12***
3.47***
0.359
4.157
12.419
−14.772
8.383
1.788
1.935
0.485
16,597
0.254
2.63***
6.79***
15.75***
−15.86***
6.68***
1.79***
4.99***
4.60***
0.158
5.142
11.044
−13.146
5.486
4.324
3.966
0.598
20,917
0.334
1.07
9.44***
12.94***
−14.45***
4.60***
3.45***
7.19***
4.69***
(MVEi,t+4m /BVEi,t ) = ˛0 + ˛1 (1/BVE)i,t + ˛2 (IBX/BVE)i,t + ˛3 (NEG IBX/BVE)i,t + ˛4 (RND/BVE)i,t + ˛5 (ADVERT/BVE)i,t + ˛6 (CAP EX/BVE)i,t + ˛7 (SALES GR/BVE)i,t + ε
where MVE = market value of equity for firm i four months after fiscal-year end t (Compustat MKVALM); BVE = book value of equity for firm i at fiscal-year end t (Compustat
#216); IBX = net income before extraordinary items for firm i at fiscal-year end t (Compustat #18); NEG IBX = net income before extraordinary items for firm i at fiscal-year
end t if < 0, otherwise = 0; RND = R&D expenditures for firm i at fiscal-year end t (Compustat #46); ADVERT = advertising expenditures for firm i at fiscal-year end t (Compustat
#45); CAP EX = capital expenditures for firm i at fiscal-year end t (Compustat #30); SALES GR = 1 year change in sales for firm i at fiscal-year end t (Compustat #12 change).
Note: t-stats are adjusted for heteroscedasticity using the White method.
**
p < 0.05.
***
p < 0.01.
5. Empirical results
5.1. Value relevance results (H1)
Following prior valuation studies, we use adjusted R2 to measure value relevance (Brown et al., 1999; Collins et al., 1997; Hand,
2005; Ryan & Zarowin, 2003). Fig. 1 provides a graphical overview
of the annual change in value relevance that has taken place from
1989 to 2006 based on the adjusted R2 results from our regression
model #1. The adjusted R2 values from the annual regressions trend
downward for all three sub-groups from 1989 to 2000, consistent
with prior research that finds a decline in value relevance over time.
However, notice that all three of the lines turn sharply upward following the collapse of the dot-com bubble in 2000 forming a U
shaped curve, which is consistent with our expectations.6
Table 2 provides a listing of the annual adjusted R2 results
by year in Panel A, with the highest and lowest values for each
column set in bold face type. For instance in the total sample column, the highest adjusted R2 is recorded in the pre-bubble period
6
Untabulated results show that when we extend our analysis to 2010, we find a
repeat of the U pattern, turning back up after the market collapse of 2008–2009.
(1989: 0.446) and the lowest during the bubble period (1998:
0.244). Similar results are found for the Hi-Tech and Lo-Tech subsamples, where the highest adjusted R2 values are found during
the pre-bubble period (Hi-Tech = 1989: 0.544 and Lo-Tech = 1992:
0.485) and lowest during the bubble period (Hi-Tech = 2000: 0.173
and Lo-Tech = 1999: 0.257). These univariate statistics support the
graphic display in Fig. 1, which reflects a reversal of the downward
trend in value relevance. The rows in Panel B present the mean
values of the annual R2 s for each of the three time periods: (1) prebubble, (2) bubble and (3) post-bubble. Note that the mean of the
adjusted R2 s for the total sample goes from 0.381 before the bubble
to 0.261 during the bubble back up to 0.382 after the bubble. The
test of significance for the total sample in Panel C shows that the
coefficients for both the BEFORE and AFTER variables are highly significantly positive (p < 0.01), indicating that the explanatory power
of the regression model is statistically lower during the dot-com
period than either the period before or after, which is further support for our first hypothesis (H1). Similarly, the mean adjusted R2
of the Hi-Tech sample goes from 0.370 to 0.230 to 0.427 and the
mean of the Lo-Tech sample goes from 0.397 to 0.284 to 0.361, all
consistent with our expectations. Panel C confirms that these differences are also significant providing additional support for our
hypothesis H1.
250
J.J. Morris, P. Alam / The Quarterly Review of Economics and Finance 52 (2012) 243–255
Table 4
Auto regression of MVE/BVE and the bubble interaction variables.
