Political climate, optimism, and investment decisions

Author’s Accepted Manuscript Political climate, decisions

optimism,

and investment

Yosef Bonaparte, Alok Kumar, Jeremy K. Page

www.elsevier.com/locate/finmar

PII: DOI: Reference:

S1386-4181(17)30115-5 http://dx.doi.org/10.1016/j.finmar.2017.05.002 FINMAR433

To appear in: Journal of Financial Markets Received date: 20 August 2014 Revised date: 4 May 2017 Accepted date: 5 May 2017 Cite this article as: Yosef Bonaparte, Alok Kumar and Jeremy K. Page, Political climate, optimism, and investment decisions, Journal of Financial Markets, http://dx.doi.org/10.1016/j.finmar.2017.05.002 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Political climate, optimism, and investment decisions∗ Yosef Bonaparte, University of Colorado at Denver Alok Kumar, University of Miami Jeremy K. Page, Brigham Young University

Abstract – We show that people’s optimism towards financial markets and the macroeconomy is dynamically influenced by their political affiliation and the current political climate. Individuals become more optimistic and perceive markets to be less risky and more undervalued when their preferred party is in power. Accordingly, investors increase allocations to risky assets and exhibit a stronger preference for high market beta, small-cap, and value stocks, and a weaker preference for local stocks. The differences in optimism and portfolio choice across political regimes are not explained by shifts in economic conditions or differential response to economic conditions by Democrat and Republican investors. Keywords: Portfolio choice, optimism, politics, sentiment, behavioral finance JEL Classifications: G11, G02

∗

Please address all correspondence to Alok Kumar, Department of Finance, 514 Jenkins Building, University of Miami, Coral Gables, FL 33124; Phone: 305-284-1882; email: [email protected]. Yosef Bonaparte can be reached at 909-263-3783 or [email protected]. Jeremy Page can be reached at 801-422-7173 or [email protected]. We would like to thank an anonymous referee, Jason Abrevaya, Brad Barber, Hank Bessembinder, Laurence Booth, Sudheer Chava, Jonathan Cohn, Russell Cooper, Richard Dusansky, Doug Emery, Will Goetzmann, John Graham, John Griffin, Jay Hartzell, Jennifer Huang, Markku Kaustia, George Korniotis, Kelvin Law, David Ng, Manju Puri, David Robinson, Sophie Shive, Tyler Shumway, Clemens Sialm, Laura Starks and seminar participants at UT-Austin, Queen’s Behavioral Finance Conference, and 2012 AFA Meetings for helpful discussions and valuable comments. We also thank Muniza Abdul for excellent research assistance.

1.

Introduction

The literature on household finance provides evidence of significant heterogeneity in the investment behavior of individual investors (e.g., Barber and Odean, 2001; Dhar and Zhu, 2006; Graham and Kumar, 2006). Participants hold different expectations about stock market performance and broad macroeconomic variables such as unemployment and inflation (e.g., Carroll, 2003; Vissing-Jorgensen, 2003; Amromin and Sharpe, 2008). However, there is relatively less research on how the heterogeneity in investors’ beliefs and expectations about the markets and the economy directly affect their portfolio decisions and trading activities. In this paper, we study whether the changing expectations of U.S. households about the behavior of financial markets and the macroeconomy affect their investment decisions. Our key innovation is to use the combination of the current political environment and the political identity of individuals to infer their expectations about the markets and the economy. The combination of political identity and political climate provides an exogenous source of variation in investors’ beliefs and expectations, which allows us to isolate the direct effects of changes in beliefs and expectations on portfolio decisions. Specifically, we conjecture that people’s expectations about the behavior of financial markets and the macroeconomy vary in a predictable manner and depends upon their political affiliation and the current political climate. This insight is based on the observation that the political identity of an individual is an important source of her degree of optimism towards the U.S. economy. Republicans are less optimistic about the domestic economy when Democrats are in power and, similarly, Democrats grow less optimistic about the economy when Republicans come to power. For example, in a 2009 Gallup poll, about 85% of Democrats believed that the economy would improve in the next 12 months, while only 50% of Republicans and 57% of Independents held an optimistic outlook about the economy (Jones, 2009).1 1

At the time of the survey, the Democrats controlled both houses of Congress and President Barack Obama (a Democrat) held office.

1

Further, media reports indicate that people’s optimism about the condition of the U.S. economy has a considerable influence on their investment choices. Even sophisticated market participants such as hedge fund managers respond to the changing political landscape (Zuckerman, 2009). In particular, during the 2008 economic downturn and the subsequent recovery, hedge fund managers with pessimistic views of the U.S. economy chose to stay out of the stock market. They lacked confidence that the Obama administration’s policies would succeed in sustaining the economic recovery. Consequently, these pessimistic managers significantly under-performed their peers and failed to benefit from the overall market recovery. Motivated by this anecdotal evidence and the evidence from the household finance literature, we conjecture that political environment and political values jointly and dynamically influence the investment decisions of U.S. households. Investors grow more optimistic about domestic financial markets and the overall U.S. economy when their own political party comes to power because they are likely to agree more with their economic policies and may feel more confident about their ability to improve the overall economy. This shift in optimism is likely to influence their perceptions of risk and reward, which in turn would affect their investment decisions and portfolio performance. In particular, because of their increased optimism levels when their own party comes to power, individuals are more likely to believe that financial assets are undervalued and would produce superior future performance. Those individuals may also perceive the markets to be less risky and would therefore exhibit a greater willingness to hold riskier portfolios. For example, investors whose own party is in power are likely to increase the market exposures of their financial portfolios. In addition, they may overweight stocks with higher market betas and exhibit a stronger preference for riskier small-cap and value styles. We test our conjectures using a large sample of UBS/Gallup survey data from 1996-2002, the National Longitudinal Survey of Youth (NLSY) data set covering 1988-2000, and portfo2

