Analyst Learning, Incentives, and Underreaction to Pension Underfunding
Xuanjuan Chen, Tong Yao, Tong Yu, and Ting Zhang*
* Chen is from Department of Finance, Kansas State University. Yao is from Department of Finance, University of
Iowa. Yu and Zhang are from College of Business Administration, University of Rhode Island. Emails:
email@example.com (Chen), firstname.lastname@example.org (Yao), email@example.com (Yu), and firstname.lastname@example.org (Zhang). We
appreciate the comments of Paul Irvine, Lawrence He, and seminar participants at Brock University, University of
Iowa, University of Rhode Island, the 2008 American Risk and Insurance Association meetings, the 2009 China
International Finance Conference meetings, the 2009 Financial Management Association meetings, and the 2009
Western Finance Association meetings. We gratefully acknowledge the contribution of Thomson Financial for
providing analyst recommendations data, available through the Institutional Brokers Estimate System. All errors are
Analyst Learning, Incentives, and Underreaction to Pension Underfunding
Corporate pension information is difficult to understand, even for sophisticated market
participants. This study uses pension funding information as a specific context to examine the
efficiency of brokerage firm analyst responses to new information. We find on average analysts
do not fully incorporate firms’ pension funding information into their earnings forecasts and
analysts under-react to pension underfunding information as a result of cognitive biases and
incentive problems. Further, analyst learning mitigates cognitive biases; analyst competition and
the presence of institutional investors help mitigate such incentive problems. Finally, evidence
suggests that analyst learning reduces market mispricing while incentive-mitigation does not
further mitigate market mispricing. It appears that the market does not anticipate analyst forecast
biases due to inexperience, but anticipates analyst biases due to incentives.
JEL Classification: G14; G32
Key words: pension underfunding, analyst underreaction, cognitive bias, learning, competition
Learning, Incentives, and Analyst Underreaction to Pension Underfunding
The (pension) funding shortfall has been the focus of analysts, the rating agencies, and regulators…But
we are not out of the woods yet. It is entirely possible that some equity analysts do not pay much attention
to pension risk when assessing P/E multiplies… and the result could be distortions of analysts’ valuations
and strategic corporate business decisions.
---- Merton (2006)
A defined benefit (DB) pension plan is considered to be underfunded when the projected
pension liabilities (the present value of total amount of employee benefits) exceeds the market
value of pension assets. The pension funding status is potentially relevant for corporate valuation
in several ways. Firms with severely underfunded pensions are required to make extra
contributions during the subsequent years, reducing cash flows to debtholders and equity holders.
In addition, reported earnings may be distorted by firms’ efforts to manipulate rate of return
assumptions for pension assets (Bergstresser, Desai, and Rauh, 2006). Finally, underfunding may
have real impact on firm operations. Rauh (2006) finds that for every dollar of mandatory
contribution to underfunded pension plan, a firm reduces its capital expenditure by $0.60 to
Whether participants in the financial market fully understand such pension funding
information is subject to debate. Some earlier studies, including Feldstein and Seligman (1981),
Feldstein and Morck (1983), Bulow, Morck and Summers (1987), and Bodie and Papke (1992),
generally conclude that the relevant pension funding information is well incorporated into stock
prices. Recently, Jin, Merton, and Bodie (2006) find that a firm’s systematic risk, or beta,
correctly reflects the risk level of its pension plans. In contrast, Coronado and Sharpe (2003)
find that investors are often confused by the complexity of pension accounting and misvalue
firms with substantial pensions. Further, Franzoni and Marín (2006) report an interesting pension
underfunding anomaly -- firms with severely underfunded pensions have lower stock returns
relative to firms with overfunded pensions up to five years after pension underfunding,
suggesting that the stock market overvalues firms with severely underfunded pension plans.
Pension funding may potentially influence other aspect of corporate activities. Rauh (2009) reports that pension
funding status affects firms’ risk-taking decision in pension investments. Franzoni (2009) shows that stock price
reaction to pension contributions is indicative of firms’ over-investment and under-investment problems.
In this study we focus on the response of brokerage firm analysts to the pension funding
information. Sell-side analysts are considered to be a group of sophisticated information
intermediary in the financial market, who specialize in analyzing corporate financial information.
However, in our baseline result, we find that sell-side analysts on average do not fully
incorporate firms’ pension funding information into their earnings forecasts. Analyst forecasts
for firms with underfunded pensions tend to exhibit significantly higher optimistic bias, relative
to those for fully or over-funded firms. Further, the more underfunded a firm’s pension is, the
stronger the optimistic forecast bias becomes. The evidence suggests that analysts under-react to
pension underfunding information, in a way consistent with the conclusion of Picconi (2006),2
and consistent with the stock market under-reaction documented by Franzoni and Marín (2006).
We use the baseline result as the launch-pad for the centerpiece issue of this study: what
mechanisms available in the financial market alleviate analyst underreaction to information?
Motivated by existing studies on analyst behavior we identify two possible causes of analyst
forecast bias in response to pension funding information – limited information processing
capacity as well as incentives to provide optimistic forecasts. Based on these potential causes, we
hypothesize that analysts’ learning to overcome their cognitive biases, as well as institutional
features of the stock market to attenuate analyst incentive problems, can improve the efficiency
of analyst reaction to pension underfunding information.
Existing studies show that analysts’ limited information processing capacity may lead to
their systematic forecast errors. 3 From this perspective the current practice of pension plan
accounting may present a particular challenge for analysts. Pension accounting is notably
complicated and opaque (Coronado and Sharpe, 2003; Picconi, 2006; Coronado, Mitchell,
Sharpe, and Nesbitt, 2008). The reported earnings for firms with pension underfunding may also
be opaque due to possible earnings management (Bergstresser, Desai, and Rauh, 2006). Finally,
However, Picconi (2006) does not directly examine the relation between pension underfunding and analyst
forecasts. He looks at delayed analyst reaction to changes in various components of pension information released in
corporate financial statements, such as changes in service cost, pension obligations, pension assets, rate of return
For example, there is evidence that analysts’ ability to incorporate available information into their forecasts is
inversely related to the information complexity (Plumlee, 2003) and information uncertainty (Hirshleifer, 2001;
Zhang, 2006), while positively related to the easiness of information accessibility and information classification
(Hopkins, 1996; and Hirst and Hopkins, 1998).
under the current accounting standard, firms disclose pension plan information in the annual
report only, making it difficult for analysts to fully assess pension funding on a timely basis.4
One potential mechanism to attenuate market participants’ cognitive biases is learning.
The effect of analyst learning is first documented by Mikhail, Walther, and Willis (1997), who
find that analyst forecast accuracy improves with their experience. Mikhail, Walther, and Willis
(2003) further show that forecasts by more experienced analysts better incorporate past earnings
and return information. In the context of pension funding information, we hypothesize that there
might exist two types of learning effects. First, more experienced analysts may be better at
processing complex pension funding information, thus reducing their bias in earnings forecasts.
Second, analysts may learn from their past experience in covering firms with pension
underfunding situation, thus reducing their forecast bias when a firm repeatedly underfunds its
Meanwhile, it is possible that even high-caliber analysts may intentionally provide
optimistically biased forecasts for incentive-related reasons. Such incentives may include
maintaining or developing investment banking business, generating trading commissions, and
gaining access to corporate management for better information (see, among others, Schipper
1991; Jackson 2003; Lim 2001). Hong and Kubik (2003) find that analysts tend to be rewarded
for optimistic forecasts in their career path. Firms with pensions tend to be large and matter most
for brokerage firms and for analyst career; therefore analyst incentive problems may be acute
when covering these firms.
The existing literature has suggested a few mechanisms through which analyst’s
incentive problems can be mitigated. A main source of analysts’ biased incentive is the pressure
from the investment banking and brokerage business. Ljunqvist, Marston, Starks, Wei, and Yan
(2007) find that demand from institutional clients for accurate forecasts counterbalances such
pressure and reduces analyst bias. Recently, Hong and Kacperczyk (2010) report that
competition among analysts also alleviates forecast biases. They point out two potential channels
through which this effect may take place. The first is that competition likely brings about
independent analysts, which in turn may generate a disciplinary pressure on other analysts. The
Insufficient disclosure of pension plan information and complexity of pension accounting have been criticized by
firms, researchers and practitioners for a long time. In September 2003, the FASB issued Exposure Draft to advocate
increasing the frequency – quarterly instead of current annual of pension disclosures. In September 2006, the FASB
released Statement 158, Employers’ Accounting for Defined Benefit Pension and Other Postretirement plans, which
requires firms to recognize the pension funding status in their financial statements.
second is that competition increase the cost of firms to influence analysts. In our specific context,
we hypothesize that for firms with higher institutional ownership and with stronger analyst
competition, analysts’ forecasts more efficiently incorporate pension funding information.
The empirical evidence supports these predictions. To start, we rely on two relatively
crude measures to provide suggestive evidence that both limited information capacity and
incentive problems are behind analyst forecast bias associated with pension underfunding. We
use average past forecast accuracy of an analyst on all firms she follows as a measure of her
information processing ability, and use her average past forecast bias for all firms she follows as
a measure of her incentive-induced optimism tendency. We find that analysts’ forecast biases
associated with pension information are negatively correlated with their average past accuracy,
and positively correlated with their average past bias.
To quantify the first learning effect, we follow Mikhail, Walther, and Willis (1997) to
measure analyst firm-specific experience by as the number of years since an analyst starts to
cover a firm. We find that as analyst experience increases, the average forecast bias associated
with pension underfunding is significantly reduced. Specifically, we double sort analyst forecasts
by firms’ pension funding status and analyst experience. The difference in average forecast bias
between the most under-funded and the least under-funded firms, and the difference in average
forecast bias between the most under-funded firms and over-funded firms, are substantially
higher among analysts in the lowest experience quartile, relative to those among analysts in the
highest experience quartile. Similar results are obtained under a refined analyst experience
measure that controls for firm size and firm age. Therefore, an analyst’s experience with the firm
she covers helps her better anticipate the negative impact of pension underfunding on firms’
We also find evidence consistent with the second learning effect: the magnitude of
forecast bias associated with pension underfunding is much higher for firms facing pension
underfunding problems for the first time, relative to firms repeatedly in the underfunding
situation. Therefore, when a firm experiences pension underfunding problems for multiple years,
analysts collectively infer from past observations of the firm when forecasting future earnings,
regardless of how long they have been following the firm. Even more interestingly, there is an
interaction between these two learning effects – the effect of analyst experience to reduce
forecast bias is stronger among firms experiencing pension underfunding for the first time. That
is, analysts’ experience matters most when they face a new situation.
Our further analysis shows that incentive-related forecast optimism is significantly
reduced by a strong presence of institutional investors as shareholders, and by competition
among analysts. Specifically, we measure institutional presence by the number shares owned by
institutional investors as a fraction of total shares outstanding (Ljunqvist et al., 2007), and
measure analyst competition by the number of analysts following the firm (Hong and
Kacperczyk 2010). Among firms with higher institutional ownership and higher analyst
coverage, forecast bias associated with pension underfunding is significantly lower. It is
interesting to note that Hong and Kacperczyk (2010) find that analyst coverage does not
significantly affect forecast bias in their control sample – firms in the highest size quartile,
although the effect is significant in their natural experiment, i.e., merging brokerage firms. In
contrast with their results, our finding on the significant effect of analyst coverage suggests that
pension underfunding may represent a situation where analysts’ incentive-related forecast bias is
more acute and hence the analyst competition effect is more visible. Hong and Kacperczyk
(2010) also point out that the effect of analyst coverage may be confounded by the analyst
selective coverage problem – analyst may drop coverage of a firm to avoid openly expressing her
negative opinion (McNichols and O’Brien 1997). To control for selective coverage, we construct
a refined measure of analyst competition by orthogonalizing analyst coverage with future
realized earnings. The results are similar. In addition, by orthogonalizing analyst coverage with
institutional ownership, we show that the effect of analyst coverage on under-funding related
forecast bias is not merely a re-incarnation of the effect of institutional ownership.
A natural question one may raise is whether the learning and incentive-attenuation effects
subsume each other. In a regression setting that control for various firm characteristics that can
be potentially linked to analyst forecast bias, we show that in general the learning effect and the
incentive-attenuation effect are jointly significant. The fact this two effects do not subsume each
other suggests that the two causes of analyst forecast bias, cognitive bias and incentives, are
largely independent of each other. On the other hand, the interaction between the two learning
effects and the interaction between the two incentive attenuation effects are both significant in
our regression analysis.
Further, we investigate whether the mechanisms identified to improve analyst response to
pension funding information are also effective to reduce market mispricing in the form of the
pension underfunding anomaly documented by Franzoni and Marín (2006). We find that
negative relation between pension underfunding and subsequent stock returns substantially
weakens with more experienced analysts and this effect is more pronounced among firms
experiencing underfunding for the first time. This result is consistent with the view that learning
improves the accuracy of analyst forecasts, reducing market mispricing. An interesting finding is
that, however, strong presence of institutional ownership and strong analyst competition do not
significantly improve subsequent stock performance.
Finally, we investigate the level of investor sophistication. Specifically, we examine how
the subsequent stock performance relates to forecast errors related to analyst learning and
incentive attenuation. The analysis involves three steps. First, we back out forecast errors related
to analyst learning and incentive attenuation based on parameters estimated from the regressions
of individual analysts’ forecast errors on firm funding ratios, variables of analyst learning,
variables of incentive attenuation, and the full set of control variables. Second, the fitted value
of analyst forecast errors from these two sources are aggregated to the firm level. Third, we
regress stock performance on the fitted values of analyst forecast errors. We find stock returns
are positively associated with learning-related forecast errors, however, uncorrelated with
incentive attenuation related forecast errors. The finding potentially suggests that investors
anticipate analyst incentive biases in reporting earnings forecasts upon pension underfunding. In
other words, the market adjusts its expectation ex ante. Consequently, incentive-related forecast
errors have little effects on stock returns in the subsequent year. On the other hand, investors do
not have better ability in anticipating the learning based forecast biases upon pension
underfunding than financial analysts. As a consequence, we observe learning has a role in
alleviating market mispricing associated with pension underfunding.
To sum up, our study extends and enriches the literature on the informational efficiency
of the financial market with respect to corporate pension funding (e.g., Rauh, 2006; Picconi,
2006 Franzoni and Marín 2006). It identifies two sets of factors that are effective in alleviating
such underreaction with respect to pension funding information – analyst learning and incentive-
attenuation mechanisms including institutional demand and analyst competition. Moreover, we
directly examine the influence of various bias attenuation process on future stock performance.
We find learning helps to mitigates market mispricing associated with pension underfunding
while analyst competition and institutional ownership do not, suggesting investors are informed
of analyst incentive biases while they do not have better ability to understand pension
underfunding than analysts.
The rest of the paper is organized as follows. Section II introduces pension underfunding
and its effects on firm balance sheet and income statement items. We also discuss analyst biases
in her forecasts. Section III describes the data and sample. Empirical results are documented in
Section IV. Section V concludes.
II. Pension Underfunding, Accounting Complexity, and Analyst Biases
In this section, we present background information on defined benefit pension plans. We
discuss pension plan underfunding, the complexity and opaqueness of pension accounting, and
the analyst biases in their earnings forecasts.
