Gender and Connections among Wall Street Analysts
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Gender and Connections among Wall Street Analysts
By LILY FANG and STERLING HUANG*
We examine how alumni ties with corporate boards affect male and female analysts’ job
performance and career outcomes differently. Connection improves men’s, but not women’s,
forecasting accuracy. Controlling for accuracy, connection still contributes to men’s, but not
women’s, likelihood of being voted by institutional investors as “star” analysts, a marker of
career success. For women, only education and forecast accuracy matter. This asymmetry is
absent from a placebo test. Our results suggest that men reap higher returns from connections
than women, and that investors are more willing to rely on soft information such as connections
to evaluate men than women.
* Fang: INSEAD, 1 Ayer Rajah Avenue, Singapore 138676. Huang: INSEAD, 1 Ayer Rajah Avenue, Singapore
138676. Special thanks go to Lauren Cohen, Andrea Frazzini, Chris Malloy, and Alok Kumar for generously sharing
their data on analyst education and gender information. We thank conference and seminar participants at the
American Finance Association, INSEAD, and University of Lugano for helpful comments. All errors and omissions
are our own.
1“I love my job [as an analyst]. The market does not care whether I am a man or a woman, only
whether I am right or wrong.”
--Kate Reddy, Sarah Jessica Parker’s
character in the comedy “I don’t
know how she does it”.
I. Introduction
Gender equality in the work place generates impassioned discussions. In the past few
decades, women have advanced significantly in education and labor market participation. In the
US, they out-number men among college graduates (Goldin, Katz, and Kuziemko, 2006), and
account for about half of the class in medical and law schools. Women are now also the majority
of the American workforce. 1 Despite these broad changes, however, the perplexing pattern
remains that women are thinly represented at the top of the business world: The ranks of women
among senior corporate officers in executive suites and board members remain in the single
digits and low teens.2 It thus appears that women’s empowerment in education and labor force
participation have not enabled them to successfully break the proverbial “glass ceiling” in the
business world. Ibbara, Carter, and Silver (2010) attribute this to the fact that men still get more
promotions than women.
This paper explores the idea that this observed “gender inequality” in career advancement
might be explained by how differentially men and women benefit from—and are evaluated
according to—their connections in the business community.
The context of our study is Wall Street analysts. We study the interplay between gender,
connections, and career outcomes in this population. Wall Street is a fascinating setting to study
these issues for at least three reasons. First, Wall Street has the reputation of being male-
dominated. Documenting the myths versus reality of the “gender inequality” in this industry is
important. Second, Wall Street is notorious clubby, and thus analysts’ connections could affect
their access to information, a key source of job performance. Explicitly establishing this value of
connections, Cohen, Frazzini, and Malloy (2008 and 2010) document that connections with
1
Economist, Dec 30, 2009.
2
According to recent Bloomberg reports, in 2012 the proportion of female CFOs in S&P 500 firms reached a record
high of 10.8%; the proportion of female CEOs was 4%, also a record high.
2companies’ boards help mutual fund managers and analysts obtain useful information to improve
their investment and recommendation performances. Yet demonstrating the gender difference in
in the relation between connections and job performance has not been done. We use the same
alum connection in Cohen, Frazzini, and Malloy (2008 and 2010) and study men and women’s
differential ability to use connections to enhance job performance.
Finally, Wall Street’s labor market is highly competitive and the stakes are high: winners
make millions and losers lose their jobs. And yet, what determines who is a “winner” are vague
and largely subjective. One of the most distinct markers of an analyst’s career success is his/her
being voted by institutional investors (mutual fund and hedge fund managers) as a star analyst. A
particularly influential ranking is conducted annually by the Institutional Investor magazine
through an opinion poll among thousands of fund managers. This poll result is featured
prominently in the October issues of the magazine each year. Winners are known as “All
American” (AA) analysts. Not only are AAs viewed as celebrities on Wall Street, but they are
also coveted by rival investment banks. As a result, the AA designation is one of the key
determinants of analyst compensation: A 2007 compensation survey among analysts indicates
that AAs on average command three times the pay of other analysts in the same bank. As an
opinion poll, the determinants of successful AA election are largely subjective, leading some
authors to argue that the AA election is a “popularity contest” (Emery and Li, 2009).
This means that the opening quote by the movie character played by Ms. Parker is only
half true. For analysts, indeed there are relatively objective job performance measures such as
earnings forecast accuracy. At the same time, career outcome measures such as the AA status is
based on largely subjective evaluation. How does connection affect both analysts’
performance—forecast accuracy, and their career outcomes—AA election results? Does
connection play a different role for men versus women?
Our finding suggests that connection improves men’s, but not women’s, forecasting
accuracy. Thus to the extent that analyst can use connections to obtain relevant information and
improve job performance (Cohen, Frazzini, and Malloy 2010), this effect appears to be driven by
male analysts. In addition, even after controlling for accuracy, connection per se still contributes
to men’s, but not women’s, likelihood of being voted by institutional investors as AA analysts.
For women, only education and forecast accuracy matter for successful election outcomes.
Interestingly, this asymmetry is absent from a placebo test. The results on AA election indicates
3that in money managers’ subjective evaluation of analysts, the analysts’ social capital is given
different weight for men versus women: Investors appear more willing to rely on soft
information such as analysts’ connections to favorably judge men than women.
This conclusion is different from Goldin and Rouse (2000), which finds evidence for sex-
biased hiring. Judging from the overall odds for female analysts to become AAs, they are not
discriminated against: Men and women’s odds of success are about the same, if not slightly in
favor of women. What we document is an asymmetry in the factors that drive their success.
Even though our study is situated in the specific context of Wall Street analysts, our
findings shed light on whether and how men and women are promoted differently, and by
inference, the phenomenon that ranks of women are thin at the top of the business world. The
analyst pool we study is a highly competitive and skilled labor pool, as is the pool for senior
corporate officers. The process by which analysts “succeed” involves both objective and
importantly also the subjective evaluations by others. Our findings suggest that women benefit
less from connections than men in both regards, and the evaluators appear to weigh social capital
more for men than for women in their judgment.
II. Gender, Connections, and Performance Evaluation
Why are women so thinly represented at the top of the business world? One reason may
be that women shun away from competition, and do less well under competitive pressure.
Niederle and Vesterlund (2007) show that among men and women with the same level of skill,
men are twice as likely as women to embrace competition by entering a tournament. Gneezy,
Niederle, and Rustichini (2003) present experimental evidence that women are likely to perform
worse than men in competitive settings, even if they perform equally well in noncompetitive
settings. Jurajda and Munich (2011) reach the same conclusion using field data on entrance to
highly selective but free universities. If reaching the top of the corporate world involves a series
of tournaments, men and women’s differential preference for competition may explain why so
few women reach the top.