lio holdings and trading data from a large U.S. discount brokerage house from 1991-1996. We use data from multiple sources because each data set has its own strengths and limitations. The Gallup data set contains accurate measures of political affiliations of individuals and responses to direct questions about optimism; however, it contains very limited information about their financial asset holdings. In contrast, the brokerage data set contains richer information about the portfolio holdings and trading activities of investors, but it does not contain information about their political affiliations. Unfortunately, the brokerage data do not contain aggregate asset holdings of stocks and bonds. The NLSY data contain the political affiliations of individuals and their aggregate holdings of stocks, bonds, and mutual funds, but they do not contain details about the individual asset positions or trading. We exploit the strengths of these data sets that span different time periods to portray a more complete picture of the impact of changes in political climate on the optimism levels and investment decisions of individual investors. First, using the Gallup data, we provide evidence of political climate-induced dynamic optimism among U.S. households and derive potential implications for changes in investment decisions. Then, using the NLSY and brokerage data sets, we test those implications and show that shifts in political environment have differential effects on the investment behavior of Democratic and Republican investors. Since we only know the locations (ZIP Codes) of brokerage investors and cannot measure their political affiliations directly, we use the voting data from the 1992 and 1996 U.S. presidential elections to indirectly infer their political identities. We identify counties with high concentration of Republicans and Democrats and assume that brokerage investors in Republican (Democratic) dominated areas are more likely to have a Republican (Democratic) political identity.2 2

We do not require that investors located in regions concentrated by Republicans (Democrats) be a Republican (Democrat). Rather, we only assume that investors in regions concentrated by Republicans (Democrats) are more likely to subscribe to the Republican (Democratic) political ideologies. Recent studies have used a similar location-based identification strategy to infer the education-level, religiosity, and race/ethnicity of investors and managers (e.g., Kumar, 2009b; Hilary and Hui, 2009; Korniotis and Kumar,

3

We identify the effect of politically-induced optimism on individuals’ investment choices using a difference-in-difference estimation strategy. Specifically, we test whether the difference between Democratic and Republican investors in reported optimism, and in turn, investment decisions, varies across political regimes. This is a critical feature of our empirical strategy, because it allows us to control for time-series variation in overall levels of investor optimism, which may be influenced by recessions, wars, and other economic and political events that may be correlated with which party is in power. Implicitly, in most of our tests we assume that economic conditions and other such external factors affect Democrats and Republicans in roughly the same way, so that we can attribute changes in relative optimism and portfolio choice to the political regime. However, we also consider and control for the possibility that Republican and Democratic investors are affected differently by broad economic conditions. To ensure that our geography-based political identification strategy has power to differentiate between investors with Republican and Democratic ideologies, we examine the stock preferences of investors located in regions dominated by Democrats and Republicans. Motivated by the evidence in Hong and Kostovetsky (2012), we assume that political values would influence the stock preferences of individual investors. Under this assumption, if our political affiliation identification strategy is effective, brokerage investors in regions with a high concentration of Democrats will underweight the politically sensitive stocks of firms whose business is inconsistent with their political values, while investors in regions with a high concentration of Republicans will overweight them. Consistent with the evidence in Hong and Kostovetsky (2012), we find that political values affect the investment decisions of individual investors.

Investors in Republican-

dominated regions overweight politically sensitive stocks, while investors located in regions with a high concentration of Democrats underweight these stocks. This evidence suggests 2011).

4

that our geography-based political affiliation identification procedure can identify investors with Republican and Democratic ideologies reasonably well. In our main empirical analysis, using the Gallup data, we show that Democrats (Republicans) become more optimistic about the stock market and the overall economy when Democrats (Republicans) come to power and there is a decline in optimism when the opposing party comes to power. The downward shift in optimism is more pronounced among Democrats when the Republican party comes to power.3 Beyond its influence on optimism, shifts in political climate influence people’s perceptions of risk and reward. Investors believe that financial markets are less risky and more undervalued when their own party is in power. This reduced assessment of the riskiness of the market and higher potential reward induces investors whose party is in power to take greater financial risks. In the second part of the paper, we first use the NLSY data to provide some preliminary evidence that individuals increase the market exposures of their financial portfolios when their preferred political party is in power. Then, using the brokerage data, we demonstrate that the systematic risks of their equity portfolios increases. The preference for stocks with higher market beta and riskier small-cap and value styles increases when the political environment is aligned with investors’ own political identity. The shifts in systematic risk exposures of investors’ portfolios occur passively as well as actively through their trading activities. We also find that when investors’ preferred party is not in power, they exhibit a stronger preference for familiar local stocks. Examining the performance implications of political climate induced risk-shifting behavior of investors, we find that investors earn higher raw portfolio returns when their own party is in power. However, this extra return can at least partially be attributed to an increase in portfolio risk associated with their increased levels of optimism. When we examine changes 3

This finding may reflect the asymmetry in political regimes between the split government during the latter part of the Clinton administration and the full Republican control of both the presidency and Congress during Bush’s term.

