Download Partisan Politics and Stock Market Performance:

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Embedded liberalism wikipedia , lookup

Transcript
Partisan Politics and Stock Market Performance:
The Effect of Expected Government Partisanship on Stock
Returns in the 2002 German Federal Election
Submission date: May 15, 2006
Receipt of referee reports: June 22, 2007
Submission of revised manuscript: July 26, 2007
Roland Füss
Michael M. Bechtel
Department of Empirical Research and
Econometrics
University of Freiburg
Platz der Alten Synagoge/KG II
79085 Freiburg im Breisgau
Germany
Department of Politics and Management
University of Konstanz
Box D86
78457 Konstanz
Germany
Email: [email protected]
We thank Gerald Schneider, Heinz Rehkugler, and Daniel Oehmann for helpful comments and suggestions on
earlier versions of this article. Also, we would like to thank Dieter Kaiser for kindly providing us with the
trading volume data. We bear responsibility for all remaining errors. The second author gratefully
acknowledges financial support by the German National Merit Foundation.
1
Abstract
Rational partisan theory and extant evidence from parties’ ideal policies suggest firms to perform
better under right- than under left-leaning governments. If investors anticipate these effects of
different parties holding office, changes in expected government partisanship should produce
distinct patterns of stock market performance, with prices reflecting the electoral prospects of the
competing parties in the pre-election time. This is the first study which investigates such
anticipated effects of expected government partisanship on stock market performance in
Germany. In accordance with rational partisan models of government we assume that increasing
public support for left- (right) leaning parties should result in decreasing (increasing) stock
market performance. We analyze the effect of expected government partisanship on small firms
in the 2002 German federal election employing GARCH(1,1) and EGARCH(1,1) volatility
models. The empirical evidence shows that overall stock performance of small German firms
was positively (negatively) linked to the probability of a right-leaning (left-leaning) coalition
winning the election. Moreover, we find increasing electoral prospects of a right-leaning
coalition to trigger volatility increases while electoral uncertainty has a volatility reducing effect.
The findings carry implications for parties’ future economic policies and show possibilities for
further research.
Keywords:
Government partisanship, stock market performance, elections,
GARCH modeling, political information, price formation
JEL-Classification: C12, G12, G38
2
1. Introduction
Political scientists and economists alike are increasingly interested in the interplay between
politics and stock markets (Schneider and Tröger 2006; Leblang and Mukherjee 2005; Jensen
and Schmith 2004). One reason for this increased attention can be seen in the opportunity to
test the explanatory power of established politico-economic models. If different parties
strategically manipulate the economy to optimally benefit their voter base, their economic
policies should produce distinct reactions by stock markets. Another reason is rooted in
research into the effects of globalization on the policy options left to parties (Boix and
Adserà 2001; Garrett 1998). If economic integration induces parties’ policies to converge,
national partisan effects on global financial markets should disappear. A final source
fostering scholarly interest is the considerable public attention drawn by stock market
developments. For example, the rapid decline of the German stock market index following
the victory of the Social Democratic Party (SDP) against the Christian Democratic Union
(CDU) in the 2002 federal election was a headline-catching event.1
We examine the systematic distributive effects of government partisanship on the
German stock market in the 2002 German federal election. We argue that the dominant
theoretical model linking government partisanship and stock market performance via
inflation does not apply to Germany, where partisan manipulation of monetary policy is
hardly possible, because the European central bank enjoys independence from the political
process (Hays et al. 2000). We base our argument on rational partisan theory (Alesina et al.
1997; Alesina 1987) and extant evidence from the analysis of party manifestos (Budge et al.
2001) with regard to parties’ preferences for economic policies. Analysis of party manifestos
has shown right-leaning parties to provide economic policies which are more beneficial to
firms’ profits than left-wing parties. Based on this assumption and the semi-strong form of
the efficient market hypothesis (Fama 1970) we can assess how investors value different
parties holding office. Indeed, if markets process information efficiently, effects of different
parties holding office in the near future should be anticipated and incorporated into today’s
prices. Changes in expected government partisanship would then produce distinct patterns of
stock market performance, with prices reflecting the electoral prospects of the competing parties:
If public support for the left- (right) leaning party coalition increases, stock market performance
3
should decrease (increase). Thus, we gain an impression of how the German stock market
expects parties’ policies to affect the performance of firms.
The effect of expected government partisanship in the 2002 German federal election is
examined by employing GARCH(1,1) and EGARCH(1,1) volatility models. The empirical
evidence shows that overall stock performance of small German firms was positively linked with
the probability of a right-leaning coalition winning the election. Moreover, we find increasing
electoral prospects of a right-leaning coalition to have triggered volatility increases while
electoral uncertainty had a volatility reducing effect. In contrast, our analysis shows no
significant effects of expected government partisanship on medium and large sized enterprises.
The findings carry important implications for parties’ future economic policies and show
possibilities for further research.
The remainder of this paper is organized as follows: The next section provides a
compact review of the literature on the effect of government partisanship on stock market
performance. Subsequently, section three develops the theoretical model and derives
hypotheses about the relationship between the stock market and expected government
partisanship. In order to account for the characteristics of financial time series data, we
employ a GARCH framework in section four to empirically evaluate our hypotheses for the
2002 German federal election. The final section summarizes and discusses the implications of
our findings.
2. Government Partisanship and the Stock Market
For more than two decades there has been constant and extensive scholarly interest in the
effects of partisanship on stock markets in the United States. This body of literature has
strongly been inspired by the partisan (business) cycle model, which traces economic
performance back to the strategic behavior of parties. Drawing on the Downsian view of
democracy (Downs 1957) Hibbs (1977, 1987) relates economic policies to parties’
ideologies. Different ideologies imply different economic policies benefiting some parts of
the electorate at the expense of others. As Hibbs points out, “governments pursue
macroeconomic policies broadly in accordance with the objective economic interests and
subjective preferences of their class-defined core political constituencies”, (1977: 1467).
1
http://news.bbc.co.uk/1/hi/business/2275359.stm; 12/29/2005.
4
Thus, as parties are assumed to be ideologically motivated and to stick to their electoral
platforms when holding office, left-leaning parties will not only try to reduce unemployment
in the pre-election period, because their voter base benefits more from low unemployment
than from low inflation. Moreover, different parties will permanently pursue policy goals in
accordance with their ideological labels with inflation being higher under left-wing than
under right-wing governments (Alesina et al. 1997; Alesina 1987; Downs 1957).
Consequently, the argument made in the literature is as follows: Since left-leaning
parties are more willing to accept higher inflation, incumbency of left parties is associated
with a decline of the real rate of returns for investors, which causes stock prices to fall.
Drawing on this theoretical linkage between parties’ preferred economic policies and
macroeconomic outcomes (Drazen 2002; Hibbs 1987), studies have tried to uncover effects
of government partisanship on stock market performance (Leblang and Mukherjee 2005;
Foerster and Schmitz 1997; Gärtner and Wellershoff 1995; Huang 1985). However, the
evidence for the United States seems inconclusive at best. Using OLS regressions on data for
20 presidential elections after 1900, Riley and Luksetich (1980) find some support for the
hypothesis that the stock market performs better under times of Republican incumbency. On
the contrary, Huang (1985) shows that for the period from 1832 to 1980, higher average
returns have been achieved during Democratic administrations than during Republican ones.
Gärtner and Wellershoff (1995) report the stock market to perform better during the second
half of a presidency irrespective of which party is holding office. Analyzing monthly data
from 1927 to 1998, Santa-Clara and Valkanov (2003) also show that there are partisan effects
with stock market returns being higher during Democratic than during Republican
presidencies. Leblang and Mukherjee (2005) find that stock dividend yields are higher under
Republican administrations and increase if the market expects the Republican party to win
the upcoming election.
Much less work has been devoted to the analysis of partisan effects on stock market
performance for countries other than the United States. Genmill (1992) scrutinizes the impact
of expected government partisanship on the FTSE 100 in the 1987 British general election.
Herron (2000) estimates that if the Labour Party had won the 1992 general election the
British stock market would have dropped 5%. Martínez and Santiso (2003) investigate the
reaction of the Wall Street to political events in the 2002 Brazilian presidential elections. In
order to estimate the impact of different presidential candidates on macroeconomic
5
performance, Jensen and Schmith (2005) use movements in the Brazilian stock market as
proxies for future expectations for the Brazilian economy. They succeed in falsifying the
hypothesis that the Brazilian presidential candidates had an effect on the mean of the stock
market.
Some have extended their analyses to the effects of politics on stock market volatility.
The evidence for the United States and Great Britain suggests that volatility is lower (higher)
in times of Democratic (Republican) and Labour (Conservative) incumbency and also if
traders expect the left-wing (right-wing) party to win the upcoming election (Leblang and
Mukherjee 2005). In contrast, Herron (2000) estimates that if the Labour Party had won the
1992 British election the British stock market would have experienced a surge in volatility.
Also in the 2002 Brazilian presidential election Jensen and Schmith (2005) find higher
volatility to be associated with the popularity of leftist candidates.
For Germany, studies are even rarer and have hitherto ignored the possibility of
partisan politics influencing volatility. Pierdzioch and Döpke (2004) examine the connection
between current government partisanship and stock returns. They apply the political business
cycle model which was originally developed with the U.S. system in mind and analyze
quarterly stock market data from 1960 to 2002. In order to assess the effect of government
partisanship they include a dummy variable in their regression equation indicating which
party holds office. However, their results suggest that government partisanship is
inconsequential for the German stock market. Approaching the question in a similar fashion
while running their estimations on monthly data, Bohl and Gottschalk (2005) reject the
hypothesis of partisan effects on the stock market in Germany. In this paper we question the
findings for the German case on both, theoretical and empirical grounds.