A. Pension Plan Underfunding
With a defined benefit pension plan, employees are entitled at retirement to receive a
certain amount of benefit primarily based on the years of service and salaries. The present value
of total amount of employees’ benefits represents a firm’s projected pension benefit obligations
(PBO).5 Required by the Employee Retirement Income Security Act (ERISA) of 1974, firms
need to earmark a certain amount of assets to meet their pension obligations. The market value of
these assets is called the fair value of pension assets (FVPA). A pension plan is overfunded if
FVPA are larger than PBO, and firms may transfer part of plan assets as firm own assets subject
to various restrict requirements. A pension plan is underfunded if FVPA is lower than PBO, and
firms have to make contributions to their pension plans.
As of the fiscal year end of 2005, about 20% of firms have defined benefit (DB) pension
plans. DB pension plans are more popular among large companies – two third of S&P 500 firms
Another measure of pension liabilities is accumulated benefit obligation (ABO). ABO is the present value of
pension benefits earned to date with benefits computed based on current compensation levels. ABO can be
considered as pension liabilities at a quit basis; that is, it represents the pension obligations that a firm owes to
employees if it ceases to run the business today. A firm’s ABO is usually less than its PBO because ABO assumes
that salaries will not rise into the future. PBO assumes salary increases, and it is more economically appealing.
SFAS 87, the FASB (Financial Accounting Standard Board) statement that mandates current pension reporting
standard specifies that pension liabilities should be stated to PBO. Given these considerations, we use PBO to
measure a firm’s pension liabilities.
provide employees with defined benefit pension plans (Coronado and Sharpe, 2003). In Panel A
of Figure 1, we show the proportion of firms with underfunded plans in all firms sponsoring DB
plans from 1988 to 2005 in COMPUSTAT database. About 36% of firms that use DB plans have
underfunded plans in 1988. The ratio rises to about 60% in early 1990s. Due to booming stock
market in late 1990s, the percentage of underfunded firms drops to around 32% in 1999.
However, pension plans funding status has deteriorated significantly since 2001 when the U.S.
stock market declined. There are more than 90% firms underfunded in 2005.
B. Complexity of Pension Accounting
The complexity of pension accounting lies in that, although sponsoring companies are
liable for the benefits promised to their employees, SFAS 87 requires that the majority of the
relevant pension information is recorded as off-balance sheet items. The only pension related
items recorded on the balance sheet are prepaid pension costs or accrued pension costs and
additional minimum liability. 6 Consequently, balance sheet and footnote provide different or
even conflicting information on pension assets and liabilities, and eventually on the funded status
of pension plans (e.g., Coronado and Sharpe, 2003; Picconi, 2006; Coronado, Mitchell, Sharpe,
and Nesbitt, 2008).
In Panel B of Figure 1, we plot the amount of actual pension funding and pension surplus
recognized on the balance sheet for each of our sample years. Pension funding surplus (deficit) is
defined as the difference between the fair value of plan assets (FVPA) and the present value of
pension obligations (PBO), where FVPA is the sum of overfunded pension plan assets (data 287)
and underfunded pension plan assets (data 296) and PBO is the sum of overfunded pension
obligations (data 286) and underfunded pension obligations (data 294).7 Following Shivdasani
and Stefanescu (2008), we compute pension surplus recognized on the balance sheet as the sum
of net amount of the accrued pension liability (data 290), prepaid pension liability (data 300),
and additional minimum pension liability (data 298).
The balance sheet records (i) the extent to which pension contributions paid were above or less than the recorded
pension expense (i.e. the prepaid pension cost or accrued pension cost), (ii) in the event of severe underfunding, an
additional minimum liability (AML) occurs. The AML does not reflect the amount of the actual pension
contribution that firms are required to make.
Effective for companies with fiscal years beginning after December 15, 1997, Statement of Financial Accounting
Standards (SFAS) 132 permits companies to combine their disclosures regarding over- and under-funded accounts
in particular circumstances. Due to this accounting change, Compustat database no longer differentiates pension
information between overfunded and underfunded firms.
We observe a huge difference between aggregate pension funding surplus (deficit) and
the amount recognized on the balance sheet, especially after 2001. Relative to a substantial
amount of pension deficit since 2002, firms have reported gradually increased pension surplus on
the balance sheet. For instance, there were totally $66 billion in pension surplus reported on the
balance sheet in 2003, much lower than $523 billion in pension deficit. The difference between
actual pension funding and the amount recognized on the balance sheet reflects the difference in
actual economic pension funding status and pension accounting practices. The actual pension
funding represents an economic definition, namely, the difference between market value of plan
assets (FVPA) and projected benefit obligations (PBO). It represents a firm’s actual pension
funding situation. The amount recognized on the balance sheet simply represents an accounting
definition, which can be significantly different from the actual pension surplus/deficit (e.g.,
Coronado and Sharpe, 2003; Picconi, 2006; Coronado, Mitchell, Sharpe, and Nesbitt, 2008).
This further shows that pension numbers reported by firms could mislead investors (and
analysts) as to the real funding status of pension plans.
Further, under SFAS 87, income statements would not instantly reflect the accurate
pension expense information due to the smoothing mechanism of recognition of pension-related
expenses. The accounting item reported in the income statement is the net periodic pension cost,
which involves four parts: service cost, interest cost, other cost, and expected returns on plan
There are issues arising from accounting treatments on pension-related expenses. First,
note that the expected return, rather than the realized return is used, according to the SFAS 87.
In a year that pension assets are realized with a substantial low return which gives rise to plan
underfunding, unrealized capital losses are not reflected in income statements. Instead, the
difference between the actual returns on plan assets and the expected returns are recorded using
Service cost is the present value of pension benefits earned by employees over the last year, which essentially is
deferred compensation. Interest cost comes from the growth in the projected pension liability over the last year.
Different from service cost, interest cost reflects the increase in the pension obligations just due to the passage of
time as employees are getting closer to receiving their pension benefits. Other cost includes actuarial gain,
amortization of transition asset and prior service, and plan amendments. The expected return on plan assets is
determined by multiplying the expected rate of return on plan assets and the market value of plan assets (FVPA).
The FVPA could be the market value of plan assets at the beginning of the year or a five-year moving average value
of plan assets.
an amortization method, which results in pension expenses actually reflecting a deferred value of
expected returns. Second, firms are required to make additional contribution to make up their
underfunding. 9 Such contribution would not be reflected in the income statement in the year
when pension underfunding has occurred. Rather, pension underfunding has effects in pension
costs in the subsequent years. Further complicating this issue, as discussed earlier, the majority
of pension assets and liabilities are off-balance sheet items. Without getting into footnotes
embedded in company financial statements, investors and financial analysts would not discover a
firm’s actual pension funding status. Taken together, the delay in recording pension expenses
may lead analysts to provide biased earnings forecasts.10 11
In the appendix, we provide a simple model to illustrate various impacts of pension
underfunding that may cause analyst forecast errors.
The minimum pension contribution that companies are required to make in a given year depends on the funding
status of the pension plan. The minimum contribution required is equal to the normal cost of the plan plus the level
of underfunding amortized over 30 years. If the plan is severely underfunded, additional mandatory contributions are
required. If pension assets are less than 90% of ABO, the plan may be liable for an additional contribution and the
company must cover the underfunded amount over 3-5 years. The Pension Equity Act 2006 sets even restrict
funding rules – beginning in 2008, sponsors are required to fund to 100% of all liabilities (including lump sum
distributions and early retirement benefits) accrued to participants and beneficiaries within seven years. We expect
more binding effect of the new funding rules on a firm’s operating performance and future cash flows.
In September 2006, the FASB released Statement 158 (effective fiscal year December 15, 2006), Employers’
Accounting for Defined Benefit Pension and Other Postretirement Plans. Statement 158 requires that firms
recognize the funding status of defined benefit pension and other postretirement benefit plans in their statements of
financial position. However, SFAS 158 has no changes in the accounting treatment of pension expenses and thus,
the same issue still exists. SFAS 158 is the outcome of Phase I of FASB’s long-term project to reconsider all aspects
of pension and other postretirement benefit accounting, with the objective to provide useful and relevant pension
accounting information to investors and creditors. We focus on SFAS 87 because our sample ends by fiscal year
A further consideration of pension underfunding is its economic effect. Pension underfunding may reduce a firm’s
capital expenditures and result in underinvestment because underfunded firms are required by law to make financial
contributions to their pension funds to meet certain funding status. Under ERISA, pension liabilities have a claim
equal to that of a federal tax lien; that is, a company’s pension liability is senior to debentures, bank loans, and the
claims of other corporate creditors (Martin and Henderson, 1983). Because of the seniority of pension liabilities,
contributions to pension plans take precedence over consumer claims, and secured and unsecured creditors’ claim
(and behind administrative expenses and employee wages) in the event of firm bankruptcy, according to the absolute
priority rule. The underfunded firms therefore face a serious “pension obligation overhang” issue. Similar to debt
overhang (Myers, 1977; Lamont, 1995), pension obligation overhang occurs when required pension contributions
caused by pension underfunding deter new investments and capital expenditures because the benefits from new
investments will go to pension beneficiaries, not new investors. As reported by Rauh (2006), these contributions
would reduce a firm’s internal capitals. If a firm is financially constrained, contribution requirements could also
affect its ability to invest in new capital, conduct research and development (R&D), and make acquisition, resulting
in a significant underinvestment.
C. Analysts’ cognitive based Bias and Incentive Based Bias
A well established fact in the analyst forecast literature is that analysts typically bias their
forecasts upwards. Barber et al. (2006) report that in mid-2000 buy recommendations accounted
for about three-fourths of total outstanding recommendations while sell recommendations totaled
only 2 percent. Such bias may be attributed to two causes. One cause for analyst optimism is that
the analyst cognitive-based argument, based on which analysts may fail to set aside preexisiting
perceptions when processing information. Barberis et al. (1998) study analyst behavior on
overall firms and model how analysts overemphasize prior information and underemphasize new
information when valuing firms. Cornell (2001) finds that analysts are reluctant to recognize
negative changes in corporate fundamentals, and Eames et al. (2002) suggest that, when
generating earnings forecasts, analysts tend to process information in a manner that biases
forecasts in the direction that supports their investment recommendation. Abarbanell and Lehavy
(2003) find that cognitive obstacles prevent analysts from revising their forecasts downward, and
Friesena and Wellerb (2006) report strong evidence that analysts are overconfident regarding the
precision of their own information and are also subject to cognitive dissonance bias.
One potential mechanism to alleviate analysts’ cognitive biases is learning. Hogarth
(1987) lists three conditions necessary for the reduction of underreaction or a cognitive bias: (1)
feedback, (2) a repetitive task, and (3) an ability to keep a record of predictions and actual
results, which are particularly applicable to a pension funding framework. The effect of analyst
learning is first empirically documented by Mikhail, Walther, and Willis (1997), who find that
analyst forecast accuracy improves with their experience. Mikhail, Walther, and Willis (2003)
further show that forecasts by more experienced analysts better incorporate past earnings and
The other cause for analysts to issue optimistic stock recommendations can be linked to
certain strategy-based incentives, including cultivating management relations to access private
information (Francis and Philbrick, 1993; Hodgkinson, 2001; Boni and Womack, 2002; Conrad
et al., 2006), generating investment banking business (Dugar and Nathan, 1995; Lin and
McNichols, 1998; Irvine, 2004; O’Brien et al., 2005; Barber et al., 2006), boosting trading
commissions (Kim and Lustgarten, 1998), and career concern (Hong and Kubik, 2003). In
conjunction with this incentives hypothesis, just as analysts have incentives to respond promptly
to good news, they also have incentives to delay their response to—or ignore—bad news.
In terms of mechanism to alleviate incentive-based analyst optimism, Ljunqvist et al.
(2007) find that demand from institutional clients for accurate forecasts counterbalances such
pressure and reduces analyst bias. Recently, Hong and Kacperczyk (2010) report that
competition among analysts also alleviates forecast biases. They point out two potential channels
through which this effect may take place. The first is that competition likely brings about
independent analysts, which in turn may generate a disciplinary pressure on other analysts. The
second is that competition increase the cost of firms to influence analysts.
It should be noted that the literature on pension underfunding predominantly focuses on
the sophistication of pension funding information. In other words, the general consensus on the
cause of under-reaction of analysts and investors is mainly on the cognitive bias. The only paper
we are aware of examining analyst response to pension underfunding is Picconi (2006). He
points out that “the persistent tardiness of analysts to incorporate this relevant and economically
significant information about earnings is surprising given that they are provided with pension
information on a repeated and timely basis.” He suggests that “this may be due to the complexity
of the information presented and the effort costs associated with incorporating this information.”
III. Data and Sample
A. Pension plan and Analyst Earnings Forecasts Data
Our data primarily come from three sources. The accounting data about defined benefit
pension plan is from Compustat annual file. Data on stock returns is from CRSP file. Analyst
earnings forecasts data come from the I/B/E/S.
The initial sample consists of all firms available on Compustat file from 1988 to 2005
that sponsor a defined benefit pension plan. SFAS 87, the FASB (Financial Accounting Standard
Board) statement that mandates current pension reporting standard, took effect for fiscal years
ending after December 15, 1986. Following Picconi (2006), we choose fiscal year 1988 as the
beginning year to ensure that all firms in the sample comply with SFAS 87. We identify a firm as
the sponsor of DB pension plans if it has pension assets and obligations items available in
Compustat file. To be included in our pension sample, firms need to satisfy several selection
criteria. First, a firm must have at least two years pension accounting data available in
Compustat. This requirement alleviates the selection bias induced by the way Compustat
constructs data (Banz and Breen, 1986). Second, to avoid market microstructure issues in
measuring returns, stock price at the end of July must be higher than $1.
We also use the I/B/E/S data to estimate analyst earnings forecast errors and forecast
revisions. We use the I/B/E/S individual analyst file in our analysis. After combining pension
sample with analyst forecast data, we have 78,480 analyst-firm-year observations from fiscal
year 1988 to 2005.
B. Funding Ratio and Analysts Forecast Error Measures
Our measure of pension funded status is the pension funding ratio. Following Franzoni
and Marín (2006), we define pension funding ratio (FRi,t) for firm i at fiscal year t as the plan
surplus (deficit) scaled by total equity market value at the end of fiscal year: 12
FVPA i ,t PBO i ,t
FR i,t = (1)
Market Value i ,t
where FVPA is the fair value of plan assets and PBO is the present value of pension obligations.
Consistent iwth Franzoni and Marín (2006) and Picconi (2006), we measure FVPA as the sum of
overfunded pension plan assets (Compustat data 287) and underfunded pension plan assets (data
296), and measure PBO as the sum of overfunded pension obligations (data 286) and
underfunded pension obligations (data 294). Note that all the above Compustat data are not
reported in balance sheets, but in footnote of financial statements. To alleviate the influence of
outliers, in each year we winsorize the top and bottom 1% of funding ratios across all firms in
We classify a firm as pension overfunded or underfunded based on funding ratio. If
funding ratio is zero or positive (FVPA ≥ PBO), a firm is viewed to have overfunded pension
plans. If funding ratio is less than zero, a firm is considered to have underfunded pension plans.