All Firms
Intercept
1/BVE
IBX/BVE
NEG IBX/BVE
RND/BVE
ADVERT/BVE
CAP EX/B/BVE
SALES GR/BVE
BUB
Lo-Tech
Estimate(t-stat)
Estimate(t-stat)
Estimate(t-stat)
Estimate(t-stat)
Estimate(t-stat)
Estimate(t-stat)
0.347
(7.56)
5.135
(66.66)
10.421
(48.57)
−12.228
(−53.92)
5.931
(48.84)
2.370
(6.95)
2.755
(24.25)
0.759
(28.30)
0.720
(11.88)
0.476
(9.58)
5.314
(62.65)
9.676
(37.61)
−11.508
(−42.56)
4.547
(31.57)
3.187
(7.71)
2.940
(19.98)
0.833
(25.68)
0.272
(3.11)
−0.854
(−4.24)
2.484
(5.34)
−2.511
(−5.05)
4.549
(17.08)
−2.289
(−3.14)
−0.486
(−2.10)
−0.213
(−3.69)
68,298
0.291
0.560
(5.29)
5.466
(32.65)
8.145
(16.23)
−9.526
(−18.18)
5.31
(27.62)
0.402
(0.42)
4.671
(14.19)
1.39
(21.69)
2.013
(14.14)
0.946
(8.29)
5.670
(30.66)
6.371
(10.47)
−7.827
(−12.36)
4.363
(18.50)
1.549
(1.33)
4.176
(9.41)
1.497
(19.74)
0.751
(3.76)
−0.735
(−1.68)
5.907
(5.49)
−5.730
(−5.04)
3.968
(8.90)
−2.662
(−1.28)
1.117
(1.68)
−0.450
(−3.17)
20,679
0.271
0.246
(5.38)
4.873
(59.80)
11.669
(53.30)
−13.875
(−59.39)
6.338
(32.77)
2.898
(8.89)
2.392
(21.94)
0.517
(19.17)
0.117
(1.99)
0.238
(4.73)
5.029
(56.03)
11.300
(43.19)
−13.501
(−48.60)
5.398
(23.19)
3.459
(8.73)
2.678
(19.07)
0.548
(16.60)
0.122
(1.38)
−0.871
(−4.06)
1.109
(2.32)
−1.258
(−2.44)
2.975
(7.11)
−1.679
(−2.41)
−0.736
(−3.29)
−0.063
(−1.09)
47,619
0.309
(1/BVE) * BUB
(IBX/BVE) * BUB
(NEG IBX/BVE) * BUB
(RND/BVE) * BUB
(ADVERT/BVE) * BUB
(CAP EX/BVE) * BUB
(SALES GR/BE) * BUB
Observations
Total R2
Hi-Tech
68,298
0.287
20,679
0.267
47,619
0.308
Auto regression of (MVEi,t+4m /BVEi,t ) on the following variables and with interaction on BUB.
MVE = market value of equity for firm i four months after fiscal-year end t (Compustat MKVALM); BVE = book value of equity for firm i at fiscal-year end t (Compustat #216);
IBX = net income before extraordinary items for firm i at fiscal-year end t (Compustat #18); NEG IBX = net income before extraordinary items for firm i at fiscal-year end
t if < 0, otherwise = 0; RND = R&D expenditures for firm i at fiscal-year end t (Compustat #46); ADVERT = advertising expenditures for firm i at fiscal-year end t (Compustat
#45); CAP EX = capital expenditures for firm i at fiscal-year end t (Compustat #30); SALES GR = 1 year change in sales for firm i at fiscal-year end t (Compustat #12 change);
BUB = dummy variable is equal to 1 during the dot-com bubble (1995–2000), otherwise is equal to 0.
t-Statistics adjusted for autocorrelation using the Yule–Walker method with lag of 2.
We further test our first hypothesis by using two alternatives
to the annual regressions. First, we divide our sample into six year
blocks that represent the six years before, during, and after the
bubble period. We then run separate regressions for these blocks
of time using the same regression model that we used for annual
regressions. The results are summarized in Table 3, and reflect a
similar pattern for the adjusted R2 values, with lower values during the bubble period than before or after. The table also shows the
coefficient estimates and t-statistics from the regressions indicating that most of the variables are statistically significant at the .01
level in all three time periods. To address possible heteroscedasticity we adjust the t-statistics using the White method.