Theoretically, linking government partisanship and stock market performance directly
via inflation seems plausible only in majoritarian democracies where parties exert control
over monetary policy. The existence of a largely independent central bank makes the
theoretical relationship between inflation and government partisanship unconvincing.2
Empirically, assessing the effect of current partisanship on stock market performance does
not reflect the prospective trading behavior of rational investors who try to anticipate partisan
2
Empirical evidence supports the presumption of no significant impact of partisanship on inflation in consensus
democracies (Hays et al. 2000) and with the introduction of a single European currency national monetary
policy is not even more controlled by national central banks.
6
effects on the economy. If markets are semi-strong form efficient (Fama 1970) expected
adverse effects on firms’ profits should be incorporated in today’s prices. Thus, to the extent
past research failed to find stock market performance to be influenced by current government
partisanship, this could mean that these effects might have already been anticipated by the
market. A second objection concerns the use of highly aggregated data which is a hindrance
to capturing reactions to changing electoral prospects of different parties competing for
office. To be more precise, in times where new polling results are disseminated through the
media at least on a weekly or even daily basis, investors are unlikely to ignore changes in
electoral prospects which occur more often than once every three months. Using quarterly
data will thus prevent researchers from detecting potentially interesting and important shortterm effects. Thirdly, OLS regression techniques are ill-suited for analyzing leptokurtotic and
volatility clustered data (Pagan and Schwert 1990; Beck 1983). Using generalized
autoregressive conditional heteroscedasticity (GARCH) models rather than ordinary
(Nofsinger 2004; Bittlingmayer 1998; Gärtner and Wellershoff 1995) or nonlinear least
squares regressions (Herron 2000) would certainly add confidence in the estimates.
Furthermore, these models would also offer the opportunity to simultaneously assess the
effect of politics on stock price volatility (see Beck 1983 for an early discussion).
We consequently deviate from past research on the interplay between partisanship and
stock market performance in Germany, as we pay attention to both the partisan differences in
economic policies and the prospective behavior of financial investors who try to anticipate
the effects of economic policies under different governments. We follow past research in that
different parties produce different patterns of economic policies on dividends which cause
distinct responses by the stock market. Extensive qualitative and quantitative analysis of
party manifestos evidences left parties to focus on demand side policies aiming at lowincome sections of society (Budge and Keman 1990). It is left parties which tend to care
about the distribution of wealth more strongly and are more likely to redistribute capital via
higher taxation of firms and high income individuals (Budge et al. 2001; Budge et al. 1994;
Garrett 1998). Our model thus links government partisanship with stock market performance
via expected economic policies (Drazen 2002; Persson and Tabellini 2000; Alesina et al.
1997; Alesina 1987).
7
3. Analytical Framework
Left- and Right-Leaning Coalition Governments in Multi-Party Systems
In political systems which use proportional representation, such as Germany, multi-party
systems emerge (Duverger’s Law) and parties have to form coalition governments. But can
we reasonably distinguish between right-leaning and left-leaning coalition governments
given that their policies reflect compromises and bargains? This question is important as it
determines to what extent the rational partisan model, which was originally developed with a
two-party system in mind, can be applied to multi-party systems such as Germany. There are
two conditions which have to be met in order to be able to reasonably speak of left- and rightleaning coalition governments. These are heterogeneity of parties’ policies (1) and
ideological homogeneity of coalitions (2). The first condition simply requires that different
parties have different preferred policies. This requirement is universal in that it needs to hold
for any political system for which a researcher wants to distinguish between left and right
parties. Interestingly, the condition is especially likely to be met in a multi-party system,
because the classical median voter theorem does not hold in non-majoritarian democracies.
This is because convergence of parties to the median is not an equilibrium behavior as a
Condorcet winning position does not exist (Adams and Merrill 2006). Therefore,
heterogeneity of parties’ policies seems especially plausible in Germany which exhibits a
multi-party system and therefore, the equilibrium prediction of the Downsian model does not
hold. Consequently, differences in parties’ ideal policies are even more pronounced, and
changes in government partisanship potentially cause even stronger policy changes and
deviations from the median voter’s ideal point than in majoritarian systems.3
Empirically, strong differences in German parties’ ideal policies can be observed,
regardless of whether ideal point estimates are based on party manifestoes (Debus 2006;
Pappi and Shikano 2004; Budge et al. 2001, 1994) or expert interviews (Benoit and Laver
2006). For the election year 2002 figure 1 shows the policy positions of the four major
German parties, the Social Democratic Party (SPD), the Greens, the Christian Democratic
Union (CDU) and the Liberals (FDP), on the classical left-right dimension ranging from 0
(extreme left) to 20 (extreme right) together with a 95 percent confidence interval. There is
strong heterogeneity in parties’ preferred policies between left and right: The ideal policies of
8
left parties like the SDP and the Greens are close together and clearly to the left of the
Christian Democrats (CDU) and the Liberals.
- figure 1 about here The second condition refers to the ideological homogeneity of possible coalitions.
While a rigorous formalization of this condition is beyond the scope of this article, it is
important to note that the zones of agreement of the two possible coalitions do not overlap.
This means that all possible policies resulting from a bargaining process within a left-leaning
coalition (SDP and Greens) are still clearly to the left of all policies which can reasonably be
expected from a right-leaning coalition (FDP and CDU).4 Thus, given the set up displayed in
figure 1, we assume that if the formateur (either the SDP or the CDU) has to form a coalition
government, he chooses the smallest and ideologically closest party which suffices to secure
a majority. This assumption is all the more plausible in our case, since in 2002 the
SDP/Greens coalition government explicitly stated their intention to continue their coalition
given the necessary majority while the CDU and the Liberals announced their aim to form a
right-wing coalition if possible. Therefore, it is justified to speak about left- and right-leaning
(coalition) governments.
Net Present Value and Expected Government Partisanship
According to the discounted cash flow (or net present value) model at time t the price of a
stock S t , depends on its expected value E [Vt ] , which equals the sum of all future dividends
discounted to the present. Thus, even if investors are primarily interested in capital gains,
because the source of the capital gains equals expected future dividends, the current market
price of an individual stock is based on the expected flow of dividends throughout the life of
a company. Given a continuous stream of cash flows, the expected value of the sum of
discounted future dividends is given by
3
For example, McGillivray (2003, 2004) finds that coalition governments in consensus democracies
redistribute across sectors more strongly than governments in Westminster systems.
4
An implicit, yet usual assumption underlying this reasoning is that parties prefer to form ideologically
homogenous, minimum winning coalitions (Martin/Stevenson 2001; Schofield/Laver 1985).
9
⎛ +∞
⎞
Et [Vt ] = Et ⎜⎜ ∫ e −δk Dk dk ⎟⎟ ,
⎝t
⎠
(1)
where Dk denotes the dividend payment at time k , δ is a discount factor composed of a
riskless interest rate rF and a risk premium RP which is appropriate given the risk of the
stock under consideration.5 As t approaches infinity, E [Vt ] resembles the stock price S t . To
see how expected economic policy is connected with the discounted cash flow model, note
that according to standard economic theory the size of a dividend payment Dk of firm i
equals i ’s profits π i divided by the number of shares (Miller and Modigliani 1961; Williams
1938). This is to say that on the micro level π i,k determines the amount of capital available to
be distributed as dividends Di ,k to shareholders.
The incumbent policymaker p j can either be right-leaning (in this case j = R ) or leftleaning ( j = L ).6 Consider a simple profit before tax function in the case of a left-leaning
coalition holding office ( p L ):
π i ,k = P · Y ( p L ) − ( L ⋅W ( p L ) + K ⋅ R( p L ))
with
(2)
∂π ∂R
∂π ∂Y
∂π ∂W
⋅ L >0,
⋅ L < 0 , and
⋅
< 0.
∂Y ∂p
∂W ∂p
∂R ∂p L
The first part of the difference equals a firm’s revenues, i.e. the product of P , which is
a vector of prices, and Y , which is a vector of output quantities.7 The second part captures
the costs associated with production, where W denotes the labor prices and L is the
quantities of labor employed in the production process. The last source of costs arises from
the amount of capital K needed for production multiplied by the costs of capital R . All three
parameters Y , W , and R are subject to policy decisions by the government p j , which can
choose between two broad classes of macro economic policy strategies. Demand side or left
In finance the capital asset pricing model (CAPM) is used to determine the appropriate discount factor δ for a
stock of firm i : δ i = rF + β i (rM − rF ) . In this equation rM is the rate of return on the market portfolio, β i is the
systematic or market risk of a security and β i (rM − rF ) is the risk premium RP.
5
6
For the German case we define a coalition between CDU, CSU (Christian Social Union) and the Liberals as
right-leaning and a coalition between SDP and Greens as left-leaning.
7
We have to assume competition to be imperfect, since otherwise there were no profits as marginal costs would
equal marginal benefits.
10
policies aim at manipulating the economy by increasing government spending and tax levels
which is supposed to stimulate the economy by increasing demand (see for example Persson
and Tabellini 2000). Thus, for policies of a left-leaning government p L the partial derivative
of π i,k with respect to output Y will be positive, because of higher demand. However, while
output increases, labor costs W increase. This is because left-leaning parties are not only
associated with higher non-wage labor costs needed for social security and redistribution, but
also strengthen labor unions’ in wage bargaining (Calmfors and Driffill 1988; OECD 2004).