A related attribute is that women are also more risk averse. Barber and Odean (2001) find
that among retail investors, men are more risk-willing in their trading behavior than women.
Huang and Kisgen (2012) and Levy, Li, and Zhang (2011) both document that women
4executives and board members are less acquisitive than men. If willingness to take risks is a
rewarded leadership trait, women’s higher risk aversion could hurt their chances of reaching
leadership positions.
Beyond the gender differences in innate characteristics, women’s endogenous career
choices and social constraints further contribute to the gender disparity among the business elite.
Women’s careers are more likely to be interrupted by child bearing and family considerations.
Bertrand, Goldin, and Katz (2010) show that female MBAs’ earnings lag males’ significantly a
decade after graduation, despite being nearly identical at the outset of their careers. And this gap
is largely explained by differences in career interruptions and weekly hours, both of which are in
turn due to motherhood.
In our setting, since we focus on Wall Street analysts—one of the most demanding and
selective professions—explanations based on the gender difference in risk-preferences and career
choices are unlikely to explain our results. Both men and women in this profession have chosen a
high competitive career. In fact, Kumar (2010) argue that self- selection means that only the
most competitive women would enter this career, leading them to outperform.
Our paper is more closely related to the literature that examines socialization as a source
of gender difference in career paths and the “glass ceiling”. Athey, Avery, and Zemsky (2000)
theorize that if senior employees are more likely to mentor junior employees of the same “type”
(e.g., gender or ethnicity), then minority employees will receive less mentoring. This could lead
to a “glass ceiling” equilibrium where the upper level remains less diverse than the entry level.
Ibarra (1992) demonstrates that men reap greater returns from networks than women. She
finds that while network positions of men and women exhibit no difference once background
characteristics are controlled for, men appear better able to use network ties to improve their
positions in organizations. Our empirical findings echo these conclusions: we find that generally
men and women are equally “connected”; but while connections are associated with better
performance and career outcomes for men, it is not the case for women. In addition, our evidence
goes a step further and suggests that investors—those who evaluate the analysts—appear more
willing to judge men favorably according to their social capital (connections) than in the case of
women. Because career advancement involves a large amount of subjective evaluation by others,
this finding indicates that the evaluators’ biases in the way they use “soft” versus “hard” signals
of capabilities could perpetuate the “glass ceiling” phenomenon.
5III. Data and Descriptive Statistics
Detailed data on analysts’ fiscal year-end earnings-per-share (EPS) forecasts and buy/sell
stock recommendations are obtained from the I/B/E/S database for the years 1993-2009. The
accuracy of the earnings forecasts, and the price impact of their recommendations are used as
analysts’ performance measures.
To identify gender, we obtain full names of AA analysts from the Institutional Investor
magazine. When in doubt, we check the articles in Institutional Investors that describe the
analysts. For non-AA analysts, we obtained and cross-check gender classification from Kumar
(2010).
For analysts’ connections, we use the same alumni tie measure as in Cohen, Frazzini, and
Malloy (2008 and 2010). This measure identifies a school tie between an analyst covering a firm
and a board member or senior officer at the firm being covered. For example, if analyst Amy
Cohen covers Apple, and if both Amy Cohen and one of Apple’s directors have gone to Stanford
University at the time when Amy Cohen provides coverage for Apple, then Amy Cohen’s
forecasts and recommendations over this periods on Apple are coded as “connected”.
To construct this school tie measure, we obtain analysts’ education information from
Cohen, Frazzini, and Malloy (2010), and officer and directors’ education information from
BoardEx. We construct the “connection” at two levels. At the coarser level, connection only
requires the analyst and the officer/director to have attended the same university. At the finer
level, we restrict connections to pairs that have attended the same degree program at the same
university. We consider 6 types of degrees—MBA, general master, PhD, medical degree, law
degree, and undergrad degree. Finally, because both analyst coverage and firms’ officers and
directors change over time, we annually update the connection variable to pick up only active
connections. For example, if analyst Amy Cohen is connected to Apple through director John
Smith who served on the Apple Board from 1999 to 2003, Amy Cohen’s forecasts and
recommendations Apple only during these years are coded as “connected”.
As the key measure for analysts’ career outcomes, we use the annual AA election results
from the October issues of the Institutional Investor magazine each year. The AA titles are given
to top analysts in roughly 60 industries, and there are four categories of awards3: 1st-place, 2nd-
place, 3rd-place, and runners-up. While the 1st and 2nd-place titles are awarded to one analyst per
3
For more details of the All American analyst election process, see, for example, Fang and Yasuda (2009, 2011).
6industry per year, the 3rd-place and runners-up titles are often shared by multiple individuals.
Thus, the 1st- and 2nd-place titles are special, representing the “cream of the crop” (Fang and
Yasuda, 2013). To distinguish between the 1st and 2nd-place titles with the rest, we classify these
two titles are high-rank, and others as low-rank.
Table 1 shows the gender distribution in our analyst sample. Between 1993 and 2009,
female analysts account for 12% of all analysts, and 14% of star analysts (AAs).4 Thus, by this
simple statistic, there is no big difference in women versus men’s overall odds of being elected
as an AA. This is consistent with evidence in Kumar (2010). Both percentages experienced some
time-variation: Female presence in the overall analyst population increased throughout the 90s,
but fell after 2000. The rise and fall is particular notable in the star analyst population (Figure 1).
We also note that after 2000, the female presence in the star analyst population has generally
exceeded female presence in the overall population; thus they are unconditionally more likely to
be elected a star in recent years.
Table 2 tabulates the number of stocks the average analyst is “connected” to, and
compares this statistics across gender and across star status (p-values for differences are
reported). For brevity, we report statistics using the coarser measure of connection—at the
university level. The finer connection measure at the degree level gives similar results. The
important observation is that there is generally no gender difference in analysts’ connections.
The p-values for the male-female differences are generally insignificant. For two years—1999
and 2001—the p-values are significant at the 10% level, but both indicate more connections
among women than men. If we instead consider number of connections as a percentage of total
number of firm covered (statistics unreported), the conclusion is similar. Thus, men do not have
more connections than women in our sample. This is important and means that the difference
between men and women we document is not because the different degrees of connectedness.
On the other hand, star analysts are almost always more connected that non-star analysts.