To analyze analysts’ response to pension information, we use the I/B/E/S detail file to
estimate individual analyst forecast errors for the forthcoming fiscal year (i.e., one-year earnings
forecast). We calculate earnings forecast errors (FCE) of analyst j for firm i for the analyst
We choose market value as a deflator because it is correlated with a firm’s future cash flow, information diffusion,
and credit constraints. A potential issue of using market value as a deflator, however, is that this ratio could correlate
to a higher B/M ratio, not necessarily implying a better funding status, as suggested by Franzoni and Marín (2006).
Other deflators include total assets, total book equity, and PBO. Our results do not change qualitatively if we use
these different deflators for pension funding ratios.
forecast made in month m of year t as the difference between the actual EPS from COMPUSTAT
and forecasted earnings covered by I/B/E/S, scaled by stock price at the beginning of the year.
Actual EPS i ,t Analyst Forecasted EPS imj ,t
FCEimj ,t =
Stock Pr icei ,t 1
Similar to Bradshaw, Richardson, and Sloan (2001), we track individual analyst EPS
forecast errors for the forthcoming fiscal year four months after the release of annual earnings
report for the most recent fiscal year, up to the month when firms report their next-year
earnings.13 Most firms file their 10K reports within three months after the fiscal year end. This
timing of earnings forecast ensures that analysts already have access to firm annual financial
statements, particularly, pension plan information, when make earning forecasts. The average
earnings forecast errors of analyst j for firm i for year t is then estimated as follows:
FCE i,j,t =
i , j ,t (3)
where M is the number of months within the tracking period.
For analysis at analyst-firm-year level, we use individual analyst forecast error defined in
equation (3) directly. Yet, for analysis at firm-year level, we average FCEi , j ,t across all analysts
covering firm i during year t. That is,
FCEi,t = FCEi , j ,t (4)
J j 1
C. Descriptive Statistics
We report descriptive statistics of the pension sample Table 1. As shown in Panel A,
about 20% of firms sponsoring DB pension plans as of the fiscal year end of 2005. DB pension
plans are more popular among large companies – two third of S&P 500 firms provide employees
with defined benefit pension plans (Coronado and Sharpe, 2003). About 34% of sponsoring
firms have underfunded plans in 1988. The ratio rises to about 60% in the middle 1990s. Due to
booming stock market in late 1990s, the percentage of underfunded firms drops to around 31% in
Bradshaw et al. (2001) track forecast errors immediately following the end of fiscal year end, while we allow for a
delay of four months after fiscal year end for the release of earning information. For most firms, our tracking period
has 6 months (e.g., May to October for fiscal year ending in December). There are fairly few earnings forecasts
observations for the forthcoming year 10 months after prior year’s earnings reports (or two months proceeding to the
next year’s earnings announcements). About 1.2% (3.1%) observations in our sample have the forecast information
after 11 (12) months of prior year’s earnings announcements.
2000. However, pension plans funding status has deteriorated significantly since 2001 when the
U.S. stock market declined. There are about 87% firms underfunded in 2005. 14 In Panel A of
Figure 1, we plot the proportion of firms with underfunded plans in all firms sponsoring DB
plans from 1989 to 2005 in COMPUSTAT database.
In Panel B of Table1, we provide the basic financial information about sponsoring firms.
Firms in the pension sample are fairly large, with the average (median) market capitalization of
$4.5 ($0.6) billion and average (median) total assets of $10 ($1.1) billion. The mean (median)
fund surplus/deficit is about $29.8 ($-3.6) million, with the mean (median) funding ratio of
3.46% (-0.34%). By breaking the sample into under- and overfunded firms, we find that
overfunded firms generally are larger than underfunded firms in market value and total assets.
The average pension funding ratio for overfunded firms is 139.31%, indicating that on average
overfunded firms have pension surplus more than their total market values. In contrast, the
average funding ratio for underfunded firms is -56.40%, which suggests that these firms’ pension
assets are short of pension obligations by more than a half of their market values. All the
variables show considerable degrees of cross sectional variation. The significant variation
exhibited in pension funding ratios is of particular interest in this study. It allows us to use sorted
portfolio approach to investigate the relation between pension funding ratios, analysts forecast
errors, and expected stock returns.
Panel C reports the extent of analyst coverage for the pension sample. There are totally
4,537 firms covered by at least one analyst, among which 899 firms have DB pension plans or
approximately 19.8% among all firms covered by analysts. On average, there are 7.8 (7.1)
analysts for one overfunded (underfunded) firm. Panel C also shows a gradual increase of analyst
coverage from 1990 to 2005, as well as an increase in the number of analysts, which is consistent
with the finding from Hong, Lim and Stein (2000).
In Panel B of Figure 1, we plot the amount of actual pension funding and pension surplus
(deficit) recognized on the balance sheet for each of our sample years. Pension funding surplus
(deficit) is defined as the difference between the fair value of plan assets (FVPA) and the present
The deterioration of pension funding status is primarily due to a combining effect of falling stock market and
decreased interest rate since 2000. On the one hand, falling stock market has caused a sharp decline of FVPA as a
majority of pension funds are predominantly invested in stocks. On the other hand, to make things even worse, the
decreased interest rate has resulted in a substantial increase of firm’s PBO because firms use interest rate to discount
their future pension obligations to get PBO. The combining effect is to significantly widen the difference between
PBO and FVPA, causing the funding status of DB plans to deteriorate dramatically.
value of pension obligations (PBO). Following Shivdasani and Stefanescu (2009), we compute
pension surplus recognized on the balance sheet as the sum of net amount of the accrued pension
liability (data 290), prepaid pension liability (data 300), and additional minimum pension
liability (data 298). We observe a huge difference between aggregate pension funding surplus
(deficit) and the amount recognized on the balance sheet, especially after 2001. For instance,
there were totally $66 billion in pension surplus reported on the balance sheet in 2003, much
lower than $523 billion in pension deficit. The difference between the pension surplus/defict and
the amount recognized on the balance sheet reflects the difference in actual economic pension
funding status and pension accounting practices. The actual pension funding represents an
economic definition, while the amount recognized on the balance sheet simply represents an
accounting definition, which can be significantly different from the actual pension surplus/deficit
(e.g., Coronado and Sharpe, 2003; Picconi, 2006; Coronado, Mitchell, Sharpe, and Nesbitt,
2008). This further shows that pension numbers reported by firms could mislead investors (and
analysts) as to the real funding status of pension plans.
A. Pension Underfunding and Analyst Forecast Errors
A.1 Baseline Results
We first examine the relation between analyst earnings forecast errors and pension
underfunding ratios. We are interested in whether analysts’ optimistic forecast bias is more
pronounced for the most underfunded firms. If analysts fully incorporate pension information
into their earnings forecasts, there should be no relationship between pension underfunding ratios
and earnings forecast errors. If analysts underreact to pension underfunding information,
earnings forecast errors would be more negative for firms with more severely underfunded
We first use a portfolio sorting method to examine the relation between analyst earnings
forecast errors and pension underfunding ratios. Following Franzoni and Marín (2006), in July of
year t, we sort all sample firms into 11 portfolios based on their pension funding ratios FR in
fiscal year t -1.15 In particular, all underfunded firms are sorted into deciles, with the first decile
representing the most underfunded and the tenth the least underfunded. All overfunded firms are
grouped into the eleventh portfolio.16 We form equally weighted portfolios in each decile, and
compute analyst earning forecast errors based on equation (4) for each portfolio, and forecast
error spreads between D1- D10 and between D1 - D11 for the one-year forecasting horizon.
The average forecast errors for the 11 sorted portfolios are reported in Table 2. Analyst
forecast errors are generally on average negative for all portfolios, consistent with the well-
documented pattern that analysts are over-optimistic with their earnings forecasts. The average
forecast error for all underfunded firms are -0.53%, significant at the 1% level. Moreover, the
portfolio with the most underfunded firms (D1) exhibits the largest negative 1-year forecast
errors of -1.81%, as opposed to forecast errors of -0.17% for the least underfunded portfolio
(D10) and -0.31% for the overfunded portfolio (D11). The difference is -1.64% (t = -5.85)
between D1 and D10, and -1.50% (t = -5.25) between D1 and D11. In figure 2, we plot the the
differences of forecast errors between D1 and D10 portfolios and D1 and D11 portfolios using 1-
year, 2-year and 3-year forecasts. The pattern clearly shows that analyst earnings forecasts are
more negative for firms with severely underfunded pension plans. 17
A.2 Previous Bias versus Accuracy
The baseline result in the previous section suggests that analysts on average do not fully
incorporate firms’ pension funding information into their earnings forecasts, as evidenced by
larger forecast errors for more severely underfunded firms. A natural question then follows: what
are the sources of analyst forecast errors associated with pension underfunding information? We
We choose July of year t as the portfolio formation date to ensure that pension plan information for the fiscal year
ending in year t-1 is available to the market (Fama and French, 1993). We obtain similar results if we use April as
the portfolio formation date.
We group all overfunded firms into one portfolio and sort all underfunded firms into decile portfolios because
pension overfunding and underfunding have an asymmetric effect on a firm’s earnings (e.g., Carroll and Niehaus
2006). That is, underfunded pension liabilities are an integrated portion of corporate liabilities while overfunded
pension asset are not entirely a corporate asset, due to strict restrictions on converting an overfunded pension plan’s
assets into unencumbered corporate assets (Jin, Merton, and Bodie 2006).
In untabulated analyses, we also partition our sample into before- and after-2000 sub-periods and repeat the above
sorting analysis. Corporate pension funding status deteriorates significantly after 2000, and we expect to observe a
stronger association between pension funding ratios and analyst earnings forecast errors. As reported in Table 2, the
1-year forecast errors spread between the most underfunded the least underfunded firms (D1-D10) is -1.03% before
2000, as opposed to -3.43% after 2000. The difference is significant at 5% level. We obtain the similar results when
examining the 2-year and 3-year analyst forecast errors.
identify two sources based on previous studies – analysts’ limited information processing
capacity and their incentives to provide optimistic forecasts.
The limited information processing is attributable to the cognitive biases. Two common
and relevant biases in the financial context have been identified by previous studies:
overconfidence and cognitive dissonance (e.g., Akerlof and Dickens, 1982; Goetzmann and
Peles, 1997). An analyst could be overconfident with her forecasting capability; and thus
overestimate the precision of her forecasting numbers. In addition, analysts are subject to
cognitive dissonance, which is characterized as a tendency to filter out information (including
pension information) that is not in accordance with her desired beliefs. Thus if an analyst
underestimates the negative effect of pension underfunding on a firms’ earnings and issues an
optimistic earnings forecast, she tends to continuously interpret subsequent information in such a
way as to support her prior belief (Friesen and Weller, 2002).
Another alternative source to analyst forecast errors for underfunded firms is related to
analysts’ incentive. Analysts are much more likely to be rewarded for their optimistic forecasts
(Hong, Lim and Stein, 2000) and considerable evidence suggests a systematic upward bias in
analysts’ earnings forecasts and stock recommendations (e.g., Fried and Givoly, 1982; Moyer et
al., 1989; O’Brien, 1988;). An important reason is analysts incentives related to maintaining or
developing underwriting relations and their incentives related to generating trading commissions
are possible causes of analysts’ systematic over-optimism (Lim, 2001; Schipper, 1991). Firms
with pension plans tend to be large and matter most for investment banking business and
brokerage firms, as well as for analyst career concern. We thus expect analyst incentive issue to
be heightened when making earnings forecast for these firms, resulting in earnings forecast
We follow Hong and Kubik (2003) to construct the forecast accuracy (ACCURACY)
measure. It is the average of the absolute value of analyst forecast errors (FCE) on all the firms
an analyst covered over the three previous years:
ACCURACY j ,t | FCEi, j ,t |
Analyst forecast bias (BIAS) is average of analyst forecast errors (FCE) on all the firms
an analyst covered over the three previous years:
BIAS j ,t FCEi, j ,t
To examine whether analyst previous forecast accuracy and bias attribute to analyst
forecast errors associated with pension underfunding information, we use a two-way sorting
method. That is, we double sort individual analyst forecast errors by pension funding ratios and
analyst forecast accuracy and forecast bias, respectively. Particularly, in July of each year t, we
first sort firms into 11 portfolios based on pension funding ratios, following the procedure
described earlier. Then, within each portfolio, we further sort all firm-analyst observations into
four quartiles based on individual analyst forecast accuracy and forecast bias. We compute the
time-series averages of cross-sectional means of individual 1-year analyst forecast errors across
44 portfolios (11 X 4), as well as forecast error differences across extreme portfolios ranked on
pension funding ratios and individual analyst forecasting accuracy and bias.
The results for the double-sorted portfolios are reported in Panel B and C of Table 2. In
Panel B, holding analyst past bias constant, we find that analyst forecast errors generally become
less negative as the rank of pension funding increases, consistent with the result in Panel A.
Holding pension funding ratios constant, we observe that analysts with the lowest forecast bias
tend to have higher forecast errors. The spread between D1 and D10 is most negative. In Panel
C, firms in Q1 group are covered by analysts with most accurate forecasts (a.k.a the lowest
forecast errors). In sum, we find that analysts’ forecast biases associated with pension
information are negatively correlated with their average past accuracy, and positively correlated
with their average past bias, suggesting one possible source of the analyst earnings forecast
Next, we perform a simple regression analysis to examine whether both limited
information capacity and incentive problems are the cause of analyst forecast errors associated
with pension underfunding. In particular, we regress analyst forecast errors (FCE) on pension
funding ratios and two interaction variables between pension funding ratios and BIAS and
ACCURACY. FCE, BIAS, and ACCURACY are defined the same as before. To capture the
unsymmetrical effect of pension underfunding and overfunding on a firm’s earnings (Carrol and
Niehaus, 1998), we use FR(+) and FR(–) for pension funding ratios. For underfunded firms,
FR(–) is equal to the difference between FVPA and PBO, scaled by a firm’s market value
(Equation 1). For overfunded firms, FR(–) is equal to zero. FR(+) is a dummy variable that is
equal to one if a firm i is overfunded at the end of fiscal year t-1, and zero if underfunded.
Including both BIAS and ACCURACY in one regression equation with interacting them with
funding ratios would help identify the marginal effect of analyst’s previous forecast accuracy and
bias on forecast errors.
The regression result is reported in Panel D. Consistent with our sorting result Panel A,
we find that FR(–) has a positive and significant coefficient (β = 2.37, t = 2.89), suggesting that
analysts have larger negative forecast errors for more underfunded firms. Further, the interaction
item FR(–)*BIAS has a significant and positive coefficient (β = 7.62, t = 2.31), and FR(–
)*ACCURACY has a significant and negative coefficient (β = -15.62, t = -2.02). The result
suggests that both analyst limited information problem and incentive problem affect analyst
earnings forecasts. This finding advances from prior understanding on underfunding effect which
mainly focuses on the information effect. We also find that analysts’ previous bias and accurate
are negatively correlated (ρ = -0.71, t = -2.60).
Overall, our findings in Table 2 show that both analysts’ limited information capacity
(measured by ACCURACY) and incentive issue (measured by BIAS) contribute to analyst
forecast errors associated with pension underfunded firms.
B. Mechanisms to Reduce Forecast Errors
B.1 Learning and Forecast Errors
So far we have reported that analysts tend to underreact to pension underfunding
information and that such underreaction is attributable to both analysts’ limited information
capacity caused by cognitive bias and incentive issue related to career concerns. One potential
mechanism to mitigate such cognitive bias is learning. In this section, we explore the analyst
learning hypothesis. In the next section, we examine several external mechanisms to attenuate
analysts’ incentive issue.