5

in risk-adjusted performance of their portfolios, the superior performance of investors whose party is in power weakens considerably. Our findings are consistent with the conjecture that investors’ optimism, and in turn their portfolio decisions, are influenced by the interaction of their own political views with the prevailing political regime. We acknowledge, however, that our results must be interpreted with some caution. The brokerage account data that we use for our main analysis is limited to two shifts in political regime: the 1992 presidential election in which President Bill Clinton succeeded President George H. W. Bush, and the 1994 midterm election in which the Republican Party gained control of both houses of Congress. As such, we cannot rule out the possibility that Democrat and Republican investors responded to these regime changes in a way that was unique to this time period. While our difference-in-difference empirical strategy largely accounts for the prevailing economic, political, and social conditions that were unique to this period, we perform additional tests to address some alternative explanations for the differential response we observe between Democrat and Republican investors. First, political regime shifts are likely to be correlated with changes in economic conditions or other external events. It is possible that the differences in preferences or investment styles of Democratic and Republican investors lead them to respond differently to the same economic conditions. If so, the variation in optimism and portfolio choice that we observe may actually reflect the differential responses of Democratic and Republican investors to changes in economic conditions rather than changes in political regime. To address this concern, our regressions include explicit controls for macroeconomic conditions as well as for the interaction between political affiliation and economic conditions. In our main analysis of both the Gallup survey data and the brokerage account data, we find that the impact of shifting political regimes on the differences in optimism and portfolio choice between Democrats and Republicans cannot be explained simply by changes in macroeconomic conditions, nor by differential responses to economic conditions by Democrats and Republicans. 6

A related possibility is that Republican and Democratic regions are subject to different economic conditions that vary over time and are negatively correlated with each other. If this were the case, and if our measure of the national political regime captured time variation in these local economic conditions, we might observe investors in Republican and Democratic areas responding in opposites ways, consistent with our results. To address this possibility, we conduct additional tests to demonstrate that the variation in investors’ portfolio choices are in fact induced by changes in the political climate and investors’ own political preference rather than simply a differential response to potential cross-sectional heterogeneity in economic conditions. Thus, while it may not be possible to completely rule out the possibility that some other unidentified local factor could produce these results, we believe that the optimism hypothesis we propose is the most plausible explanation for the observed patterns in market risk exposure and trading behavior. Collectively, these results suggest an economically meaningful link between political climate and investment decisions, where the impact of political climate on investment decisions depends upon the political affiliations of individuals. These findings contribute to a recent and growing literature in finance on interaction between optimism and financial decisions. Researchers have shown that portfolio decisions are influenced by people’s expectations about the performance of the aggregate economy and the stock market. For example, Strong and Xu (2003) show that home bias increases when investors grow more optimistic about the domestic economy. In another related study, Puri and Robinson (2007) show that optimistic individuals take greater financial risks. They are more likely to participate in financial markets and conditional upon participation they choose riskier securities. However, previous research does not examine whether the level of optimism and portfolio decisions are influenced by changes in the political environment, which is the main focus of our paper. In addition, unlike previous studies, we have detailed data on the stock-level holdings and trading activities of investors instead of aggregate asset-class level data (i.e., stocks, bonds, 7

etc.). The richer data allow us to characterize the potential investment “mistakes” that arise from changes in perceived uncertainty induced by a changing political climate.4 Our paper also extends the growing literature on politics and finance. Kaustia and Torstila (2011) show that stock market decisions of Finnish investors are influenced by their political preferences, where individuals with right-wing ideologies are more likely to participate. Hong and Kostovetsky (2012) find that political values influence the portfolio decisions of even relatively more sophisticated market participants, such as mutual fund and hedge fund managers. In particular, Democratic managers exhibit a stronger preference for socially responsible firms, while managers who contribute more to the Republican party overweight socially irresponsible industries (e.g., tobacco or guns). Instead of focusing on the unconditional relation between political values and investments, we study how political values and political environment jointly and dynamically influence the investment decisions of U.S. households. The rest of the paper is organized as follows. In the next section, we describe the data sources. In Section 3, we examine the relation between political climate and optimism. In Section 4, we examine the impact of changes in political climate on investment decisions. We consider a few alternative explanations of our findings in Section 5 and conclude in Section 6 with a brief discussion.

2.

Data and summary statistics

We use data from two main sources. The first data source is the UBS/Gallup Investor Optimism Survey.5 The survey is conducted by the Gallup organization, and it includes a randomly selected national cross-section of heads of household or spouses in households with 4

Other related studies such as Vissing-Jorgensen (2003), Puri and Robinson (2007), and Amromin and Sharpe (2008) use survey data to measure investors’ beliefs at a given point in time and have limited information about portfolio holdings and trading activities. 5 Previous studies such as Vissing-Jorgensen (2003) and Graham et al. (2009) have used the Gallup survey data to study the impact of optimism and competence on portfolio choice.

8

total savings and investments from stocks, bonds, or mutual funds in an investment account, or in a self-directed IRA or 401(k) retirement account of $10,000 or more.6 Gallup required interviewees to be at least 18 years old, and conducted approximately 1,000 interviews via telephone each month during the first two weeks of the month. The monthly polls started in October 1996 and have been conducted since then. Although the survey is not a panel, cohort analysis is possible due to the large number of investors interviewed each month.7 The UBS/Gallup poll provides qualitative responses about optimism or pessimism regarding the stock market and other macroeconomic variables, including expectations about portfolio returns, the aggregate economy, stock market, inflation, income, and unemployment. The data set also contains information about household asset holdings, income, and self-reported realized portfolio returns, as well as demographic variables such age, education, race, and gender. The data are organized in one “big file” that includes questions only on political affiliation, demographic information, and investor optimism for the period from October 1996 to December 2002. There are 57,428 respondents in this file. About 29% of respondents report that they consider themselves a Republican and approximately 20% identify themselves as Democrats. In addition, 28% of respondents report that they are independent, while the rest support other parties.8 Some of the responses from the Gallup survey are recoded. Specifically, we define a race binary variable that takes the value of one if the respondent is White and zero otherwise. We recode the education level by assigning a new variable that takes a value of 9 if the respondent is a high school graduate or less and a value of 14 if the respondent has attended 6

According to Vissing-Jorgensen (2003) and based on the 1998 Survey of Consumer Finances, households with $10,000 or more in financial assets owned more than 99% of all household financial wealth in the U.S. 7 Not all the monthly polls have the same set of questions. For example, in 1996, no questions were posed about realized portfolio returns or future forecasts of portfolio returns. Also, in a few instances, answers to questions are not publicly available. For example, answers to the question about political affiliation are not available for the year 1999. 8 Among the independent investors, about 35% lean toward the Republican Party and 37% lean toward the Democratic Party.