Because of higher labor costs prices will increase, which causes the independent central bank
with the main objective to maintain price stability to intervene by setting higher key interest
rates. This triggers costs of capital to rise and thus results in a negative effect on profits, i.e.
the partial derivative
∂π ∂R
⋅
is smaller than 0. Generally, these two adverse effects
∂R ∂p L
together should exceed the gains from increased demand causing firms to perform worse
under left-wing administrations. In contrast, supply side or right-leaning policies p R focus
on the incentive structure within the economy. Lowering taxes as well as welfare expenditure
combined with a policy of wage-restraint are assumed to create a climate attracting
investment and increasing incentives to take up work, thereby enhancing economic growth.
Thus, we hypothesize a positive effect of right-leaning governments on firms’ profits.
However, before profits can be distributed, they have to be taxed with rate τ . Profits
after tax π iτ,k can be defined as:
(
)(
π iτ, k = f π i.k p j , Z i ⋅ 1 - τ k p j
)
(3)
The first component is determined by factors conditional on a government’s economic
policy ( π i.k p j ), because the party or coalition of parties holding office permanently sets
macro economic key parameters in a way beneficial to their voter base (Alesina 1987; Hibbs
1977). Profits before tax are also influenced by factors Z specific to a firm, e.g. its product
innovations, technological progress, or the quality of management, all of which are assumed
to be independent from a government’s economic policy and k , respectively. Finally, profits
of company i after tax depend on the corporate tax rate τ k , which is directly set by the party
or party coalition controlling government with left parties preferring higher tax rates than
right parties.
11
Having defined profits this way, we can easily see how the impact of parties’ different
economic policies is transmitted to the stock market. According to our empirically supported
assumption that right-wing parties enact policies more beneficial to firms’ profits, for a trader
in the pre-election time there are two possible states of the world after the upcoming election.
With probability PrtL ∈ [0,1] a left party or coalition wins the election while the right-wing
party or coalition wins with PrtR = (1 − PrtL ) . In order to see how the expected value varies
with expected government partisanship we use equation 1:
⎛ +∞ −δk
⎞
⎛ +∞ −δk
⎞
R
R
⎜
⎟
E t [Vt ] = Pr ⎜ ∫ e Dk p dk ⎟ + (1 − Prt )⎜⎜ ∫ e Dk p L dk ⎟⎟
⎝t
⎠
⎝t
⎠
R
t
(5)
Rational expectations lead investors to value future dividends as the sum of two expected
values. The first part equals the net present value of future dividends under a right-leaning
government times the probability that the right-wing party or coalition will win the upcoming
election. The second part is the net present value of all future dividends under a left-leaning
government multiplied by the probability of a left-leaning coalition winning the election.
After some simple algebraic transformation we obtain:
[
]
+∞
+∞
Et [Vt ] = ⎛⎜ ∫ e −δk Dk p L dk ⎞⎟ + PrtR ⎛⎜ ∫ e −δk Dk p R − Dk p L dk ⎞⎟
⎝t
⎠
⎝t
⎠
(6)
This equation has a very intuitive interpretation. While there is a minimum net present value
of a stock equal to the first integral, i.e. in a world in which a left-leaning coalition governs,
this value rises in the probability of a right-wing coalition winning the election times the
surplus in profits achieved under a right-wing government. In contrast, the expected value is
reduced if a victory becomes less likely.8
Volatility and Price Behavior
In linking the expected value of shares – which is dependent on expected government
partisanship – with investors’ trading behavior in order to obtain predictions for the mean and
the volatility of stock prices we rely on the work of Glosten and Milgrom (1985) as it has
been reformulated and modified by Leblang and Mukherjee (2005) in game-theoretic terms.
8
Setting up the equation using PrtL instead of PrtR shows that the stock price decreases as PrtL increases.
12
However, instead of restating the full model we restrict ourselves to a non-formal description
of the relevant causal mechanisms.
Trade is assumed to take place in the form of a sequential game played by a risk-averse
trader with homogeneous expectations who takes prices as given and a risk neutral market
maker (specialist) who quotes binding stock prices in order to ensure the liquidity and
viability of the market.9 The designated dealer is able to transfer (buy) the demanded
(offered) stock amount to (from) the trader at each time interval, which causes prices to
adjust to changes in supply or demand, respectively. Prior to the election traders acquire
information, which enables them to form expectations about the probability of a certain party
winning office. The trader then chooses the optimal demand of stocks in accordance with his
expectations about future economic policies, which are dependent on his expectations about
which party will win the upcoming election. The market maker adjusts prices accordingly, so
that these reflect the change in the trader’s expectation. Consider the equation
Δ t = E[Vt ] - Pt ,
(7)
where Δ t denotes the spread which equals the difference between the expected value of the
stock E[Vt ] and the quoted price Pt . With an increase in E[Vt ] as caused by an increase in
the probability of a right-leaning coalition winning the upcoming election, Δ t will rise. If the
“true” expected value of the stock is different from the currently quoted price, because beliefs
have been revised according to the electoral prospects of the political parties, the market
maker will adjust quotes and the market will converge to the new equilibrium. When the
trader rebalances his portfolio in response to a positive Δ t , the number of shares traded
increases. In order to equilibrate demand and supply the market maker optimally adjusts
prices and volatility. To abate demand he not only sets higher prices, but also increases
volatility, because this reduces the demand of risk-averse traders (Karpoff 1986; Andersen
1996). In other words, whenever demand for stocks increases, because stocks gain
attractiveness, higher trading volume is associated with an increase in volatility. In case Δ t is
9
For the Frankfurt Stock Exchange market makers are usually assigned to support trading in inactive stocks and
are known as designated sponsors. Designated dealers operate only in the continuous trading of the Xetra
system, where they must be admitted as participant. For this they have to “insure that there is additional
liquidity in its stock by posting quotes in the system, either at the sponsor’s own initiative, at the request of a
market
participant
(quote
request),
or
trading
auctions.”
http://deutscheboerse.com/dbag/dispatch/en/kir/gdb_navigation/trading_members/30_Market_ Making?; 31st March 2006.
13
negative, because a left-leaning government is expected after the election, the expected value
of stocks decreases. As investing in stocks looses attractiveness, demand falls. Again, the
market maker responds by choosing price and volatility optimally. To achieve balance
between demand and supply he lowers the price, and in order to create incentives for the riskaverse trader to buy or at least hold his stocks, volatility is set at low levels. This implies
different reactions of trading volume to different types of new information (Lisenfeld 1998;
Edington and Lee 1993). While good news, such as increased electoral prospects of the rightwing coalition, have a positive effect on trading volume which in turn increases volatility,
downward movements lower the number of shares traded and consequently volatility
decreases. We thus expect to find the following relationship:
Hypothesis 1: If the electoral probability of a right-wing (left-wing) coalition increases,
trading volume increases (decreases) which causes the mean and the volatility
of the stock market to rise (fall).
This hypothesis reflects that investors anticipate the effects of economic policies on future
dividend payments. On the aggregate level, the expectation of economic policies more
beneficial to firms’ profits should result in higher stock market prices and higher volatility.
However, there is a second source which might trigger higher volatility. With the expected
election outcome becoming closer and closer, it becomes increasingly difficult to predict the
expected value of a stock. The market maker consequently tries to equilibrate demand and
supply by quick price changes. This causes stock prices to strongly fluctuate in times where
uncertainty about the election outcome is high (Leblang and Mukherjee 2005; see also
Fowler 2006). Consequently, we expect that
Hypothesis 2: If electoral uncertainty increases (decreases), volatility increases (decreases).
Table 1 summarizes our empirical implications.10 Note that hypothesis 1 implies an
interaction effect between the expectation of a right-wing government and trading volume. If
a right-leaning coalition is expected, stocks become more attractive, which results in higher
demand, i.e. trading volume increases. Higher trading volume leads to a higher mean and
10
The second hypothesis captures an idea central to financial analysts where volatility is viewed as a measure of
risk. For example, it enters the Black-Scholes-Formula used to determine the price of an asset. It seems already
intuitively plausible then, that for volatility to be a measure of risk it should be correlated with (political)
uncertainty.
14
volatility of stock prices. In case a left-leaning coalition is expected to hold office after the
upcoming election, trading volume falls and this reduces prices and their volatility.
- table 1 about here -
4. Data and Methodology
To test our hypotheses we use a sample of daily stock prices and survey data for the nine
months preceding the 2002 German federal election from Forsa, one of Germany’s leading
polling institutes.11 In order to assess the impact of expected government partisanship we
choose the SDAX (small-cap German stock market index) as our dependent variable (figure
1). The SDAX is a stock market index designed to represent the overall performance of socalled small caps, i.e. firms with comparatively low market capitalization.12
- figure 2 about here There are several reasons which justify our decision to select the SDAX instead of the
DAX (Deutscher Aktienindex) or the MDAX (medium-cap stock market index) as our
dependent variable. First, national economic performance hinges crucially on the well-being
of small enterprises: They account for 40 percent of net investment in Germany, 70 percent
of all jobs and 80 percent of all trainees (see Deutscher Bundestag 2002). This alone might be
reason enough to examine SDAX returns. Second, because the DAX (MDAX) reflects the
performance of global (semi-global) firms which generate most of their turnover outside
Germany, we do not see a reason why their stocks should notably respond to national
politics. DAX (MDAX) companies realize 75% (62%) of their turnover outside Germany.
