Corporate Governance Indices and Firms' Market Values:
Time Series Evidence from Russia
BERNARD S. BLACK
University of Texas at Austin
INESSA LOVE
The World Bank
ANDREI RACHINSKY
New Economic School, Russia
first draft November 2005
this draft August 2006
(forthcoming, Emerging Markets Review, 2007)
European Corporate Governance Institute
Finance Working Paper No. xx/2005
University of Texas School of Law
Law and Economics Working Paper No. 66
University of Texas, McCombs School of Business
Working Paper No, FIN-05-05
This paper can be downloaded without charge from the
Social Science Research Network electronic library at:
http://ssrn.com/abstract=866988
1
Corporate Governance Indices and Firms' Market Values:
Time Series Evidence from Russia+
BERNARD S. BLACK*
University of Texas at Austin
INESSA LOVE**
The World Bank
ANDREI RACHINSKY***
New Economic School, Russia
Draft August 2006
ABSTRACT
There is increasing evidence that broad measures of firm-level corporate governance predict higher
share prices. However, almost all prior work relies on cross-sectional data. This work leaves open
the possibility that endogeneity or omitted firm-level variables explain the observed correlations. We
address the second possibility by offering time-series evidence from Russia for 1999-present,
exploiting a number of available governance indices. We find an economically important and
statistically strong correlation between governance and market value both in OLS and in fixed effects
regressions with firm-index fixed effects. We also find large differences in coefficients and
significance levels, including some sign reversals, between OLS and fixed effects specifications.
This suggests that cross-sectional results may be unreliable. We also find significant differences in
the predictive power of different indices, and in the components of these indices. How one measures
governance matters.
Key words: Russia, corporate governance, corporate governance index, law and finance, firm
valuation, disclosure, emerging markets
JEL classification: G32, G34
+
We thank the World Bank for financial support. We thank Woochan Kim, Kate Litvak, and Sheridan
Titman, and workshop and conference participants at McCombs School of Business, University of Texas at
Austin, and the Workshop on Financial Markets Development in Central and Eastern European Countries for
comments on earlier drafts. We also thank Edward Al-Hussainy, Rei Odawara and Feng Lui for excellent
research assistance.
*
Hayden W. Head Regents Chair for Faculty Excellence, University of Texas Law School, and
Professor of Finance, Red McCombs School of Business, University of Texas. Tel: (+1) 512-471-4632, fax:
(+1) 512-232-1767, e-mail: [email protected]
**
Senior Economist, The World Bank for Reconstruction and Development, 1818 H St., N.W., Washington
D.C., USA. Tel: (+1) 202-458-0590, fax: (+1) 202-458-0590. e-mail: [email protected]
***
Economist, Center for Economic and Financial Research at the New Economic School, Nakhimovsky
prospekt 47, Moscow 117418, Russia Tel: (+7) 495-105-50-02, fax: (+7) 495-105-50-03, e-mail:
[email protected]
2
1. Introduction
There is evidence that broad measures of firm-level corporate governance predict higher share
prices in emerging markets. This evidence comes from both single-country studies (Black,
2001 on Russia; Black, Jang and Kim, 2006 on Korea; Gompers, Ishii and Metrick, 2003 on the
U.S.) and multicountry studies (Durnev and Kim, 2005; Klapper and Love, 2004). However,
most prior work relies on cross-sectional data. This leaves open the possibility that endogeneity
or bias due to omitted firm-level variables explain the observed correlations. Here, we address
the omitted variable bias issue by offering time-series evidence from Russia for 1999-2005.
The importance of corporate governance to investors in Russian firms has spawned a number
of efforts to measure good governance.
The Brunswick Warburg investment bank rated
governance from 1999-2003; the Troika Dialog investment bank has done so since 2000,
Standard and Poor's has published overall "corporate governance" ratings since 2001 and
"disclosure" ratings since 2002; the Institute for Corporate Law and Governance (ICLG) rated
governance from 2000-2003, and the Russian Institute of Directors (RID) since 2004. Our
study exploits these measures. We present results for each measure and for an overall measure
that aggregates information from each.
We find an economically important and statistically strong relationship between governance
and firm market value (proxied by Tobin's q and, in robustness checks, market/book and
market/sales) in fixed effects regressions for a combined governance index (with firm-index
fixed effects) and for the Brunswick, Troika, S&P disclosure, and ICLG indices (with firm fixed
effects). This work strengthens the case for a causal association between governance and firm
market value, by ruling out some (though not all) of the non-causal explanations for this
association.
We find large differences between OLS and fixed effects specifications in the estimated
coefficients on our combined governance index and the individual indices, with some sign
reversals. For example, the Brunswick index is insignificant in OLS but positive and highly
significant with firm fixed effects. In contrast, the RID index is positive and highly significant
-1-
in OLS but insignificant and negative with firm fixed effects. This suggests that prior crosssectional results may be unreliable. We also find large differences in the predictive power of
different indices, and the components of these indices. The strength of particular indices, as
predictors of firm market value, typically comes primarily from a limited subset of the
components included in the index. How one measures governance matters.
For our combined governance index, we estimate that, with firm fixed effects, a two standard
deviation increase in governance predicts an 0.14 increase in ln(Tobin's q), or about a 21%
increase in share price (all share price change estimates are for a firm with median Tobin’s q and
median leverage). A worst-to-best improvement in governance predicts a 0.45 change in
ln(Tobin's q), or about an 81% increase in share price. OLS estimates for the combined index
are more than twice times as large as firm fixed effects estimates.
Estimated economic effects are larger than this for several individual indices. For the
Brunswick index, which has the largest coefficient with firm fixed effects, a two standard
deviation change in governance predicts an 0.34 increase in ln(Tobin's q) or about a 57%
increase in share price; a worst-to-best change predicts an 0.70 increase in ln(Tobin's q), or about
a 143% change in share price.
Share prices are the trading prices for minority shares.
Our study cannot show whether
higher share prices reflect higher value for all shareholders, lower private benefits for controlling
shareholders, or some of both.
Put differently, we cannot test whether our sample firms were
initially out-of-equilibrium and could increase firm value through governance changes, or in an
equilibrium in which firm value was already maximized and gains to outside shareholders come
at controlling shareholders' expense.
However, the voluntary governance improvements by
many firms during our study period suggest that the initial situation was out-of-equilibrium, and
that the gains to minority shareholders were only partly offset by reduced opportunities for selfdealing by insiders.
This paper is organized as follows. Section 2 reviews prior literature on the connection
between firm-level governance and firm value or performance. Section 3 describes our data
-2-
sources and how we construct our governance indices. Section 4 covers methodology. Section
5 presents our main results. Section 6 concludes.
2.
2.1.
Literature Review:
New Steps in This Paper
Governance Studies in General
This paper addresses whether firm-level variation in overall firm-level corporate
governance practices predicts firms' market values, in both cross-section and time series.
A
large literature studies the link between specific aspects of corporate governance (such as audit
committee, independent directors, and takeover defenses, and minority shareholder protections)
and firms' market value or performance.
A separate large literature explores the connection
between country-level rules affecting corporate governance and firm behavior and the strengths
of securities markets.
Morck, Wolfenzon and Yeung (2005) provide a recent review.
Work
on whether firm-level variation in overall corporate governance predicts firms' market value or
performance is more limited, and most of this work relies on cross-sectional results.
Several studies find a connection between a measure of firm-level governance and share
price in a single country. Related papers studying emerging markets include Black (2001)
(Russia); Black, Jang and Kim (2006) (Korea); Black, Kim, Jang and Park (2006) (Korea).
Another strand of this literature finds similar results on a cross-country basis (Durnev and Kim,
2005; Klapper and Love, 2004). The positive share price reaction to cross-listing (e.g., Doidge,
Karolyi and Stulz, 2004b) also suggests that governance can predict share price. At the same
time, we understand fairly little about why firms make the governance choices they do. There is
a large role for compliance with country norms (Doidge, Karolyi and Stulz, 2004a) and for
idiosyncratic choice (Black, Jang and Kim, 2006a).
We also know fairly little about the
channels through which governance affects value.