9

a college or has received any vocational training. Respondents who are college graduates are assigned a value of 15 and those with postgraduate degrees are assigned a value of 17. We also redefine the categorical income variable, where the new income variable is the mid-point of the categorical income bracket reported in the survey. Similarly, the new asset holdings variable is assigned the mid-value of the asset bracket reported in the survey. The highest income bracket is “$100,000 or greater” and the highest asset holdings bracket is “$1 million or greater.” We recode the top bracket by multiplying the reported values by 1.5. Our second data set comes from a large U.S. discount brokerage house and covers the 1991 to 1996 period. This data set contains the trades and monthly portfolio positions of a sample of 62,387 retail investors who hold stocks. An average investor in the sample holds a four-stock portfolio (median is three) with an average size of $35,629 (median is $13,869). For a subset of households, the brokerage data also contain demographic characteristics, including age, income, location (ZIP Code), occupation, marital status, gender, etc. These demographic characteristics of investors are measured a few months after the end of the sample period (June 1997) and are provided by Infobase, Inc.9 We enrich the brokerage database using ZIP Code-level demographic data from the U.S. Census and county-level voting data. We obtain racial and ethnic compositions as well as educational attainment levels for each ZIP Code using data from the 1990 U.S. Census. We assign to each investor the racial and ethnic characteristics of the ZIP Code in which the investor resides. We also also assign to each investor the educational attainment level of his or her ZIP Code, under the assumption that investors who live in ZIP Codes with higher proportions of educated people tend to be more educated.10 Next, to identify the political affiliation of each investor, we obtain the county-level voting 9

See Barber and Odean (2000) for additional details about the brokerage data. Although these demographic proxies are noisy, we use them to have the same set of demographic controls in the Gallup and brokerage data sets. Our results are very similar when we only use the demographic variables available in the brokerage sample as control variables. 10

10

data from the 1992 and the 1996 U.S. presidential elections in the U.S.11 We assume that investors who live in counties where people voted strongly for the Republican (Democratic) party are more likely to be a Republican (Democrat). Specifically, for each county, we compute the proportion of total votes to the Democratic and the Republican parties during the 1992 and 1996 presidential elections. We obtain the average values of these two measures across the two presidential elections and use them as proxies for the political affiliation of investors in the brokerage sample. For example, if 60% of all votes in a county were in favor of the Democratic party during the two elections, we assign a Democrat score of 0.60 to investors who live in that county. In some of our tests, we use data from the National Longitudinal Survey of Youth (NLSY) conducted by the Bureau of Labor Statistics. This data set contains information from annual interviews covering a variety of topics, mostly related to labor market outcomes. For our purposes, we focus on information from the surveys on individuals’ net worth and the value of their financial holdings in stocks, bonds, and mutual funds. The value reported in the survey is an aggregation of those three asset classes, and so we are only able to consider aggregate market participation, rather than allocation between stocks and bonds. In addition to the high-level information on assets, survey participants are also asked about their political party affiliation (Democrat, Republican, Independent, or Other), and the strength of their affiliation (Strong or Not Very Strong). This information allows us to examine the relative market exposures of individuals based on both their personal political preferences and the current political climate. The NLSY sample period spans from 1980 to 2000 and thus includes the period of the brokerage data with additional years before and after. Beyond these data sets, we obtain stock price, return, and trading volume data from the Center for Research on Security Prices (CRSP). We obtain characteristic-based performance 11

The county-level presidential voting data are available at http://www.uselectionatlas.org/.

11

benchmarks from Russell Wermers’ website.12 The Appendix provides a brief summary of the key variables. Table 1 reports their summary statistics.

3.

Political climate and optimism

In this section, we use the Gallup survey data to establish how changes in political regimes differentially affect the optimism levels of Republicans and Democrats. We also study whether these optimism shifts affect people’s expectations about future market performance and its riskiness.

3.1.

Gallup survey evidence

Figure 1 depicts the difference in optimism between Democratic and Republican survey respondents for the October 1996 to December 2002 period. We define optimism as the proportion of respondents that are “somewhat” or “very” optimistic about the stock market performance, economic growth, income, employment, investment goals, or inflation during the subsequent 12 months. Figure 1 shows a large shift in the optimism levels of Republicans and Democrats between the time when election results were announced in November 2000 and when President George W. Bush took office in January 2001. Before the election results were announced, Democrats are slightly more optimistic than Republicans. However, soon after the announcement of the election results, the difference in the proportions of Republicans and Democrats who are optimistic about the economy (i.e., the optimism gap) widens to about 40%. For example, about 62% of Democrats are optimistic about the stock market in the year 2000 and that number drops to about 36% in 2001. The optimism about the overall economy is affected in a similar manner. The optimism related to employment drops from 69% to 31%, while the optimism about economic growth falls from 70% to 35% during the same 12

The web site is http://www.smith.umd.edu/faculty/rwermers/ftpsite/Dgtw/coverpage.htm.