Enterprises represented in the DAX, such as Daimler-Chrysler, Siemens or BASF, enjoyed
record profits for years, but hardly paid taxes in Germany. Also in production these
companies are anything else than fully dependent on national economic policies: More than
11
The raw polling data is accessible online, see: http://www.wahlrecht.de/umfragen/forsa/2002.htm; 18th March
2006. It is also available at the Central Archive for Empirical Social Research (ZA), University of Cologne.
12
Our choice to analyze partisan effects in the 2002 election is driven by theoretical and data availability
considerations. First, we wanted to assess partisan effects on stocks of small, medium, and large enterprises.
Since the small-cap stock index series starts at the end of 2001, only the 2002 or the 2005 election were left in
the set of possible choices. Second, we wanted an election which was clear in terms of expectations about
parties’ coalition preferences. Third, the electoral race should have been close as this adds variance to the
electoral probabilities and therefore facilitates efficient estimation of anticipated partisan effects. These
restrictions made the 2002 election the only rational choice.
15
50% (40%) of their employees are located in countries other than Germany, which leaves
these firms hardly sensible to changes in labor and non-wage labor costs within Germany.
Finally, the higher amount of resources available to mid- and large-sized companies makes it
much easier for them to use exit options, i.e. evade adverse changes in national economic
policies (Kurzer 1993; Garrett 1998; Hymer 1979). Thus, if there are any anticipated partisan
effects on the stock market we should find them in the SDAX data.13
Turning to our key explanatory variable, the probability of party or coalition j winning
the upcoming election, we face different operationalization possibilities. Brander (1991) uses
the vote share a party received in the most recent survey to measure the probability of a
party’s victory. Similarly, Jensen and Schmith (2005) pool public opinion data from different
pollsters and fill in missing values with the last polling result. A second alternative is the
market model where odds on elections provided by bookmakers are used to reflect “the
acquisition of new information on the relative standing” (Herron 2000: 331) of parties
(Genmill 1992; Roberts 1990). Third, it is possible to calculate electoral probabilities from
polling results which account for the time left to the upcoming election and the variance in
polled vote shares (“electoral option model” see Alesina et al. 1997: 114-116). We decided to
apply the electoral option model, which is becoming increasingly popular in the literature
(Leblang and Mukherjee 2005; Hays et al. 2000; Alesina et al. 1997). Electoral probabilities
are preferable to other measures as they reflect two important facts: First, if polling results
are very volatile, a winning margin does not contain as much information as it does if polling
results do not change much in the pre-election time. Second, if a party is leading in the polls,
this is more likely to result in an electoral victory if there are only a few days left until the
election.14
The electoral option model is appropriate, since in 2002 there were two opposing
groups of parties (SDP and Greens; CDU and Liberals) which intended to form ideologically
different coalition governments. Therefore, the situation was comparable to what is known
from two-party systems in the sense that it was a race between two ideologically different
coalitions. Thus, we can sum up the polled vote shares of the respective parties in order to
13
However, we re-estimated all models with DAX and MDAX as dependent variables. The models were jointly
insignificant. More on this below.
14
A third reason which eliminates the second operationalization possibility is that book market data was not
available. Also note that political stock market data provides a fourth, potential measure. However, such data
was not available for the 2002 pre-election period.
16
generate a measure which indicates the probability of one group of potential coalition
partners winning the votes necessary in order to form a coalition government. Consequently,
our measure of the probability that a right-wing coalition formed by the CDU and the
Liberals (FDP), will receive the majority in the upcoming election at time te is:
⎡ Q R + μm - 50 ⎤
PrtR = Φ ⎢ t
⎥,
σ m
⎣
⎦
(8)
where Φ is the cumulative standard normal distribution, QtR denotes the proportion of
citizens who intended to vote for the right-wing coalition (CDU/CSU and Liberals) relative
to those who intended to vote for the left-wing coalition (SPD and Greens) at time t, and m is
the number of days left until the election. μ is the sample mean of changes in this
proportion, and σ is the sample standard deviation in daily changes. This delivers a measure
which tells us how likely it is that a right-wing coalition will hold office after the upcoming.
Thus, PrtR is used as a proxy for investors’ expectation of a right-leaning government after
the election. The probability of a left-wing coalition winning office is calculated as
PrtL = 1 − PrtR .
In order to assess the effect of electoral uncertainty we capitalize on the detailed polling
data available to create a measure of electoral uncertainty et (υ t ) , where we use the
proportion of voters who have not decided which party to vote for. Because in the preelection time the number of waverers equals the potential votes left for the competing
coalition to make up the winning margin, it becomes increasingly difficult to catch up in the
electoral race the more voters have indicated a clear preference for a party. To evaluate
whether this operationalization is not a matter of taste and the results simply statistical
artifacts, we re-estimated all models with an alternative measure of electoral risk used in the
literature which is based on the electoral probabilities:
et (Prt j ) = 1 − 4(Prt j − 0.5) 2 ,
(9)
where et denotes entropy at time t and Prt j is the probability of coalition j ∈ {R, L} winning
the election. As can easily be verified, et (Prt j ) is an inverse U-shaped function which reaches
its maximum 1 if Prt j = 0.5 . Since Prt j ∈ [0,1] the function reaches its minima for Prt j = 0
17
and Prt j = 1 . This reflects that uncertainty is minimal if the probability of a victory is either
very high ( Prt j ≈ 1 ) or very low ( Prt j ≈ 0 ). The intuition behind this measure is that the
smaller the difference in the winning probabilities, the less certain are expectations of
government partisanship (Leblang and Mukherjee 2005).
- table 2 about here –
Table 2 presents descriptive statistics for our main variables. First note that the SDAX
returns, our dependent variable, is not only strongly skewed to the left, but also leptokurtotic.
Unsurprisingly, the Jarque Bera test soundly rejects the null hypothesis of normally
distributed returns. While excessive kurtosis is surely present in the dependent variable, we
still need to test for volatility clustering.15 The Lagrange multiplier test rejects the null of no
volatility clustering (ARCH effects) at the 1% significance level, hence we must indeed
assume squared residuals to be correlated across time. Finally, autocorrelation diagnostic
tests were computed and the Ljung-Box test of squared returns for 5 and 25 lags indicates the
presence of volatility clustering. Unit root tests clearly show that the SDAX return series is
stationary.16
Suited to the characteristics of our dependent variable we employ a GARCH
(Generalized Autoregressive Conditional Heteroscedasticity) framework to test the empirical
implications of the model. This technique is especially suited for the analysis of data which
exhibits leptokurtosis and volatility clustering (Bollerslev 1986; Engle 2001, 1982) as is the
case with most financial time series. The basic idea behind GARCH models is to view the
variance to be conditional on its own history, i.e. the series is modeled as a linear function of
its lagged values (autoregressive). While GARCH models are standard in empirical finance
(Engle 2001; Bollerslev et al. 1988) and although Beck (1983) pointed towards their great
potential for research in political science more than two decades ago, political scientists
hitherto have largely missed to apply this technique. We thus find it appropriate to briefly
review the GARCH model as well as its extension EGARCH which we also intend to use in
our analysis.
15
Kurtosis of our SDAX series equals 7.419. This clearly exceeds 3 which we would expect from a normal
distribution.
18
Let Pt be the stock market price observed at time t, and rt = ln( Pt ) − ln( Pt −1 ) is the
continuously compounded return on the stock market index from t-1 to t. If ψ t −1 is the
information set given at time t-1, the conditional mean and the conditional variance of rt can
be defined as
μt = Et [rt ψ t -1 ] and ht = Et [(rt - μ t ) 2 ψ t -1 ] .
(10)
Typically, Ψt-1 consists of all linear functions of the past returns. In a GARCH(1,1)
specification, the conditional mean is:
rt = μ t + ξ Lt + ε t ,
(11)
where μt is a constant, Lt is a vector of exogenous variables, and ε t ψ t -1 ~ N (0, ht ) . The
conditional variance for the standard GARCH(1,1) model with exogenous variables is
ht = ω + αε t2−1 + βht -1 + λi X i ,t .
(12)
The conditional variance consists of four terms: a constant (ω), the prior shocks ε t2-1 (ARCH
term), the past variance ht -1 (GARCH term), and a set of exogenous volatility regressors X i ,t .
The ARCH term captures the tendency of volatility clustering, i.e. if large (small) price
changes are followed by other large (small) price changes albeit of unpredictable sign. A high
α coefficient signalizes that large surprises induce relatively large revisions in future
volatility. The persistence in volatility is measured by the sum of α and β . If the coefficient
of the GARCH term, β, is much larger than α this suggests that the market has a memory
that lasts longer than one period, and that volatility is more sensitive to its own history (its
own lagged values) than to new innovations. The last term in equation (12) captures the
impact of exogenous (e.g. economic, political) variables on volatility. The model can be
estimated via maximum likelihood where the distribution of the innovations ε t -1 is assumed
to be Gaussian.17
16
Note that our dependent variable is not fractionally integrated. By definition, a variable is called fractionally
integrated if it is neither integrated nor stationary. Since the return series is I(0), fractional integration is not an
issue here.