Previous studies document cross-sectional variations in the degree of analyst
underreaction (e.g., Ali, Klein, and Rosenfeld, 1992; Jacob and Lys, 1999; Butler and Lang,
1991). In particular, Mikhail, Walther, and Willis (2003) find that analysts underreact to prior
earnings information less as their experience increases, providing one reason why analysts
forecast earnings more accurately with experience. Motivated by the studies on cross-sectional
variations in analyst underreaction, we examine whether analysts underreact to pension
underfunding information to a less extent as they accumulate more firm-specific research
experience through learning, thus alleviating forecast errors. Two types of learnings are
explored: i) analysts’ own experience and ii) companies’ prior history of underfunding.
B.1.1 Learning by Doing
Hogarth (1987) lists three conditions necessary for the reduction of underreaction or a
cognitive bias: (1) feedback, (2) a repetitive task, and (3) an ability to keep a record of
predictions and actual results, which are particularly applicable to a pension funding framework.
As discussed by Mikhail, Walther and Willis (2003), analysts receive feedback through
comparing their earnings forecasts with a firm’s actual earnings. Analysts also perform a
repetitive task – they analyze pension information and make earnings forecasts for the same
firms on a continuous quarterly (or yearly) basis. Further, analysts maintain an extensive record
of their forecasting work because they are evaluated by their employer based on the earnings
forecast record (e.g., Hong and Kubik, 2003). As predicted by the “Learning By Doing” model
(Arrow, 1962), we expect analysts to digest pension information more efficiently as they
accumulate more learning and forecast experience with the firms they follow, thus improving
their forecast accuracy with repetition and feedback.
We measure the experience (EXPi,j,t) of analyst j at time t as analyst tenure, or the number
of years, prior to the current forecast, for which an analyst j has issued a forecast report for a
firm i. A potential problem of this measure of analyst experience is that it ignores analyst
forecasting experience prior to 1983, which is the year that the I/B/E/S forecast data is first
available. Although the I/B/E/S dataset does not record how long an analyst has worked in the
profession, we could count how many years an analyst has existed in the database. We thus first
estimate analyst experience using all observations in the I/B/E/S database from 1983 onward;
and then we set the cutoff time as the year of 1988, corresponding to our beginning period of
sample (1988 to 2005). In this way, all analysts who produced forecasts in 1983 will be
experienced under our definition no matter how many years they were in the profession before
In untabulated analysis, we first compute a few descriptive statistics on analyst
experience. There is a wide dispersion in the level of analyst experience. Analysts have an
average (median) research experience of 3.5 (3.0) years, with a maximum of 14.0 years. To get
the average analyst forecasting experience of all analysts covering an individual firm i, we
calculate the mean experience across all analysts. On average, a firm is covered by analysts with
an average (median) research experience of 2.5 (2.4) years, with a maximum of 6.3 years.
Analyst experience may be correlated with firm age and market capitalization. It is likely
that analysts tend to follow old firms and large-cap firms for a long time since these firms are
likely to release more information to the public, which facilitates analyst earnings forecast. To
control for these firm specific effects, we construct an adjusted experience measure controlling
for firm life and market capitalization. Specifically, we regress analyst experience on firm age,
computed as the number of years in the Compustat database, and the logarithm of firm market
capitalization. We use the regression residual as the adjusted analyst experience measure.
We now use a two-way sorting procedure to examine whether analysts produce less
forecast errors for underfunding firms as they accumulate more research experience. The sorting
process is similar to that discussed in the previous section. In July of each year t, we first sort
firms into 11 portfolios based on pension funding ratios. Then, within each portfolio, we further
sort all firm-analyst observations into 4 quartiles based on individual analyst forecast experience.
That is, if five analysts make earnings forecasts for a firm in a particular year and their
experience differs, the five forecast errors will be classified into different analyst experience
deciles. We compute the time-series averages of cross-sectional means of individual 1-year
analyst forecast errors across 44 portfolios (11 X 4), as well as forecast error differences across
extreme portfolios ranked on pension funding ratios and individual analyst forecasting
The results for the double-sorted portfolios are presented in Table 3. In Panel A, we
report results for double-sorting individual analyst forecast errors on pension funding ratios and
raw analyst experience. Holding analyst experience constant, we observe that the most
underfunded firms (D1) exhibit the larger forecast errors than the least underfunded firms (D10)
and the overfunded firms (D11). Next, holding firm pension funding ratios constant, we find that
analyst forecast errors generally become less negative as the rank of analyst experience
increases. For example, within the most underfunded portfolio (D1), the average forecast errors
are -2.78% for the least experienced analysts and -1.12% for the most experienced analysts.
Within the least underfunded firms (D10), the average forecast errors for the least experienced
analysts and the most experienced analysts are -0.54% and -0.09%, respectively. The difference-
in-difference between D1-D10 and Q1-Q4 is -1.21 (t = 2.41), suggesting that as analysts
accumulate more experience through learning, their earnings forecast errors become smaller.
Panel B of Table 3 reports results for double-sorting analyst forecast errors on pension
funding ratio and adjusted analyst experience. The results are similar to those reported in Panel
A. In sum, our double-sorting results indicate that analysts underreact less to pension funding
information as they gain more forecasting experience through continuous learning, thus
producing earnings forecasts with smaller forecast errors. As the results are consistent when we
sort them into quartile groups based on the raw experience and adjusted experience, our
remaining analysis focus on the adjusted experience measure.
B.1.2 Analyst Experience and Forecast Errors: First versus Multiple Underfunders
In this section, we explore the second learning effect: company’s prior history of
underfunding occurrence. We test whether financial analysts’ learning differs across firms with
different frequency of pension underfunding emerge. By categorizing firms into first-time
underfunders and multiple time underfunders, we repeat the analysis in Section B.1.1 by
examining the relationship between pension underfunding occurrence and forecast errors.
We first estimate the average 1-year individual analyst forecast errors at each of the
portfolios double sorted by firm funding ratios and firm underfunding occurrence. A firm is first-
time underfunded when it is underfunded in year t-1 but overfunded in the prior two years. A
firm is multiple-time underfunded when it is underfunded in year t-1 and also underfunded in
either year t-2 or t-3. We use the I/B/E/S detail file and estimate individual analyst forecast errors
according to equation (2). As previously described, in each year t, we first sort all observations
into 11 portfolios based on firm pension funding ratios reported in year t -1. Within each
portfolio, we further sort observations into two groups based on whether a firm is first-time
underfunded or multiple-time underfunded. We estimate individual analyst forecast errors for
these 22 (11 x 2) portfolios. As reported in Panel A of Table 4, we find that analyst forecast
errors are generally less negative for firms with multiple-time underfunding occurrence across all
11 portfolios sorted on pension funding ratios. For example, within the most underfunded firms
(D1), the forecast errors are -2.83% and -0.79% for first-time underfunding firms and multiple-
time underfunding firms, respectively. As to the least underfunded firms (D10), the forecast
errors are -0.37% for first-time underfunding firms and -0.07% for multiple-time underfunding
firms. The difference-in-difference between (D1-D10) and (first-multiple underfunders) is -
1.74% (t = -2.55).
In Panel B and C of Table 4, within first-time underfunders and multiple-time
underfunders, we examine D1 and D10 forecast error differences across most experienced and
least experienced analyst groups, after sorting on firm pension funding ratios. The sorting
methodology is the same as described in Section B.1.1. With the first-time underfunding group,
we find that analysts forecast errors are generally less for the most experienced analysts, holding
pension funding ratio constant, consistent with the results in Table 3. For example, within the
most underfunded portfolio (D1), the average forecast errors are -3.84% for analysts with the
least adjusted experience and -1.74% for analysts with the most adjusted experience. For the
least underfunded firms (D10), the average forecast errors for the least adjusted experienced
analysts and most adjusted experienced analysts are -0.61% and -0.19%, with the difference-in-
difference between D1-D10 and Q1-Q4 of -1.63 (t=-2.35). Moreover, for the multiple-time
underfunding firms group, we observe that analyst forecast errors are generally less than the
firm-time underfunding group. In fact, the difference-in-difference for the multiple-time
underfunding group is -0.39 (t = -1.36), much less than -1.63% for the first-time underfunding
Overall, the evidence in Table 4 is consistent with the two types of learning effects. As
analysts accumulate more research experience and as they learn from firms’ pension
underfunding emerge, their forecasting ability gradually increases, resulting in less forecast
errors. More interestingly, we report an interaction between these two learning effects – the
effect of analyst experience to reduce forecast bias is stronger among firms experiencing pension
underfunding for the first time. That is, analysts’ experience matters most when they face a new
B.2 Analyst Incentives and Analyst Forecast Errors
In this section, we examine whether the presence of external mechanisms, including
institutional ownership and analyst competition are able to attenuate analysts’ incentive issues,
thus reducing earnings forecast errors for pension underfunded firms.
We measure institutional presence by the number shares owned by institutional investors
as a fraction of total shares outstanding (Ljunqvist et al., 2007). To be specific, at the end of each
year, we compute the institutional ownership by calculating the percentage of stocks owned by
institutional investors covered in the 13f institutional holding database. We follow Hong and
Kacperczyk (2010) to measure analyst competition by the number of analysts following the firm.
To adjust for the selection bias, we regress the number of analysts following a firm by the
logarithm of firm size and firm actual earnings in year t. We expect that incentive-related
forecast optimism will be significantly reduced by a strong presence of institutional investors as
shareholders, and by competition among analysts.
Similar to the previous tests, we first use a two-way sorting method to examine the
relationship between analyst forecast errors and institutional ownership and analyst coverage.
The results are reported in Table 5. Panel A shows the result when we sort stocks sequentially
based on funding ratios then by institutional holdings. Panel B shows the results when we sort
stocks sequentially based on funding ratios then by analyst coverage. We find that after holding
firm pension funding ratios constant, analyst forecast errors generally become less negative as
the ranks of institutional holdings and analyst coverage increase. For example, as shown in Panel
A, within the most underfunded firms, the forecast errors are -3.49% and -1.07% for the firms
with the lowest (Q1) and the highest (Q4) institutional ownership, respectively. As to the least
underfunded firms (D10), the forecast errors are -0.45% and -0.29% for these two groups. The
difference-in-difference between (D1-D10) and (Q1-Q4) is -2.17% (t = -2.35).
We then proceed to conduct a three-way sorting – we first sort analyst forecast errors by
pension funding ratios, then by institutional holdings, and finally, by analyst coverage. The
results are reported in Panel C of Table 5. With both bottom 1/3 and top 1/3 of institutional
holding groups, we find that analysts forecast errors are generally less negative as the rank of
analyst competition increases, holding pension funding ratio constant. Further, we find that
analyst forecast errors for each portfolio in the top 1/3 of institutional ownership are generally
much lower than those in corresponding portfolios in the bottom 1/3 of institutional ownership.
In fact, the difference-in-difference for the top 1/3 of institutional holding is -1.08% (t = -1.74),
much lower than -3.12% (t = -3.27) for the bottom 1/3 of institutional holding. This finding
suggests that there is an interaction effect between analyst competition and institutional
ownership. That is, analyst competition plays a stronger role in reducing forecast bias associated
with pension underfunding among the firms with higher institutional ownership, Such interaction
effect can be potentially linked to the two channels, as pointed out by Hong and Kacperczyk
(2010), through which analyst competition reduces forecast bias. In the first channel, objective
forecasts provided by independent analysts exert pressure on other analysts. Such pressure can be
stronger when the demand for objective forecasts from institutional clients is stronger. In the
second channel, analyst competition increases firms’ influence cost. The influence cost can be
higher when the demand from institutional clients offset firms’ influence on analysts.
B.3 Regression Analysis of Forecast Errors
So far, we find evidence supportive to both the learning and incentive-attenuation effects.
Yet, two issues remain unaddressed. First, can analyst optimism come from other factors
correlated to pension underfunding? Second, do the two effects subsume each other? In this
section we answer the two questions using panel regressions.
As to the first issue, we utilize a number of other independent variables that have shown
to be related to forecast error by prior studies. Richardson, Teoh, and Wysocki (2004) and Core,
Guay, and Rusticus (2006), show that firm Beta, market value, book-to-market ratio, and
momentum explain variations in forecast errors. Bradshaw, Richardson, Scott and Sloan (2001)
show accounting accruals are positively associated with forecast errors. As a measure of
information uncertainty, Zhang (2006) demonstrates that forecast dispersions affect forecast
errors. Moreover, Ali, Klein, and Rosenfeld (1992) argue that lagged period forecast error for the
same firm made by the same analyst may reflect analysts’ behavioral bias in making forecasts
due to the persistence of earnings forecast errors. Finally, we include the amount of pension asset
liability recognized on balance sheet as a control variable.
To address the second question, we perform a panel regression with both sets of
independent variables that proxy for learning and incentive-attenuation effects. The full model
with all key explanatory variables and control variables is shown as below:
FCE i , j ,t =+1FR(+)i,t-1+2FR(-)i,t-1+3 FR(-)i,t-1*EXP i,j,t-1 +4 FR(-)i,t-1*FIRST i,t-1
+5 FR(-)i,t-1*EXP i,j,t-1 *FIRSTi,t-1+6 FR(-)i,t-1*INST i,j,t-1
+7 FR(-)i,t-1*COVG i,t-1 +8 FR(-)i,t-1*INST i,t-1 *COVGi,t-1
+9 *FR_BS i,t-1 +10 FCE 1,yr,t 1 +11 DISP i,t-1 + 11 logSIZE i, t-1
+12logBM i,t-1 +13 Betai,t-1 +14 MOM i,t-1 +15ACC i,t-1 + i,j, t (7)
FCE i , j ,t = 1-year forecast errors for firm i made by analyst j in year t, calculated
based on equation (6) using the I/B/E/S detail file.
FR(–)i,t-1 = Pension funding ratio for firm i at the end of fiscal year t-1, calculated as
the difference between fair value of pension assets (FVPA) and present
value of pension benefit obligations (PBO), scaled by a firm’s market
value at year t (equation 1). For underfunded firms, FR(–) is equal to its
original value; for overfunded firms, FR(–) is equal to zero.
FR(+)i,t-1 = Dummy variable that is equal to one if a firm i is overfunded at the end
of fiscal year t-1, and zero if underfunded.
EXPi,j,t-1 =Analyst experience, measured as the number of prior years for which
analysts j have issued an earnings forecast report for firm i prior to year t.
We use the natural logarithm of one plus EXPi,j,t-1 in the regression to
reflect the non-linear learning function.
FIRSTi,t-1 = Dummy variable that is equal to one if the pension plan of firm i is
underfunded at the end of fiscal year t-1 and overfunded in both year t-2
and t-3; it equals to zero if otherwise.
INSTi,t-1 = the ratio of the aggregated number of shares held by institutional
investors to the total number of shares outstanding of firm I at the closest
quarter end before fiscal year t-1 (in percentage). We use Thomson 13F
data to calculate the aggregated number of shares held by institutional
COVGi,t-1 =the number of analysts covering firm i in fiscal year t-1.