12

time period. These results indicate that people’s optimism levels vary around political regime changes and, more importantly, those optimism shifts depend upon the political affiliation of individuals. Table 2 reports the average optimism scores of Democratic, Independent, and Republican survey respondents for two distinct political regimes: (i) 1996 to 1998, when President Clinton (a Democrat) held office, and (ii) 2001 to 2002, during which President Bush (a Republican) was in office. We exclude the year 1999 from our analysis because the political affiliation data are not available for this year and exclude the year 2000 because the horizon of optimism questions overlaps the two presidential regimes. In columns (1) to (3), we present the proportions of individuals within these three political categories who report being optimistic, while in column (4) we report the difference in the proportion of optimistic responses between Republicans and Democrats. Responses are considered optimistic if respondents answered 4 or 5 on a 5-point scale (i.e., “somewhat” or “very” optimistic). In columns (5) to (7), we report the mean optimism scores of those investor groups, where a score of 5 means very optimistic and a score of 1 means very pessimistic. We present these statistics for both the Optimism Index, which is aggregated from several survey questions regarding various aspects of the economy, as well as statistics from each of the individual optimism questions that comprise the index. While the degree of optimism in our sample is lower during the George W. Bush presidency for respondents from all political categories, the relative optimism between Democrats and Republicans flips between the two political regimes. Democrats are more optimistic than Republicans during the Clinton presidency, while Republicans are more optimistic than Democrats during the Bush presidency. Specifically, the mean optimism difference between Republicans and Democrats increases from −0.075 during the Clinton presidency to 0.390 during the Bush presidency.

13

3.2.

Optimism regression estimates

To ensure that the optimism differences in the univariate analysis do not simply reflect differences in the demographic attributes of Republicans and Democrats, we estimate a series of regressions in which one of the optimism measures is the dependent variable. The optimism regression results are reported in Table 3, Panel A. In column (1), we report estimates from a regression using the continuous composite optimism index (OPTIDX ) as the dependent variable. The set of independent variables includes the political affiliation indicators, political affiliation and political climate interaction terms, and various demographic control variables (age, education, race, gender, income, and wealth). The main variables of interest are the interactions between political affiliation (Republican, Democrat, and Independent 13 ) and D in Power dummy, which takes the value of one during the Democratic presidency. Our coefficients on these interaction terms (combined with the direct effects captured by the D in Power, Democrat, and Republican dummies) indicate that, relative to the Independents, Democratic investors’ responses as measured by the continuous optimism index are higher by 0.119 + 0.027 = 0.146 during the Democratic presidency and lower by 0.079 during the Republican presidency. In contrast, Republican responses are lower by 0.195 − 0.027 = 0.168 during the Democratic presidency and higher by 0.156 during the Republican presidency. These estimates suggest a significant shift in optimism levels across political regimes, consistent with the graphical evidence in Figure 1. The coefficient estimates of various control variables are also consistent with the prior evidence. In particular, similar to Jacobsen et al. (2008), we find that men are more optimistic. Individuals with higher income and wealth levels are also more optimistic, while older individuals are relatively less optimistic. In the next specification, for robustness, we use a composite optimism dummy (OP13

We exclude respondents who indicated a political affiliation other than Democrat, Republican, or Independent from the sample.

14

TIDXD), which is set to one if an individual is optimistic with respect to any one of its seven components. The marginal probabilities from the probit estimation are presented in column (2) of Table 3. These estimates are qualitatively similar to the estimates in column (1) and indicate that optimism levels are sensitive to the political environment. In columns (3) to (8), we report the results of additional regressions in which we use the individual components of the optimism index as dependent variables. We find that the interactions between political affiliation and political climate are significantly negative across all specifications. One concern in interpreting our results is that shifts in political regime are likely to be correlated with changes in economic conditions. It is possible that the differences we find in reported optimism reflect differences in the way innovations in real economic conditions are perceived by Democrat versus Republican respondents. To address this concern, we repeat the analysis from Panel A of Table 3 with explicit controls for macroeconomic conditions. Specifically, we construct an index following Korniotis and Kumar (2013) that captures macroeconomic innovations, with components reflecting income growth, relative unemployment, and the housing collateral ratio. We include this index and its interactions with the political affiliation dummies to control both for the direct effects of economic conditions and for any differences between Democratic and Republican investors in their response to economic conditions. Panel B of Table 3 reports the results of these regressions. We find that the coefficient estimates of the Political Affiliation × D in Power interactions remain largely significant when we explicitly control for economic conditions. Also of note, we find that the Macro Index interaction tends to be significant for Democrats across specifications, but is generally weak and insignificant for Republicans. This evidence suggests that while variation in economic conditions helps explain some of the variation in Democrats’ optimism, the shift in political power remains an important determinant of optimism for Democrats and is particularly important for Republicans. 15

3.3.

Political identity and perceptions of risk and reward

Next, we investigate whether people’s perceptions about risk and reward are influenced by their political affiliation and the current political climate. For this question, we use data from the Gallup surveys in 2002 because the Gallup surveys in other years do not contain questions about perceptions of market risk and undervaluation. The sample period, which covers March 2002 through December 2002, is during the George W. Bush presidential tenure. Because the sample period does not include a change in political regime, the data on perceptions of market risk and valuation are merely suggestive and cannot provide specific support for the influence of political climate. Table 4 reports the estimates from risk and undervaluation regressions. The dependent variable in these regressions is either a categorical measure of perceived risk or a dummy variable that is set to one if the respondent feels that the market is undervalued. We find that Republicans perceive the markets to be less risky and they are more likely to believe that the market is undervalued. The coefficient estimate of the Republican dummy variable is −0.261 (t-statistic = −3.18) in the risk regression (column (1)) and 0.081 (z-statistic = 3.20) in the undervaluation regression (column (3)). In contrast, the Democrat dummy has insignificant coefficient estimates in both specifications. In the market risk perception regression, the coefficient estimate of the Republican variable weakens and becomes insignificant when we add the optimism variable in the regression specification (see column (2)). The market undervaluation regressions show a similar but weaker pattern. This evidence indicates that the optimism index at least partially reflects the optimism generated by changes in the political climate. Overall, these results obtained using the Gallup data support our key hypothesis, which posits that individuals are more optimistic about the economy when their preferred political party is in power. The optimism regression estimates provide consistent evidence of a dy-

16

namic relation between political affiliation and optimism that is influenced by the political environment. The evidence on perception of market risk and valuation is also consistent with our optimism conjecture, but the limited sample period prevents any specific conclusions about the impact of political climate on perceptions of risk and reward.

4.