Although the adjusted R2 results in both Table 2 and Table 3
indicate that the model explains less of the variation in market
value during the bubble period than before or after, the contribution of the coefficients are more difficult to measure because
most of them are highly significant. To examine this impact, we
modify our model to add a variable to represent the bubble period
(BUB), which we set equal to (1) for observations during the bubble period and (0) for all other observations. We then run an auto
regression using SAS PROC AUTOREG on the entire sample, and an
expanded model with interaction terms for the interaction of the
independent variables with the bubble variable. Our expectation
is that if the coefficient is less relevant during the bubble period,
the interaction coefficient will be negative. Table 4 provides the
results of these regressions for the total sample and both the HiTech and Lo-Tech sub-samples. In the total sample, we find that
five out of our seven interaction terms have the expected significant (p < 0.05) negative coefficients: the reciprocal of book value,
negative income indicator, advertising, capital expenditures, and
sales growth. However, contrary to our expectations, the coefficients on the interaction with income and R&D expenditures are
significantly positive. The results for our Lo-Tech sample are similar to the total sample, only the sales growth coefficient is no
longer significant. Our Hi-Tech sample however, has only two of
the interaction terms with significant negative coefficients: negative income and sales growth and one (reciprocal of book value)
that is only marginally significant (p < 0.10). Advertising is not significant and capital expenditures are marginally significant with a
positive sign. Income and R&D expenses remain significantly positive. Although these results are not in themselves conclusive, taken
as a whole with the R2 analysis in Tables 2 and 3, they strengthen
our argument that the value relevance of accounting information
to market value did decline during the dot-com bubble period,
but reversed course and increased following the collapse of the
bubble.
J.J. Morris, P. Alam / The Quarterly Review of Economics and Finance 52 (2012) 243–255
251
Table 5
Comparison of e-loading factors.
Total
Hi-Tech
Lo-Tech
Panel A: Mean values by year
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
0.277
0.261
0.234
0.255
0.252
0.292
0.229
0.327
0.322
0.354
0.204
0.303
0.261
0.294
0.294
0.258
0.107
0.342
0.298
0.322
0.394
0.379
0.371
0.371
0.522
0.485
0.448
0.404
0.600
0.505
0.584
0.544
0.524
0.223
0.259
0.251
0.207
0.211
0.214
0.267
0.180
0.256
0.257
0.314
0.123
0.151
0.119
0.134
0.165
0.120
0.049
Panel B: Mean values by period
Pre-bubble period (1990–1994)
Bubble period (1995–2000)
Post-bubble period (2001–2003)
0.262
0.290
0.243
0.351
0.472
0.476
0.235
0.214
0.118
Exp
Total
Sign
Coef.
p-Value
Coef.
p-Value
Coef.
p-Value
0.295
−0.036
−0.049
17
0.179
<0.001
0.226
0.130
0.474
−0.124
0.008
17
.328
<0.001
0.042
0.894
0.219
0.014
−0.106
69589
0.014
<0.001
0.612
0.003
Panel C: Test of significance
(+)
Intercept
(−)
BEFORE
(−)
AFTER
Observations
2
Total R
Hi-Tech
Lo-Tech
Values in bold represent highest and lowest values for each column.
e-Loading factors are the coefficients on AQfactor from the following CAPM regression model:
Rj,t − RF,t = ˛j,t + ˇj,T (RM,t − RF,t ) + ej,T AQfactort + εj,t
where T = index for the number of trading days in year T; Rj,t = firm j’s return on day t; RF,t = the risk free rate on day t; RM,t = the market return on day t; AQfactort = accrual
quality factor on day t.
Test of significance is based on auto regression of e-loading factors using SAS PROC AUTOREG with the Yule–Walker method and lag of 2 with the following model:
e-Loading = ˇ0 + ˇ1 (BEFORE) + ˇ2 (AFTER) + ε
where Intercept = mean value during the bubble period; BEFORE = dummy variable is equal to 1 before the dot-com bubble, otherwise is equal to 0; AFTER = dummy variable
is equal to 1 after the dot-com bubble, otherwise is equal to 0.
5.2. Earnings quality results (H2)
Fig. 2 provides an overview of the change in earnings quality that has taken place from 1989 to 2005 using the e-loading
factor as a proxy. We expect the pattern here to be the inverse
of the value relevance graphic with an inverted U shape because
a higher e-loading value is an indication that investors perceive
lower earnings quality. Therefore, if investors are using earnings
quality as a rational explanation for the declining value relevance
of accounting information, then this factor should be highest during the bubble period, when value relevance is the lowest. The total
sample does not reflect the expected pattern, remaining relatively
flat across all three time periods. Lo-Tech results show relatively little change from the pre-bubble to the bubble period, but decrease
sharply in 1999 and remain low through 2005, indicating higher
perceived earnings quality. The Hi-Tech results increase during the
bubble period as expected, but remain high even after the collapse,
indicating that the market may have continued to perceive poor
earnings quality for that segment, at least through 2004.
Table 5 Panel A provides e-loading factors by year, showing that
the highest value for the total sample occurs during the bubble
period (1998: 0.354); and lowest during the post-bubble period
(2005: 0.107) indicating that perceived earnings quality was lower
during the bubble period, and that it improves in the post-bubble
period. Panel B shows that the mean values during each of the periods for the total sample follows the expected pattern, with the
highest value in the bubble period (0.290 bubble vs. 0.262 prebubble and 0.243 post-bubble), however as indicated in Panel C,
the change is not significant.