17
Bollerslev and Wooldridge (1992) show that in a GARCH model the maximum likelihood estimates of the
parameters are consistent even if the true distribution of the innovations is not Gaussian. However, the usual
standard errors of the estimators are not consistent if the assumption of Gaussian errors is violated. In this case
19
There is strong evidence that changes in prices at time t − 1 and the conditional
variance of returns at t are negatively correlated (e.g. Black 1976). Hence, volatility reacts
more strongly to negative than to positive information (leverage effect). Unlike the linear
GARCH(1,1) specification in equation (12), which assumes a symmetric effect on positive
and negative errors on volatility, asymmetric ARCH models take into account the sign of the
innovation. Indeed, asymmetric ARCH models have been found to significantly outperform
their symmetric counterparts (Nelson 1991; Glosten et. al 1993). There are several ways to
capture asymmetric volatility responses. We use the exponential GARCH(1,1) (also called
EGARCH(1,1)) model proposed by Nelson (1991) to analyze the volatility dynamics
specified by:
⎛ ε
ε
ln ht = ω + α ⎜ t −1 − E t −1
⎜ ht −1
ht −1
⎝
⎞
⎟ + γ ε t −1 + β ln h + λ X ,
t -1
i
i ,t
⎟
ht −1
⎠
(13)
where α captures the volatility clustering and γ the leverage effect. Thus, volatility is driven
by both the size and the sign of a shock. If the shock is positive (“good news”), i.e.
ε t −1
ht −1
ε t −1
ht −1
> 0 , the impact of the innovation ε t −1 is (α + γ ) ⋅
< 0 it is (α − γ ) ⋅
ε t −1
ht −1
ε t −1
ht −1
. For “bad news”, i.e.
and a price decrease yields a subsequent increase in volatility.
If γ = 0 , ln ht responds symmetrically to
ε t −1
ht −1
. If a leverage effect exists, γ is negative.
Moreover, the fact that the EGARCH process is specified in terms of log-volatility implies
that ht is always positive and therefore, there are no restrictions on the sign of the model
parameters contained in the variance equation.
semi-robust standard errors can be used. These standard errors are robust to deviations in the normality of the
residuals.
20
5. Empirical Results
Before we can interpret the results a possible objection has to be addressed. In the 2002
election there were two events which could potentially bias our results. The first event is the
massive flooding in East Germany in August 2002 which caused devastation all along the
Elbe river. This opportunity was effectively used by the federal government, consisting of a
coalition between the Social democrats and the Greens, to enormously regain sympathy when
it mobilized the armed forces not only to help with sandbagging against widespread flooding,
but also for clean-up and reconstruction work in the affected regions. The victims of the
flooding were even promised compensation payments.18 The second major event was
Chancellor Schröder’s (SDP) denouncement not to let German soldiers take part in the
planned war on Iraq. This decision met the preference of the vast majority of German voters,
while the Christian democrats (CDU/CSU) did not clearly rule out this possibility. However,
these major events are captured by according changes in polling results and thus, in the
electoral probabilities. As we are interested in the systematic effect of expected government
partisanship on stock market performance, these single events simply add to the variance of
our main explanatory variable. Potentially confounding international events relevant to the
stock market are controlled for by including the lagged Dow Jones Industrial return in our
estimations (Schneider and Tröger 2006).
We estimated three alternative specifications for each of the two electoral probabilities
accounting for potential heteroscedasticity in the residuals by applying Bollerslev and
Wooldridge (1992) semi-robust standard errors. Turning to the interpretation of the results
for the mean equation of our GARCH specification (table 3), we can see in models I, III, and
IV that the interaction term of the logged differences in trading volume and the electoral
probability Pr R has a positive sign and is significant.19 This is in line with the predictions of
our model where demand for stocks is supposed to rise in anticipation of a right-leaning
government which leads to higher stock market returns. With regard to stock market
volatility, which is modeled in the variance equation of the GARCH specification, the
18
Up to 30.000 soldiers were employed at the Elbe region in August and September 2002 with an extra 10.000
on the alert.
19
Note that the estimators for the unconditional effect of the electoral probabilities and trading volume should
not be interpreted as representing real world relationships, as they assume the other variable to be zero, which is
never the case in our sample and also seems not interesting viewed from the perspective of the theory (see e.g.
Braumoeller 2004).
21
significant ARCH coefficient α̂ suggests that yesterday’s deviations have a considerable
impact on changes of the following day. According to the hypotheses, the coefficient of the
interaction term suggests that if the market expected a right-leaning coalition to win the
upcoming 2002 German federal election, trading volume increased which resulted in higher
stock market volatility. The estimate stays significant across different specifications and
changes only slightly in substance. These estimators also remain robust against confounding
influences from the so-called Monday effect, a well known market anomaly.20
- table 3 about here The opposite picture emerges if we focus on how the electoral prospects of a leftleaning government influenced stock market performance (II, IV, VI). The coefficient of the
interaction term has a negative marginal effect on SDAX returns. With the market expecting
a left-wing government, also volatility decreased. For the electoral uncertainty measure the
coefficient is only significant in models II and III where it has a negative sign, by that
contradicting the theory which predicts higher uncertainty to trigger higher volatility.
However, because the coefficient is not robust, we should refrain from drawing any
inferences.
Further commenting on the results for our variance specifications we can see that all α̂
coefficients are significant, meaning that ARCH effects in the SDAX time series indeed exist.
The magnitude of the ARCH terms shows the effect of an unexpected shock on the volatility
of the following day. High values suggest an instable expected volatility or a disproportionate
response from market participants to past price innovations ε t2−1 . The values range from 0.3
and 0.4. Normally, those values lie between 0.1 to 0.2. However, this finding is unrelated to
the election in 2002 and can be interpreted as a specific feature of the SDAX returns as
estimations of pure GARCH models clearly show. The conditional variance at t-1 is of minor
relevance in the parsimonious models (I and II) given the low βˆ coefficients. In contrast, the
persistence of the volatility, calculated as the sum of ARCH and GARCH terms, increases
from 0.769 in the baseline specification (I) up to 0.915 (model VI) once we fully specify by
adding inflation, interest rate, and a Monday dummy to the mean and the variance equation,
20
The impact of the treasury bill rate on the volatility process of SDAX returns is only significant in GARCH
models V and VI. The positive sign indicates that a high interest rate raises the cost of borrowing for
companies, and hence increases the level of equity return volatility. This result nicely fits to the finding of
Glosten et al. (1993) who show that the treasury bill rate and stock return volatility are positively correlated.
22
respectively. Note that for the electoral prospects of the left-wing coalition (II, IV, VI)
persistence is nearly ten percent higher than in models I, III, and V. As the sum of the
GARCH terms is substantially smaller than one, all models show a mean-reverting
behavior.21
The ARCH-LM test for each of the six specifications fails to reject the null of no
clustering in the residuals, which means that the volatility dynamics of our SDAX returns is
successfully modeled. As can be seen at the bottom row of table 3, also autocorrelation
diagnostic tests were computed. The Ljung-Box Q(5) statistic for each model indicates that
serial correlation in the mean still exists, while this is not the case once we account for 25
lags. The standardized squared residuals in the GARCH models are uncorrelated, suggesting
that the GARCH models have adequately captured the persistence in the variance of returns.
Finally, the Jarque-Bera statistic suggest that deviations from normality in the standardized
residuals, mainly caused by the excess kurtosis, could be reduced substantially.
In a second step we re-estimated all specifications using an EGARCH model in order to
test the robustness of the estimators once we account for asymmetric effects of past
deviations on volatility. Table 4 shows the results. None of the estimators we are interested in
substantially deviates from those of the GARCH specification. Note however, that accounting
for asymmetric effects of past deviations turns out to have been the correct decision: Once we
fully specify the model γˆ , which captures the effect of negative changes on volatility, is
significantly smaller than 0, which shows that there is a leverage effect in the SDAX returns.
- table 4 about here The information criteria reported in the bottom part of table 4 confirm the impression of
the EGARCH model fitting slightly better to the data. Moreover, the diagnostic tests show
that our standardized residuals do not suffer from ARCH effects and are also normally
distributed. Also, the Ljung-Box statistics leads us to accept the null hypothesis of no serial
correlation in the standardized residuals in the mean and in the squared standardized
residuals. All these diagnostic tools emphasize the reliability of our results. Furthermore,
since the GARCH and EGARCH specifications are estimated assuming normally distributed
residuals, we tested whether our results remain robust against the assumption of Student-t and
21
We re-estimated all models varying the sample length. The results did not change.
23
generalized error distributions (GED). However, our substantial conclusions remain
unchanged.
As discussed above, the electoral uncertainty variable did not have a robust effect on
volatility in the 2002 election. To test whether this result is due to our operationalization we
re-estimated both, GARCH and EGARCH models each with the six specifications using the
alternative measure of electoral risk based on the electoral probabilities et (Prt j ) . The
measure is designed to reflect that uncertainty is minimal if the probability of an electoral
victory is either very high ( Prt j ≈ 1 ) or very low ( Prt j ≈ 0 ). Table 5 reports the coefficients
for electoral risk only in order to economize on space. The estimators of the GARCH model
indicate no significant impact of electoral uncertainty on changes in stock market returns.
The EGARCH estimates however do suggest that volatility decreases if electoral uncertainty
increases. Thus, increased closeness of the electoral race had a volatility reducing effect in
the 2002 German federal election. This finding is surprising but not new to the literature. For
the 2000 U.S. presidential election Leblang and Mukherjee (2004: 312-313) also find a
volatility reducing effect of electoral uncertainty using the same measure.
- table 5 about here Although the volatility-reducing effect of electoral uncertainty on stock returns is not
new, it still awaits an explanation. A first attempt has been made by Freeman et al. (2000). As
in the case of exchange rates, which are affected by political uncertainty in major economies
like Australia, United Kingdom and the United States, but not in Germany, our results too
could be attributed to the institutional characteristics of Germany’s political system. More
specifically, in systems where proportional representation fosters coalition governments,
significant policy changes become less likely with increasing closeness of the electoral race,
because this might force ideologically different parties to form a grand coalition with the
coalition agreement being a very moderate, balanced version of parties’ policies. Thus,
higher electoral uncertainty could be interpreted as a signal of relative future economic policy
stability, which would imply less risk resulting in lower volatility.