While we focus here on emerging markets, related U.S. work includes Gompers, Ishii and
Metrick (2003), Core, Guay and Rusticus (2005), Gillan, Hartzell and Starks (2003), and
Bebchuk, Cohen and Ferrell (2004).
-3-
Most studies of firm-level governance have two important limitations.
Except for
concurrent work in Korea by Black, Kim, Jang and Park (2005), all are limited to cross-sectional
analysis. Outside the U.S. panel data has not been available. In the U.S., governance changes
too slowly to make a firm fixed effects approach feasible. The cross-country studies also have
available only limited control variables.
These limitations raise the potential for omitted
variable bias, in which unobserved firm heterogeneity predicts both governance and market
value, leading to a spurious correlation between the two. We report evidence below that this
bias is likely to be important.
A second problem with existing studies is the potential for the link between a governance
index and share prices to reflect endogeneity, in which higher-valued firms choose better
governance, rather than the other way around.
Most studies, ours included, lack a good
instrument to address this issue (Black, Jang and Kim (2006) is an exception).
2.2. Prior Research on Russian Corporate Governance
There are two related studies of Russian corporate governance.
Black (2001) studies a
small sample of 21 firms in 1999, with very limited control variables, but reports a strong
correlation between the Brunswick index and the market value of Russian firms, as a percentage
of their theoretical market value if priced at Western multiples (as estimated by Troika Dialog).
He finds a correlation between ln(market value/theoretical value) and governance of r = 0.90,
and a worst to best governance change (from Gazprom and subsidiaries of Yukos at the low end
to Vimpelcom at the high end) predicts a factor of 700 change in market value.
This article extends this prior research on Russia in several dimensions.
As compared to
Black (2001), our sample includes 99 firms (instead of 21); our sample period spans 1999-2005
(instead of being limited to a single point in time in late 1999), we use six corporate governance
indices (instead of one), we have access to extensive control variables (to address omitted
variable bias), and we use a replicable measure of value (Tobin's q, market/book, and
market/sales) instead of a measure of value that depended in part on the judgment of the Troika
Dialog investment bank, and could not be readily replicated.
-4-
The stronger data let us explore whether the previously observed correlation between
governance and firm market value is similar in cross-section and in time series with fixed effects
(we find large differences), and whether different governance indices produce similar results (we
again find large differences).
The second prior Russia study is Goetzmann, Spiegel and Ukhov (2002). They offer
governance explanations for the price differences between Russian preferred and common shares,
and find evidence consistent with a sharp improvement in Russian corporate governance since
1999.
2.3. Why Study Russia?
Russia is an especially suitable laboratory for studying the effect of firm-level governance on
firm value. It combines a fair sized capital market, including many large, formerly state-owned
enterprises that were privatized during the 1990s, with notably bad governance at both firm and
country levels, especially in 1999, at the start of our sample period. Russian governance has
improved substantially since 1999. Many leading Russian companies have now raised both
equity and debt capital in international financial markets, prompting them to improve their
corporate governance. Likely not coincidentally, Russian share prices soared during this 19992005 period we study. While Russia’s real GDP grew at about 5% during this period, Russia’s
stock market grew at a compound annual rate of over 50% (see Figure 2).
Yet most Russian companies remain undervalued relative to their western competitors. For
example, Gazprom, the world's largest oil and gas company based on reserves, had a market
capitalization in October 2005 of only about $1 per barrel of proven reserves, compared to $18
for major Western oil companies such as Exxon Mobil and Royal Dutch Shell (Economist, 2005).
Part of this discount reflects domestic Russian energy price controls; part reflects political risk
(as Russia's de facto expropriation of Yukos reminds us); but much reflects firm-level
governance. Indeed, among Russian oil and gas companies, Black (2001) finds a strong crosssectional correlation in 1999 between governance and the market value of oil and gas companies
per barrel of reserves.
-5-
In developed countries, governance at most firms is quite stable over time. This has thus far
precluded time-series research on governance. The improvements in Russian governance over
the 1999-2005 period of our study provide enough time-variation to let us test the relationship
between corporate governance and market value in a fixed effects framework. In effect, our
research exploits the likely out-of-equilibrium nature of Russian corporate governance, due to
Russia's continuing transition to a market economy and its recovery from a 1998 financial crisis.
The fixed effects approach lets us address omitted variable bias arising from unobserved
firm-level heterogeneity that is time-invariant.
We also employ a reasonably extensive set of
control variables, which can address some potential sources of time-varying firm characteristics.
Our results thus rule out some (though not all) of the non-causal explanations for the observed
relationship between governance and firm market value.
A second advance of this study draws on the existence of multiple governance indices,
which have significant differences in emphasis (see Table 1), and cover an often overlapping set
of firms in the same country over roughly the same time period.
This lets us investigate
whether it matters how one measures governance, or whether one measure is pretty much as
good as another.
market values.
We find that the indices vary substantially in their ability to predict firm
The variability in predictive power extends to the subindices which make up
each index.
3.
3.1.
Data
Governance data
Russian market participants well understand the importance of governance in valuing
Russian firms.
responded.
They have demanded governance information, and a variety of sources have
There are six available corporate governance indices, from five different providers,
covering different portions of our sample period:
two major investment banks, one rating
agency (which produces two different products); and two nonprofit Russian organizations.
-6-
Table 1 summarizes the available corporate governance indices, the periods they cover, the
subindices they use, and the weights given to each subindex.
Brunswick UBS Warburg is a well known international investment company with strong
interest in Russia. Its research department was the first to measure corporate governance in
Russian companies, beginning in 1999 and continuing through 2003, after which Brunswick
apparently informally abandoned this effort.
Black (2001) used their initial 1999 scores, and
found that they strongly predict firms' market values.
Brunswick reports an overall governance
score on a 0 ~ 72 scale (higher numbers indicate worse governance), which is the sum of the
scores on 8 subindices, for transparency, share dilution risk, asset transfer and transfer pricing
risk, merger and restructuring risk, bankruptcy risk, ownership restrictions, corporate governance
initiatives, and registrar risk.
Brunswick scores are available for 25 firms for which we also
have the financial and other data needed to construct the dependent and other independent
variables we use in our regression analysis; we can include all firms in firm-fixed effects
specifications because we have two or more scores available for each firm.
Some of the Brunswick "governance" elements may seem odd from a Western perspective,
but make sense within Russia.
Consider bankruptcy risk, for example.
In Russia, a
bankruptcy filing, often by a fully solvent company, is a common means through which
controlling shareholders squeeze out minority shareholders for minimal consideration; it is also a
favored means for a hostile takeover, in which an outsider uses the bankruptcy process to
squeeze out the former controlling owners.
restructuring for similar purposes.
A controlling shareholder can use a merger or
Registrar risk involves the risk that the share registrar
(whose records are the only official proof of ownership) will lose or freeze ownership if so
requested by someone with influence (a controlling shareholder or an outsider seeking to acquire
control).
Transfer pricing is a common means for siphoning profits out of a public company
into an offshore affiliate that is wholly owned by the firm's controlling shareholders.
Share
dilution through a large private offering of shares to the controlling shareholder or its (often
undisclosed) affiliates at a fraction of true value was a major risk prior to 2002.
-7-
In 2002,
Russian company law was amended to require preemptive rights during an offering of common
shares or securities convertible into common shares.
Troika Dialog is Russia’s largest and oldest investment bank.
Its research department
began to measure corporate governance in 2000 and has continued to do so since, roughly
annually.
It reports scores on five measures:
oversight and control structure; management
and investor relations; corporate conduct; and information disclosure and financial discipline,
and ownership structure and transparency.
overall governance score, on a 5 ~ 15 scale.1
We weight these subindices equally to produce an
Troika reports cover 76 firms for which we have
data on control variables; 65 have two or more reports and thus are available with firm fixed
effects.
The Institute of Corporate Law and Governance (ICLG) is a nonprofit institute, launched in
2000 to develop an investor protection system and upgrade the corporate governance culture in
Russia.
Its principal founder, Dmitri Vasiliev, was the first Chairman of the Russian Federal
Commission for the Securities Market and was known for his active support of investor rights.