FR_BSi,t-1 =the amount of pension assets (liability) recognized on the balance sheet
for firm i in fiscal year t-1. It is calculated as the net amount of the accrued
pension liability, prepaid pension liability, intangible assets and additional
minimum liability (data290+data300-data298)
DISP i,t-1 = Forecast dispersion for firm i in year t-1, measured as the standard
deviation of analyst forecasts made four months prior to fiscal year-end,
scaled by the prior year-end stock price.
logSIZE i,t-1 = Log of firm i market capitalization at the end of fiscal year t-1. Market
capitalization is calculated as the closing price at the end of year t-1
multiplied by common shares outstanding at the end of year t-1.
logBM i,t-1 = Log of book-to-market ratio for firm i at the end of fiscal year t-1, where
book value of equity is the book value of stockholders’ equity item 216),
plus balance sheet deferred taxes and investment tax credit (item 35, if
available), minus the book value of preferred stock (item 56: preferred
stock redemption value; item 10: preferred stock liquidating value; or item
130: preferred stock carrying value, in the sequence according to data
Beta i,t-1 = Beta is calculated by regressing the stock’s daily return on the value-
weighted market return using ordinary least square and 100 trading days
of returns data ending in December of year t-1.
MOM i,t-1 = Previous 6-month stock returns for firm i ending in December of year t-
ACC i,t-1 = Accounting accruals for firm i at the end of fiscal year t-1. It is
computed strictly following Sloan (1996).
We begin by estimating a panel regression of analyst forecast errors on firm pension
funding ratio only. We add control variables in the second regression. Next, we test for the
learning effect and incentive-attenuation effect separately. Our final regression jointly tests the
two effects. For all regressions, we include firm dummies and year dummies. Standard errors are
clustered by both firm and year.18
The results are reported in Table 6. In column (1), we only include FR(+) and FR(-) in
the regression. The dependent variable is the average forecast errors made by all analysts
covering a given firm for a fiscal year. The coefficient on pension overfunding dummy FR(+) is
-0.16 (t = -1.07), indifferent from zero. The coefficient on pension underfunding ratio FR(–) is
2.29 (t = 1.90), suggesting analysts on average make more optimistic earnings forecasts for firms
with more severe pension underfunding. This result is consistent with findings in Table 2, and
suggests that analysts fail to fully incorporate the negative effect of a firm’s pension
underfunding on future earnings in their forecasts.
From column (2), we start to use individual analyst forecasts as the dependent variable
and add all control variables in the regressions. As shown in column (2), several other factors are
significantly related to forecast errors. The coefficient on analyst forecast dispersion is 8.60
(t=1.75), the coefficient on firm size is -0.14 (t=2.66), the coefficient on book-to-market ratio is -
0.34 (t=-3.53), and the coefficient on momentum is -2.54 (t=2.29). The effects of these variables
For all panel regressions performed in this paper, if not specified otherwise, we include fixed firm and year effects
in the regression and adjust standard errors clustering by both firm and year.
are consistent with prior studies. More importantly, the coefficient on FR(–) is robust after
controlling for the impact of other factors.
In columns (3) and (4), we look at the learning effect. Column (3) shows that the
coefficient on FR(–)*EXP is -0.23, significant at the 5% level (t = -2.25). This result indicates
that analysts are able to better incorporate pension underfunding information into future earnings
forecasts as they gain more firm-specific experience, thus leading to reduced forecast errors. It is
consistent with a general improvement in analysts forecast rationality with experience (e.g.,
Mikhail, et al. 1997, 2003). Another key explanatory variable, FR(–)*FIRST, has a significantly
positive coefficient of 3.69 (t = 1.92), suggesting analysts make more optimistic forecasts when a
covered firm becomes underfunded for the first time. In column (4), we include the interaction of
analyst experience and whether a covered firm is first time underfunded or has been repeatedly
underfunded. The coefficient on FR(–)*EXP*FIRST is -0.39 (t=-1.98). The result suggests that
when analysts have the same experience, they tend to make larger forecast errors when a covered
firm becomes underfunded for the first time.
The results on the incentive-attenuation effect are provided in columns (5) and (6).
Column (5) shows that the coefficient on FR(–)*INST is -1.34 (t=-2.13) and that of FR(–
)*COVG is -1.11 (t=-1.82). That is, both institutional ownership and analyst competition can
reduce forecast errors related to firm underfunding ratios. In column (6), we add the interaction
between institutional ownership and analyst coverage. The coefficient on FR(–)*INST*COVG is
significantly positive (β = 1.34, t = 1.91). The result demonstrates that analyst competition plays
a more important role when institutional investors have smaller voice.
In the last column in Table 6, we jointly test the two hypotheses. We include learning
effect and the governance effect in the same model, as well as all control variables. We find our
results in model 2 to 6 preserve: both learning and governance can significantly reduce analyst
optimistic forecast errors related to firm funding ratios.
C. Pension Underfunding, Analyst Behavior, and Stock Returns
Franzoni and Marín (2006) report an interesting pension underfunding anomaly -- firms
with severely underfunded pensions have lower stock returns relative to firms with overfunded
pensions up to five years subsequent to the first emerge of pension underfunding. In this section,
we check if the mechanisms identified to alleviate analyst forecast biases on pension funding
information are also effective to reduce market mispricing. We also check whether forecast
errors associated with analyst learning and their incentive mitigation have the same effects on
stock returns. By doing this, we could draw inference on investors’ ability on handling different
sources of analyst biases.
C.1 Stock Returns of Underfunded Firms: Uninformed versus Incentive Bias
Underfunded firms tend to have worse future operating performance, which if not
realized by investors, will cause lower stock return for these firms. Why underfunded firms may
perform worse in their operation? We argue this is due to the negative economic consequences of
pension underfunding. First, pension underfunding has negative effects on employees’ work
incentives and productivity. The underfunded plans could potentially reduce workers efforts and
productivity, and increase the turnover rate (e.g., Lazear, 1979; Hutchens, 1989). 19 Second,
pension underfunding may reduce a firm’s capital expenditures and result in underinvestment.
This is because underfunded firms are required by law to make financial contributions to their
pension funds to meet certain funding status. Third, pension underfunding could also decrease a
firm’s debt ratings and thus increase a firm’s cost of debt capital. Martin and Henderson (1983),
Maher (1987), and Carroll and Niehaus (1998) find that underfunded firms have lower bond
We perform panel regressions to examine the relation between pension underfunding and
Reti,t=+1FR(+)i,t-1 +2FR(–)i,t-1+3 FR(–)i,t-1*EXPi,t-1 +4FR(–)i,t-1*FIRSTi,t-1
+5FR(-)i,t-1*EXPi,t-1*FIRST,t-1 +6FR(-)i,t-1*INSTi,t-1 +7FR(-)i,t-1*COVGi,t-1
+8FR(-)i,t-1*COVGi,t-1*INST +9FR_BSi,t-1+10DISP i,t-1+11logBMi,t-1
+12BETAi,t-1 +13MOMi,t-1+14SUEi,t-1+15ACCi,t- +i,t (8)
Reti,t is the 1-year buy-and-hold stock returns for firm i from July of year t to June of year t+1
where firm pension funding ratio is evaluated at the end of fiscal year t-1. Fixed firm and year
effects are included in the model and standard errors are adjusted for two-way clustering by firm
and year. Independent variables are defined in the same way as (7) except for that i) EXP is
Lazear (1983) and Hutchens (1989) find that workers with pensions are more productive than workers without
measured in the firm level, and ii) we additionally include SUEi,t-1, the standardized unexpected
earnings for firm i, to measure a firm’s earnings surprises.20
EXPi,t is defined as the average experience of all analysts covering firm i in year t,
i , j ,t (9)
where EXPi,j,t is the adjusted experience for analyst j issuing forecast reports in year t. J is the
number of analysts following firm i in year t.
The results are reported in Table 7. In model (1), FR(–) has a significantly positive
coefficient of 2.29 (t = 2.90). This is consistent with the result in Franzoni and Marin (2006)
where stocks in more underfunding groups underperform. The coefficient on the FR(+) dummy
is positive but insignificant at a conventional level, suggesting that overfunded firms as a group
do not earn significantly different average returns.
We add control variables in model (2) and find consistent results on coefficients of FR(-)
and FR(+). The results on control variables are interesting and deserve elaborations. The
coefficient on FR_BS (firm funding ratio based on items reported on the balance sheet) is
significantly negative. It indicates that investors may be misled by balance sheet information.
Consistent with previous studies (Fama and French, 1992), we find that beta fails to explain the
cross-sectional stock returns in both models. There is an insignificant relation between size and
stock returns during our sample period. The book-to-market ratio positively relates to stock
returns, confirming a well-documented value effect. The coefficient for SUE is insignificant.
In model (3), we additionally interact FR(-) with EXP and FIRST. The results on FR(+)
and FR(-) preserve. The coefficient on FR(–)*EXP is significantly negative (β = -6.65, t = -1.90).
The result shows that the negative return predictability of firm underfunding ratio is substantially
lower for firms covered by more experienced analysts. We also find that the coefficient on FR(-
)*FIRST is 67.89 (t = 2.08), indicating investors underreact to pension underfunding information
in a greater magnitude for first-time underfunders. In model (4), we further include FR(-
)*EXP*FIRST in the regression. The coefficient is -49.11 (t = -1.76), significant at the 10 percent
Following Bernard and Thomas (1990), we compute SUE as (Eit – Ei,t-4 – ci,t)/σi,t. Ei,t is the quarterly earnings
(COMPUSTAT quarterly data item 8) reported during quarter t, Ei,t-4 is earnings reported four quarters ago. ci,t and
σi,t are the time series mean and standard deviation, respectively, of (Ei,t – Ei,t-4) over the preceding eight quarters,
with a minimum of four quarters required for the calculation to be valid. The earnings announcement dates are from
the COMPUSTAT quarterly file. If the announcement date is missing, we assume that earnings are reported two
months after fiscal quarter-end.
level. It indicates that effect of analyst experience to reduce mispricing is stronger for first-time
underfunders than for multiple-time underfunders. Overall, the results from model (3) and (4)
show that the learning effect that lowers analyst forecast errors also reduces the
underperformance of underfunded firms.
In models (5) and (6), we look at if analyst coverage and institutional ownership reduces
the underfunding effect on future stock returns. We include FR(-)*INST and FR(-)*COVG in
column (5) and further include FR(-)*COVG*INST in column (6). Quiet interesting, all the
coefficients are indifferent from zero. It appears that these incentive-attenuation mechanisms,
though work for analysts, do not affect the mispricing due to pension underfunding. In other
words, investors seem to understand analyst incentive bias and somehow do not include their
bias when make investment decisions. Finally, the regression reported in Column (7) includes
the full set of regressors, including fund ratios, variables on analysts learning, and variables
mitigating the incentive problem. The result is consistent with the prior columns.
Taken together, the evidence presented in this subsection confirms the pension
underfunding anomaly found by Franzoni and Marin (2006). In addition, what is new here is that
we find, though both learning and incentive-attenuation effects reduce analyst forecast errors,
only learning effect affects future stock returns of underfunded firms. How to explain the
different effects on forecast errors and future stock returns? We address the question in the next
C.2 Investor Response and Analyst Forecast Errors
In this section, we propose that because investors selectively follow analyst forecasts,
future stock returns respond similarly to some analyst characteristics, and respond differently to
other characteristics at the same time. We test this by directly linking stock returns and analyst
forecast errors related to pension underfunding.
We use two steps to perform the analysis. First, we perform the full regression specified
in equation (7) with firm and year fixed effects to obtain the fitted value of FCEi,j,t :
ˆ ˆ ˆ ˆ ˆ ˆ
FCEi , j ,t 1 FR() i ,t 1 2 FR() i ,t 1 + 3 FR() i ,t 1 * EXPi , j ,t 1 4 FR() i ,t 1 * FIRSTi ,t 1
ˆ ˆ ˆ
5 FR() i ,t 1 * EXPi , j ,t 1 * FIRSTi ,t 1 6 FR() i ,t 1 * INSTi ,t 1 7 FR() i ,t 1 * COVGi ,t 1
8 FR() i ,t 1 * INSTi ,t 1 * COVGi ,t 1 k * CONTROLS i ,t 1
k 10 (10)
We also estimate three different components of FCEi , j ,t :
ˆ ˆ ˆ
FCEiFR,t 1 FR() i ,t 1 2 FR() i ,t 1 (11)
^ ^ ^
FCE iEXP 3 FR() i ,t 1 * EXPi ,t 1 4 FR() i ,t 1 * FIRSTi ,t 1 4 FR() i ,t 1 * FIRSTi ,t 1 * EXPi ,t 1 (12)
, j ,t
^ ^ ^
FCEiINC 3 FR()i ,t 1 * INSTi ,t 1 4 FR()i ,t 1 * COVGi ,t 1 4 FR()i ,t 1 * INSTi ,t 1 * COVGi ,t 1
, j ,t
We define the difference between actual FCE and the above three components of fitted
FCE iOTH FCE i , j ,t FCE iFR,t FCE iEXP FCE iINC
, j ,t ,j , j ,t , j ,t
Second, we calculate the average fitted values across all analysts that covering a given
firm in a given year. Finally, we perform regression of one-year stock returns on FCEi ,t 1 ,
ˆ ˆ ˆ ˆ
FCE FR i ,t 1 , FCE EXPi ,t 1 , FCE INC i ,t 1 , and FCE OTH i ,t 1 , separately and jointly. If stock returns on
average are significantly correlated to FCEi ,t 1 , investors probably follows analyst
recommendation blindly and do not figure out their optimism related to pension funding status
and other identified factors. If this is the case, regressing stock returns on different components
of FCEi ,t 1 could tell us which part of FCEi ,t 1 investors fail to interpret correctly.
We report the results in Table 8. In column (1), we regress stock return on FCEi ,t 1 only.
The coefficient is 1.94 (t=2.16). It suggests that overall investors do not see through analysts
optimism in their forecasts. In column (2), we regress stock returns on FCE FR i ,t 1 . The
coefficient of 1.46 is significant at the 5% level (t=1.97). It appears that while analysts do not
react correctly to pension funding ratio correctly, investors are not aware of their bias and make
similar mistake as analysts in their investment. Column (3) tests whether investors are aware of
the marginal role of analyst experience in their earnings forecast related to pension underfunding.
We find a significant coefficient of 2.70 (t=2.16), which indicates that investors do not
understand the role of analyst experience. Interestingly, in Column (4), the coefficient of -0.26
on FCE INC i ,t 1 is indifferent from zero (t=-0.43). That is, investors do realize incentive problems
faced by sell side analysts and correctly discount analysts forecast bias stemmed from incentives
in their stock investment decisions. In column (5), we find investors also do not correctly
respond to analyst forecast bias due to other known factors. In column (6), we jointly test the
various bias sources and the result is similar to separate regressions.
The sophisticated nature of pension information is a natural experiment to see how
sophisticated market participants respond to major corporate events and to find out specific
market mechanisms improving information efficiency. In this study, we find that analysts
underreact to the information content of pension underfunding. We also find that analyst
learning, competitions among analysts, and the presence of institutional investors help mitigate
such biases and lead to lower forecast errors. Interestingly, investors exhibit different levels of
sophistication in reacting to the information problem and the incentive problems. Specifically,
analyst learning reduces market mispricing while incentive-mitigation does not further mitigate
market mispricing. It appears that the market does not anticipate analyst forecast biases due to
inexperience, but anticipates analyst biases due to incentives.