In 1993, stars and non-stars had 2.65 and 1.42 connections respectively, and in 2009 the numbers
4
The 12% figure for the overall population is slightly lower than previous reports. Green et al. (2007) find that females account
for 16% of analysts in top-tier brokerages and 13% in other brokerages. Kumar (2010) reports female to represent 16% of his
sample. One reason for the difference is that our data is a merged set between the I/B/E/S analyst file and analysts’ education
data, which is compiled from websites. The discrepancy suggests that female analysts’ education profiles are slightly under
sampled in the websites. Our figures on the female presence on star analysts (14%), however, match more closely with prior
evidence, suggesting that the under-sampling is less of a problem with more visible female analysts.
7are 4.27 versus 3.11. Both differences are significant. This is expected, and indicates that
connection is a valuable social and human capital for analysts.
Figure 2 graphs the fraction of analysts who attended Ivy League schools. The graph
indicates a clear downward trend in the percentages of analysts with Ivy League education.
Among female analysts, this fraction decreased from 42% in 1993 to 23% in 2009. This could be
due to the increasing popularity of other career options. But throughout the sample period, a
higher percentage of the female analysts hold degrees from Ivy League schools than male. If
education is used as a signal of skill (Spencer, 1973), female analysts appear at least as
competent as their male counterparts. This is also consistent with Kumar (2010).
Table 3 compares a number of other statistics between men and women. For brevity,
averages over the entire period are compared. We find that women tend to be slightly less
experienced than men. They also work less than men by covering fewer firms and fewer
industries. But work intensity seems about equal for each firm covered: Women and men issue
the same number of forecasts and recommendations per firm per year. These patterns are
expected and consistent with observations from Bertrand, Goldin, and Katz (2010).
In summary, while there is some differences in work-pattern between genders, on average
female analysts are as connected, and at least as well-educated, as their male colleagues.
IV. Connections, Performance, and Career Outcomes
A. Forecast Accuracy
In this section, we examine how connection differentially affects male and female
analysts’ forecast accuracy. We define forecast accuracy as the absolute difference between an
analyst’s forecast of a firm’s earnings per share (EPS) and the actual reported EPS, scaled by
stock price. Since this is a percentage forecast error measure, the small it is, the more accurate
(and better) the forecast is.
Forecast data set is highly heterogeneous. Analysts cover different firms that vary in
forecasting difficulty. Thus, a simple percentage forecast error may not be comparable across
firms. For example, whereas a 5% forecast error may be quite good for a complex and volatile
technology company, it may be very large for a stable and simple utility business. At the same
time, a firm is typically covered by a number of analysts. Among analysts who provide coverage
8for the same firm, it is the relative accuracy of one analyst versus another that matters. From
investors’ perspective, it is also natural for them to compare analysts this way. For instance,
while it is natural to compare all analysts providing coverage for IBM with one another, it is not
reasonable to directly compare IBM analysts with GE analysts by their average forecast errors.
Thus, following existing literature (Clement and Tse, 2005), we compute a standardized
forecast accuracy as follows:
Raw Accuracy i, j, t min( Raw Accuracy j,t )
(1) S tan dardized _ Accuracy i, j, t
max(Raw Accuracy j,t ) min( Raw Accuracy j,t )
where Raw Accuracyi, j, t is the scaled forecast error measure defined above based on the forecast
made by analyst i for firm j in year t, and min(.) and max(.) are the minimum and maximum of
the Raw Accuracy measures exhibited by all analysts covering the same firm j in the same period
t, respectively.5
This standardization thus converts the simple percentage error measure into a ranking:
All analysts covering IBM in the same time period are ranked relative to one another. Thus it
removes the heterogeneity in forecast errors across firm-year combinations, and the resulting
measure is comparable across analysts and firms. But to calculate the measure, we require that
the firm is covered by at least 5 analysts in a given year. As a result, thinly covered firms are
dropped in our final sample.
Table 4 provides summary statistics for the raw (unstandardized) forecast error, and the
standardized error. The table shows that standardization is important. Panel A shows that the
mean raw forecast error is 0.02 (or 2%). That is, the average forecast error as a percentage of the
prevailing stock price is 2%. The minimum is 0 (the analyst’s forecast matched the actual
reported EPS exactly), and the maximum is slightly over 75%. Notably, the mean raw forecast
error for men is 2%, slightly larger than women’s 1.9% (the difference is statistically significant
at 1%). Turning to standardized errors, we find its mean is 0.391 (39.1%), which means that the
distance between the average forecast and the best forecast (minimum raw forecast error) is just
under 40% of the distance between the best and the worst forecasts, for the same firm during the
same period. Notably, for women the average is 0.402, slightly larger than men’s 0.389 (the
difference is statistically significant at 1%).
5
We repeated our analysis using a different scaling method, and found qualitatively the same results. The alternative
scaling converts measures to z-scores by subtracting the variable’s mean and dividing it by it standard deviation.
9Because the standardized error measure provides a ranking of all analysts covering the
same firm in the same year, it is more informative about the relative performance among analysts.
We will focus on this measure in the remainder of the paper (rank-based measure has become the
standard, see, for example, Clement and Tse (2005), Hong et al. (2000)).
Table 5 tests the difference in standardized forecast error between genders, and between
connected versus non-connected forecasts. To ensure we capture differences due to connection
rather than among analysts, we keep only analysts who are connected to at least one of the stocks
they cover in this test. Thus all analysts in this test have some connected forecasts and some non-
connected forecasts.
First, on a standardized basis, men appear more accurate than women. Men’s average
standardized error is 0.387, versus women’s 0.402. Though statistically significant at 1%, the
difference is economically small. It means that men on average are 3% ((0.402-0.387)/0.40)
closer to the “best” forecast than women. The difference is significant at 10% for only 7 of the
17 year sample period. Thus there is no overwhelming performance difference between genders.
On the other hand, there is a clear difference associated with connection. The
standardized error for connected forecasts is 0.378, 4% lower than the 0.394 figure for non-
connected forecasts, the difference being significant at 1%. Over the 17 years, this difference is
significant at 10% in all but 2 years. This result on the relation between connection and forecast
accuracy is consistent with evidence from Cohen, Frazzini, and Malloy (2008, 2010), and
suggests that connections, which are important links for informal communication, contributes
positively to information discovery.
Panel B of the table further examines the connection-related difference in the male and
female sub-samples. In the male sample, the conclusion mirrors the whole sample: connection is
associated with significantly smaller forecast errors. The average standardized error for
connected forecasts is 0.375, compared to 0.392 for non-connected forecasts, a 5% economic
difference and significant at 1%. In sharp contrast, we fail to see a connection-related advantage
in the female sample: the average standardized error for connected and non-connected forecasts
made by female analysts are 0.399 and 0.404 respectively, with a t-stat of 1.37 for the difference.