The ICLG corporate governance assessment includes components for information disclosure,
ownership structure, board of directors and management structure, shareholder rights,
expropriation risk, and corporate governance history.
The score is understood to draw in part
on the OECD principles of Corporate Governance (OECD, 2004).
ICLG does not disclose its
methodology, nor the weights assigned to each component. It produced quarterly governance
scores, on an 100 point scale, from 2000 through 2004, but then apparently ran out of funding.
ICLG reports cover 36 firms for which we have data on control variables; 34 firms are available
with firm fixed effects.
Standard and Poor’s (S&P) is a leading international rating agency, which provides a variety
of credit and other ratings.
In some countries, including Russia, it provides "transparency and
1
Each subranking runs from 1-3 (often reported as below average, average, or above average). We sum the
five subrankings to obtain an overall ranking, which runs from 5 to 15. Each Troika report also contains an "overall
ranking" on a 1-3 scale. We do not rely on this overall ranking, which appears to be a rough average of the
individual rankings.
-8-
disclosure" scores for major firms based on public information, plus governance scores for
individual firms which pay S&P to rate their overall governance.
S&P publishes its disclosure
scores annually since year 2002; these scores are based on 89 potential disclosures, each gets one
point on an overall 89 point scale.
S&P disclosure scores are available for 52 firms for which
we have data on control variables, of which 51 remain available with firm fixed effects.
In contrast, the S&P Governance scores are published irregularly, and become publicly
available only if a company hire S&P to prepare them and then chooses to reveal its score.
These scores are on a 1 ~ 10 scale, with pluses and minuses.
They are available for only 13
firms, all of which remain available with firm fixed effects.
The Russian Institute of Directors (RID) is a nonprofit entity founded in 2001 by a group of
major Russian companies with the goal of improving Russian corporate governance and making
investments in Russian companies more attractive.
In 2004 RID produced its first corporate
governance scores, prepared in cooperation with the Russian financial rating agency Expert.
The scores run from 0 (SD) to 10 (A+++).
They are based partly on public information and
partly on a survey of Russian companies, for those firms which responded to the RID survey.
Overall score is determined by number of indicators combined into 4 groups: shareholders rights,
government and control bodies, information disclosure, stakeholders’ interest and corporate
social responsibility. Weights of each category in overall score and values of subindices are not
disclosed.
The RID index covers 116 companies for which we have data on control variables,
substantially more than the other indices.
For 78 of these companies, its governance score is
based solely on public information; for the remaining 38 firms, the score is also based on survey
responses.
In effect, RID provides two different scores, computed differently for responding
firms and non-responding firms.
For the responding firms, the RID score predicts Tobin's q in
cross-sectional, pooled OLS regressions, but not with firm fixed effects.
For non-responding
firms, the RID score does not predict Tobin's q even in cross-section. Below, we report results
only for the 38 responding firms; of these, 29 are available with firm fixed effects.
-9-
In
robustness checks, we obtain similar results for the combined index if we include all RID
observations, or none.
Overall, we have 964 observations of 99 firms for cross-sectional regressions.
In fixed
effects regressions with firm-index clusters, we have 907 observations of 76 firms with two or
more observations of the same firm and index.
Table 2 shows availability and average
governance scores for each index by quarter.
For fixed effects and random effects
specifications, we are limited to an unbalanced panel, because we do not have quarter-by-quarter
data for individual firms spanning the entire sample period.
Governance scores are produced with a lag between data collection and publication.
specifies the date as of which the scores were compiled; other ratings sources do not.
ICLG
To match
the effective dates of other scores with financial and share price data, we assumed that a ranking
published in the first half of quarter t relates back to quarter t-1.
Thus, we treat a report issued
on Feb. 1, 2003 (in the first half of the 1st quarter of 2003) as relating back to the fourth quarter
of 2002, but treat a report issued on March 1, 2003 (in the second half of this quarter) as relating
to the current quarter (1st quarter of 2003), and so on.
frequencies:
Indices are produced with different
quarterly, semi-annual or annual.
We construct our panel data with quarterly
Each governance index uses a separate scale.
In most, better governance leads to a higher
frequency.
score; for Brunswick, better governance produces a lower score.
To make the indices
comparable, we convert them to a standard normal distribution (mean 0 and standard deviation
of 1), with higher scores indicating better governance.
indices cover different companies at different times.
Note, however, that in part, different
Thus, a company with an average score
on one index might have an above (or below) average score on another index because of
differences in sample composition.
We use these standardized governance scores to construct a "pooled" governance index
which contains all observations of each index, for each quarter for which a governance score is
available.
In some cases, we have governance scores for the same company in the same
- 10 -
quarter for more than one index.
In regressions using the combined index, we treat these as
separate observations (in which only governance score differs but all other variables are the
same) and we use dummy variables for each index.
In robustness checks, we obtain similar
results with a "quarter averaged" index in which we average the available scores for each
company in each quarter.
Figure 1 plots the mean score for each index by quarter.
Individual indices are produced at
irregular intervals and may cover different firms at different dates.
To generate continuous
lines in this figure we used linear interpolation to generate governance scores for firm-quarters
without a score which fall in between two quarters with a score for that same firm on the same
index.2 The Brunswick and S&P Disclosure scores increase significantly over time (based on
Cuzick's nonparametric test for trend (Cuzick, 1985, implemented as "nptrend" in Stata); scores
on the other indices do not show a significant trend.
In Figure 2 we plot the pooled governance index over time.
To address the problem that
different indexes are produced in different quarters and cover somewhat different samples of
firms, we begin with interpolated standardized scores for each index, as described above.
We
then compute the average change in governance score compared to the previous quarter.
The
base period is the third quarter of 1999, which is the first quarter for which we have any
governance scores.
composition.
This approach lets us control for the variance over time in sample
As Figure 2 shows, the average score on the combined index increases over time,
roughly in parallel with the Russian Trading System (RTS) index (the principal Russian stock
index).
Anecdotal evidence suggests that some of the dramatic increase in the RTS index over
our study period likely reflects improvements in firm governance.
3.2.
Financial and Stock Market Data
We supplement our governance data with financial, industry, and stock market data.
We
use end-of-quarter data for information taken from balance sheets and income statements, and
2
Thus, if a firm was ranked on a particular index in quarter t with score Rt and was next ranked in quarter t+k
with score Rt+k, we assign the firm a score in quarter t+i (0 < i < k) using linear interpolation: Rt+i= [(k-i)*Rt +
i*Rt+k]/k.
- 11 -
quarterly average data for market values (many Russian firms' shares do not trade daily, and the
trade(s) closest to quarter end may not reflect market values).
We compute the quarterly average market value of equity for each firm using RTS trade data,
from www.rts.ru). If the number of outstanding shares is constant during the quarter, we multiply
the trade-weighted quarterly average price by the number of outstanding shares at the end of the
quarter to obtain market value of equity.3
If the number of outstanding shares changes during
the quarter, we multiply the daily average price by the number of shares outstanding on that day
to obtain a daily market value of equity, and sum across days to get a trade-weighted quarterly
average.
If a firm does not trade in a particular quarter (17 observations), we interpolate
market value of equity based on the prior and subsequent quarters.
We obtain financial data
from System of Complex Revelation of Information (SCRIN), at www.scrin.ru. SCRIN contains
firm’s quarterly income statements and balance sheets in Russian accounting standards.
Of the
1148 firm-quarters with governance scores available, we exclude 184 due to lack of other data,
producing the final sample of 964 observations.
We classify firms into industry sectors based on International Standard Industrial
Classification (ISIC) codes. Most of our observations come from the communication (256
observations), utilities (256 observations) and extraction (225 observations) sectors; these three
sectors represent 77% of our sample. Utilities include large energy producers and regional energy
companies. Most communication firms are regional traditional telephone companies, the rest are
mobile telephone companies. The extraction sector is dominated by oil and gas firms, which
account for 211 of the 225 observations in this sector.
We group the remaining firms (219
observations) into an "other" sector, of these 106 are in transportation or transportation
equipment.