Hi-Tech firms have their highest e-loading during the bubble
period (2000: 0.600) and their lowest value is in the post-bubble
period (2005: 0.223). Panel B shows that Hi-Tech firms reflect an
increase in the mean e-loading factors during the bubble period
from 0.351 to 0.472 (p = 0.042), consistent with our expectations,
but they remain high after the collapse. In fact, the mean values
actually increase again from 0.472 to 0.476 between the bubble
and post-bubble periods, although the change is not significant
(p = 0.894).
An interesting pattern is presented for Lo-Tech firms, where the
highest e-loading factor occurs during the bubble period (1998:
0.314) and the lowest occurs in the post-bubble period (2005:
0.049), consistent with the other two samples. However, the mean
value decreases from the pre-bubble period to the bubble period
(0.235–0.214) although the amount of change is not significant
252
J.J. Morris, P. Alam / The Quarterly Review of Economics and Finance 52 (2012) 243–255
Fig. 3. Comparison of forecast accuracy over time.
Fig. 2. Comparison of e-Loadings over time.
Table 6
Comparison of forecast accuracy ratios.
Total
Hi-Tech
Lo-Tech
Panel A: Mean values by year
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
1.122
1.195
1.348
1.247
1.035
1.065
1.014
1.159
1.042
1.146
1.055
1.045
1.063
0.984
0.906
0.920
0.981
1.024
1.198
0.978
1.332
1.226
1.007
1.010
0.871
1.110
1.020
0.957
1.002
0.822
0.940
0.864
0.777
0.919
0.923
0.904
1.101
1.255
1.352
1.253
1.044
1.083
1.061
1.176
1.051
1.225
1.077
1.147
1.128
1.046
0.969
0.921
1.009
1.078
Panel B: Mean values by period
Pre-bubble period (1990–1994)
Bubble period (1995–2000)
Post-bubble period (2001–2003)
1.169
1.077
0.980
1.125
0.964
0.888
1.181
1.123
1.025
Exp
Total
Sign
Coef.
p-Value
Coef.
p-Value
Coef.
p-Value
1.066
0.117
−0.090
18
0.633
<0.001
0.015
0.050
0.964
0.171
−0.086
18
0.562
<0.001
0.003
0.093
1.118
0.071
−0.096
18
0.461
<0.001
0.187
0.083
Panel C: Test of significance
Intercept
(+)
BEFORE
(−)
(−)
AFTER
Observations
2
Total R
Hi-Tech
Lo-Tech
Values in bold represent highest and lowest values for each column.
Forecast accuracy is based on the following: median analyst forecast earnings per share/actual earnings per share.
Test of significance is based on auto regression of forecast accuracy ratios using SAS PROC AUTOREG with the Yule–Walker method and lag of 2 with the following model:
Forecast accuracy ratio = ˇ0 + ˇ1 (BEFORE) + ˇ2 (AFTER) + ε
where Intercept = mean value during the bubble period; BEFORE = dummy variable is equal to 1 before the dot-com bubble, otherwise is equal to 0; AFTER = dummy variable
is equal to 1 after the dot-com bubble, otherwise is equal to 0.
J.J. Morris, P. Alam / The Quarterly Review of Economics and Finance 52 (2012) 243–255
(p = 0.612) but the change is significant following the collapse of
the bubble, decreasing to 0.118 (p = 0.003). These results provide
some support for our alternative explanation that the decrease in
value relevance may be a rational reaction of the market to a decline
in earnings quality; therefore hypothesis H2 is partially supported.
Overall the results document a decline in perceived earnings quality for Hi-Tech firms during the dot-com bubble, which remained
low through 2004. They also suggest that the perceived quality of
earnings for Lo-Tech firms improved after the collapse.
5.3. Forecast accuracy results (H3)
Fig. 3 shows that all three of the forecast accuracy ratios for each
of three samples were higher in the pre-bubble period than the bubble period or the post-bubble period. In fact, the ratios for Hi-Tech
firms reflect a much less aggressive pattern in the bubble period,
dropping below 1.0 for three of the years. If financial analysts were
contributing to the irrational exuberance, by ignoring basic fundamentals as suggested by Healy and Palepu (2003), we would expect
an increase in this ratio during the bubble period. In fact, we would
expect to see the same inverted U pattern that we expected in the
earnings quality graphic, with the most aggressive forecasts (ratios
greater than 1.0) appearing during the bubble period and more conservative forecasts (ratios closer to or less than 1.0) before and after
the bubble period. This is not the pattern we see in Fig. 3. Instead we
see a downward trend, with the most aggressive forecasts before
the bubble and the least aggressive after the bubble for each of the
samples.