Connection is associated with marginally significantly better forecast accuracy in only 1 out of
the 17 years—2007. In the two other years when the difference is significant—2003 and 2006,
the direction is wrong: Connection is associated with worse (larger) forecast errors.
10In sum, Table 5 indicates that connected forecasts are more accurate than non-connected
forecasts. However, the connection effect is present only among men but not women. The fact
that connection is associated with higher accuracy is consistent with Cohen, Malloy, and Frazzini
(2008) and (2010), and indicates that connections can be a potentially important advantage for
analysts. But our result suggests that there is a gender asymmetry: The positive link between
connection and accuracy only exists for men but not women.
Table 6 corroborates this result using regression analysis. The dependent variable is
standardized forecast error. The key independent variables are the male dummy, connection, and
the interaction between the two. We continue to include only analysts who are connected to at
least one of the stocks he/she covers, to ensure that the model identifies variations due to
connection, rather than due to differences across analysts. Thus the resulting sample includes
only analysts who are connected to at least one of the stocks he/she covers. Model (1) and (2) use
the two levels of connection measures respectively. The panel regression includes a host of
analyst, forecast, and firm characteristics that have been identified in prior literature to affect
forecast accuracy. Analyst characteristics include an Ivy League dummy, an All-star dummy, the
analyst’s general as well as stock-specific experience, the size of the broker she/he works for, etc.
Forecast characteristics include days since last forecast (by the same analyst for the same firm),
forecast horizon, and forecast frequency. Firm characteristics include firm size, book-to-market
ratio, and past returns. We include joint firm-year fixed effects, and cluster the standard errors by
analyst.
Results in Table 6 confirm the univariate results in Table 5. The gender coefficient is
negative (marginally significant in Model 2), indicating that there is not a large gender difference
in performance (accuracy). The coefficient on Connect1 is positive and marginally insignificant
(Model 1) and that on Connect 2 is positive and significant at the 5% (Model 2). These variables
by themselves capture the effect of connection for female analysts. Their positive sign indicates
that connections do not result in more accurate forecasts for female analysts. On the other hand,
the interaction term between both connection measures and the gender dummy is negative and
significant, meaning that connection is related to more accurate forecasts for male analysts.6
6
It is possible that women benefit more from same-gender connections: female analysts may be more comfortable with female
executives and may be able to obtain useful soft information through informal communications. In unreported analysis, we focus
on same-sex networks and did not find it to play a stronger role for women.
11B. Recommendation Impact
Table 7 investigates how connections differentially affect the price impact of male and
female analysts’ buy/sell stock recommendations. Price impact is the abnormal stock returns
around the days when an analyst issues a new buy/sell recommendation. This measure reflects
the market’s perceived informativeness in the analyst’s recommendation, and like forecast
accuracy, it is another important—and relatively objective measure—of analysts’ performance.
Following extensive literature, we calculate the Cumulative Abnormal Returns associated
with a recommendation using the Daniel, Grinblatt, Titman, and Wermers (1997) (DGTW)
characteristics-based benchmarks as follows:
CAR[t , T ] ri , Bi , ,
T
(2)
t
where [t, T] is the relevant event window, ri , is the return for stock i on date , Bi , is stock i’s
benchmark return on date . Subtracting the contemporaneous benchmark return from the stock
return removes expected stock movements associated with stocks’ size, book-to-market ratio,
and momentum, leaving only firm-specific abnormal returns.7
We examine two event windows. The CAR[0,+1] (Panel A) is the immediate one-day
window after the analyst issues a recommendation. Extensive prior literature documents that the
market is efficient with respect to public information such as analyst recommendations; any
information contained in analyst recommendations is quickly incorporated into prices within one
day (Barber et al. (2001), Green (2006), Fang and Yasuda (2011), among others). Thus price
impact should be concentrated in this window. The CAR[+1, +30] window is the subsequent 30-
day window. While we do not expect additional price impact in this window, it is important to
look at this window for signs of price reversal. The idea is that if the price impact in day 1 is due
to true information (rather than over-reaction), then we should not see return reversal in the
subsequent days. But if the initial price impact is due to market over-reaction (to analyst status,
or gender, for example), such movement would be spurious and reversed in subsequent days.
Table 7 indicates that for male analysts, connection is associated with significantly higher
price impact. For connected recommendations made by male analysts, CAR[0, +1] is 1.58%, 24
basis points higher than the 1.33% price impact of non-connected male recommendations. The
7
The DGTW benchmarks are available via http://www.smith.umd.edu/faculty/rwermers/ftpsite/Dgtw/coverpage.htm. Details of
the DGTW benchmark construction is discussed in Daniel, Grinblatt, Titman, and Wermers (1997) and Wermers (2004).
12difference is significant at the 5% level. For sells, connected recommendations by male analysts
have an average CAR[0, +1] of -2.55%, 26 basis points bigger than the -2.29% figure exhibited
by non-connected recommendations. These effects are economicly large: 24 basis points
difference in daily returns implies an annual return difference of 60%. For female analysts, on
the other hand, connection is not associated with stronger price impact in either buys or sells.
The results in the CAR[+1, +30] window (Panel B) indicate that there is generally no
reversal in the subsequent period, as returns continue to drift in the positive (negative) direction
after buy (sell) recommendations. This indicates that the initial price impact reflects market’s
perceived information content in the recommendations, rather than over-reaction. 8 Thus the
positive association between connection and price impact in male analysts’ recommendation and
the absence of this pattern for females indicate that market perceives male analysts to be able to
obtain useful information through their connections but female analysts do not.
Summarizing Sections A and B, we find that analysts’ connections with company boards
are associated with both more accurate forecasts and more impactful recommendations for male
analysts but not for female analysts. There is a gender difference in the conclusion that
connections can be important sources of information and improve analyst performance (Cohen,
Frazzini, Malloy, 2010). The fact that such a positive link exists for men but not for women
indicates that men obtain more benefits from connections than women.
C. Career Outcomes – Star Status by Investor Voting
We now turn our attention to career outcomes, which we proxy with the AA status that is
the result of fund managers’ subjective voting. We again focus on the differential effect
connection has in male and female analysts’ odds of being voted.