Our main dependent variable is Tobin’s q, defined as (Market value of assets)/(Book value of
assets). Market value of assets is estimated as [market value of common stock + market value of
3
RTS reports trading volume in both dollars and shares. We use this information to determine the trading
price per share in dollars, which we convert to rubles based on end-of-quarter exchange rates.
- 12 -
preferred stock + book value of debt].
In alternative specifications we also use market/sales
ratio (market value of equity/sales) and market/book ratio (market value of equity/book value of
equity).
Most Russian firms which have preferred shares have trading in both common and
preferred shares, where the preferred shares are similar to low-voting common shares
(Goetzmann, Spiegel and Ukhov, 2002).
We have no information on non-traded preferred
shares, so if any firms have issued these shares, we will underestimate their Tobin's q.
However, the underestimate should be roughly constant over time and thus should have little
effect on results with firm fixed effects.
To reduce the influence of outliers, we take logs of these variables, and also winsorize at the
1% and 99% levels.
In robustness checks, we obtain similar results without winsorizing, and
similar but somewhat weaker results if we neither winsorize nor take logs (especially for
market/book, which contains some very high values due to low book value of equity).
We use the following control variables in all regressions, in an effort to control, subject to
data availability, for the principal factors which could predict Tobin's q and might also correlate
with governance (compare Black, Jang and Kim, 2006; Durnev and Kim, 2005; and Klapper and
Love, 2004).
•
Variables are measured at quarter-end except as indicated.
Ln(RTS):
movements.
ln(RTS index value).
This variable controls for general share price
If overall price movements, in part, reflect general improvements in
governance, this will bias against our finding an association between governance and
Tobin's q (or other dependent variables).
•
Ln(assets):
ln(book value of assets).
governance and Tobin's q.
Firm size is potentially correlated with both
However, the actual correlation between governance and
ln(assets) is only 0.05.
•
Liquidity:
(quarterly trading volume of common shares in rubles)/(market value of
common shares).
If both common and preferred shares trade, liquidity = (quarterly
trading volume of common and preferred shares in rubles)/(market value of equity).
In robustness checks, we obtain similar results for our governance indices with an
- 13 -
alternate measure of liquidity, ln(number of trades of common and preferred shares
during the quarter).
•
Leverage:
•
Annual (i.e. year-over-year) real sales growth, as a proxy for growth prospects.
book value of debt/(book value of assets), winsorized at 1% and 99%.
To
preserve sample size, we set sales growth equal to zero for 125 observations (47 distinct
firms) with missing sales growth data, and include a "missing growth" dummy variable,
which equals one for these observations.
The coefficient on the dummy variable will
capture the average effect of sales growth for firms with missing sales growth data.
•
Net income/assets:
net income/book value of assets, winsorized at 1% and 99%, as a
measure of profitability.
To preserve sample size, we set net income/assets equal to
zero for 25 observations with missing income data, and include a "missing net income"
dummy variable, which equals 1 for these observations.4
We use net income/assets,
rather than the potentially preferred measure of operating income/assets, because we have
net income for almost all firm-quarters in our sample, but operating income for only 83%
of firm-quarters.
•
MSCI Index dummy:
this dummy is equal to one if the firm is included in the Morgan
Stanley Capital International Index in July 2005 (we did not have access to historical data
on this index).
We include this variable because these firms may attract more
international investor interest, and therefore trade at higher prices.
We lose this variable
with firm fixed effects because it is not time-varying.
•
Year-sector dummies:
a family of dummy variables, each equal to one for a particular
year and industry sector, using the four sectors discussed above. These dummies capture
sector-specific time trends.
•
Index dummy variables Dj one for each governance measure other than RID (indexed by
j), set = 1 for observations involving that index.
4
We use index dummy variables in OLS
Vimpelcom, which was traded on the New York Stock Exchange from 1999, but actively traded on RTS only
from 2005, represents 16 firm-quarters with missing sales growth and 9 firm-quarters with missing net income data.
- 14 -
and random effects regressions involving the combined index; they are not available in
fixed effects regressions.
In robustness checks, we also control for capital intensity, measured as ln(fixed assets/sales)
which may be associated with both governance and performance (Klapper and Love, 2004),
with similar results. We omit this variable in our main specification because we have fewer
observations for it than for our other control variables.
Table 3 defines our dependent and control variables.
statistics for our principal variables.
governance indices.
size.
Table 4, Panel A presents summary
Table 4, Panel B presents correlation coefficients for our
We use interpolated governance scores in this table to preserve sample
Most pairwise correlations are positive and significant.
Panel C presents correlation
coefficients for our principal variables.
4. Empirical Methodology
To study the effect of corporate governance on firm valuation and performance we
estimate the following model for the combined index:
Yi , t = α + β * Govi , j , t + γ * Xi, t +
∑
(λ j * Dj ) + ε i , j , t
indices j
Here i indexes firms, t indexes time by quarter, j indexes governance measures, Yit is a
market value measure, Govijt is the pooled governance index, Xit is a vector of control variables
described above, Dj are dummy variables for individual indices (with RID as the omitted index),
ei,j,t is the error term, and α, β, γ, and λ are regression coefficients or coefficient vectors. The
control variables include year-sector dummies, which allow for an unspecified time variation in
firm market value, which can vary by sector. We use ln(Tobin's q) as our main measure of
market value. We estimate this model separately for each of the six governance indices and for
the combined index. We omit the index dummies in regressions using a particular index j
(rather than the combined index).
Because our ranking data comes from six different sources, we often have more than one
ranking available for the same firm in the same quarter. To account for the possibility that the
- 15 -
same firm may be placed differently in rankings from different sources, in regressions with
combined indices, we allow the error term to have a firm-index specific component. Thus, our
composite error is given by:
ε i, j, t =
∑ν
i, j
+ ui , j , t
indices j
where vi,j is the unobserved firm-index specific effect and ui,j,t is the idiosyncratic error term. In
regressions with individual governance indices, this specification reduces to firm fixed effects.
We estimate this model using several different assumptions about the composite error
term. The OLS estimate assumes that there is no correlation between Govi,j,t or Xi,t and the
composite error term ei,j,t, and is inconsistent if this assumption is violated. Even if this
assumption holds, the composite errors for each firm will be correlated, due to the presence of vi,j
in each time period. We therefore use standard errors estimated with firm clusters for regressions
involving a single index, and with firm-index clusters for regressions involving the combined
index. This allows for an unspecified correlation of the errors for each firm, within each index.
The Breusch-Pagan Lagrangian multiplier test rejects the assumption that the variance of
vi,j is equal to zero, suggesting that firm-index specific effects are important and therefore OLS is
inconsistent.
The random effects model provides efficient estimates under the restrictive assumption
that the unobserved firm-index specific effect is uncorrelated with Govi,j,t or Xi,t. However, it is
likely that the firm-index specific effects vi,j are correlated with governance or the X’s, and
therefore the OLS and random effects estimates will be biased. The fixed effects estimator is a
consistent estimator under this assumption.
A Hausman test performed on a simplified
specification (replacing industry-year dummies with industry dummies) rejects random effects in
favor of fixed effects for the combined index and most individual indices. Therefore, we use
fixed effects estimators as our main estimators and report OLS results for comparison with the
prior literature.
- 16 -
5. Principal Results
5.1.
Results for Pooled Governance Index
Figure 3 present a simple scatter plot of ln(Tobin's q) versus our pooled governance index,
with a fitted regression line from a simple regression of ln(Tobin's q) on constant term and
pooled governance index.
There is a significant positive coefficient on the combined index:
Ln(Tobin's q) = 0.050 (t = 0.83) + 0.171 (t = 3.26) * (Combined index)
Table 5 reports our main results for the combined index.5
The index is highly
significant in all three specifications -- OLS with firm-index clusters, firm-index random effects,
and firm-index fixed effects.
This is our first main result -- the connection between governance
and firm market value survives with random effects and fixed effects specifications.
We place
principal weight on the fixed effects specification, which is the preferred specification for our
sample.
As noted above, prior studies of the connection between firm-level governance and
market value, with the exception of Black, Kim, Jang and Park (2006) (below, BKJR), rely on
OLS, which is likely to be misspecified (as discussed in the previous section).