Table 6 Panel A provides the year by year mean values reflected
in the graph. Notice that the highest forecast accuracy ratios occur
in the pre-bubble period (1991) for all three groups: total (1.348),
Hi-Tech (1.332), and Lo-Tech (1.352). The lowest ratios occur in
the post-bubble period for all three groups; total (2003: 0.906),
Hi-Tech (2003: 0.777), and Lo-Tech (2004: 0.921). The mean values in Panel B reflect similar results with the highest ratios in the
pre-bubble period and the lowest in the post-bubble period. The
decrease from the pre-bubble to the bubble period is significant
for the total sample (p = 0.015) and the Hi-Tech sample (p = 0.003),
but not for the Lo-Tech sample (p = 0.187). Panel C shows that the
decrease from the bubble period to the post-bubble period is significant for the total sample (p = 0.050), and marginally significant
for the other two sub-samples (Hi-Tech: p = 0.093) and (Lo-Tech:
p = 0.083). Therefore, our third hypothesis H3 is not supported,
indicating that overly aggressive analyst forecasts do not provide
a more rational explanation for the decline in value relevance of
accounting information for those firms.
6. Conclusions
During the dot-com bubble of the 1990s, there was talk about
traditional accounting and financial information losing its value relevance with respect to serving as a proxy for expected future cash
flow. As a result, some called for changes in the way that accounting information was reported. Others argued that we were entering
into a new economy period, and that non-accounting factors were
more important in value estimation than traditional accounting
measures.
This study provides evidence that during the dot-com bubble
period, the value relevance of accounting information as measured
by adjusted R2 values from a valuation regression model was significantly lower than for the previous six years, consistent with
prior research. However, we also document that in the six years
after the dot-com bubble collapsed, the adjusted R2 values are significantly higher than during the dot-com bubble period, and are
253
comparable to the period leading up to the bubble. This trend
occurred in a broad cross-section of all firms during that period as
well as for a sub-sample of high technology firms that were considered to be typical of the “new economy” firms. These results tend
to support the argument that during the dot-com bubble period
the market may have behaved in a less rational manner than it did
before or after.
We examine two alternative explanations for the decline in
value relevance. The first is that the decline may have been due
in part to a decline in the quality of financial reporting. Using
a proxy for perceived earnings quality, we find that the perception of earnings quality has not changed significantly during the
dot-com bubble period for a broad cross section of firms. However, we did find a significant decline in perceived earnings quality
between the pre-bubble and bubble period for Hi-Tech firms but
not for Lo-Tech firms and that after the collapse of the bubble
only the perceived quality of Lo-Tech firms has improved significantly. In fact, the perceived quality of earnings for Hi-Tech
firms continued to be low, suggesting that the investing public
may still have been apprehensive about the quality of financial
reporting by these firms. Therefore, we conclude that a decline
in earnings quality can only explain part of the decline in value
relevance.
We also examine whether the decline in value relevance may be
due to over-aggressive forecast estimates by financial analysts. We
show that prior to and during the dot-com bubble period, analysts’
forecasts consistently exceeded actual results, and that after the
collapse, forecasted EPS have been lower than actual EPS, especially
for Hi-Tech firms. However, contrary to our expectations, the forecast accuracy ratios during the bubble period are not higher than
the pre-bubble or post-bubble periods. Therefore, our results do
not support an alternative suggestion that overly aggressive analyst
forecasts may have contributed to the decline in value relevance.
Our results are also contrary to prior research that finds financial
analysts were partly to blame for the dot-com bubble (Liu & Song,
2001; O’Brien & Tian, 2006). Our findings, on the other hand, show
that analysts’ forecasts were fairly close to actual earning numbers
during the bubble period. Given the mixed results between prior
studies and our findings, we recommend future studies should further investigate the role of financial analysts during the dot-com
period. One avenue of future research could be the investigation
of the stock recommendations of sell-side analysts during the dotcom period.
In summary, we found only some support for the earnings quality explanation and no support for the aggressive analyst forecast
explanation, which leaves open the possibility that irrational exuberance is responsible for the decline in value relevance. Although
the lack of support for our alternative explanations does not in
itself provide support for the irrational exuberance explanation,
we cannot exclude it as a possibility. Finally, our results show
that whatever caused the decline in value relevance, the trend
reversed following the collapse and fundamental accounting measures regained most of the value relevance they had prior to the
dot-com bubble period.