We define three favorable career outcomes based on AA election results and use them as
dependent variables in probit analysis. “Promotion1” refers to cases where an analyst who was a
non-AA star in last year and becomes and AA this year. “Promotion2” are cases where an
8
We examined whether same-sex connection is more “useful” for female analysts. We find some evidence of this,
but the effect is not statistically significant. In unreported analysis, we find that CAR[0, +1] for buy
recommendations from female analysts with same-sex connection is 1.44%, 40 basis points bigger than the non-
connected female recommendations, albeit the difference is not significant at the 10% level, possibly due to a small
sample. For sells, female recommendations with same-sex connections exhibit CAR[0, +1] of -2.93%, 77 basis
points larger (in magnitude) than non-connected female recommendations, but one again the effect is not statistically
significant.
13analyst who was a low-rank AA (3rd-place or runner-up title holder) last year and becomes a
high-rank AA (1st or 2nd-place winner) this year. “All star” simply indicates whether an analyst
is elected as a star. Because AA election results tend to persist from year to year, when we use
“All Star” as the dependent variable, the lagged AA variable is used as a control for this
persistence. Explanatory variables include measures of analyst work quality and characteristics
such as analyst’s experience, the number of stocks and industries he/she covers, his/her track
recordof past forecast accuracy, and education—whether he/she has attended Ivy League school.
Table 8 reports the estimation results. In Panel A, we find that gender is generally
insignificant in determining the election outcomes. This is consistent with uni-variate evidence in
Table 1 which indicates that overall men and women’s odds of being elected are about the same;
there is no “gender inequality” in the rates of “promotion” to AA status. In the whole sample
connections alone and its interaction with the male dummy are also not significant. In contrast,
investors value experience—and especially firm-specific experience—when evaluating analysts.
One striking observation is low magnitudes of the pseudo-R2 in the regressions. For the
“promotion” regressions, the pseudo-R2 is below 10%; for the All-star equation, it is over 50%
but this is mainly due to the inclusion of the lagged star status and the persistence of election
results from year to year (Fang and Yasuda 2009, 2013). These results indicate that observable
analyst characteristics explain a very small fraction of analysts’ promotions; much of what
determines star election outcomes are unknown and subjective.
Panel B examines the probit regression in the male and female sub-samples separately, to
see whether different factors matter for the two genders’ election outcomes. Results here reveal
an interesting asymmetry. In the male sample, connection per se is highly valued by investors.
The connection indicator has coefficients in the range of 0.15-0.17, indicating that being
connected improves the odds by roughly 15% after other effects are controlled for. Importantly,
we have controlled for forecast accuracy in these regressions. So this effect is not because men
can use connections to improve their performance better than women.
In contrast, connection per se is not valued in the female sample by investors. In fact, the
variable has a negative sign in all equations. For women, two other variables matter: Ivy League
education, and past forecasting accuracy. Ivy League has a coefficient ranging from 0.33 to 0.54
in the different equations, indicating a huge economic effect. Past forecast error has a
significantly negative coefficient in the promotion from non-star to star, and from the promotion
14from the low-rank star status to the high-rank star status. Thus, inaccurate forecasts are penalized
(alternatively stated, accurate forecasts are rewarded) in the female sample. Interestingly, the
impact of forecast quality is especially significant for the promotion from non-star to star status,
but less so for the promotion from low-rank star to high-rank star, suggesting that quantifiable
competence plays a particularly important role in the promotion of relatively novice stars.
Notably, neither Ivy League education nor forecast accuracy is significant in determining men’s
odds of promotion or being elected a star.
These results reveal that investors value male and female analysts differently: While
connections is valued by investors and affects career outcomes positively for men, for women, it
is measurable characteristics—education and forecast accuracy—that matter. Thus, while male
and female analysts have roughly the same odds of being voted as stars, the factors that increase
their odds are different. Since we have shown that connections potentially improve analysts’ job
performance such as earnings forecast accuracy (especially for men), it is important that
accuracy is controlled for in the regression. The contrast in the relative importance of the two
factors in the two samples show that, fund managers, when subjectively evaluating and voting
for who they consider as “the best” analysts, are more willing to weigh signals such as the
analysts’ connectedness positively for men than for women; for women, performance rather than
such signals are favorably weighted.
D. A Placebo Test
One important question is whether the asymmetry in the factors that affect men and
women’s chance of being elected truly reflects a human bias in the way people (fund managers
in this context) evaluate men and women. An alternative is that connections are more strongly
correlated with some unobserved skill in men than in women, and therefore it is more important
in the male sample regression than in the female sample. The validity of these two alternative
explanations has important implications on how we interpret the evidence. If the asymmetry
reflects human bias, then our finding is relevant in understanding the “glass ceiling”
phenomenon; it shows that men appear to be judged more on their social capital (connections)
and women on their performance. But if the asymmetry reflects an omitted variable, then the
result only indicates investors’ efficient use of potential skill-relevant information in connections.
15To differentiate between these two possibilities, we use an alternative, computer-based
“best analyst” ranking as a placebo test. Since 1992, the Wall Street Journal publishes its own
annual “Best on the Street” list of top analysts for the year. But unlike the Institutional Investor
AA list which is based on human voting, Wall Street Journal’s ranking is data-based and
computerized. A research company named FactSet Research Systems collects, verifies
underlying data on stock recommendations made by analysts, and computes an aggregate
numerical score for each analyst’s performance made through the past 12 months, taking into
account analysts’ buy/hold/sell calls. While the exact algorithm is not disclosed, the Wall Street
Journal’s own description of the process emphasizes its “objectivity, accuracy, and fairness”.
We obtained Wall Street Journal “Best on the Street” rankings for the period 1999-2009.
The idea is that if the asymmetry in the factors that affect success in the AA voting reflects
biases in the way people make subjective evaluations, then we should not observe the same type
of asymmetry when the analysis is repeated on the Wall Street Journal ranking. In other words,
we should not see connections matter differentially for men and women in this setting.
We repeat the probit analysis in this setting, and report the results in Table 9. Results here
indicate that the aforementioned asymmetry in the factors influencing men and women’s odds of
becoming a star does not exist in this alternative ranking. Notably, connections per se do not
matter for either population. The contrast between this result and that in Table 8 suggests that the
asymmetrical impact of connections on men and women’s career outcomes is unique in
subjective evaluation.
E. Young versus old
If investors rely on connections to evaluate analysts, this may matter more for young
analysts with little track record than for older analysts. To examine this, we split the sample into
the “young” versus “old” population, by the median of experience (6 years). We repeat the
election results regression for the two sub-samples, and report the results in Table 10. We find
that the previously documented asymmetry exists primarily in the young analyst population, but
largely disappears for the older analyst population. Thus, young male analysts directly benefit
from connections in career outcomes while young female analysts do not.