The fixed effects results rule out some of the potential non-causal explanations for a
correlation between governance and firm value.
explanations are:
The principal remaining non-causal
(i) a reverse causal relationship, in which firm value predicts governance;
and (ii) unobserved, time-varying firm-level heterogeneity, where the unobserved sources of
variation predict both governance and firm value.
We also find evidence that OLS coefficients can be biased.
The coefficient on the
combined index drops substantially as we move from OLS (coefficient = 0.150) to random
effects (coeff. = 0.094) to firm fixed effects (coeff. = 0.067).
fixed effects coefficient is .150/.067 = 2.24.
The ratio of the OLS to the firm
The fixed effects coefficient is well outside the
95% confidence interval derived from OLS, of [.103, .197]; similarly, the OLS coefficient is well
outside the 95% confidence interval derived with fixed effects of [0.038, 0.096].
5
This suggests
In Table 5, we suppress the coefficients on index dummy variables, missing variable dummies, and yearsector dummies. The index dummies and missing variable dummies are insignificant with firm-index fixed effects.
- 17 -
that the fixed effect is capturing firm-level heterogeneity, which is not captured by our control
variables and therefore our OLS model is misspecified. Compare BKJR, who study the power of
a governance index to predict ln(Tobin's q) for Korean firms, and find a 1.56 ratio between the
OLS and firm fixed effects coefficients on their governance index.
result:
This is our second main
in a "governance to value" study of the ability of a governance index to predict firm
market values, OLS results are likely to be unreliable.
In robustness checks, we obtain similar results with a “quarter-averaged” aggregate index,
in which we average all rankings available for each firm in the same quarter, instead of treating
different observations of the same firm (using different indices) as separate observations.
With firm-index fixed effects, a two standard deviation increase in the combined index
predicts an 0.14 increase in ln(Tobin’s q), or about a 21% increase in share price for a firm with
median Tobin’s q (1.05) and median leverage (debt/assets = 0.31). A worst-to-best improvement
in the combined index predicts an 0.45 change in ln(Tobin's q), or about an 81% change in share
price.
These results are economically important, but fall well short of Black's (2001) prior study,
in which a worst-to-best change in governance, measured by the 1999 version of the Brunswick
index, predicted a factor of 700 increase in market capitalization, as a fraction of hypothetical
Western capitalization.
This difference deserves explanation.
In hindsight, 1999 was a time
when Russian corporate governance was at a low point -- with large variation between firms and
a large fraction of firm value at stake through a firm's governance.
The gap between market
capitalization and theoretical Western market capitalization has substantially narrowed since then
for most Russian firms, with the worst governed firms showing the largest gains.
For example,
Gazprom has gone from being valued at 0.002 of Western value (2.6 cents per barrel of
reserves!) to being valued today at about 0.05 of Western value ($1 per barrel of oil-equivalent
reserves).
LukOil has gone from 0.028 of Western value ($0.36 per barrel of reserves) to 0.13
of Western value ($2.50 per barrel of reserves).
- 18 -
The combined index shows an improvement over time in governance (see Figure 2), but
not nearly this dramatic.
do not capture.
Apparently, there are important aspects of governance that the indices
Also, most of our data points come from later years (only 98 of 964
observations are from 1999 and 2000), when governance may have been less salient.
5.2
Results for Control Variables
Many of the control variables are significant in predicting Tobin's q.
results briefly, focusing on fixed effects results.
We note differences from BKJR, which is the
only comparison paper which uses firm fixed effects.
as one might expect.
We discuss these
The RTS index is significantly positive,
Ln(assets) is negative and highly significant effects, suggesting that firms
suffer declining Tobin's q as they grow.
ln(sales) as a measure of firm size.
small and insignificant.
In robustness checks, we obtain similar results with
We expected liquidity to take a positive coefficient, but it is
An alternate measure of liquidity, ln(number of trades), is significant
and positive, but is correlated with firm size and may not be a good measure of liquidity.
The coefficient on leverage is strongly positive.
In robustness checks, we obtain similar
results with alternate definitions of leverage (book value of debt/market value of assets, and book
value of debt/market value of equity).
This result is somewhat puzzling.
Whatever the
reasons for an OLS correlation between Tobin's q and leverage, it is not apparent why higher
leverage predicts higher Tobin's q with firm fixed effects.
Perhaps an increase in leverage is
connected to firm value because it suggests that the firm is “good enough” (or well-connected) to
be able to obtain debt financing, or that the firm has growth prospects sufficient to justify
borrowing, which are not capture by our sales growth variable.
As expected, sales growth and profitability (measured by net income/assets) are both
significantly positively related to valuation.
In robustness checks, we obtain similar results if
we omit sales growth and profitability altogether, and with alternate definitions of sales growth
(based on quarterly rather than annual growth) and profitability (operating income/assets and
operating income/sales).
- 19 -
The MSCI index is unavailable with fixed effects, but is significant and positive with
random effects.
Firms in the MSCI index may be more visible and enjoy better analyst
coverage.
5.3
Results for Individual Governance Indices and Alternative Measures of Firm Value
Table 6 reports our results for individual indices.
Each table cell reports the coefficient
on a particular index in a regression similar to Table 5; coefficients on control variables are
suppressed.
variable.
Consider first the firm fixed effects column with ln(Tobin's q) as dependent
There are large differences between the indices.
Brunswick to -0.041 for S&P Governance.
Coefficients range from 0.171 for
Brunswick, Troika, S&P Disclosure, and ICLG are
positive and significant; while S&P Governance and RID have insignificant negative coefficients.
This is our third main result:
How one measures governance matters.
Some indices
have significant ability to predict the market values of Russian firms; others do not.
Recall too
that the RID measure, for firms for which it was based solely on public information, had no
predictive value in either OLS or firm fixed effects; we therefore excluded these firms from this
study.
More sophisticated indices do not necessarily perform better.
In particular, the S&P
Disclosure index is a "useful" measure of governance, in that it predicts Tobin's q, and therefore
may correspond to the elements of governance that investors care about.
It outperforms, in this
sense, the more complex S&P Governance index produced by the same rating agency.
The results for individual indices also reinforce our second main result -- the importance
of controlling for firm fixed effects and the potential unreliability of OLS estimates of the
relationship between a governance index and firm market value.
The Troika, S&P Disclosure,
and ICLG indices are positive and significant in both OLS and firm fixed effects specifications,
but there are large differences in coefficients for S&P Disclosure and ICLG..
The RID index is
significant and positive in OLS, and has the largest OLS coefficient of any of our indices, but
becomes insignificant and negative with firm fixed effects.
In contrast, the Brunswick index is
insignificant in OLS, but with firm fixed effects it becomes significant and has the largest
- 20 -
coefficient of any index.
The S&P Governance index is insignificant in both OLS and fixed
effects specifications.6
Table 6 also presents robustness checks with two alternative measures of firm market
value -- market/book and market/sales.
Results for market/book are similar to those for Tobin's
Results for market/sales are similar for the combined index, Brunswick, and Troika, but
q.
weaker for S&P Disclosure and ICLG, which lose significance with firm fixed effects.
5.4.
Results for Subindices
In Table 7, we examine the subindices of the Brunswick, Troika, and S&P Disclosure
indices.
These are the indices which are significantly related to governance in Table 6, and for
which we have information on subindex scores.
See Table 1 for details on subindices.
Each
column of Table 7 reports the coefficients on the subindices of a single index, from a regression
similar to those reported in Table 6, except (i) we replace the index with separate independent
variables for each subindex; and (ii) we use non-normalized subindex scores (with Brunswick
scores reversed so higher scores imply better governance).
We focus here on firm-fixed effects results.
comparison.
Table 7 also reports OLS results for
The differences between OLS and firm fixed effects coefficients for subindices
again shows combined indexthat there are often large differences between OLS and firm fixed
effects coefficients.
For Troika, for example the "information disclosure and financial
discipline" subindex is positive and highly significant in OLS, but negative and insignificant
with firm fixed effects.