Political climate and investment decisions

In this section, we examine whether these shifts in people’s expectations have a direct impact on their investment decisions. Our key conjecture is that, due to higher levels of optimism, investors would hold riskier portfolios when their own party is in power. To test this conjecture, we focus on multiple dimensions of investing, including broad portfolio allocation decisions, style preferences, risk-shifting behavior, and local stock holdings. We also study the impact of these investment decisions on portfolio performance. In all these tests, we ensure that our results reflect the effects of shifts in the political environment rather than the effects of changes in economic environment that vary with the political climate.

4.1.

Political climate and broad portfolio allocation decisions

To begin, we study the broad asset allocation decisions of investors. Specifically, we examine whether shifts in investor optimism due to changing political climate affects their broad portfolio allocation decisions. Using data from the NLSY over the 1988 to 2000 period, we compute the aggregate market exposure of each investor portfolio, which is defined as the total value of allocations to stocks, bonds, and mutual funds as a proportion of total net worth. Panel A of Table 5 reports the average market exposures of Democratic and Republican portfolios during DCONTROL (1993-1994) and split government periods. The sample includes respondents who indicated a “Strong” affiliation with either the Democratic or the Republican Party. To control for potential variation in regional economic conditions or other unobserved regional variables, we regionally-adjust the risky portfolio allocations

17

by subtracting the mean value from the respondent’s Census region. In reporting the mean values in Table 5, we add in the national mean to make the magnitudes more meaningful without affecting the differences between Democrats and Republicans. We find that, unconditionally, Republican respondents expose their portfolios to higher market risks. The point estimate of the difference-in-difference between the market exposures of Democratic and Republican respondents across political regimes is positive, which suggests that investors’ market exposures are relatively higher when their preferred political party is in power. Unfortunately, with this somewhat crude measure of political climate, the difference between the market exposure estimates is not statistically significant. Given the relatively longer sample period of the NLSY data, we are able to consider an alternative and stricter definition of political environment. According to this new definition, political environment is captured by the number of branches out of the Presidency, the Senate, and the House of Representatives, that are controlled by the Democratic Party. Using this definition of political climate, we find that the difference-in-difference between Democratic and Republican investors across political regimes is positive and significant (see Panel B of Table 5). In economic terms, when the Democratic Party is in power (i.e., there is a Democratic president and a Democratic Congress), Democratic investors increase the market exposure of their portfolios by 1.24% more than Republican investors. Overall, while these results are not very strong, they are consistent with our conjecture that investors take greater financial risks when their preferred political party is in power and they feel more optimistic about the economy.

4.2.

Political affiliation and stock preferences

Next, we use the brokerage data to compare the stock preferences of investors located in highly Republican and Democratic regions. We use the brokerage data set for this analysis because it contains detailed information about both portfolio holdings and trading activi-

18

ties of investors. Because the brokerage data does not identify the political affiliations of investors, we use county-level voting data matched to each investor’s ZIP Code of residence to proxy for individual political affiliation. To assess the role of political values in the portfolio choices of individual investors, we follow Hong and Kostovetsky (2012) and define politically sensitive stocks as those belonging to tobacco, guns and defense, and natural resources industries. Hong and Kostovetsky find that Republican mutual fund managers tend to give greater weight in their stock portfolios to politically sensitive industries relative to Democratic fund managers. The tobacco industry includes producers and sellers of tobacco products (SIC codes 0132, 21xx, 5194, and 5993). Guns and defense industry includes manufacturers of firearms and ammunition (SIC codes 348x), manufacturers of military vehicles and guided missiles (SIC codes 376x and 3795), and major defense contractors.14 The natural resources category includes logging, forestry, and mining industries (SIC codes 0800-1499 and 2411). We create a politically sensitive stock dummy that equals one if the stock belongs to any of these three industries, and zero otherwise. During the 1991-1996 brokerage sample period, politically sensitive stocks represent 7.17% of the aggregate market. To characterize the stock preferences of Republican and Democratic investors, we estimate fixed effects panel regressions in which the dependent variable is the excess weight (relative to the stock’s weight in the market portfolio) assigned to a stock in the aggregate group portfolio. The aggregate group portfolio is constructed by combining the portfolios of all investors within the group. For example, the aggregate Republican portfolio is constructed by combining the portfolios of all investors who are in the bottom quintile of the % Democrat measure. The main independent variable is the politically sensitive stock dummy. Other independent variables include market beta, firm size, book-to-market ratio, past 114

This set includes well-known firms such as Lockheed Martin, Boeing, Northrop Grumman, Raytheon, General Dynamics, United Technologies, SAIC, TRW, L-3 Communications, Honeywell, Hughes Electronics, Rockwell International, and Textron.

19

month and 12-month stock returns, stock price, idiosyncratic volatility, firm age, indicators for whether the stock belongs to the S& P 500 or pays dividends, monthly share turnover, and return skewness. The panel regression estimates presented in Table 6 indicate that the politically sensitive stock dummy has a positive coefficient estimate when we consider the stock weights in the aggregate Republican portfolio (columns (1) and (4)). In contrast, the politically sensitive stock dummy is significantly negative when we consider the stock weights in the aggregate Democratic portfolio (columns (2) and (5)). This evidence indicates that the excess weight allocated to politically sensitive firms does not merely reflect investors’ known preferences for other stock attributes. Examining the coefficient estimates of other stock attributes, we find that investors in Republican areas hold riskier stocks. In particular, relative to the Democratic investors, they exhibit a stronger preference for firms that are smaller, younger, have higher idiosyncratic volatility, and higher skewness. This evidence indicates that politically “conservative” investors do not hold “conservative” stock portfolios. In economic terms, we find that Republican investors in our sample overweight politically sensitive stocks by 0.090 (column (4)). Relative to the mean excess weight of 0.923 for all stocks in the aggregate Republican portfolio, this represents a 9.75% increase. In contrast, investors in Democratic areas underweight politically sensitive stocks by 0.032 (column (5)), which reflects a 4.56% decrease relative to the mean excess weight of 0.701 for all stocks in the aggregate Democratic portfolio. Overall, these stock preference regression estimates indicate that political values affect the stock preferences of individual investors. Our results are consistent with the evidence in Hong and Kostovetsky (2012), who show that political values influence the stock preferences of mutual fund and hedge fund managers. Using a different political identity identification strategy (local voting patterns instead of political contributions), a different group of investors (individuals investors instead of money managers), and a different time period, we 20

show that political values affect investment decisions. Most importantly, these results suggest that our geography-based political affiliation identification procedure works reasonably well.