16V. Conclusions
Using a large sample of Wall Street analysts, we analyze how connections (social capital)
affect job performance and career outcomes (human capital) differently for men and women. Our
findings support the notion that men reap higher returns from their social capital than women.
The male and female analysts in our sample do not exhibit differential amount of social capital;
but connections are related to performance and career outcomes in different ways for the two
genders. For men, connections are associated with more accurate earnings forecasts, more
impactful buy/sell recommendations—both of which are objective job performance measures.
And importantly, after controlling for these factors, connections per se still contribute positively
for men’s odds of being voted as a “Star” analyst by institutional investors, a key marker of
career success. None of these benefits of connections appear strong among female analysts.
Instead, important factors for female analysts’ career success include education background and
a track record of accurate forecasts.
These findings echo those in Ibarra (1992) that men reap higher benefits (or “returns”)
from connections than women. But our findings suggest two distinct reasons for this to happen.
First, men themselves are better at extracting useful information and use it to improve their
performance—earnings forecast accuracy and recommendation impact. But the second reason is
that the evaluators of the analysts—institutional money managers who use the analysts’ research
to make investment decisions—appear more willing to put weight on signals such as social
connections to favorably evaluate men, while preferring to rely on “hard data” such as education
and track record to appraise women.
Why might this be the case? It may be that when faced with the task of evaluating
performance under imperfect information, people feel more comfortable to rely on soft signals
such as a worker’s social ties to make a judgment when that worker is of a “familiar type”: There
is less asymmetric information and people can more reliably infer the value of the signals than
when evaluating unfamiliar types. Thus such behavior is completely rational on the part of the
evaluator, with no intent to discriminate against the unfamiliar type. Yet nonetheless it can
reinforce the “glass ceiling” equilibrium as the “familiar” type—which nearly by definition
would be the majority—appear to receive more benefits from their social connections. Indeed we
find that asymmetry in the way connections affect career outcomes for men and women is
particularly pronounced for young analysts.
17Even though our evidence is based on the Wall Street context, it is broadly relevant. As in
the case of analysts, whose career success depend largely on the subjective voting by money
managers, the career advancement of lower level employees depends heavily on the subjective
judgment by their superiors. If the “glass ceiling” phenomenon is reinforced by biases in the way
superiors judge subordinates of familiar versus unfamiliar types, then its fix would involve both
a natural evolution whereby the unfamiliar type becomes more numerous and more familiar to
the evaluators, as well as a top-down approach whereby the diversity at the top (among the
evaluators) “chips away” (Matsa and Miller, 2011) the glass ceiling.
18References
Athey, S., C. Avery, and P. Zemsky, 2000, Mentoring and Diversity, American Economic
Review, 765-786.
Barber, B., and T. Odean, 2001, Boys will be boys: Gender, over-confidence, and common stock
investment, Quarterly Journal of Economics 116, 261-292.
Bertrand, M., C. Goldin, and L. Katz, 2010, Dynamics of the Gender Gap for Young
Professionals in the Finance and Corporate Sectors, American Economic Journal Applied
Economics 2, 228-255.
Clement, M. and S. Tse, 2005, Financial analyst characteristics and herding behavior in
forecasting. Journal of Finance 60, 307–341.
Cohen, L., A. Frazzini, and C. Malloy, 2008, The small world of investing: Board connections
and mutual fund returns, Journal of Political Economy 116, 951-979.
Cohen, L., A. Frazzini, and C. Malloy, 2010, Sell-side School Ties, Journal of Finance, 65, 1409-
1437.
Daniel, K., M. Grinblatt, S. Titman, and R. Wermers, 1997, Measuring mutual fund performance
using characteristic-based benchmarks, Journal of Finance 52, 1035-1058.
Emery, Douglas, and Xi Li, 2009, Are Wall Street Analyst Rankings Popularity Contest?,
Journal of Financial and Quantitative Analysis, 44, 411-437.
Fang, L. and A. Yasuda, 2009, The Effectiveness of Reputation as a Disciplinary Mechanism in
Sell-side Research, Review of Financial Studies 22, 3735-3777.
Fang, L. and A. Yasuda, 2013, Are stars’ opinions worth more? The Relation between Analyst
Reputation and Recommendation Values, working paper.
Gneezy, U., M. Niederle, and A., Rustichini, 2003, Performance in Competitive Environments:
Gender Differences, Quarterly Journal of Economics 118(3), 1049-1074.
Goldin, C., L. Katz, and I. Kuziemko, 2006, The Homecoming of American College Women:
The Reversal and the College Gender Gap, Journal of Economic Perspectives 20, 133-156.
Goldin, C. and C. Rouse, 2000, Orchestrating Impartiality: The Impact of “Blind” Auditions on
Female Musicians, American Economic Review 90(4), 715-741.
Green, C. 2006, The Value of Client Access to Analyst Recommendation, Journal of Financial
and Quantitative Analysis 41, 1–24.
Hong, H., J. Kubik, and A. Solomon, 2000, Security Analysts’ Career Concerns and Herding of
Earnings Forecasts, Rand Journal of Economics 31, 121-144.
19Huang, J., D. Kisgen, 2012, Gender and Corporate Finance, forthcoming, Journal of Financial
Economics.
Ibarra, H., 1992, Homophily and differential returns: Sex difference in network structure and
access in an advertising firm, Administrative Science Quarterly 37, 422-447.
Ibarra, H., N. Carter, and C. Silva, 2010, Why Men Still Get More Promotions Than Women,
Harvard Business Review, September, 80-85.
Jurajda, S., and D. Munich, 2008, Gender Gap in Performance under Competitive Pressure:
Admissions to Czech Universities." American Economic Review 101(3), 514-18.
Korkki, P., 2011, For women, parity is still a subtly steep climb, New York Times, Oct 8.
Kuhnen, C., 2009, Business networks, corporate governance and contracting in the mutual fund
industry, Journal of Finance 64, 2185-2220.
Kulich, C., G. Trojanowski, M. Ryan, S. A. Haslam, and L. Ronneboog, 2010, Who gets the
carrot and who gets the stick? Evidence of gender disparities in executive remuneration,
Strategic Management Journal, forthcoming.
Kumar, A., 2010, Self-selection and the forecasting abilities of female equity analysts, Journal of
Accounting Research 48, 393-435.
Levy, M., K. Li, and F. Zhang, 2011, Men are from Mars, Women are from Venus: Gender and
Mergers and Acquisitions, working paper.