An Illustrative Model on the Pension Underfunding Effect
To facilitate our understanding, we present a simple model to see how forecast errors are
formed and how such errors relate to corporate pension underfunding. The central assumption
made here is that analysts make earnings forecasts for year 1 (Y1) using a linear extrapolation of
firm earnings in year 0 (Y0).
Suppose that a firm is underfunded in its pension plans in Y0 and the underfunding is
driven by lower-than-expected asset returns during Y0. Projected pension obligations (PBO) stay
the same over time. Y0 is the first year that the firm experiences pension underfunding. Also, for
simplicity, we assume the realized capital gain/loss from pension assets to be zero. The
information related to the revelation of underfunding is provided in the following graph:
1. Pension underfunded for the first time 1. Service costs are increased due to additional
2. Underfunding is not recognized in the balance contribution to underfunded plans.
sheet at the end of Y0. 2. The amortized unrealized losses in pension
3. Underfunding does not affect the income investment in Y0 pension assets are recorded in year
statement in Y0. 1 income statement
3. Expected return on pension assets is reduced in Y1
given the shrinkage in Y0 pension assets.
We denote various components of net periodic pension cost as below: service cost (s),
interest cost (i), other cost (o), and expected return on plan assets (e). The firm’s operation profit
is denoted as π.
In Y0, the actual earnings for an underfunded firm are:
E0 0 [ s0 o0 i0 e0 ] (A1)
Suppose that there is no change in the firm’s pension obligations (PBO) during Y1.21
Following SFAS 87, at the end of Y1, we have following expression for the actual realized
earnings for an underfunded firm:
E1 ( 0 1 12 ) [s0 s1 o0 o1 i0 (e0 e1 )]
where: 1 is the additional operation profit in Y1 without the influence of pension
underfunding. - 12 is the reduction in operation profit in Y1 arising from pension underfunding.
Δs1 is the additional contribution in Y1 to make up Y0 pension underfunding. 22 Δo1 is the
We assume that firms use constant pension obligation discount rate to estimate the present value of future pension
benefits. Under the SFAS 87, firms are required to determine the discount rate base on prevailing interest rate for
30-year U.S. Treasury Bill. SEC’s ruling and other legislation have largely reduced managers’ discretion in setting
the pension discount rate. Effective for the fiscal year ending after December 15, 2006, the SFAS 158 states that
firms should consider the returns on high quality bonds when determining the discount rate. Our assumption is thus
plausible given that firms have less discretion in setting their pension discount rate.
The minimum contribution is contingent upon the funded status of the plan. Generally, it is equal to the pension
obligation earned by employees (service costs) during the year 1 plus the level of underfunding amortized over 30
years. Here we set the minimum contribution equal to service costs (Δs1); while leaving the rest amortization part
amortized unrealized losses in Y0 pension assets. Δe1 is the reduction in expected return on
pension assets due to the shrinkage of Y0 pension plan assets.
Let’s assume analysts linearly extrapolate Y0 earnings components to forecast Y1
earnings. More specifically, analysts consider the firm’s operating profit would grow by g and
pension cost stays at Y0’s level. Then the forecasted earnings (FE) for Y1 would be:
FE1 ( 0 1 ) [s0 o0 i0 e0 ]
As a result, we have the following expression for the forecast error of Y1 earnings:
ERROR1 (s1 o1 e1 12 ) (A4)
To sum up, analyst forecast errors are driven by i) the additional contribution to
make up Y0 pension underfunding; ii) the amortized unrealized losses in Y0 pension assets; and
iii) the reduction in expected return on pension assets due to the shrinkage in Y0 pension assets.
These items can be viewed as mechanical effects of pension underfunding coming from complex
and opaque accounting treatment related to pension expenses. These mechanical effects are
consistent with those identified in Coronado and Sharpe (2003), Pocconi (2006), and Coronado
et al. (2008). Moreover, the last item, 12 , is analysts’ under-reaction to economic consequence
of pension underfunding, consistent with Rauh (2006).
Abarbanell, Jeffery S., and Reuvan Lehavy, 2003, Biased forecasts or biased earnings? The role
of reported earnings in explaining apparent bias and over/underreaction in analysts’ earnings
forecasts, Journal of Accounting and Economics 36, 105-146.
Akerlof, George A., and Dickens, William T, 1982, The Economic Consequences of Cognitive
Dissonance," American Economic Review 72(3), 307-319.
Ali, A., Klein, A., Rosenfeld, J., 1992, Analysts' use of information about permanent and
transitory earnings components in forecasting annual EPS, Accounting Review 67, 183–198.
Arrow, K. J., 1962, The economic implications of learning by doing, Review of Economic
Studies 29, 155-173.
Banz, R., Breen, W.J., 1986, Sample-dependent results using accounting and market data: some
evidence. Journal of Finance 41, 779-793.
Barber, B., Lehavy, R., McNichols, M, Trueman, B., 2006, Buys, holds, and sells: The
distribution of investment banks’ stock ratings and the implications for the profitability of
analysts’ recommendations. Journal of Accounting and Economics 41, 87–117.
Barberis, N., Shleifer, A., Vishny, R., 1998, A model of investor sentiment, Journal of Finance
Bergstresser, Daniel, Mihir Desai, and Joshua Rauh, 2006, Earnings manipulation, pension
assumptions and managerial investment decisions, Quarterly Journal of Economics 121 (1), 157-
Bernard, V. and J. Thomas, 1990, Evidence that Stock Prices Do Not Fully Reflect the
implications of current earnings for future earnings, Journal of Accounting and Economics 13,
Bodie, Z., and Papke, L., 1992, Pension fund finance, In Bodie, Z., Munnell, A. eds: Pension and
the Economy, University of Pennsylvania Press, Philadelphia.
Boni, Leslie, and Kent L. Womack, 2002, Solving the sell-side research problem: Insights from
buy-side professionals, Working paper, Tuck School of Business, Dartmouth.
Bradshaw, Mark T., Richardson, Scott A., and Richard G. Sloan, 2001, Do analysts and auditors
use the information in accruals? Journal of Accounting Research 39, 45-74.
Bulow, Jeremy, R. Mock, and L. Summers, 1987, How does fully reflect the implications of
current earnings for future earnings, Journal of Accounting and Economics 13, 305-341.
Butler, K. C. and L. H. Lang. 1991, The Forecast Accuracy of Individual Analysts: Evidence of
Systematic Optimism and Pessimism, Journal of Accounting Research, 29, 150-156.
Carroll, Thomas J., and Greg Niehaus, 1998, Plan funding and corporate debt ratings, Journal of
Risk and Insurance 65, 427–443.
Core, John E., W. Guay, and T. Rusticus, 2006, Does weak governance cause weak stock
returns? An examination of firm operating performance and investors’ expectations, Journal of
Finance 61, 655-687.
Cornell, B., 2001. Is the response of analysts to information consistent with fundamental
valuation? The case of Intel, Financial Management 30, 113–136.
Coronado, J. L. and S. A. Sharpe, 2003, Did pension plan accounting contribute to a stock
market bubble? Brookings Paper on Economy Activity, 323-359.
Coronado, J.L. and O. S. Mitchell, S. A. Sharpe, S. B. Nesbitt, 2008, Footnotes Aren’t Enough:
The Impact of Pension Accounting on Stock Values, NBER Working Paper.
Conrad, J., Cornell, B., Landsman, W., Rountree, B., 2006. How do analyst recommendations
respond to major news? Journal of Financial and Quantitative Analysis 41, 25–49.
Daniel, K.D., D. Hirshleifer, and A. Subrahamanym, 1998, Investor psychology and security
market under- and over-reaction, Journal of Finance 53, 1839-1886.
Dugar, Amitabh, and Siva Nathan, 1995, The effect of investment banking relationships on
financial analysts’ earnings investment recommendations, Contemporary Accounting Research
Eames, M., Glover, S., Kennedy, J., 2002, The association between trading recommendations
and broker-analysts’ earnings forecasts. Journal of Accounting Research 40, 85–104.
Fama, Eugene F. and Kenneth R. French, 1992, The cross-section of expected stock returns,
Journal of Finance 47, 427-465.
Fama, E. F., and K. R. French, 1993, Common risk factors in the returns on stocks and bonds,
Journal of Financial Economics 49, 283-306.
Financial Accounting Standards Board (FASB), 1985. Employers’ Accounting for Pensions.
Statement of Financial Accounting Standard No. 87. Stamford, CT: FASB.
Financial Accounting Standards Board (FASB), 2006, Employers’ accounting for defined benefit
pension and other postretirement plans. Statement of Financial Accounting Standard No. 158.
Stamford, CT: FASB.
Francis, Jennifer, and Donna Philbrick, 1993, analysts’ decisions as products of a multi-task
environment, Journal of Accounting Research 31, 216–230.
Franzoni, Francesco and Marín, Jose, 2006, Pension plan funding and stock market efficiency,
Journal of Finance 61, 921-956.
Franzoni, Francesco, 2009, Underinvestment vs. overinvestment: Evidence from price reactions
to pension contributions, Journal of Financial Economics 92, 491-518.
Feldstein, Martin, and R. Morck, 1983, Pension funding decisions, interest rate assumptions and
share prices, in Zvi Bodie and John B. Shoven, eds.: Financial Aspects of the U.S. Pension
System (Chicago: UCP).
Feldstein, Martin, and S. Seligman, 1981, Pension funding, share prices, and national savings,
Journal of Finance 36, 801-824.
Fried, D. and D. Givoly, 1982, Financial Analysts’ Forecasts of Earnings: A Better Surrogate for
Market Expectations.” Journal of Accounting and Economics, 82-107.
Friesena, G., Wellerb, P., 2006. Quantifying cognitive biases in analyst earnings forecasts.
Journal of Financial Markets 9, 333–365.
Goetzmann, William N. and Nadav Peles, 1997, Cognitive dissonance and mutual fund investors,
Journal of Financial Research, 145-158.
Hiershleifer, D., 2001, Investor psychology and asset pricing, Journal of Finance 56, 1533-1597.
Hirst, D. E., and P. E. Hopkins, 1998, Comprehensive income reporting and analysts' valuation
judgments, Journal of Accounting Research 36, 47-75.
Hodgkinson, L., 2001. Analysts’ forecasts and the broker relationship, Journal of Business
Finance & Accounting 28, 943–961.
Hogarth, R.M., 1987, Judgement and choice: The psychology of decision, 2nd ed. New York:
Hong H., and J. Kubik, 2003, Analyzing the analysts: Career concerns and biased earnings
forecasts, Journal of Finance 58, 313-351.
Hong and Kacperczhy, 2010, Competition and Bias, Quarterly Journal of Economics,
Hong, H, T. Lim, and J. Stein, 2000, Bad news travels slowly: size, analyst coverage, and the
profitability of momentum strategies, Journal of Finance 59, 265-295.
Hopkins, P. E., 1996, The effect of financial statement classification of hybrid financial
instruments on financial analysts' stock price judgments, Journal of Accounting Research 34, 33-
Hutchens, Robert. 1989, Seniority, Wages, and Productivity: A Turbulent Decade, Journal of
Economic Perspectives 3(4): 49-64.
Irvine, P., 2004, Analysts’ forecasts and brokerage-firm trading, The Accounting Review 79,
Jin, Li, Merton, Robert C., and Bodie, Zvi, 2006, Do a firm’s equity returns reflect the risk of its
pension plan? Journal of Financial Economics 81, 1-26.
Jackson, Andrew R., 2005, Trade generation, reputation, and sell-side analysts, Journal of
finance 60, 673-717.
Jacob, J. and Lys, T., 1999, Determinants and implications of the serial-correlation in analysts'
earnings forecast errors, Working Paper, University of Colorado and Northwestern University.
Kim, C., Lustgarten, S., 1998. Broker-analysts’ trade-boosting incentive and their earnings
forecast bias. Working paper, Queens College of the City University.
Lamont, O., 1995, Corporate-Debt Overhang and Macroeconomic Expectations, American
Economic Review 85, 1106-1117.
Lazear, Edward. 1979, Why Is There Mandatory Retirement? Journal of Political Economy 87,
Lim, Terence, 2001, Rationality and analysts’ forecast bias, Journal of Finance 56, 369–385.
Lin, Hsiou-wei, and Maureen F. McNichols, 1998, Underwriting relationships, analysts’ earnings
forecasts and investment recommendations, Journal of Accounting and Economics 25, 101–
Ljungqvist, A., F. Marston, L.T. Starks, K. D. Wei and H. Yan, 2007. Conflicts of interest in
sellside research and the moderating role of institutional investors, Journal of Financial
Economics 85, 420-456.
Martin, L. J., and G. V. Henderson, Jr., 1983, On Bond Ratings and Pension Obligations: A
Note, Journal of Financial and Quantitative Analysis, 463-470.
Maher, John. 1987. Pension Obligations and the Bond Credit Market: An Empirical Analysis of
Accounting Numbers, Accounting Review 62: 785- 798.
McNichols, M., P. O’Brien, 1997, Self-selection and analyst coverage, Journal of Accounting
Research 35, 167–199.
Mikhail, Michael B., Beverly R. Walther, and Richard H. Willis, 1997, Do security analysts
improve their performance with experience? Journal of Accounting Research 35, 131-157.
Mikhail, Michael B, Beverly R. Walther, and Richard H. Willis, 2003, The Effect of Experience
on Security Analyst Underreaction, Journal of Accounting and Economics 35, 101-116.
Moyer, R. - C., Chatfield, R. - E., Sisneros, P. - M., 1989. Security Analyst Monitoring Activity:
Agency Costs and Information Demands. The Journal of Financial and Quantitative Analysis 24,
Myers, S., 1977, Determinants of Corporate Borrowing, Journal of Financial Economics, 5, 147-
O’Brien, Patricia C., 1988, Analysts’ forecasts as earnings expectations, Journal of Accounting
and Economics 10, 53-83.
O’Brien, P., McNichols, M., Lin, H., 2005. Analyst impartiality and investment banking
relationships, Journal of Accounting Research 43, 623–650.
Picconi, Marc, 2006, The Perils of Pensions: Does pension accounting lead investors and
analysts astray? Accounting Review 81, 925-955.
Plumlee, M. A., 2003, The effect of information complexity on analysts' use of that information,
Accounting Review 78, 275-296.
Rauh, Joshua D., 2006, Investment and financing constraints: evidence from the funding of
corporate pension plans, Journal of Finance 61, 33-71.
Rauh, Joshua D., 2006, Investment and financing constraints: Evidence from the funding of
corporate pension plans, Journal of Finance 61, 33-72.
Richardson, S., Teoh, S., Wysocki, P., 2004. The walk-down to beatable analyst forecasts: the
roles of equity issuance and inside trading incentives. Contemporary Accounting Research 21,
Schipper, K. 1991. Analysts’ forecasts. Accounting Horizons 5: 105-132.
Shivdasani, Anil and Irina Stefanescu, 2009, How do pensions affect corporate capital structure
decisions? Working Paper, University of North Carolina and Indiana University.
Sloan, Richard G., 1996, Do Stock Prices Fully Reflect Information in Accruals and Cash Flows
about Future Earnings? Accounting Review 71, 289-315
Zhang, X. F., 2006, Information uncertainty and stock returns, Journal of Finance 61, 105-137.