Finally, we have to report on our argument made above, that we should not expect
stocks of large and medium-sized enterprises operating on a global scale, which can easily
use exit options, to show reactions to changes in electoral prospects of national political
24
parties. We re-estimated all models for the DAX and MDAX series.22 Because they are very
similar, we discuss them together. The estimators of the interaction term were insignificant in
almost all specifications. But more importantly, even the null hypothesis of the variables
having no joint effect on the indices for large and medium sized enterprises could not be
rejected. This strongly supports the presumption of national politics having no systematic
impact on internationalized firms (anymore).
How do these findings compare with the results from previous studies? Pierdzioch and
Döpke (2004) examine the relation between current government partisanship and stock
market returns in Germany. Their results show no robust effect of government partisanship
on stock market performance. Also Bohl and Gottschalk (2005) refute the conjecture of
government partisanship affecting the German stock market. Our results demonstrate that the
stock market is far from being immune to partisan politics, but rather seems to incorporate
these expected effects into current prices. This not only shows that parties matter, but also
lends support to the semi-strong form of the efficient market hypothesis. More broadly, our
results contain another message if viewed against the finding that output growth tends to be
higher under left- than under right-leaning governments (Alesina et al. 1997). Our
estimations show that stock returns of small firms are higher if a right-leaning government is
becoming more likely. This supports our assumption that left-leaning governments stimulate
growth, but also trigger a disproportionate increase in production costs, which in the end
leads investors to expect lower profits.
6. Conclusion
Hitherto most studies examining the effect of government partisanship on stock market
performance failed to pay attention to the prospective trading behavior of rational investors.
In this paper we examined the systematic distributive effects of expected government
partisanship on the stock market in the 2002 German federal election. Our argument is based
on rational partisan theory (Alesina 1987; Alesina et al. 1997) and extant evidence from the
analysis of party manifestos (Budge et al. 2001) with regard to parties’ preferences for
22
The results are available upon request.
25
economic policies. We assume different parties to differently manipulate the economic key
parameters demand, labor costs, costs of capital and corporate tax, all central to firms’
profits. Investors will anticipate the impact of changes in future economic policy on expected
dividend payment. In case the left-wing coalition wins the upcoming election, traders expect
a minimum dividend, while as the electoral prospects of the right-wing coalition increases,
investing in a share becomes more attractive. In the stylized world of our market
microstructure these changes in the expected value of an asset cause shifts in the mean and
the volatility of stock prices. Higher (lower) net present values increase (decrease) trading
volume which results in a higher (lower) mean and volatility of stock prices. Additionally,
electoral uncertainty should be reflected by higher stock market volatility.
Results from GARCH(1,1) and EGARCH(1,1) volatility models support the rational
partisan hypotheses. However, the empirical evidence shows that in the 2002 election only
overall stock performance of small German firms was positively linked with the probability of a
right-leaning coalition winning the election, while medium-sized and large firms were not
systematically affected by expected government partisanship. For small firms we find increasing
electoral prospects of a right-leaning coalition to cause volatility increases. This not only shows
that partisan politics still matter and that future political developments are incorporated into
today’s prices, which lends support to the semi-strong form of market efficiency (Fama 1970).
Moreover, this could explain why past research failed to find significant influences of current
government partisanship on stock market behavior in Germany (Pierdzoch and Döpke 2004;
Bohl and Gottschalk 2005). Surprisingly, higher electoral uncertainty caused volatility to
decrease. This clearly is at odds with the theory, although past research has reported similar
results for the 2000 U.S. Presidential Election (Leblang and Mukherjee 2004).
This paper provides impetus and shows possibilities for further research in four directions.
First, it seems promising to examine whether the sensitivity of the German stock market varies
across time and across sections. It would be interesting to see whether and how stock market
reactions to expected government partisanship in the case of stocks of large and medium
companies decreased gradually. Also across sections the sensitivity to politics needs attention as
it could well be conditional on economic sectors and their varying success in lobbying different
parties for protection (Grossman and Helpman 2001; McGillivray 2004). For example, Herron
et al. (1999) identify 15 economic sectors where stock prices vary significantly with changes
in expectations about the winner of the 1992 U.S. presidential election. Second, we need
26
cross country evidence on the impact of expected government partisanship on stock markets
to assess the effect of domestic political institutions. The third direction of further research
efforts concerns the puzzling uncertainty-volatility nexus. We presume that in political
systems in which the use of proportional representation fosters coalition governments,
electoral closeness is more likely to measure the probability of a grand coalition. These
coalition policies will either conserve the status quo or will be watered down versions of the
coalition parties’ policies. Thus, higher electoral uncertainty might be interpreted as a signal
of relative future economic policy stability, which would imply less risk resulting in lower
volatility. But certainly, the finding of higher political uncertainty decreasing stock market
volatility necessitates a more thorough theoretical analysis.
Finally, these findings carry important implications for the development of parties’
future economic policies. In the literature on the effects of globalization on democracy and
governments’ spending behavior it has been repeatedly argued that increased economic
openness diminishes the scope of left-wing parties to implement their ideal policies, because
investors as well as firms can use exit options to evade adverse domestic economic policies
(Kurzer 1993; Garrett 1998; Cusack 1999). For Germany our evidence suggests that reality is
more than just one step ahead. While stocks of small firms were still responsive to changes in
expected government partisanship in the 2002 German federal election, the performance of
medium sized and large firms was not. This let us suspect that these companies have
successfully diversified their risk and are no longer vulnerable to changes in national
government. The second part of this story is that for parties which prefer to maintain a strong
welfare state there might be ever less left from which to extract resources needed for
redistribution.
27
Literature
Alesina, Alberto (1987): Macroeconomic Policy in a Two-Party-System as a Repeated Game, in:
Quarterly Journal of Economics 102: 651-78.
Alesina, Alberto/Roubini, Nouriel/Cohen, Gerald (1997): Political Cycles and the Macroeconomy.
Cambridge, Mass.: MIT Press.
Andersen, Torben G. (1996): Return Volatility and Trading Volume: An Information Flow
Interpretation of Stochastic Volatility, in: Journal of Finance 51(1): 169-204.
Beck, Nathaniel (1983): Time-Varying Parameter Regression Models, in: American Journal of
Political Science 27 (3): 557-600.
Benoit, Kenneth/Laver, Michael (2006): Party Policy in Modern Democracies. London: Routledge.
Bittlingmayer, George (1998): Output, Political Uncertainty, and Stock Market Fluctuations:
Germany, 1890 - 1940, in: Journal of Finance 53 (6): 2243-2257.
Black, Fischer (1976): Studies in Stock Price Volatility Changes, Proceedings of the 1976 Meeting of
the Business and Economic Statistics Section, American Statistical Association: 177-181.
Boardman, Anthony/Vertinsky, Ilan/Whistler, Diana (1997): Using Information Diffusion Models to
Estimate the Impacts of Regulatory Events on Publicly Traded Firms, in: Journal of Public
Economics 63: 283-300.
Bohl, Martin T./Gottschalk, Katrin (2005): Steht der deutsche Aktienmarkt unter politischem
Einfluss? Working Paper; http://opus.zbw-kiel.de/volltexte/2005/3770/pdf/p12005.pdf;
10.04.2006.
Boix, Carles/Adserà, Alicia (2002): Trade, Democracy, and the Size of the Public Sector: The
Political Underpinnings of Openness, in: International Organization 56 (2): 229-262.
Bollerslev, Tim (1986): Generalized Autoregressive Conditional Heteroscedasticity, in: Journal of
Econometrics 31: 307-27.
Bollerslev, Tim/Jubinski, Dan (1999): Equity Trading Volume and Volatility: Latent Information
Arrivals and Common Long-Run Dependencies, in: Journal of Business and Economic
Statistics 17(1): 9-21.
Bollerslev, Tim/Wooldridge, Jeffrey M. (1992): Quasi-Maximum Likelihood Estimation and
Inference in Dynamic Models with Time-Varying Covariances, in: Econometric Review 11:
143-172.
Bollerslev, Tim/Engle, Robert F./Wooldridge, Jeffrey M. (1988). A Capital Asset Pricing Model with
Time-Varying Covariances, in: Journal of Political Economy 96: 116-31.
Brander, James A. (1991): Election polls, free trade, and the stock market: evidence from the 1988
Canadian general election, in: Canadian Journal of Economics 24 (4): 827-43.
Braumoeller, Bear F. (2004): Hypothesis Testing and Multiplicative Interaction Terms, in:
International Organization 58: 807-820.
Budge, Ian/Klingemann, Hans-Dieter/Volkens, Andrea/Bara, Judith/Tanenbaum, Eric (2001):
Mapping Policy Preferences. Estimates for Parties, Electors and Governments 1945-1998.
Oxford: Oxford University Press.
Budge, Ian/Hofferbert, Richard/Klingemann, Hans-Dieter (1994): Parties, Policies, and Democracy.
Boulder, San Francisco, Oxford: Westview Press.
Budge, Ian/Keman, Hans (1990): Parties and Democracy. Coalition Formation and Government
Functioning in Twenty States. Oxford: Oxford University Press.
28
Calmfors, Lars/Driffill, John/Honkapohja, Seppo/Giavazzi, Francesco (1988): Bargaining Structure,
Corporatism and Macroeconomic Performance, in: Economy Policy 3 (6): 14-61.