Matsa, D, and A. Miller, 2011, Chipping Away at the Glass Ceiling: Gender Spillovers in
Corporate Leadership, American Economic Review 101(3), 635-639.
Niederle, M, and L. Vesterlund, 2007, Do women shy away from competition? Do men compete
too much? Quarterly Journal of Economics 122, 1067-1101.
Spence, M., 1973, Job market signaling, Quarterly Journal of Economics 87, 355-374.
Wermers, R., 2004, Is money really smart? New Evidence on the Relation Between Mutual Fund
Flows, Manager Behavior, and Performance, working paper.
20Figure 1. Gender Distribution
This figure plots the fraction of females in the general analyst population and the star analyst (AA) population. The
sample of analyst is from our merged sample between I/B/E/S file, the analyst education file, and the BoardEx file.
Star analysts are identified from the October issues of the Institutional Investor magazine.
Percentage of Female in Analyst
Population
25%
20%
15%
10%
5%
0%
1993
1994
1995
1996
1997
1999
2000
2001
2002
2003
2004
2006
2007
2008
2009
1998
2005
Among all analysts Among star analysts
21Figure 2. Comparing Education
This figure plots the fraction of male and female analysts who have ever attended an Ivy League school. Star
analysts are identified from the October issues of the Institutional Investor magazine.
Proportion of analysts who attended
Ivy League
45%
40%
35%
30%
25%
20%
15% Male
10%
5% Female
0%
Proportion of star analysts who
attended Ivy League
90%
80%
70%
60%
50%
40%
30%
20% Male
10% Female
0%
22Table 1. Gender Distribution
This table reports the number of male and female analysts in the general population and the star analyst population.
Star analysts are identified from the October issues of the Institutional Investor magazine.
All analysts Star analysts
year Male Female % Female Male Female % Female
1993 215 24 10.04% 48 4 7.69%
1994 264 34 11.41% 55 3 5.17%
1995 302 42 12.21% 57 5 8.06%
1996 364 50 12.08% 53 7 11.67%
1997 432 70 13.94% 55 7 11.29%
1998 506 72 12.46% 66 8 10.81%
1999 554 84 13.17% 66 11 14.29%
2000 610 91 12.98% 71 13 15.48%
2001 642 88 12.05% 51 14 21.54%
2002 681 92 11.90% 51 13 20.31%
2003 762 104 12.01% 49 11 18.33%
2004 859 108 11.17% 45 10 18.18%
2005 937 127 11.94% 44 8 15.38%
2006 832 109 11.58% 48 7 12.73%
2007 722 92 11.30% 45 8 15.09%
2008 633 76 10.72% 52 9 14.75%
2009 548 64 10.46% 33 5 13.16%
Average 580 78 11.85% 52 8 13.76%
23Table 2. Comparing Connections
This table compares the number of connections between male and female analysts, and between star and non-star
analysts. A connection means the analyst shares a school tie—he or she has attended the same university—with an
officer or director of a firm he/she covers. Star analysts are identifies from the October issues of the Institutional
Investor magazine. *, **, *** denotes statistical significance of the difference at the 10%, 5%, and 1% levels based
on two-tailed tests, respectively.
Panel A: Comparing connections
p -value p -value
year Male Female Star Non-star
(diff.) (diff.)
1993 1.73 1.54 0.73 2.65 1.42 0.00 ***
1994 1.59 1.35 0.62 2.59 1.28 0.00 ***
1995 1.69 1.64 0.91 2.96 1.34 0.00 ***
1996 1.61 1.74 0.75 3.16 1.31 0.00 ***
1997 1.52 1.60 0.81 3.22 1.22 0.00 ***
1998 1.46 1.94 0.14 2.92 1.27 0.00 ***
1999 1.68 2.23 0.06 * 3.48 1.46 0.00 ***
2000 1.85 2.33 0.12 3.77 1.62 0.00 ***
2001 2.05 2.63 0.07 * 4.48 1.84 0.00 ***
2002 2.16 2.40 0.44 4.17 1.95 0.00 ***
2003 2.25 2.01 0.43 4.08 2.05 0.00 ***
2004 2.41 2.22 0.57 4.41 2.25 0.00 ***
2005 2.46 2.40 0.83 5.08 2.30 0.00 ***
2006 2.78 2.94 0.66 4.72 2.65 0.00 ***
2007 3.08 3.32 0.55 4.78 2.95 0.00 ***
2008 3.03 3.22 0.63 4.08 2.93 0.01 ***
2009 3.21 3.25 0.93 4.27 3.11 0.03 **
Panel B: Comparing conenctions in star and non-star sub-samples
Star Analysts Non-star Analysts
p -value p -value
year Male Female Male Female
(diff.) (diff.)
1993 2.72 1.75 0.59 1.41 1.50 0.86
1994 2.63 2.20 0.79 1.29 1.21 0.86
1995 2.82 4.13 0.33 1.38 1.06 0.43
1996 2.94 5.14 0.18 1.33 1.19 0.68
1997 2.99 4.80 0.21 1.24 1.07 0.54
1998 2.59 5.50 0.02 ** 1.25 1.37 0.70
1999 3.20 5.42 0.04 ** 1.42 1.69 0.32
2000 3.40 5.47 0.03 ** 1.63 1.61 0.95
2001 4.23 5.35 0.32 1.83 1.97 0.64
2002 3.73 6.13 0.03 ** 1.99 1.68 0.33
2003 4.00 4.43 0.66 2.10 1.63 0.14
2004 4.41 4.42 1.00 2.28 1.95 0.34
2005 4.94 6.00 0.44 2.31 2.16 0.60
2006 4.52 6.11 0.22 2.65 2.65 1.00
2007 4.71 5.25 0.73 2.93 3.13 0.62
2008 3.89 5.30 0.23 2.93 2.91 0.96
2009 4.02 5.75 0.21 3.13 2.89 0.65
24Table 3. Summary Statistics
This table reports demographic and work pattern statistics of male and female analyst samples. Ivy League is an
indicator variable that equals 1 if the analyst has ever attended an Ivy League school and 0 otherwise. Ever was star
is an indicator variable that equals 1 if the analyst has ever been elected as a star analyst and 0 otherwise. Star
analysts are identified from the October issues of the Institutional Investor magazine. Connect1 is an indicator
variable that equals 1 if the analyst covering a firm has attended the same university as one of the active senior
officer and directors of the firm and 0 otherwise. Connect2 is an indicator variable that equals 1 if the analyst
covering a firm has attended the same degree program in the same university as one of the active senior officers and
directors of the firm and 0 otherwise. Number of firms connected to is the sum of Connect1 among all firms
covered by an analyst. % of firms connected to is the number of connections an analyst has, defined above, divided
by the total number of stocks the analyst covers. No. of firms covered is the number of firms for which an analyst
provides earnings per share (EPS) forecasts. No. of industries covered is the number of industries, according to the
Fama French 48 industries classification, represented by the firms that the analyst covers. No. of earnings forecast
made per firm per year is the number of year-end EPS forecasts made by an analyst for a firm in a given year. No.
of recs issued per firm per year is the number of buy/sell stock recommendations that an analyst issues for a firm in
a given year. Years of experience measure the number of years an analyst appears in the I/B/E/S database. *, **,
*** denotes statistical significance of the difference at the 10%, 5%, and 1% levels based on two-tailed tests,
respectively.