Table 1: Summary Statistics
Panel A reports summary statistics of firms sponsoring defined benefit (DB) pension plans from 1988 to 2005.
Assets are Compustat data 6. Size is measured by firm market capitalization at the end of fiscal years (data 199
times data 25). The fair market value of plan assets (FVPA) equal to the sum of data 287 and data 296. The
projected pension obligations (PBO) equal to the sum of data 286 and data 294. Fund surplus/deficit is equal to the
different between FVPA and PBO. Funding ratio (FR) is defined as the fund surplus/deficit divided by firm
capitalization at the end of fiscal year. A firm is underfunded (UF) if FR < 0, and overfunded (OF) if FR 0. Panel
B shows the average analyst coverage for the combined datasets of Compustat pension sample and I/B/E/S data, and
we report analyst coverage across the sample period as well as the coverage in 1990, 1995, 2000, and 2005. To
calculate the mean and median of the above variables, we first compute the cross-sectional mean (median) in each
year and then estimate the averaged mean (median) statistics across the sample period.
Panel A: Pension Overfunded Firms (OF) vs. Underfunded Firms (UF)
All Firms OF Firms UF Firms
(N=29,899) (N=13,086) (N=16,813)
Mean Median Mean Median Mean Median
Assets ($ mil.) 10,003.06 1,119.42 12,620.53 1,458.20 8,828.70 935.95
Size ($ mil.) 4,496.34 643.88 5,147.48 805.62 4,119.62 534.15
FVPA ($ mil.) 770.14 65.99 1,095.62 114.02 534.42 43.44
PBO ($ mil.) 799.99 73.02 931.219 94.62 688.63 56.86
Fund surplus/deficit ($ mil.) 29.84 -3.60 164.33 12.70 -154.21 -9.99
Funding ratio (FR) (%) 3.46 -0.34 139.31 1.94 56.4 -1.85
Panel B: Analyst Coverage
1988 - 2005 1990 1995 2000 2005
# Firms covered by analysts 4,537 3,381 5,039 5,226 5,114
# Firms w/DB plans covered by
899 902 967 820 805
# Analysts 2,212 1,908 2,142 2,345 2,256
# Analysts covering per firm 7.0 7.2 6.5 6.6 6.1
# Analysts covering per OF firm 7.8 7.5 6.8 7.0 6.4
# Analysts covering per UF firm 7.1 7.0 6.6 7.1 6.4
Table 2: Analyst Earnings Forecast Errors, Pension Funding Ratios and Analyst Biases
Panel A reports average earnings forecast errors (FCE) for one-year earnings forecast across 11 pension funding
ratio portfolios for the full sample period. Sample firms are sorted into eleven portfolios in July of year t based on
pension funding ratios FR in fiscal year t -1. We use the I/B/E/S detail file and estimate FCEi,t based on equation (4).
We first calculate the average forecast errors across firms in each portfolio in each year, and then calculate the time-
series average of forecast errors at each portfolio, as well as forecast error spreads between extreme portfolios. Panel
B reports the time-series averages of cross-sectional means of individual analyst FCE for double-sorted portfolios by
funding ratios and by analyst bias. Here, we use individual analyst FCE i,j,t calculated by equation (3). Analyst bias is
the average forecast error made by an analyst for all covering firms in the prior three years. In July of each year t, we
first sort firms into 11 portfolios based on pension funding ratios. Then, within each portfolio, we further sort all
firm-analyst observations into four quartiles based on individual analyst forecast bias. Panel C reports the time-
series averages of cross-sectional means of individual analyst FCE for double-sorted portfolios by funding ratios and
analyst forecast accuracy. Analyst accuracy is the average of absolute value of forecast error made by an analyst for
all covering firms in the prior year years. Panel D shows the results of a panel regression of FCE i,j,t on firm funding
ratio, analyst bias, and analyst accuracy. We include fixed firm and fixed year effects in the model. Standard errors
are adjusted for clustering by firm and year following Petersen (2009). t-statistics are reported in parenthesis. The
symbols ***, ** and * indicate the significance level at 0.01, 0.05, and 0.10, respectively.
Panel A: Forecast Errors sorted by Firm Funding Ratios
FR Rank Forecast Errors
1(most UF) -1.81
10 (least UF) -0.17
D11 (OF) -0.31
All UF -0.53***
D1 - D10 -1.64***
D1 - D11 -1.50***
Panel B: Forecast Errors sorted by Firm Funding Ratio and Panel C: Forecast Errors sorted by Funding Ratios
Previous Analyst Forecast Biases and Previous Analyst Forecast Accuracy
Analyst Prior Forecast Bias Analyst Prior Forecast Accuracy
Q1 Q4 Q4
FR Rank 2 3 Q1 (low) 2 3
(low) (high) (high)
1(most UF) -2.49 -1.99 -1.55 -1.15 -1.02 -1.78 -1.92 -2.53
2 -1.41 -0.76 -0.82 -0.68 -0.66 -0.79 -0.89 -1.34
3 -0.73 -0.54 -0.48 -0.35 -0.28 -0.58 -0.48 -0.76
4 -0.88 -0.72 -0.56 -0.35 -0.33 -0.63 -0.71 -0.84
5 -0.56 -0.36 -0.35 -0.3 -0.3 -0.38 -0.48 -0.4
6 -0.74 -0.58 -0.43 -0.36 -0.17 -0.46 -0.66 -0.81
7 -0.62 -0.45 -0.33 -0.34 -0.24 -0.39 -0.45 -0.65
8 -0.47 -0.3 -0.21 -0.23 -0.21 -0.21 -0.32 -0.46
9 -0.53 -0.39 -0.23 -0.29 -0.16 -0.24 -0.45 -0.57
10 (least UF) -0.25 -0.17 -0.13 -0.11 -0.09 -0.13 -0.19 -0.26
D11 (OF) -0.38 -0.38 -0.27 -0.16 -0.18 -0.26 -0.37 -0.45
D1 - D10 -2.24 -1.82 -1.42 -1.04 -0.93 -1.65 -1.73 -2.27
(t – stat) (-2.13) (-2.06) (-1.91) (-1.71) (-1.43) (-1.94) (-1.89) (-2.35)
D1 - D11 -2.11 -1.61 -1.28 -0.99 -0.84 -1.52 -1.55 -2.08
(t – stat) (-1.98) (-2.04) (-1.80) (-1.57) (-1.56) (-1.93) (-1.84) (-2.13)
D1-D10 -1.20 1.34
(t-stat) (-1.87) (2.25)
D1-D10 -1.12 1.24
(t-stat) (-1.91) (2.19)
Panel D: Regression of Forecast Errors on Pension Funding Ratios, Previous Forecast Biases and Accuracy
Number of observations 69,872
Adjusted R2 0.19
Table 3: Analyst Forecast Errors, Pension Funding Ratios and Analyst Experience
This table reports the average of individual analyst forecast for double-sorted portfolios by firm funding ratios and
analyst experience. We use the I/B/E/S detail file to calculate individual analyst forecast errors (FCEi,j,t) for one-year
earnings forecasts. The double-sort process is the same as in Table 2. Panel A uses raw analyst experience in double-
sorts. Raw forecast experience is the number of years for which an analyst has issued earnings forecasts for a firm.
Panel B uses adjusted analyst experience in double-sorts. Adjusted analyst experience is estimated by the residual
terms in the annual cross sectional regression of raw experience made by an analyst for a firm on the age of the firm
and the logarithm of market capitalization of the firm. The t-statistics for the forecast error differences are reported
in the parentheses.
Panel A: Forecast Errors sorted by Firm Panel B: Forecast Errors sorted by Firm Funding
Funding Ratio and Analyst Raw Experience Ratio and Analyst Adjusted Experience
Analyst Raw Experience Analyst Adjusted Experience
Q1 Q4 Q4
FR Rank 2 3 Q1 (low) 2 3
(low) (high) (high)
1(most UF) -2.78 -1.86 -1.57 -1.12 -2.67 -1.97 -1.56 -1.04
2 -1.38 -0.91 -0.81 -0.54 -1.32 -1.19 -0.61 -0.52
3 -0.81 -0.67 -0.44 -0.14 -1.07 -0.54 -0.32 -0.23
4 -0.91 -0.7 -0.56 -0.29 -0.8 -0.81 -0.65 -0.29
5 -0.43 -0.58 -0.36 -0.17 -0.79 -0.46 -0.24 -0.21
6 -1.31 -0.52 -0.17 -0.11 -1.01 -0.65 -0.21 -0.3
7 -0.93 -0.44 -0.18 -0.12 -0.89 -0.33 -0.36 -0.19
8 -0.84 -0.2 -0.12 -0.08 -0.84 -0.2 -0.12 -0.08
9 -0.99 -0.31 -0.19 -0.05 -0.88 -0.29 -0.26 -0.11
10 (least UF) -0.54 -0.19 -0.16 -0.09 -0.54 -0.19 -0.16 -0.13
D11 (OF) -0.51 -0.41 -0.12 -0.19 -0.53 -0.41 -0.23 -0.15
D1 - D10 -2.24 -1.67 -1.41 -1.03 -2.13 -1.78 -1.4 -0.91
(t – stat) (-2.44) (-2.04) (-1.70) (-1.47) (-2.67) (-2.14) (-1.69) (-1.35)
D1 - D11 -2.27 -1.45 -1.45 -0.93 -2.14 -1.56 -1.33 -0.89
(t – stat) (-2.38) (-2.14) (-1.85) (-1.38) (-2.86) (-2.07) (-1.84) (-1.18)
D1-D10 -1.21 -1.22
(t-stat) (-2.41) (-1.97)
D1-D10 -1.34 -1.25
(t-stat) (-3.49) (-2.37)
Table 4: Analyst Forecast Errors across, Firm Underfunding Experience and Analyst Experience
This table reports the average individual forecast errors at each of the portfolios sorted by firm funding ratios, firm
underfunding experience, and analyst experience. We use the I/B/E/S detail file to calculate individual analyst
forecast errors for one-year earnings forecasts. Panel A reports the results based on double-sortt by firm funding
ratios and pension underfunding experience. A firm is first-time underfunded when it is underfunded in year t-1 but
overfunded in the prior two years. A firm is multiple-time underfunded when it is underfunded in year t-1 and also
underfunded in either year t-2 or t-3. We first calculate the average forecast errors at firm level, then at portfolio
level and finally cross time. Panel B further sorts analysts into quartiles based on their adjusted experience. We first
calculate the average forecast error at portfolio level and then across time. The t-statistics for the forecast error
differences are reported in the parentheses.
Panel A: Forecast Errors: First- vs. Multiple-time Underfunders
FR Rank First Multi
1(most UF) -2.83 -0.79
2 -1.55 -0.33
3 -0.96 -0.23
4 -0.93 -0.31
5 -0.72 -0.23
6 -0.81 -0.18
7 -0.75 -0.16
8 -0.37 -0.18
9 -0.78 -0.14
10 (least UF) -0.37 -0.07
D1 - D10 -2.46 -0.72
(t – stat) (-2.70) (-1.71)
D1 - D10 -1.74
(t – stat) (-2.55)
Panel B: Forecast Errors Double-sorted for First-time Panel C: Forecast Errors Double-sorted for
Underfunders Multiple-time Underfunders
Analyst Adjusted Experience Analyst Adjusted Experience
Q1 Q4 Q1 Q4
FR Rank 2 3 2 3
(low) (high) (low) (high)
1(most UF) -3.84 -3.32 -2.44 -1.74 -0.98 -0.89 -0.76 -0.54
2 -2.41 -1.94 -1.08 -0.79 -0.45 -0.36 -0.3 -0.21
3 -1.48 -0.97 -0.91 -0.55 -0.42 -0.32 -0.14 -0.05
4 -1.47 -0.95 -0.82 -0.48 -0.47 -0.21 -0.29 -0.27
5 -1.48 -0.68 -0.53 -0.22 -0.34 -0.26 -0.24 -0.11
6 -1.05 -0.86 -0.85 -0.46 -0.34 -0.24 -0.11 -0.04
7 -0.95 -0.81 -0.79 -0.38 -0.36 -0.18 -0.02 -0.11
8 -0.63 -0.45 -0.21 -0.23 -0.33 -0.22 -0.09 -0.07
9 -0.84 -0.81 -0.82 -0.56 -0.31 -0.16 -0.06 -0.03
10 (least UF) -0.61 -0.43 -0.26 -0.19 -0.11 -0.07 -0.04 -0.06
D1 - D10 -3.23 -2.89 -2.18 -1.55 -0.87 -0.82 -0.72 -0.48
(t – stat) (-3.12) (-2.67) (-2.36) (-2.18) (-1.91) (-1.96) (-1.69) (-1.38)
D1-D10 -1.68 -0.39
(t-stat) (-2.35) (-1.36)
Table 5: Analyst Forecast Errors, Institutional Ownership and Analyst Coverage
This table reports the average individual forecast errors at each of the portfolios sorted by firm funding ratios, firm
institutional ownership, and analyst coverage. We use the I/B/E/S detail file to calculate individual analyst forecast
errors for one-year earnings forecasts. Panel A reports the results based on double-sorting by firm funding ratios and
institutional ownership. Institutional ownership of a firm is calculated as the percentage of the number of shares
owned by institutional investors of the total number of shares outstanding of the firm for a year. We first calculate
the average forecast errors at firm level, then at portfolio level and finally cross time. Panel B reports the results
based on double-sorting by firm funding ratios and analyst coverage. Analyst coverage is the residual term in the
regression of the number of firms covered by an analyst in a year on the logarithm of the market capitalization of the
firms and the realized earnings per share of the firms in the following year. Panel C reports the results based on
three-way sorts by firm funding ratio, firm institutional ownership, and analyst coverage. We first calculate the
average forecast error at portfolio level and then across time. The t-statistics for the forecast error differences are
reported in the parentheses.