4.3.

Results from difference-in-difference tests

In this subsection, we examine changes in the stock preferences of Democratic and Republican investors across political regimes. Specifically, we test whether the stock portfolios of Democratic investors exhibit higher market betas when the political regime switches from Republican to Democratic, and vice versa for Republican investors. Such risk-shifting behavior by Republicans and Democrats would arise naturally if investors’ perceptions of risk and reward depend upon their political affiliation and the current political climate. For each investor group, Democrat and Republican, we compute an ex ante portfolio beta as the linear combination of stock betas in the group’s aggregate portfolio, weighted according to each stock’s weight in the aggregate portfolio. We include the stock holdings of households in the highest quintile of the Democrat measure, defined as the fraction of votes cast for the Democratic presidential candidate in the county where the household resides, in the aggregate Democrat portfolio. Similarly, we form the aggregate Republican portfolio from the stock holdings of households in the lowest quintile of the Democrat measure. We divide the sample period into a DCONTROL period from December 1992 to November 1994, during which the Democratic Party held the Presidency and controlled both houses of Congress, and a split government period covering January 1991 to November 1992 and December 1994 to November 1996.15 This definition of political regime is based on the 15 The sample period for the brokerage data is from January 1991 to November 1996. George H. W. Bush, a Republican, served as President during 1991 and 1992, followed by Bill Clinton, a Democrat, who took office in January of 1993 and presided through the end of the sample period. Meanwhile, both houses of Congress were controlled by the Democratic Party through 1994, after which the Republican party held a majority in both houses through the end of the sample period. Thus, the Democratic Party controlled both the Presidency and Congress during 1993 and 1994, while the periods before and after this interval where characterized by a split government (i.e., a Republican president and Democratic Congress during 1991-1992, and a Democrat president and Republican Congress during 1995-1996).

21

assumption that both branches of government are likely to influence an investor’s level of economic optimism and perception of economic uncertainty. However, it is possible that the political party of the President is more salient in the minds of voters than the party that controls the Congress. We find qualitatively similar results when we consider a political regime measure based solely on the political party of the sitting President. For each portfolio (Democrat and Republican), we compute the mean portfolio beta in the DCONTROL period and in the split government period. Further, we compute the difference in average portfolio beta between Democrat and Republican households during each regime and also obtain the difference-in-difference estimate. Table 7 shows the results of these difference-in-difference tests. We find that the average portfolio beta of Democratic households during the DCONTROL period is 0.982, compared to 0.957 for Republicans. During the split government period, Democratic investors held portfolios with an average beta of 0.944 compared to 0.930 for Republican households. The difference-in-difference estimate is positive (= 0.012) and significant (t-statistic = 4.08), indicating that Democratic investors maintain relatively higher market risk exposure during the DCONTROL period.16 We observe similar effects when we examine the four factor portfolio betas with respect to the RMRF, SMB, HML, and UMD factors. In each case, the difference-in-difference estimate is positive and significant, consistent with the conjecture that investors have higher systematic risk exposures when their own party is in power. The exception is the momentum (UMD) betas, where the difference-in-difference estimate is weakly negative but statistically insignificant.

4.4.

Additional graphical evidence

To provide further economic intuition, we illustrate the portfolio beta shifts graphically. We use the four-factor model to estimate the factor exposures of investor portfolios and our 16

Following the recommendations of Bertrand et al. (2004), we compute the standard errors using the Newey and West (1987) correction where we allow for all lags to be potentially important.

22

proxy for the political identity of an investor is the proportion of the county-level population that voted for the Democratic party. We compute the portfolio betas for each household portfolio separately when Democrats are in control (DCONTROL = 1) and when there is a split government (DCONTROL = 0). To facilitate meaningful comparisons across the various risk measures, we standardize the portfolio beta measures separately for the two periods (mean is set to zero and the standard deviation is one). Figure 2 shows the mean factor exposure differentials for political identity sorted household groups. The mean factor exposure differences across the two political regimes are shown in the plot. We find that across all four portfolio risk measures, there is an almost monotonic relation between our political affiliation proxy and the beta difference across the two political regimes. Investors in Republican areas exhibit lower portfolio risk, while those in regions with high concentration of Democrats exhibit higher portfolio risk, when Democrats come to power. The cross-sectional difference is also evident distinctly when we examine its time series. Figure 3 shows the time series of the mean difference in the portfolio betas of Democratic and Republican investors. For brevity, we only show the results for market beta and HML beta differential measures, which provide the strongest patterns. The lighter line shows the time series using the raw data and the darker line shows the five-month moving average of this series. The time series plots show that both the market beta and the HML beta differences between Democrats and Republicans are higher during the period of Democratic control (the shaded region).

4.5.