For Brunswick, several subindices are significantly related to market value, while a
couple take negative coefficients. For Troika, the coefficients are positive for all subindices; the
power of the index is spread out over a number of subindices, with positive coefficients and tstatistics above 1 on all but "information disclosure and financial discipline", and significant
6
Some of the weakness of S&P Governance could reflect small sample size – only 46 observations of 13 firms. Some
could reflect the ability of firms to obtain ratings first, and then decide whether the ratings should be made public. Some of the
weakness of RID with fixed effects could reflect small sample size and the short period for which the index is available – with
fixed effects, we have only 60 observations of 29 firms.
- 21 -
coefficients on "oversight and control structure" and "management and investor relations."
For
S&P Disclosure, in contrast, the power of the index comes from the "financial and operational
information" subindex; the other two subindices are insignificant. The differences between
subindices reinforce our third main result -- it matters how one measures governance.
6. Conclusion
Most studies of the connection between firm-level corporate governance and share prices
are limited to cross-sectional data and simple OLS specifications. We study the connection
between the firm-level governance of Russian firms and their market values over 1999-2004, a
period of dramatic change in Russian corporate governance, using OLS, firm random effects, and
firm fixed effects specifications.
The importance of corporate governance to investors in Russian firms has spawned no
less than six different indices, of which four remain active today. We draw on all six here. A
combined index is economically and statistically strong across all specifications: pooled OLS
with firm-index clusters; firm-index random effects, and firm-index fixed effects; and is robust to
choice of firm-value variable (Tobin's q, market/book, or market/sales). This work strengthens
the case for a causal association between firm-level governance and firm market value, by ruling
out some (though not all) of the non-causal explanations for this association.
We report three principal new findings: (1) At least in Russia, governance predicts firm
value in a firm fixed effects framework. (2) There are large differences between OLS and fixed
effects results. This casts doubt on OLS results, and thus on most prior work on the connection
between governance and firm value (including some of our own work). (3) How one measures
governance matters.
The Brunswick, Troika, S&P Disclosure, and ICLG indices have
significant predictive power with firm fixed effects; the S&P Governance and RID indices do not.
- 22 -
References
Bebchuk, Lucian Arye, Alma Cohen and Allen Ferrell (2004), What Matters in Corporate Governance?, working
paper , at http://ssrn.com/abstract=593423.
Black, Bernard (2001), “The Corporate Governance Behavior and Market Value of Russian Firms,” Emerging
Markets Review, Vol. 2, pp. 89-108.
Black, Bernard S., Hasung Jang, and Woochan Kim (2006), “Does Corporate Governance Affect Firms’ Market
Values? Evidence from Korea,” Journal of Law, Economics, & Organization, forthcoming, working paper at
http://ssrn.com/abstract=311275
Black, Bernard S., Hasung Jang, and Woochan Kim (2006a), “Predicting Firms’ Corporate Governance Choices:
Evidence from Korea,” Journal of Corporate Finance Vol. 12, pp. 660-691.
Black, Bernard S., Woochan Kim, Hasung Jang and Kyung-Suh Park (2006), "Does Corporate Governance Predict
Firms' Market Values: Time-Series Evidence from Korea", available at http://ssrn.com/abstract=844744.
Core, John E., Wayne R. Guay and Tjomme O. Rusticus (2005), Does Weak Governance Cause Weak Stock
Returns? An Examination of Firm Operating Performance and Investors' Expectations, working paper,
http://ssrn.com/abstract=533582
Cuzick, Jack (1985), "A Wilcoxon-Type Test for Trend", Statistics in Medicine 4: 87-90.
Doidge, C., A. Karolyi, and R. Stulz (2004a), Why Do Countries Matter So Much for Corporate Governance,
Working Paper, Ohio State University.
Doidge, Craig, G. Andrew Karolyi, and Rene Stulz (2004b), “Why Are Foreign Firms Listed in the U.S. Worth
More?” Journal of Financial Economics, Vol.71, pp.205-38
Durnev, Artyom, and E. Han Kim (2005), “To Steal or Not to Steal: Firm Attributes, Legal Environment, and
Valuation,” Journal of Finance Vol. 60, pp. 1461-1493.
Economist (2005), Gazprom, Russia’s energetic enigma, October 6, 2005.
Gillan, Stuart L., Jay C. Hartzell and Laura T. Starks (2003), "Explaining Corporate Governance: Boards, Bylaws,
and Charter Provisions", working paper., at http://ssrn.com/abstract=442740
Goetzmann, William N., Matthew Spiegel and Andrey Ukhov (2002), "Modeling and Measuring Russian Corporate
Governance:
The Case of Russian Preferred and Common Shares", working paper,
http:/ssrn.com/abstract=305494
Gompers, Paul, Joy Ishii, and Andrew Metrick (2003), “Corporate Governance and Equity Prices,” Quarterly
Journal of Economics, Vol.118, pp.107-55
Klapper, Leora F. and Inessa Love (2004), “Corporate Governance, Investor Protection, and Performance in
Emerging Markets,” Journal of Corporate Finance Vol.10, pp.287-322.
Morck, Randall, Daniel Wolfenzon and Bernard Yeung (2005), "Corporate Governance, Economic Entrenchment,
and Growth", Journal of Economic Literature Vol. 43, 655-720 (2005)
OECD (2004), Principles of Corporate Governance.
Wooldridge, Jeffrey (2001), Econometric Analysis of Cross-Section and Panel Data, The MIT Press.
- 23 -
Table 1: Corporate governance indices and subindices
Each governance index, time period covered, scale used (higher score implies better governance for all indices
except Brunswick), subindices, and weights on each subindex. We convert each index to standard normal form
(mean = 0, σ =1), with higher scores indicating better governance. We combine all 6 standardized indices together to
get the combined index. Brunswick changed its methodology after 1999. We recalculated the 1999 scores using its
post-1999 methodology. Troika provides five subindices but no overall score; we assigned equal weight to each
subindex.
Corporate governance rankings and their categories
Subindex weights
Brunswick UBS Warburg (Brunswick)
3rd qtr 1999 – 3rd qtr 2003
Transparency
Scale = 0 (best to 72 (worst)
Dilution
Asset transfers / transfer pricing
Mergers / restructuring
Bankruptcy
Ownership restrictions
Corporate governance initiatives
Registrar
Troika Dialog (Troika)
3rd qtr 2000 – 4th qtr 2005
Ownership structure and transparency
Scale = 5 to 15
Oversight and control structure
Management and investor relations
Corporate conduct
Information disclosure and financial discipline
S&P Corporate Governance (S&P Governance)
1st qtr 2001 -- 4th qtr 2005
Ownership structure and influence
Scale = 1 to 10
Financial stakeholder rights and relations
Financial transparency and information disclosure
Board structure and process
S&P Transparency and Disclosure (S&P Disclosure)
3rd qtr 2002 – 3rd qtr 2005
Ownership structure and investor relations
Scale = 0 to 89
Financial and operational information
Board and management structure and process
Institute of Corporate Law and Governance (ICLG)
4th qtr 2000 – 4th qtr 2003
Information disclosure
Scale = 0 to 100
Ownership structure
Board of directors and management structure
Shareholder rights
Expropriation risk
Corporate governance history
Russian Institute of Directors / Expert (RID)
2nd qtr 2004 – 3rd qtr 2005
Shareholders rights
Scale: 1 to 10 (SD to A+++)
Government and control bodies
Information disclosure
Stakeholders’ interest and corporate social responsibility
Combined index. Union of all six indices. If two or more indices cover the same firm in
the same quarter, we treat these as separate observations.
- 24 -
19.4%
18.1%
13.9%
13.9%
16.7%
4.2%
12.5%
1.4%
20%
20%
20%
20%
20%
25.0%
25.0%
25.0%
25.0%
33.3%
33.3%
33.3%
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
n = 964
Table 2. Corporate governance rankings and average standardized corporate governance scores by quarter
Number of firms and average score in each quarter for the Brunswick, Troika, S&P Governance, S&P Disclosure, ICLG and RID governance indices.
Indices are defined in Table 1. Each index is converted to standard normal form (mean = 0; σ = 1). For each index we report the number of firms covered
and mean normalized score for each quarter. The last column provides the sum of all observations for all indices for each quarter. The last two row provide
(i) the total number of observations (firms) for OLS and (ii) the effective number with firm fixed effects (excluding firms with only one observation).
year
quarter
3
1999
4
1
2
2000
3
4
1
2
2001
3
4
1
2
2002
3
4
1
2
2003
3
4
1
2
2004
3
4
1
2
2005
3
4
Observ. (firms)
Fixed effects obs.