These results have implications for at least four groups: academics, financial professionals, investors, and regulators. For
academics, it confirms prior research on the decline in value relevance of accounting and financial information that was taking
place in the 1990s, and it documents a reversal in this trend following the collapse of the bubble. To our knowledge, no other
study had documented this reversal. Our study also explores various explanations as to what may have contributed to the decline,
other than “irrational exuberance.” Future behavioral and earnings
quality research may be useful in providing additional insights into
this phenomenon.
254
J.J. Morris, P. Alam / The Quarterly Review of Economics and Finance 52 (2012) 243–255
For financial professionals such as auditors, financial analysts,
mutual fund managers and corporate accountants, these results
suggest that investors do value the financial information provided
to them, especially during non-bubble times. As for investors, these
results suggest that the long history of using fundamental accounting and financial information for investment decision-making has
merit. Additionally, our findings provide warning signs to help the
investor spot future bubbles in that when the relation between
fundamental accounting variables and stock prices begin to reach
levels never before seen, a red flag should be raised.
Finally, for regulators and other standard setters, this study confirms that GAAP and other reporting regulations serve a purpose in
the process of communicating between publicly traded corporations and the investing public. They should proceed with caution
when pressure mounts to replace the historical accounting model
with “modern” approaches to financial reporting.
Appendix A. Construction of AQfactor from
http://faculty.fuqua.duke.edu/∼fecker/EFKOS 2006.htm
A Returns-Based Representation of Earnings Quality
Ecker, Francis, Kim, Olsson and Schipper
(The Accounting Review, July 2006)
Construction of AQfactor
The construction of AQfactor starts with identifying all firms
with the necessary data to estimate the underlying accruals quality
metric, developed by Dechow and Dichev (2002) and modified by
McNichols (2002). Requiring a minimum of 20 firms per industryyear, we run annual cross-sectional regressions of total current
accruals on past, present and future cash flows from operations,
as well as on gross property, plant and equipment and the change
in sales revenues, separately for each of the 48 Fama–French (1997)
industries. Accruals quality at the end of year T (AQ) is the standard
deviation of the five firm- and year-specific residuals obtained from
the regressions in Years T-5 to T-1. Lagging AQ by one year accounts
for the fact that the industry regressions contain the leading cash
flow from operations.
We assign firms to AQ deciles using a dynamic portfolio technique that allows for differences in firms’ fiscal year ends as well
as over-time changes in accruals quality. Specifically, we further
lag the AQ metric by three months after fiscal year end to ensure
public availability of the accounting data and then form deciles on
the first day of each month based on the firm’s most recent value
of AQ. If the AQ signal for the following fiscal year is missing due to
insufficient data, the firm is excluded from this portfolio formation
after twelve months (but allowed to re-enter the portfolio later).
Finally, AQfactor is defined as the equal-weighted daily return of
the four deciles of firms with the highest (=poorest) AQ less the
equal-weighted daily return of the four deciles of firms with the
lowest (=best) AQ.
The file ‘AQfactor 1970–2003.xls’ contains the 8586 daily portfolio returns from 1970 to 2003. For a more detailed description
about the procedure, including Compustat variable numbers, etc.,
please refer to the published article.
Note: In addition to the file described above, an updated file is also
available from the site, “AQfactor 1970-2005.xls” which has 9,090 daily
portfolio returns from 1970 to 2005. This is the data used in this paper.
References
Abarbanell, J., & Bushee, B. (1997). Fundamental analysis, future earnings, and stock
prices. Journal of Accounting Research, 35, 1–24.
Abarbanell, J., & Bushee, B. (1998). Abnormal returns to a fundamental analysis
strategy. The Accounting Review, 73, 19–45.
Aharon, D. Y., Gavious, I., & Yosef, R. (2010). Stock market bubble effects on mergers
and acquisitions. The Quarterly Review of Economics and Finance, 50, 456–470.
Anderson, K., Brooks, C., & Katsaris, A. (2010). Speculative bubbles in the S&P 500:
Was the tech bubble confined to the tech sector? Journal of Empirical Finance,
17, 345–361.
Armstrong, C., Davila, A., Foster, G., & Hand, J. R. M. (2011). Market-to-revenue multiples in public and private capital markets. Australian Journal of Management,
36, 15–57.
Ashton, R. (2005). Intellectual capital and value creation: A review. Journal of
Accounting Literature, 24, 53–134.
Balachandran, S., & Mohanram, P. (2011). Is the decline in the value relevance of
accounting driven by increased conservatism? Review of Accounting Studies, 16,
272–301.
Basu, S. (1997). The conservatism principle and the asymmetric timeliness of earnings. Journal of Accounting & Economics, 24, 3–37.
Best, B. (2006). Financial statements in the new economy. In an essay extracted
from: http://www.benbest.com/business/newecon.html (21.02.06).