Male Female t-stat
Ivy league 21.98% 26.97% 1.72 *
Ever was star 12.83% 14.11% 0.55
% of firms connected to - University (Connect1) 20.21% 23.86% 19.07 ***
% of firms connected to - Degree (Connect2) 11.49% 13.35% 12.23 ***
Number of firms connected to 1.89 1.96 0.39
Avg no. of firms coverred 9.79 8.58 -3.26 ***
Avg no. of industries 2.63 2.29 -3.41 ***
Avg no. of earnings forecasts made per firm per year 3.29 3.28 -0.09
Avg no. of recs issued per firm per year 1.36 1.35 0.50
Avg years of experience 4.73 4.32 4.21 ***
25Table 4. Forecast Error Statistics
This table reports summary statistics of the forecast errors of analysts’ earnings per share (EPS) forecasts. Raw
Forecast Error is the absolute difference between an analyst’s forecast of the company’s year-end EPS and the firm’s
actual reported year-end EPS, scaled by the prevailing stock price in the quarter before the release of the actual EPS.
Standardized Forecast Error is calculated using Equation (1). Specifically, it is the Raw Forecast Error minus the
minimum Raw Forecast Error among all forecasts issued by all analysts for the same firm in the same year, divided
by the difference between the maximum and minimum Raw Forecast Error among all forecasts issued by all analysts
for the same firm in the same year.
Summary statistics of forecast errors
mean sd min p25 p50 p75 max
Raw Forecast Error
Female 0.019 0.054 0.000 0.001 0.004 0.013 0.743
Male 0.020 0.058 0.000 0.001 0.004 0.014 0.747
All 0.020 0.058 0.000 0.001 0.004 0.014 0.747
Standardized Forecast Error
Male 0.389 0.337 0.000 0.083 0.304 0.657 1.000
Female 0.402 0.338 0.000 0.093 0.327 0.676 1.000
All 0.391 0.337 0.000 0.084 0.307 0.659 1.000
26Table 5. Univariate Comparisons of Standardized Forecast Error
This table compares Standardized Forecast Errors between male and female analysts, and between connected and
non-connected analysts. Only analysts who are connected to at least one stock they cover are included in this test.
Standardized Forecast Error is calculated using Equation (1). A forecast is made by a “Connected” analyst if at the
time of the forecast, the analyst shares a school tie—has attended the same university—with one of the active senior
officers and directors of the firm. *, **, *** denotes statistical significance of the difference at the 10%, 5%, and
1% levels based on two-tailed tests, respectively.
Panel A Overall gender and connection effect
Male Female t-stat(Diff) Connected Not connected t-stat(Diff)
1993 0.410 0.464 3.740 *** 0.3893 0.4201 2.9020 ***
1994 0.402 0.405 0.214 0.3816 0.4066 2.4616 **
1995 0.399 0.410 1.055 0.3948 0.4023 0.8316
1996 0.416 0.424 0.758 0.3966 0.4228 3.0539 ***
1997 0.406 0.418 1.255 0.3893 0.4124 2.8773 ***
1998 0.406 0.418 1.392 0.3964 0.4110 2.0697 **
1999 0.399 0.420 2.489 ** 0.3930 0.4053 1.8349 *
2000 0.399 0.411 1.416 0.3897 0.4053 2.4745 **
2001 0.383 0.395 1.641 0.3744 0.3891 2.7554 ***
2002 0.390 0.408 2.536 ** 0.3820 0.3957 2.6470 ***
2003 0.367 0.376 1.276 0.3612 0.3704 1.9344 *
2004 0.387 0.397 1.530 0.3781 0.3927 3.4669 ***
2005 0.383 0.377 -0.967 0.3705 0.3874 4.2615 ***
2006 0.381 0.405 4.035 *** 0.3698 0.3893 4.9976 ***
2007 0.398 0.411 2.247 ** 0.3941 0.4011 1.7310 *
2008 0.399 0.419 3.169 *** 0.3987 0.4018 0.7524
2009 0.340 0.358 2.725 *** 0.3316 0.3459 3.4540 ***
All years 0.387 0.402 8.282 *** 0.3777 0.3930 11.9215 ***
Panel B Effects of connection in gender subsamples
Male Female
Male Female t-stat(Diff) Connected Not connected t-stat(Diff)
1993 0.385 0.415 2.726 *** 0.470 0.434 0.972
1994 0.383 0.406 2.204 ** 0.412 0.369 1.213
1995 0.391 0.401 1.007 0.408 0.418 -0.397
1996 0.397 0.422 2.694 *** 0.432 0.395 1.501
1997 0.388 0.411 2.628 *** 0.425 0.398 1.328
1998 0.393 0.410 2.257 ** 0.418 0.417 0.078
1999 0.387 0.403 2.290 ** 0.419 0.424 -0.312
2000 0.385 0.405 2.876 *** 0.409 0.414 -0.276
2001 0.372 0.388 2.762 *** 0.399 0.388 0.759
2002 0.378 0.394 2.960 *** 0.408 0.409 -0.056
2003 0.362 0.369 1.263 0.388 0.354 2.402 **
2004 0.375 0.393 3.999 *** 0.394 0.403 -0.715
2005 0.370 0.388 4.418 *** 0.378 0.376 0.134
2006 0.367 0.387 4.866 *** 0.411 0.392 1.660 *
2007 0.390 0.401 2.592 *** 0.404 0.427 -1.888 *
2008 0.394 0.401 1.439 0.413 0.430 -1.340
2009 0.328 0.345 3.784 *** 0.357 0.359 -0.167
All years 0.375 0.392 12.507 *** 0.404 0.399 1.377
27You can also read