(firms)
Brunswick
obs. mean
11
-0.67
13
-0.28
13
-0.23
14
17
18
23
Troika
S&P Governance
obs. mean
obs.
mean
obs.
ICLG
mean
RID
obs.
mean
all indices
sum of obs.
11
13
15
23
0.08
0.23
46
0.06
-0.20
-0.03
0.47
0.46
S&P Disclosure
obs.
mean
49
0.09
47
-0.09
49
109
(25)
57
286
109
(25)
275
-0.28
0.57
-0.30
-0.34
0.09
0.38
1.10
1.62
0.31
0.44
-0.98
0.57
0.83
-0.71
0.44
(13)
46
0.56
162
(52)
292
(13)
161
(51)
290
-0.03
(76)
(65)
46
-0.09
38
-0.26
42
25
21
0.13
20
0.30
(36)
69
(38)
(34)
60
(29)
941 (76)
-0.35
-0.27
-0.25
-0.12
0.12
0.10
0.11
0.01
0.07
0.09
0.36
2
-0.56
-0.31
21
27
30
34
23
22
23
22
23
22
22
-0.34
-1.26
0.05
36
28
0.26
1
1
1
1
3
1
3
4
1
2
6
1
7
1
4
5
2
46
28
46
15
68
28
30
53
41
108
26
23
26
134
23
2
34
92
28
1
4
71
59
964 (99)
23
0.10
Table 3. Definitions of Control Variables
Values are measured at the end of each quarter except as indicated. Market values of shares for each quarter
are the average value for that quarter.
Variable
Ln(Tobin’s q)
Ln(Market/Book)
Ln(Market/Sales)
Ln(RTS)
Ln(Assets)
Liquidity
Leverage
Sales Growth
Missing Growth
Dummy
Net Income/Assets
Missing Income
Dummy
Description
Log of [Market value of assets / Book value of assets]. Market value of assets is estimated as
[market value of common stock + market value of preferred stock + book value of debt],
winsorized at 1% and 99%.
Log of [Market value of common stock + market value of preferred stock/Book value of
assets] , winsorized at 1% and 99%.
Log of [Market value of common stock + market value of preferred stock/Sales], winsorized
at 1% and 99%.
Log of Russian Trading System (RTS) index at each quarter end.
Log of Book value of assets.
Ratio of quarterly trading volume in rubles on common shares to market value of common
shares. If both common and preferred shares trade, liquidity = (trading volume in rubles for
common and preferred shares)/(market value of equity). Set = 0 if trading data is missing.
[Book value of debt/Book value of assets], winsorized at 1% and 99%.
Real annual growth rate of sales over the current year relative to previous calendar year,
winsorized at 1% and 99%.
Dummy variable, set =1 for firm-quarters with missing growth data.
Net income over total assets, winsorized at 1% and 99%.
Dummy variable, set =1 for firm-quarters with missing income data.
1 if firm is included in the Morgan Stanley Capital International Index at July 2005 (we were
not able to obtain historical data); 0 otherwise.
Dummy variables, =1 for an observation of a particular governance index and 0 otherwise.
Index Dummies
Thus, Brunswick dummy =1 for observations with Brunswick as the governance index.
Dummy variables, =1 for an observation in a particular industry sector and year. Sectors are
Sector-year dummies
communication, utilities, extraction, and other.
MSCI Index Dummy
Table 4. Panel A. Summary Statistics
Selected summary statistics for principal variables. Variables are defined in Table 3.
Variable
Ln(Tobin’s q)
Ln(Market/Book)
Ln(Market/Sales)
Brunswick
Troika
S&P Governance
S&P Disclosure
ICLG
RID/Expert
Combined Index
Ln(RTS)
Ln(Assets)
Liquidity
Leverage
Sales Growth
MSCI Index Dummy
Net Income/Assets
No. of obs.
964
964
942
109
286
46
162
292
69
964
964
964
964
964
839
964
939
Mean
0.05
0.10
1.43
0.02
0.01
0.00
0.00
-0.01
0.00
0.00
6.03
24.18
0.01
0.38
0.06
0.27
0.02
Min
-1.31
-2.48
-1.21
-2.90
-2.17
-1.52
-2.18
-2.58
-1.97
-2.90
4.42
19.08
0.00
0.00
-0.71
0.00
-0.04
26
median
0.02
0.03
1.32
0.15
0.05
-0.08
-0.01
0.01
-0.71
0.05
5.89
24.08
0.00
0.31
0.07
0.00
0.01
Max
1.65
3.19
6.91
2.17
2.73
2.67
2.45
3.51
1.81
3.51
7.03
28.92
0.11
1.00
0.79
1.00
0.16
Std. Dev.
0.67
1.14
1.41
0.99
1.01
1.00
1.00
1.00
1.00
1.00
0.59
1.47
0.02
0.24
0.22
0.45
0.03
Table 4. Panel B. Correlation coefficients for governance rankings
Correlation coefficients between governance indices, based on interpolated governance scores, defined as in
Figure 1. Number of observations for each pair is reported underneath each correlation coefficient. Boldface
indicates significance at 5% level.
Index
[1]
1 Brunswick
1
2 Troika
3 S&P Governance
4 S&P Disclosure
5 ICLG
6 RID
0.514
198
0.721
32
0.286
87
0.217
168
0
[2]
[3]
[4]
[5]
1
0.411
121
0.534
463
0.593
294
0.642
53
1
0.201
109
0.772
33
-0.174
31
1
0.616
121
0.293
46
1
0
Table 4. Panel C. Correlation coefficients for selected variables
Correlation coefficients for selected variables. Variables are defined in Table 3.
significance at 5% level. Sample size varies from 824 to 964.
1
2
3
4
5
6
7
8
9
10
11
Ln(Tobin’s q)
Ln(Market/Book)
Ln(Market/Sales)
Combined Index
Ln(RTS)
Ln(Assets)
Liquidity
Leverage
Sales Growth
MSCI Index Dummy
Net income/Assets
[1]
1
0.943
0.622
0.257
0.142
0.046
-0.005
0.471
-0.027
0.278
0.345
[2]
1
1
0.603
0.239
0.107
0.069
0.003
0.531
-0.029
0.229
0.280
[3]
[4]
[5]
1
0.351
0.209
0.165
0.115
-0.034
-0.053
0.297
0.124
1
0.087
0.053
0.116
-0.002
0.098
0.264
0.056
1
0.096
-0.140
-0.046
0.069
-0.014
-0.025
27
[6]
[7]
1
0.401
1
0.025 -0.055
-0.011 0.106
0.336 0.429
0.118 0.072
Boldface indicates
[8]
[9]
[10]
1
-0.174
-0.049
0.011
1
0.071
0.065
1
0.202
Table 5. Results for Combined Index
Coefficients on combined index and control variables for firm-index fixed effects, firm-index random effects,
and ordinary least squares regressions of ln(Tobin's q). Combined index and other variables are defined in
Tables 1 and 3. Index dummies and sector-year dummies are included in regressions but not reported. All
regressions use robust standard errors; pooled OLS regressions use firm-index clusters. t- or z-statistics are
reported in parentheses. *, **, and *** indicate significance at 10%, 5%, and 1% levels. Significant results
(at 5% level or better) are shown in boldface.