Bharath, T. S., & Viswanathan, S. (2006). Is the internet bubble consistent with rationality? University of Michigan and University of Maryland. (Working paper)
Binswanger, M. (2004a). How do stock prices respond to fundamental shocks?
Finance Research Letters, 1, 90–99.
Binswanger, M. (2004b). How important are fundamentals? Evidence from a structural VAR model for the stock markets in the US, Japan and Europe. Journal of
International Financial Markets, Institutions and Money, 14, 185–201.
Binswanger, M. (2004c). Stock returns and real activity in the G-7 countries: Did the
relationship change in the early 1980? The Quarterly Review of Economics and
Finance, 44, 237–252.
Bonson, E., Cortijo, V., & Escobar, T. (2008). The role of XBRL in enhanced business
reporting (EBR). Journal of Emerging Technologies in Accounting, 5, 161–173.
Brown, S., Lo, K., & Lys, T. (1999). Use of R2 in accounting research: Measuring
changes in value relevance over the last four decades. Journal of Accounting &
Economics, 28, 83–115.
Collins, D. W., Maydew, E. L., & Wiess, I. S. (1997). Changes in the value-relevance of
earnings and equity book values over the past forty years. Journal of Accounting
& Economics, 24, 39–67.
Cooper, M., Khorana, A., Osobov, I., Patel, A., & Rau, R. (2005). Managerial actions
in response to a market downturn: Valuation effects of name changes in the
dot.com decline. Journal of Corporate Finance, 11, 319–335.
Core, J. E., Guay, W. R., & Van Buskirk, A. (2003). Market valuations in the New Economy: An investigation of what has changed. Journal of Accounting & Economics,
34, 43–67.
Dechow, P., & Dichev, I. (2002). The quality of accruals and earnings: The role of
accural estimation errors. The Accounting Review, 77, 35–59.
Demers, E., & Lev, B. (2001). A rude awakening: Internet shakeout in 2000. Review
of Accounting Studies, 6, 331–359.
Ecker, F., Francis, J., Kim, I., Olsson, P., & Schipper, K. (2006). A returns-based representation of earnings quality. The Accounting Review, 81, 749–780.
Ely, K., & Waymire, G. (1999). Accounting standard-setting organizations and earnings relevance: Longitudinal evidence from NYSE common stocks, 1927–93.
Journal of Accounting Research, 37, 293–318.
Fama, E. F., & French, K. R. (1997). Industry costs of equity. Journal of Financial
Economics, 43, 153–193.
Francis, J., & Schipper, K. (1999). Have financial statements lost their relevance?
Journal of Accounting & Economics, 39, 295–327.
Glaum, M., & Friedrich, N. (2006). After the bubble: Valuation of telecommunications
companies by financial analysts. Journal of International Financial Management
and Accounting, 17, 160–174.
Greenspan, A. (1996). The challenge of central banking in a democratic society. In a speech before the American Enterprise Institute
for
Public
Policy
Research.
Washington,
DC.
Available
from:
http://www.federalreserve.gov/boarddocs/speeches/1996/19961205.htm
(2006).
Hand, J. R. M. (2005). The value relevance of financial statements. The Accounting
Review, 80, 613–648.
Hao, S., Jin, Q., & Zhang, G. (2011). Investment growth and the relation between
equity value, earnings, and equity book value. The Accounting Review, 86,
605–635.
Hayn, C. (1995). The information content of losses. Journal of Accounting & Economics,
20
Healy, P. M., & Palepu, K. G. (2003). How the quest for efficiency corroded the market.
Harvard Business Review, 81, 75–85.
Henderson, B., Kobelsky, K., & Richardson, V. J. (2010). The relevance of information
technology expenditures. Journal of Information Systems, 24, 39–77.
Jegadeesh, N., & Titman, S. (1993). Returns to buying winners and selling losers:
Implications for stock market efficiency. The Journal of Finance, 48, 65–91.
Jegadeesh, N., & Titman, S. (2001). Profitability of momentum strategies:
An evaluation of alternative explanations. The Journal of Finance, 56,
699–720.
Jenkins, D. S., Kane, G. D., & Velury, U. (2006). Earnings quality decline and the effect
of industry specialist auditors: An analysis of the late 1990. Journal of Accounting
and Public Policy, 25, 71–90.
Jeon, S., & Kim, J. (2011). The role of R&D on the faluation of IPO firms. Journal of
International Business Research, 10, 39–57.
Keating, E. K., Lys, T., & Magee, R. P. (2003). Internet downturn: Finding valuation
factors in spring 2000. Journal of Accounting & Economics, 34, 189–236.