Panel A: Double Sort by Funding Ratio and Institutional Panel B: Double Sort by Funding Ratio and
Ownership Analyst Coverage
Institutional Ownership Analyst Coverage
Q1 Q4 Q4
FR Rank 2 3 Q1 (low) 2 3
(low) (high) (high)
1(most UF) -3.49 -1.52 -1.16 -1.07 -2.17 -2.05 -1.58 -1.47
2 -1.74 -0.99 -0.5 -0.39 -1.35 -0.92 -0.71 -0.65
3 -0.92 -0.35 -0.49 -0.3 -0.78 -0.66 -0.42 -0.43
4 -0.91 -1.04 -0.35 -0.41 -0.83 -0.71 -0.61 -0.34
5 -0.78 -0.23 -0.08 -0.43 -0.68 -0.48 -0.21 -0.15
6 -0.86 -0.65 -0.28 -0.27 -0.79 -0.57 -0.38 -0.31
7 -0.54 -0.33 -0.32 -0.46 -0.46 -0.42 -0.41 -0.35
8 -0.46 -0.45 -0.08 -0.24 -0.33 -0.31 -0.35 -0.26
9 -0.95 -0.31 -0.19 -0.09 -0.41 -0.45 -0.34 -0.28
10 (least UF) -0.36 -0.12 -0.11 -0.11 -0.16 -0.27 -0.17 -0.08
D11 (OF) -0.45 -0.23 -0.27 -0.29 -0.34 -0.43 -0.29 -0.21
D1 - D10 -3.13 -1.40 -1.05 -0.96 -2.01 -1.78 -1.41 -1.39
(t – stat) (-3.27) (-1.94) (-1.76) (-1.47) (-2.46) (-2.13) (-1.92) (-1.87)
D1 - D11 -3.04 -1.29 -0.89 -0.78 -1.83 -1.62 -1.29 -1.26
(t – stat) (-2.86) (-1.76) (-1.50) (-1.37) (-2.05) (-1.86) (-1.74) (-1.82)
D1-D10 -2.17 -0.62
(t-stat) (-2.35) (-1.71)
D1-D10 -2.26 -0.57
(t-stat) (-2.12) (-1.86)
Panel C: Triple Sorts of Funding Ratios, Institutional Holding, and Analyst Coverage
Analyst Coverage Groups of bottom Analyst Coverage Groups of Top Tercile
Tercile Group of Institutional Holding Group of Institutional Holding
FR Rank Q1 (low) 2 Q1 (low) 2
1(most UF) -4.54 -1.24 -1.37 -2.08 -1.16 -0.87
2 -1.81 -0.75 -0.37 -1.16 -0.73 -0.46
3 -1.17 -0.56 -0.26 -0.83 -0.22 -0.26
4 -0.71 -0.72 -0.31 -0.64 -0.51 -0.16
5 -0.93 -0.32 -0.55 -0.61 -0.23 -0.19
6 -0.56 -0.41 -0.23 -0.43 -0.35 -0.22
7 -0.39 -0.31 -0.48 -0.38 -0.37 -0.57
8 -0.42 -0.57 -0.43 -0.32 -0.45 -0.34
9 -0.55 -0.32 -0.18 -0.36 -0.26 -0.14
10 (least UF) -0.21 -0.43 -0.16 -0.15 -0.24 -0.12
-0.35 -0.26 -0.39 -0.32 -0.15 -0.22
D1 - D10 -4.33 -0.81 -1.21 -1.93 -0.92 -0.75
(t – stat) (-3.46) (-2.08) (-2.35) (-2.63) (-1.87) (-1.86)
D1 - D11 -4.19 -0.98 -0.98 -1.76 -1.01 -0.65
(t – stat) (-3.58) (-2.31) (-1.76) (-2.32) (-1.84) (-1.75)
D1-D10 -3.12 -1.08
(t-stat) (-3.27) (-1.74)
D1-D10 -3.21 -1.11
(t-stat) (-2.84) (-1.98)
Table 6: Regressions of Analyst Forecast Errors
This table reports the results of panel regressions of earnings forecast errors on firm funding ratio, analyst
experience, firm underfunding experience, firm institutional ownership, analyst coverage, and control variables. The
dependent variable is the 1-year average forecast error for a firm (column 1) or individual analyst forecast errors for
a firm (other columns). FR(+) is a dummy variable equal to one if funding ratio is no less than zero and zero
otherwise. FR(–) equals to funding ratio if funding ratio is less than 0 and zero otherwise. EXP is adjusted analyst
experience. FIRST is a dummy variable for a firm that equals one if the firm is a first-time underfunder and zero
otherwise. INST represents institutional ownership of a firm. COVG represents analyst coverage. FR_BS is the
funding ratio is the aggregate pension funding reported on the balance sheets (data 290 + data 300 – data 298) scaled
by the firm’s market capitalization. FCEt-1 is the lagged year forecast error made by the same analyst for the same
firm. DISPt-1 is the forecast dispersion for firm i in year t-1, measured as the standard deviation of analyst forecasts
made four months prior to fiscal year-end, scaled by the prior year-end stock price. LogSIZEt-1 is the log value of
firm i market value (the closing price at the end of year t-1 multiplied by common shares outstanding at the end of
year t-1) at the end of fiscal year t-1. LogBMt-1 is the log value of Book-to-market ratio for firm i at the end of fiscal
year t-1, where book value of equity is the book value of stockholders’ equity (data 216), plus balance sheet deferred
taxes and investment tax credit (data 35, if available), minus the book value of preferred stock (data 56, data 10, or
data 130). Betat-1 is calculated by regressing the stock’s daily return on the value-weighted market return using
ordinary least square and 100 trading days of returns data ending on December 31 of year t-1. MOMt-1 is the
previous 6-month stock return for firm i by the end of year t-1. ACCt-1 is accounting accruals for firm i in year t-1. It
is computed following Sloan (1996). We include fixed firm and fixed year effects in the model. Standard errors are
adjusted for clustering by firm and year following Petersen (2009). The t- statistics are reported in parenthesis. The
symbols ***, ** and * indicate the significant at 0.01, 0.05, and 0.10 level, respectively.
(1) (2) (3) (4) (5) (6) (7)
INTERETP -0.62** 0.41 0.44 0.44 0.41 0.41 0.43
(-2.25) (1.08) (1.19) (1.19) (1.07) (1.06) (1.17)
FR(+)i,t-1 -0.16 -0.14 -0.17 -0.17 -0.15 -0.14 -0.16
(-1.07) (-1.02) (-1.35) (-1.35) (-1.07) (-1.07) (-1.09)
FR(-)i,t-1 2.29* 1.73* 1.58* 1.58* 1.62** 1.71** 1.74**
(1.90) (1.87) (1.91) (1.91) (2.23) (2.28) (2.25)
FR(-)i,t-1*EXPi,j,t -0.23** -0.21** -0.22**
(-2.25) (-2.10) (-2.04)
FR(-)i,t-1*FIRSTi,t-1 3.69* 3.40* 3.16*
(1.92) (1.90) (1.89)
FR(-)i,t-1*EXPi,j,t * FIRSTi,t-1 -0.39** -0.36**
FR(-)i,t-1*INSTi,t-1 -1.34** -1.26** -0.84**
(-2.13) (-2.08) (-2.07)
FR(-)i,t-1*COVGi,t-1 -1.11* -1.28** -1.04*
(-1.82) (-2.35) (-1.86)
FR(-)i,t-1*INSTi,t-1*COVGi,t-1 1.34* 1.27*
FR(-)i,t-1*EXPi,j,t * INSTi,t-1
FR(-)i,t-1*EXPi,j,t * COVGi,t-1
FR_BSi,t-1 0.12 0.13 0.13 0.02 -0.02 0.01
(0.20) (0.21) (0.20) (0.03) (-0.02) (0.01)
FCEi,j,t-1 0.05 0.05 0.05 0.05 0.05 0.05
(1.10) (1.00) (1.00) (1.09) (1.07) (0.98)
DISPi,t-1 8.60* 8.30* 8.31* 8.67* 8.60* 8.35*
(1.75) (1.69) (1.69) (1.77) (1.76) (1.71)
LogSIZEi,t-1 -0.14*** -0.14*** -0.14*** -0.14** -0.14*** -0.14***
(-2.66) (-2.72) (-2.72) (-2.57) (-2.59) (-2.64)
LogBMi,t-1 -0.34*** -0.34*** -0.34*** -0.33*** -0.33*** -0.33***
(-3.53) (-3.50) (-3.50) (-3.32) (-3.29) (-3.25)
BETAi,t-1 0.03 0.03 0.03 0.03 0.03 0.03
(0.21) (0.20) (0.20) (0.24) (0.22) (0.20)
MOMi,t-1 -2.54** -2.51** -2.50** -2.53** -2.51** -2.48**
(-2.29) (-2.28) (-2.28) (-2.29) (-2.31) (-2.30)
ACCi,t-1 -0.11 -0.08 -0.08 -0.11 -0.07 -0.04
(-0.15) (-0.12) (-0.11) (-0.15) (-0.10) (-0.06)
Firm Dummies Yes Yes Yes Yes Yes Yes
Year Dummies Yes Yes Yes Yes Yes Yes
Num of Obs 12,404 69,872 69,872 69,872 69,872 69,872 69,872
Adjusted R2 0.218 0.291 0.312 0.332 0.311 0.312 0.353
Table 7: Regressions of Stock Returns on Analyst and Firm Characteristics
This table reports the results of panel regressions for firm stock performance on pension funding ratios (FR) and
various analyst and firm characteristics. Firm funding ratios are evaluated at the end of fiscal year t. The dependent
variable is the 1-year buy-and hold stock return from July of year t to June of year t+1 subsequent to fund ratio
evaluation year t-1. The independent variables are defined the same as in Table 6 except: EXP is the average
adjusted analyst experience across all analysts covering a firm for a year; SUE is the standardized unexpected
earnings for a firm, estimated following Bernard and Thomas (1990). We include fixed firm and fixed year effects in
the model. Standard errors are adjusted for clustering by firm and year following Petersen (2009). The t- statistics
are reported in parenthesis. The symbols ***, ** and * indicate the significant at 0.01, 0.05, and 0.10 level,
(1) (2) (3) (4) (5) (6) (7)
INTERETP 13.34*** 12.95 13.00 12.90 12.96 12.82 12.69
(4.87) (1.57) (1.59) (1.58) (1.60) (1.60) (1.60)
FR(+)i,t-1 1.08 0.70 0.59 0.63 0.68 0.63 0.56
(1.11) (0.63) (0.55) (0.58) (0.60) (0.54) (0.48)
FR(-)i,t-1 25.59** 35.12** 30.08** 29.84** 54.00** 55.22** 49.07**
(2.03) (2.42) (1.99) (2.21) (1.98) (2.03) (1.98)
FR(-)i,t-1*EXPi,t -6.65* -6.84* -6.44*
(-1.90) (-1.86) (-1.90)
FR(-)i,t-1*FIRSTi,t-1 67.89** 79.91** 78.86**
(2.08) (2.07) (2.31)
FR(-)i,t-1*EXPi,j,t * FIRSTi,t-1 -49.11* -82.59**
FR(-)i,t-1*INSTi,t-1 -49.33 -36.77 -31.61
(-1.29) (-1.15) (-1.14)
FR(-)i,t-1*COVGi,t-1 -4.11 -2.35 -1.59
(-1.34) (-1.32) (-1.22)
FR(-)i,t-1*INSTi,t-1*COVGi,t-1 5.75 13.59
FR(-)i,t-1*EXPi,j,t * INSTi,t-1
FR(-)i,t-1*EXPi,j,t * COVGi,t-1
FR_BSi,t-1 -5.53* -5.81* -5.87* -5.32* -5.32* -5.76*
(-1.78) (-1.80) (-1.68) (-1.67) (-1.67) (-1.69)
DISPi,t-1 -48.35 -51.34 -50.80 -41.93 -41.84 -44.96
(-1.37) (-1.45) (-1.44) (-1.19) (-1.20) (-1.27)
LogSIZEi,t-1 0.38 0.40 0.40 0.37 0.40 0.43
(0.57) (0.59) (0.59) (0.56) (0.61) (0.65)
LogBMi,t-1 2.81*** 2.86*** 2.87*** 2.84*** 2.84*** 2.88***
(2.61) (2.66) (2.67) (2.71) (2.70) (2.75)
BETAi,t-1 -0.79 -0.80 -0.81 -0.81 -0.85 -0.85
(-0.22) (-0.22) (-0.23) (-0.23) (-0.23) (-0.23)
MOMi,t-1 20.51 21.42 20.96 21.03 21.04 21.45
(0.60) (0.63) (0.61) (0.62) (0.61) (0.62)
SUEi,t-1 0.16 0.18 0.17 0.19 0.19 0.18
(0.35) (0.39) (0.37) (0.43) (0.42) (0.40)
ACCi,t-1 -38.04*** -37.81*** -37.72*** -36.68*** -36.64*** -36.41***
(-3.16) (-3.13) (-3.14) (-3.63) (-3.26) (-3.24)
Firm Dummies Yes Yes Yes Yes Yes Yes
Year Dummies Yes Yes Yes Yes Yes Yes
Num of Obs 21,890 11,194 8,538 16,938 21,097 8,449
Adjusted R 0.094 0.078 0.080 0.106 0.095 0.118
Table 8: Stock Performance and Analyst Forecast Errors
This table reports the results on regressions of the future 1-year stock return on various predicted forecast errors in
the lagged year based on our hypotheses. The dependent variable is the 1-year buy-and-hold return from July of year
t to June of year t+1subsquent to firm funding evaluation year t-1. We perform panel regressions based on equation
(7) and estimate the fitted FCEs according to equation (10) to (14). We include fixed firm and fixed year effects in
the model. Standard errors are adjusted for clustering by firm and year following Petersen (2009). The t-statistics for
the forecast error differences are reported in the parentheses.
(1) (2) (3) (4) (5) (6)
INTERETPT 13.25*** 11.84*** 16.22*** 11.64*** 13.72*** 13.76***
(4.38) (3.88) (4.15) 3.65 (4.24) (4.92)
FCEi ,t 1 1.94**
i ,t 1 1.46** 1.28*
i ,t 1 2.70** 2.92**
FCE INC i ,t 1 -0.26 -1.33
FCE OTH i ,t 1 2.11*** 2.17***
Firm Dummies Included Included Included Included Included Included
Year Dummies Included Included Included Included Included Included
Number of Obs 10,338 10,338 10,338 10,338 10,338 10,338
Adjusted R2 0.047 0.049 0.049 0.047 0.059 0.065
Figure 1: Defined Benefit Pension Plans in Sample
Panel A shows the proportion of firms with underfunded pension plans among all firms sponsoring DB pension
plans. Panel B reports aggregate pension plan surplus (deficit). We classify firms as sponsors of DB pension plans if
they have relevant pension plan information available in the Compustat database. In particular, we set the fair value
of plan assets (FVPA) as the sum of overfunded pension plan assets (data 287) and set underfunded pension plan
assets (data 296) and the present value of pension obligations (PBO) as the sum of overfunded pension obligations
(data 286) and underfunded pension obligations (data 294). Firms are identified as sponsors of DB pension plans if
they have above pension data in the Compustat. Pension funding surplus (deficit) is defined as the difference
between FVPA and PBO. In Panel B, the amount recognized on the balance is calculated as the sum of net amount
of the accrued pension liability (data 290), prepaid pension liability (data 300), and additional minimum pension
liability (data 298). The pension sample includes 29,901 firm-years observations between fiscal year 1988 to 2005,
with the average of 1,759 firms in each year. The fiscal year with the maximum (minimum) number of firms is 1996
(2004) with 1,875 (1,460) firms.
Panel A: Percentage of Firms with Underfunded Pension Plans
1988 1990 1992 19941996 1998 2000 2002 2004
Panel B: Pension Funding: Actual Amount versus Balance-sheet Recognized Amount
Plan Surplus/Deficit ($ bil)
Balance-sheet Recognized Amount
1988 1990 1992 1994 1996 1998 2000 2002 2004
Figure 2: Forecast Error Spreads
This table reports the spreads of analyst earnings forecast errors between D1 (most underfunded) and D10 (least
underfunded) as well as D1 and D11 (overfunded). Sample firms are sorted into 11 portfolios in July of year t based
on pension funding ratios FR in fiscal year t -1. We compute analyst earning forecast errors for each portfolio,
forecast error spreads between D1- D10, and errors spread between D1 - D11 over various forecasting horizons.
FCE1yr, FCE2yr, and FCE3yr refer to cross-sectional average of 1-year, 2-year, and 3-year analyst forecast errors.