(1)
Dependent variable
Firm-Index
Fixed Effects
Combined index
Ln(RTS)
Ln(Assets)
Liquidity
Leverage
Sales Growth
MSCI Index Dummy
Net Income/Assets
Index dummy variables
Year-sector dummies
Missing growth, income dummy
variables
Number of Observations
Number of firms
R2
0.067***
[4.61]
0.362***
[5.74]
-0.226***
[7.03]
-1.498*
[1.66]
0.375***
[4.27]
0.248***
[4.87]
(2)
(3)
ln(Tobin’s q)
Firm-Index
Pooled OLS (firmRandom Effects
index clusters)
0.359
[1.24]
0.094***
[6.98]
0.343***
[5.28]
-0.131***
[7.14]
-1.503*
[1.83]
0.453***
[6.07]
0.190***
[3.77]
0.375***
[5.88]
0.787***
[2.71]
0.150***
[6.39]
0.321***
[4.45]
-0.087***
[2.99]
-0.447
[0.22]
0.768***
[6.70]
0.022
[0.21]
0.221***
[2.96]
3.365***
[5.64]
yes
yes
yes
Yes
Yes
yes
yes
Yes
964
99
0.46
964
99
0.63
964
99
0.68
- 28 -
Table 6. Results for Individual Indices and Alternate Firm Value Measures
Coefficients on different governance indices for firm fixed effects regressions and ordinary least squares
regressions of ln(Tobin's q), Ln(Market/Book) and Ln(Market/Sales). Each cell gives the coefficient from a
separate regression. Control variables are the same as in Table 5. Governance indices and other variables are
defined in Tables 1 and 3. Number of observations (firms) is for Ln(Tobin’s q) and Ln(Market/Book)
regressions; observations for Ln(Market/Sales) are slightly less. Control variables are the same as in Table 5,
except index dummy variables are used only in OLS regressions with combined index. All regressions use
robust standard errors; OLS regressions with individual indices use firm clusters. Fixed effects and OLS
regressions with combined index use firm-index fixed effects and firm-index clusters, respectively. tstatistics are reported in parentheses. *, **, and *** indicate significance at 10%, 5%, and 1% levels.
Significant results (at 5% level or better) are shown in boldface.
dependent variable
Combined index
Brunswick
Troika
S&P Governance
S&P Disclosure
ICLG
RID
(1)
(2)
Ln(Tobin’s q)
No. of obs. Firm-Index
(No. of firms) Fixed Effects
964
0.067***
(99)
[4.61]
Firm Fixed
Effects
109
0.171***
(25)
[4.11]
286
0.122***
(76)
[3.83]
46
-0.041
(13)
[0.61]
162
0.107**
(52)
[2.44]
292
0.084***
(36)
[2.88]
69
-0.038
(38)
[0.46]
Pooled
OLS
0.150***
[6.39]
0.068
[0.61]
0.162***
[4.81]
0.048
[1.36]
0.313***
[5.59]
0.229***
[5.22]
0.343***
[3.86]
(3)
(4)
(5)
(6)
Ln(Market/Book)
Ln(Market/Sales)
Firm-Index Pooled
Firm-Index
Pooled
Fixed Effects
OLS
Fixed Effects
OLS
0.077***
0.278***
0.134***
0.426***
[2.67]
[6.69]
[3.77]
[5.69]
Firm Fixed
Firm Fixed
Effects
Effects
0.226***
0.064
0.484***
0.144
[3.33]
[0.35]
[5.28]
[0.50]
0.135**
0.320***
0.249***
0.538***
[2.02]
[5.70]
[3.61]
[4.03]
0.022
0.097*
-0.202
0.533***
[0.18]
[1.83]
[1.34]
[3.24]
0.325***
0.479***
0.133
0.880***
[3.08]
[4.85]
[1.26]
[4.39]
0.099*
0.396***
0.108
0.472***
[1.96]
[4.80]
[1.39]
[3.39]
0.078
0.597***
-0.096
0.469*
[0.33]
[5.01]
[0.25]
[1.91]
- 29 -
Table 7. Results for Subindices.
Dependent variable is ln(Tobin's q) regressed on subindices of each index (as separate independent variables).
We use non-normalized subindex scores (with Brunswick scores reversed so higher scores indicate better
governance), so coefficients are not directly comparable to Table 6. Each regression includes all subindices of a
particular index. The coefficient on the registrar subindex of Brunswick is not reported because its weight in the
index is mere 1.4%. Control variables are the same as in Table 5, but omit index dummy variables. Governance
indices and other variables are defined in Tables 1 and 3. All regressions use robust standard errors; pooled OLS
regressions use firm clusters. t-statistics are reported in parentheses. *, **, and *** indicate significance at
10%, 5%, and 1% levels. Significant results (at 5% level or better) are shown in boldface.
Dependent variable
ln(Tobin's q)
Specification
Brunswick
Transparency
Dilution
Asset transfers / transfer
pricing
Mergers / restructuring
Bankruptcy
Ownership restrictions
Corporate governance
initiatives
Troika
Ownership structure and
transparency
Firm Fixed Effects
Pooled OLS
0.033*
[1.83]
0.01
[0.64]
0.061***
[2.77]
-0.014
[0.8]
0.080**
[2.65]
0.379**
[2.44]
-0.015
[0.53
Oversight and control
structure
Management and investor
relations
Corporate conduct
Information disclosure and
financial discipline
-0.032*
[1.74]
-0.025
[1.24]
-0.069
[1.56]
-0.031*
[1.73]
0.141***
[3.45]
0.189
[1.36]
0.107**
[3.04]
0.037
0.022
[1.91]
[0.30]
0.115***
0.095
[3.06]
[1.55]
0.058
0.052
[1.55]
[1.13]
0.042
-0.014
[1.11]
[0.33]
0.014
0.208***
[0.35]
[4.84]
S&P Disclosure
Ownership structure and
investor relations
0.001
-0.002
[0.32]
[0.33]
Financial and operational
information
0.006*
0.019***
[1.96]
[4.84]
Board and management
structure and process
-0.002
-0.002
No. of observations
No. of firms
R2
[0.61]
109
284
162
[0.37]
109
284
162
25
76
52
25
76
52
0.77
0.59
0.6
0.83
0.72
0.74
- 30 -
Figure 1: Russian Corporate Governance Indices over Time
Average interpolated standardized scores, by quarter, for each governance index. All indices are standardized
with mean =0; σ = 1, and positive scores indicating better governance. If a firm was ranked in quarter t with
score Rt and next ranked in quarter t+k with score Rt+k, we assign a score in quarter t+i (for 0 < i < k) using
linear interpolation: Rt+i= [(k-i)*Rt + i*Rt+k]/k. The figure shows the mean interpolated standardized score
for each index and each quarter.
Brunswick
Troika
S&P Governance
S&P Disclosure
ICLG
RID/Expert
2005q4
2005q3
2005q2
2005q1
2004q4
2004q3
2004q2
2004q1
2003q4
2003q3
2003q2
2003q1
2002q4
2002q3
2002q2
2002q1
2001q4
2001q3
2001q2
2001q1
2000q4
2000q3
2000q2
2000q1
1999q4
1999q3
0
1999q2
0.5
1999q1
Average corporate governance score
1
-0.5
-1
- 31 -
Figure 2: Russian Corporate Governance and RTS Index over Time
We calculate accumulated average corporate governance growth as follows. All rankings are standardized
with mean =0; σ = 1. We interpolate scores within each index as described in Figure 1. We then compute,
for the firms with scores in both quarter t and quarter t-1, the mean increase in score in quarter t. The figure
shows the cumulative increase in mean governance score (left scale). Shaded area indicates RTS index value
(right scale).
1600
0.8
0.7
800
0.6
400
RTS index
0.5
0.4
200
0.3
0.2
100
0.1
2005q4
2005q3
2005q2
2005q1
2004q4
2004q3
2004q2
2004q1
2003q4
2003q3
2003q2
2003q1
2002q4
2002q3
2002q2
2002q1
2001q4
2001q3
2001q2
2001q1
2000q4
2000q3
2000q2
2000q1
1999q4
1999q3
1999q2
0
1999q1
Accumulated average corporate governance growth
0.9
- 32 -
Figure 3. Scatter Plot of Ln(Tobin's q) Versus Combined Index
-2
-1
Ln (Tobin's q)
0
1
2
Scatter plot of ln(Tobin's q), winsorized at 1% and 99% versus standardized governance scores (pooled across
all governance indices). The regression line is given by (t-statistics estimated with firm clusters in
parenthesis):
Ln(Tobin's q) = 0.050 (t = 0.83) + 0.171 (t = 3.26) * (Combined index)
-4
-2
0
2
Standardized Governance Score
4
- 33 -