J.J. Morris, P. Alam / The Quarterly Review of Economics and Finance 52 (2012) 243–255
Kothari, S. P., & Shanken, J. (2003). Time-series coefficient variation in valuerelevance regressions: A discussion of Core, Guay, and Van Buskirk and new
evidence. Journal of Accounting & Economics, 34, 69–87.
Krugman, P. (2004). The great unraveling: Losing our way in the new century. New
York, NY.
Leger, L., & Leone, V. (2008). Changes in the risk structure of stock returns: Consumer
confidence and the dotcom bubble. Review of Financial Economics, 17, 228–244.
Lev, B., & Thiagarijan, R. (1993). Fundamental information analysis. Journal of
Accounting Research, 31, 190–215.
Lev, B., & Zarowin, P. A. (1999). The boundaries of financial reporting and how to
extend them. Journal of Accounting Research, 37, 353–385.
Lewellen, J. (2003). Discussion of the internet downturn: Finding valuation factors
in spring 2000. Journal of Accounting & Economics, 34, 237–247.
Liu, Q., & Song, F. (2001). The rise and fall of Internet stocks: Should financial analysts
be blamed? School of Economics, University of Hong Kong. (Working paper).
Louis, R. J., & Eldomiaty, T. (2010). How do stock prices respond to fundamental
shocks in the case of the United States? Evidence from NASDAQ and DJIA. The
Quarterly Review of Economics and Finance, 50, 310–322.
Maines, L., Bartov, E., Fairfield, P., Hirst, D. E., Iannaconi, T., Mallett, R., et al. (2003).
Implications of accounting research for the FASB’s initiatives on disclosure of
information about intagible assets. Accounting Horizons, 17, 175–185.
McNichols, M. (2002). Discussion of: The quality of accruals and earnings: The role
of accrual estimation earros. The Accounting Review, 77, 61–69.
Morck, R., Shleifer, S., & Vishney, R. (1998). Management ownership and market
valuation, an empirical analysis. Journal of Financial Economics, 20.
Murgulov, Z., & Bornholt, G. (2009). Seasoned equity offerings by new economy companies in Australia. Accounting Accountability and Performance, 15,
1–32.
O’Brien, P., & Tian, Y. (2006). Financial analysts’ role in the 1996–2000 internet bubble.
University of Waterloo. (Working paper).
Ofek, E., & Richardson, M. (2003). Dotcom mania: The rise and fall of internet stock
prices. Journal of Finance, 58, 1113–1137.
255
Ohlson, J. A. (1995). Earnings, book values, and dividends in equity valuation. Contemporary Accounting Research, 11, 661–687.
Pastor, L., & Veronesi, P. (2006). Was there a NASDAQ bubble in the late 1990? Journal
of Financial Economics, 81.
Penman, S. H. (2003). The quality of financial statements: Perspectives from the
recent stock market bubble. Accounting Horizons, 17, 77–96.
Rajgopal, S., Kothari, S. P., & Venkatachalam, M. (2000). The relevance of web traffic
for internet stock prices. University of Washington. (Working paper).
Ryan, S. G., & Zarowin, P. A. (2003). Why has the contemporaneous linear returnsearnings relation declined? The Accounting Review, 78, 523–553.
SEC. (2001). Cautionary advice regarding the use of pro-forma financial information
in earnings releases. Release 33-8039, 34-45124, FR-59, December 4, 2001.
Shah, S., & Akbar, S. (2008). Value relevance of advertising expenditure: A reveiw of
the literature. International Journal of Management Reviews, 10, 301–325.
Simpson, A. (2008). Voluntary disclosure of advertising expenditures. Journal of
Accounting Auditing and Finance, 23, 403–436.
Sinha, N., & Watts, J. (2001). Economic consequences of the declining relevance of
financial reports. Journal of Accounting Research, 39, 663–681.
Skinner, D. (2008). Accounting for intangibles – A critical review of policy recommendations. Accounting and Business Research, 38, 191–204.
Stiglitz, J. (2003). The roaring nineties: A history of the world’s most prosperous decade.
New York, NY.
Trueman, B., Wong, M. H. F., & Zhang, X. J. (2000). The eyeballs have it: Searching for
the value in internet stocks. Journal of Accounting Research, 38, 137–162.
Trueman, B., Wong, M. H. F., & Zhang, X. J. (2003). Anomalous stock returns around
internet firms’ earnings announcements. Journal of Accounting & Economics, 34,
249–271.
Wang, L., & Alam, P. (2007). Information technology capability: Firm valuation,
earnings uncertainty, and forecast accuracy. Journal of Information Systems, 21,
27–48.
Xu, B., Magnan, M., & Andre, P. (2008). The stock market valuation of R&D information in biotech firms. Contemporary Accounting Research, 24, 1291–1318.