Cusack, Thomas R. (1999): Partisan politics and fiscal policy, in: Comparative Political Studies 32
(4): 464-486.
Deutscher Bundestag (2002): Schlussbericht der Enquete-Kommission “Globalisierung der
Weltwirtschaft
–
Herausforderung
und
Antworten;
http://www.bundestag.de/gremien/welt/glob_end/glob.pdf; 12.04.2006.
Downs, Anthony (1957): An Economic Theory of Democracy. New York: Harper.
Drazen, Allan (2002): Political economy in macroeconomics. 2. print. and 1. paperback print.
Princeton: Princeton University Press.
Edington, Louis H./Lee, Jae Ha (1993): How Markets Process Information: News Releases and
Volatility, in: Journal of Finance 48: 1161-1191.
Engle, Robert F. (1982): Autoregressive Conditional Heteroscedasticity with Estimates of the
Variance of U.K. Inflation, in: Econometrica 50 (4): 987-1008.
Engle, Robert F. (2001): GARCH 101: The Use of ARCH/GARCH Models in Applied Econometrics,
in: Journal of Economic Perspectives 15 (4): 157-68.
Fama, Eugene F. (1970): Efficient Capital Markets: A Review of Theory and Empirical Work, in:
Journal of Finance 25 (2): 383-417.
Foerster, Stephen R./Schmitz, John J. (1997): The Transmission of U.S. Electoral Cycles to
International Stock Returns, in: Journal of International Business Studies 28 (1): 1-27.
Fowler, James (2006): Elections and Markets: The Effect of Partisanship, Policy Risk, and Electoral
Margins on the Economy, in: Journal of Politics 68 (1): 89-103.
Freeman, John/Hays, Jude C./Stix, Helmut (2000): Democracy and Markets: The case of exchange
rates, in: American Journal of Political Science 44 (3): 449-468.
Gärtner, Manfred/Wellershoff, Klaus W. (1995): Is there an Election Cycle in American Stock
Returns?, in: International Review of Economics and Finance 4 (4): 387-410.
Gallant, Ronald A./Rossi, Peter E./Tauchen, George (1992): Stock Prices and Volume, in: Review of
Financial Studies 5 (2): 199-242.
Garrett, Geoffrey (1998): Partisan Politics in the Global Economy. Cambridge.
Genmill, Gordon (1992): Political Risk and Market Efficiency: Tests based in British Stock and
Options Markets in the 1987 Election, in: Journal of Banking and Finance 19 (1): 211-31.
Glosten, Lawrence P./Jagannathan, Ravi/Runkle, David E. (1993): On the Relation between the
Expected Value and the Volatility of the Nominal Excess Return on Stocks, in: Journal of
Finance 48 (5): 1779-1801.
Glosten, Lawrence P./Milgrom, Paul R. (1985): Bid, Ask and Transaction Prices in a Specialist
Market with Heterogeneously Informed Traders, in: Journal of Financial Economics 14: 71100.
Grossman, Gene M./Helpman, Elhanan, 2001: Special Interest Politics. Princeton: MIT Press.
Guidolin, Massimo/La Ferrara, Eliana (2005): The Economic Effects of Violent Conflict: Evidence
from Asset Market Reactions, Working Paper.
Hays, Jude C./Stix, Helmut/Freeman, John F. (2000): The Electoral Information Hypothesis
Revisited; http://www.polisci.umn.edu/faculty/freeman/eihr.pdf; 23.12.2005.
Herron, Michael C. (2000): Estimating the Economic Impact of Political Party Competition in the
1992 British Election, in: American Journal of Political Science 44 (2): 320-31.
29
Herron, Michael C./Lavin, James/Cram, Donald/Silver, Jay (1999): Measurement of Political Effects
in the United States Economy: A Study of the 1992 Presidential Elections, in: Economics and
Politics 11 (1): 51-79.
Hibbs, Douglas (1987): The American Political Economy. Macroeconomics and Electoral Politics.
Cambridge (Mass.)/London: Harvard University Press.
Hibbs, Douglas (1977): Political Parties and Macroeconomic Policy, in: American Political Science
Review 71 (4): 1467-1487.
Huang, Roger (1985): Common Stock Returns and Presidential Elections, in: Financial Analysts
Journal 41 (2): 24-39.
Hymer,
Stephen (1979): The Multinational Corporation.
Cambridge/London/New York: Cambridge University Press.
A
Radical
Approach.
Jensen, Nathan M./Schmith, Scott (2005): Market Responses to Politics. The Rise of Lula and the
Decline of the Brazilian Stock Market, in: Comparative Political Studies 38 (10): 1245-70.
Karpoff, J. M. (1986): A Theory of Trading Volume: Journal of Finance 41: 1069-1088.
Kurzer, Paulette (1993): Business and Banking: Political Change and Economic Integration in
Western Europe. Ithaca/New York: Cornell University Press.
Leblang, David/Mukherjee, Bumba (2005): Government Partisanship, Elections, and the Stock
Market: Examining American and British Stock Returns, in: American Journal of Political
Science: 49 (4): 780-802.
Leblang, David/Mukherjee, Bumba (2004): Presidential Elections and the Stock Market: Comparing
Markov-Switching and (FIE)GARCH Models of Stock Volatility, in: Political Analysis 12
(3): 296-322.
Lijphart, Arend (1999): Patterns of Democracy: Government Form and Performance in Thirty-six
Countries. New Haven: Yale University Press.
Lipset, Seymour M./Rokkan, Stein (1967): Cleavage Structures, Party Systems, and Voter
Alignments. An Introduction, in: Lipset, Seymour/Rokkan, Stein: Party Systems and Voter
Alignments, New York: Free Press: 1-64.
Lisenfeld, Roman (1998): Dynamic Bivariate Mixture Modesl: Modeling the Behavior of Prices and
Trading Volume, in: Journal of Business and Economic Statistics 16 (1): 101-109.
Martínez, Juan/Santiso, Javier (2003): Financial Markets and Politics: The Confidence Game in Latin
American Emerging Economies, in: International Political Science Review 24 (3): 363–95.
McGillivray, Fiona (2004): Privileging Industry: The Comparative Politics of Trade and Industrial
Policy. Princeton: Princeton University Press.
McGillivray, Fiona (2003): Redistributive Politics and Stock Price Dispersion, in: British Journal of Political
Science 33 (3): 367-395.
Miller, Merton H./Modigliani, Franco (1961): Dividend, Policy, Growth, and the Valuation of Shares,
in: Journal of Business 34 (4): 411-33.
Nelson, Daniel B. (1991): Conditional Heteroscedasticity in Asset Returns: A New Approach, in:
Econometrica 59: 347-70.
Nofsinger, John R. (2004): The Stock Market and Political Cycles. Unpublished Manuscript.
OECD 2004: Wage-setting Institutions and Outcomes, OECD Employment Outlook, Paris.
Pagan, Adrian R./Schwert, William G. (1990): Alternative Models of Stock Volatility, in: Journal of
Econometrics 45 (2): 267-90.
30
Persson, Torsten/Tabellini, Guido (2000): Political Economics. Explaining Economic Policy.
Cambridge (Mass.)/London: MIT Press.
Pierdzioch, Christian/Döpke, Jörg (2004): Politics and the Stock Market: Evidence from Germany, in:
European Journal of Political Economy (forthcoming).
Riley, William B./Luksetich, William A. (1980): The Market Prefers Republicans: Myth or Reality,
in: Journal of Financial and Quantitative Analysis 15 (3): 541-60.
Roberts, Brian E. (1990): Political Institution, Policy Expectations, and the 1980 Election, in:
American Journal of Political Science 34 (2): 289-310.
Santa-Clara, Pedro/Valkanov, Rossen (2003): The Presidential Puzzle: Political Cycles and the Stock
Market, in: Journal of Finance 58 (5): 1841-72.
Schneider, Gerald/Tröger, Vera (2006): War and the World Economy: Stock Market Reactions to
International Conflicts, 1990-2000, in: Journal of Conflict Resolution (forthcoming).
Schofield, Norman/Laver, Michael (1985): Bargaining Theory and Portfolio Payoffs in European
Coalition Governments 1945-83, in: British Journal of Political Science 15 (2): 143-164.
Suominen, Matti (2001): Trading Volume and Information Revelation in Stock Markets, in: Journal
of Financial and Quantitative Analysis 36(4): 545-565.
Williams, John B. (1938): The Theory of Investment Value. Cambridge (Mass.): Harvard
University Press.
31
Figure 1: German Parties’ Ideal Points on the Economic (Left-Right) Policy Dimension
in 2002
Greens SPD
FDP CDU
Zone of agreement left-wing coalition
0
5
Zone of agreement right-wing coalition
10
15
20
Economic Policy: Left-Right
SPD
Greens
CDU
FDP
Ideal point estimates with 95-percent confidence intervals shown. Data source: Benoit/Laver 2006: Appendix B.
32
0
-.01
-.02
-.03
-.04
SDAX Returns
.01
.02
Figure 2: SDAX Returns in the 2002 German Federal Election
15jan2002
15mar2002
15may2002
15jul2002
15sep2002
15feb2002
15apr2002
15jun2002
15aug2002
33
Table 1: Hypothesized Effects of Expected Government Partisanship and Electoral
Uncertainty on the Mean and the Volatility of the Stock Market
Independent Variable
Effect on mean
Effect on volatility
1a
Right-wing government expected * trading volume
+
+
1b
Left-wing government expected * trading volume
-
-
2
Electoral uncertainty
+
34
Table 2: Descriptive Statistics
Mean
Min
Max
St.dev.
Skewness
Kurtosis
JB test
SDAX Returns
-0.136
-3.460
2.073
0.662
-0.991
7.419
179.837***
Dow Jones Returns
-0.119
-4.752
6.154
1.516
0.466
4.910
34.632***
Trading Volume
339,284.1
82,595.0
1,528,029
215,307.9
2.315
11.419
707.725***
Treasury Bill Rate
3.098
2.600
3.500
.0292
-0.408
1.899
14.384***
Inflation Rate
1.378
0.980
2.080
0.359
0.737
2.016
24.085***
PrtR
59.517
9.666
86.596
12.295
-1.007
5.965
98.496***
PrtL
40.483
13.404
90.334
12.295
1.007
5.965
98.496***
et (υ t )
14.737
11.448
17.937
1.636
-0.065
1.916
9.145**
e t (PrtR )
68.122
10.921
99.828
22.508
-0.823
2.784
21.109***
N = 184. Mean, min, max, and standard deviation (St.dev.) in percentages except for trading volume. ***, **,
and * denote statistical significance at the .01, .05 and .10 level. JB = Jarque-Bera test for normal distribution.
35
Table 3: GARCH Models for Logged Changes in SDAX Returns (N = 184)
Parameters
Mean equation
Δ(PrR)
I
0.027**
(0.012)
Δ(PrL)
ΔLog(Trading Volume)
ΔLog(Trading Volume) x PrR
-0.010*
(0.005)
0.016**
(0.008)
III
IV
0.028***
(0.007)
-0.027**
(0.012)
0.006**
(0.003)
0.138***
(0.026)
-0.016**
(0.008)
0.138***
(0.026)
-0.041
(0.032)
0.387**
(0.158)
0.382***
(0.107)
-0.909*
(0.554)
ΔLog(Trading Volume) x PrL
ΔLog(Dow Jonest-1)
II
-0.007*
(0.005)
0.013*
(0.007)
-0.027*
(0.015)
0.007**
(0.003)
0.151***
(0.023)
0.786
(0.542)
-0.041
(0.032)
-0.037
(0.030)
-0.043
(0.030)
0.387**
(0.158)
0.383***
(0.107)
0.387***
(0.135)
0.408***
(0.107)
-0.593
(0.390)
0.308***
(0.119)
0.551***
(0.115)
DMonday
Constant
VI
0.032***
(0.009)
-0.018**
(0.008)
0.144***
(0.040)
0.824
(0.843)
Δ(Inflation)
V
-0.028***
(0.009)
0.008***
(0.002)
-0.010*
(0.005)
0.017**
(0.007)
-0.019**
(0.007)
0.150***
(0.022)
0.532*
(0.288)
0.178**
(0.081)
-0.062*
(0.033)
0.144***
(0.021)
0.401
(0.276)
0.151*
(0.083)
-0.060*
(0.034)
Variance equation
α̂
β̂
Δ(PrR)
Δ(PrL)
ΔLog(TradingVolume)
ΔLog(TradingVolume) x PrR
-0.926*
(0.550)
1.547**
(0.726)
-2.556
(1.447)
0.480**
(0.229)
0.480**
(0.229)
-3.336**
(1.446)
0.298
(0.319)
0.582**
(0.237)
1.683
1.892
-142.790
9.589***
0.964
13.404**
29.191
4.295
20.453
1.683
1.892
-142.787
9.583***
0.964
13.403**
29.193
4.296
20.444
1.666
1.911
-139.306
6.101**
0.856
11.385**
27.490
2.511
17.200
Δ(Interest Rate)
Constant
AIC
SIC
LogL
J.B. test
ARCH LM(1) test
Q(5)
Q(25)
Q2(5)
Q2(25)
-0.931
(0.609)
1.559**
(0.799)
-1.553**
(0.728)
-2.557*
(1.446)
ΔLog(Trading Volume) x PrL
et (υ t )
0.909*
(0.553)
0.623***
(0.192)
0.464
(0.742)
0.613
(0.470)
0.328***
(0.116)
0.550***
(0.099)
-0.510*
(0.305)
0.323***
(0.110)
0.592***
(0.095)
0.436
(0.287)
0.542***
(0.125)
-0.724**
(0.352)
1.313***
(0.476)
-1.297
(1.907)
-3.013
(2.877)
0.322
(0.475)
0.520
(0.462)
-1.556
(0.991)
0.332*
(0.186)
0.288*
(0.162)
-1.137***
(0.362)
-1.364
(0.920)
0.345**
(0.144)
0.250*
(0.150)
1.725
1.970
-144.709
40.483***
0.003
10.283*
27.520
1.205
11.731
1.636
1.898
-135.524
5.027*
0.514
10.094*
21.006
3.182
16.496
1.618
1.880
-133.872
4.500
0.373
10.347*
21.388
3.020
16.621
Coefficients shown with Bollerslev and Wooldridge semi-robust standard errors in parentheses.
denote statistical significance at the .01, .05 and .10 level.
*** **
,
, and
*
36
Table 4: EGARCH models for Logged Changes in SDAX Returns (N = 184)
Parameters
Mean equation
Δ(PrR)
I
0.027***
(0.010)
Δ(PrL)
ΔLog(Trading Volume)
ΔLog(Trading Volume) x PrR
-0.009*
(0.005)
0.016**
(0.008)
III
IV
0.027***
(0.009)
-0.027***
(0.010)
0.007**
(0.003)
0.145***
(0.022)
-0.016**
(0.008)
0.145***
(0.022)
-0.069**
(0.031)
0.298**
(0.128)
-0.135
(0.092)
0.843***
(0.093)
-0.019
(0.023)
ΔLog(Trading Volume) x PrL
ΔLog(Dow Jonest-1)
II
-0.010*
(0.005)
0.017**
(0.008)
-0.027***
(0.009)
0.007**
(0.003)
0.149***
(0.021)
0.209
(0.386)
-0.069**
(0.031)
-0.067**
(0.032)
-0.067**
(0.032)
0.298**
(0.128)
-0.135
(0.092)
0.843***
(0.093)
0.339**
(0.144)
-0.158*
(0.098)
0.824***
(0.106)
-0.024
(0.026))
0.347**
(0.143)
-0.158*
(0.097)
0.822***
(0.108)
DMonday
Constant
VI
0.029***
(0.009)
-0.016**
(0.008)
0.150***
(0.021)
0.200
(0.383)
Δ(Inflation)
V
-0.029***
(0.009)
0.007***
(0.002)
-0.007
(0.005)
0.014**
(0.007)
-0.014**
(0.007)
0.149***
(0.022)
0.370
(0.320)
0.214***
(0.078)
-0.109***
(0.036)
0.149***
(0.022)
0.370
(0.320)
0.214***
(0.078)
-0.109***
(0.036)
Variance equation
α̂
γˆ
β̂
Δ(PrR)
Δ(PrL)
ΔLog(TradingVolume)
ΔLog(TradingVolume) x PrR
-0.006
(0.014)
0.031
(0.023)
0.019
(0.023)
0.024**
(0.010)
-0.012
(0.026)
0.039*
(0.023)
-0.003***
(0.001)
-0.031
(0.023)
-0.003***
(0.001)
Constant
-0.451**
(0.001)
-0.451**
(0.176)
-0.002***
(0.001)
0.014
(0.028)
-0.507**
(0.201)
AIC
SIC
LogL
J.B. test
ARCH LM(1) test
Q(5)
Q(25)
Q2(5)
Q2(25)
1.589
1.816
-133.170
4.172
0.000
9.502*
26.460
1.185
21.518
1.589
1.816
-133.170
4.172
0.000
9.503*
26.461
1.185
21.518
1.600
1.862
-132.161
4.419
0.043*
9.721*
25.762
0.987
21.444
ΔLog(Trading Volume) x PrL
et (υ t )
Δ(Interest Rate)
0.025
(0.026)
0.027**
(0.011)
0.374***
(0.143)
-0.204**
(0.100)
0.827***
(0.085)
-0.026
(0.023)
0.374***
(0.143)
-0.204**
(0.100)
0.827***
(0.085)
0.026***
(0.023)
0.029***
(0.009)
-0.014
(0.014)
0.043*
(0.022)
-0.039*
(0.023)
-0.003***
(0.001)
0.014
(0.028)
-0.517
(0.203)
-0.003***
(0.001)
0.010
(0.024)
-0.543***
(0.171)
-0.043*
(0.022)
-0.003***
(0.001)
0.010
(0.024)
-0.543***
(0.171)
1.600
1.862
-132.168
4.449
0.126
9.739*
25.680
0.975
21.733
1.566
1.845
-128.060
1.972
0.213
8.718
19.612
2.086
18.928
1.566
1.845
-128.060
1.972
0.214
8.716
19.607
2.085
18.931
Coefficients shown with Bollerslev and Wooldridge semi-robust standard errors in parentheses.
denote statistical significance at the .01, .05 and .10 level.
*** **
,
, and
*
37
Table 5: Results for Alternative Uncertainty Measure ( et (PrtR ) )
GARCH
EARCH
I
-0.051
(0.053)
-0.023**
(0.009)
II
-0.053
(0.054)
-0.023**
(0.009)
III
-0.100*
(0.055)
-0.023**
(0.010)
IV
-0.089*
(0.054)
-0.022**
(0.010)
V
-0.006
(0.042)
-0.021**
(0.009)
VI
-0.006
(0.042)
-0.021**
(0.009)
Coefficients shown with Bollerslev and Wooldridge semi-robust standard errors in parentheses. ***, **, and *
denote statistical significance at the .01, .05 and .10 level.
38