Household-level panel regressions

The difference-in-difference tests and additional graphical evidence are based on aggregate Democrat and Republican portfolios, which has the advantage of accounting for potential cross-sectional dependence among investors. To control for individual household character-

23

istics which might influence the observed shifts in portfolio risk exposures, we also estimate household-level fixed effects panel regressions. The primary dependent variable is an ex ante measure of portfolio beta, which we compute as the weighted average of the CAPM betas of the individual stocks in the household’s portfolio, estimated using monthly returns over the prior 48 months. We also consider the market, SMB, HML, and UMD betas of household portfolios, estimated using a four-factor regression over the same window. The primary independent variables are measures of political affiliation (Democrat, High Democrat dummy, and High Republican dummy) interacted with DCONTROL.In several specifications, we include additional investor demographic characteristics as control variables, each of which are also interacted with DCONTROL. In these instances, the main effects of the investor characteristics, including political affiliation measures, drop out because of the household fixed effects. These investor characteristics include the age and gender of the head of household as well as household income, which we obtain directly from the brokerage data. We also include proxies for the household’s education level and race based on ZIP Code-level census data corresponding to the household’s residence. Intuitively, the coefficient estimates in our fixed effect regression specifications capture the difference in the average portfolio choice of a certain group of investors (e.g., Republican or older investors) during the December 1992 to November 1994 period of full Democrat control (i.e., when DCONTROL is one) and outside of that period (i.e., when DCONTROL is zero). Specifically, the DCONTROL × High Republican and DCONTROL × High Democrat interaction terms are used to test whether after controlling for other demographic characteristics, the change in the average portfolio variable across the two political regimes is significantly stronger for Republican or Democratic investors, respectively. Our sample consists of a large panel with repeated household observations over time, and as such, the observations in the sample may not be entirely independent. In particular, because most households do not adjust their portfolios frequently and household preferences 24

are likely persistent, measures of portfolio risk are likely correlated across time for a given household. Further, because investors with the same political preferences tend to hold similar types of stocks (as shown in Table 6), there may be cross-correlation in systematic risk measures among households with the same political orientation in a given month. We account for these potential sources of dependence among the observations in the sample by clustering the standard errors both by household and by party-month.

4.6.

Baseline regression estimates

The regression estimates reported in Panel A of Table 8 support the hypothesis that investors maintain higher market risk exposures when their own political party is in power. The positive and significant coefficient on DCONTROL × Democrat indicates that investors in strongly Democratic areas increase the riskiness of their portfolios relative to investors in Republican areas when the Democratic Party fully controls the federal government. In economic terms, the portfolio beta regression estimates indicate that, relative to other investors in the sample, an investor who lives in an area that is in the 90th percentile of the % Democratic measure (= 0.694) increases the portfolio beta by an average of 6.083 × 0.694/100 = 0.042 during the period of Democratic control (see column (4)). Relative to the mean beta estimate of 0.985, this represents an average beta differential of 2.60%. In contrast, the portfolio beta of an investor who lives in a highly Republican region is on average −0.016 lower than other investors in the sample (see column (6)), which represents an average beta differential of 1.61% relative to the mean. Thus, consistent with our conjecture, the mean spread between the portfolio betas of Democratic and Republican investors widens when Democrats are in power. To better understand the risk-shifting behavior of investors, we examine the shifts in other attributes of portfolio risk. Specifically, we study whether investors’ preferences for riskier small-cap and value styles become stronger when the political climate is aligned with

25

their political preferences. Like the portfolio beta measure used above, we estimate the factor exposures of each stock using a four-factor model that includes the market factor (RMRF ), the size factor ( SMB ), the value factor (HML), and the momentum factor (UMD). Using these factor exposure estimates, we obtain the market, size, value, and momentum tilts of each investor portfolio during both political regimes. In Panel B of Table 8, we summarize the results from the household fixed effects regressions where one of these four portfolio factor exposures is the dependent variable. With the exception of the UMD momentum factor, we find that political affiliation and regime play a similar role with respect to these measures of systematic risk exposure. For example, the coefficient estimates in column (5) indicate that the average portfolio HML beta of investors in highly Democratic regions is 0.015 higher (relative to other investors) when Democrats are in control. In contrast, column (6) estimates indicate that the average HML beta of investors in highly Republican regions is 0.023 lower relative to other investors when Democrats are in control. Relative to the mean HML beta of −0.045, these beta shifts are economically significant and support our main conjecture. Collectively, the household-level portfolio beta regression estimates provide support for our hypothesis that political affiliation and the existing political climate jointly influence investors’ optimism, which in turn affect their portfolio risk exposures and style preferences. Specifically, the greater economic optimism that investors feel when their own political party is in power leads them to assume relatively higher exposures to market risk, as well as smallcap and value styles.

4.7.

Results using alternative specifications

For robustness, we consider two alternative regression specifications. The first alternative specification uses time fixed effects rather than household fixed effects. This specification tests whether the portfolio choice of Democrats relative to Republicans varies significantly

26

across political regimes. The second alternative procedure collapses the monthly decisions into a single measure per household. This single household-level measure is the difference in the mean portfolio choice between the full Democratic control period and the split government periods. We then estimate cross-sectional regressions of these portfolio choice differences on the political affiliation measures and other household characteristics. The results are summarized in Table 8, Panel B. Typically, our results with time fixed effects models are qualitatively similar to and often stronger than the results using our main specifications that use household fixed effects. The results are also qualitatively similar when we estimate cross-sectional regressions but the statistical significance of our estimates weaken mainly due to reduced sample size. Because all households are not present in the sample for the full six-year period, for many households we are unable to estimate their portfolio decisions during both political regimes.17 In those cases, the portfolio choice difference measure across the two political regimes cannot be computed.

4.8.

Shifts in political optimism or economic optimism?

One concern in the interpretation of our results is that shifts in political climate are not exogenous. Political regime shifts are likely to be correlated with innovations in economic conditions and other external events. It is possible that differences in preferences or investment styles of Democratic and Republican investors lead them to respond differently to the same economic conditions. Thus, the variation in portfolio decisions that we observe may actually reflect differential responses of Democratic and Republican investors to changes in economic conditions rather than differences in confidence based on the current political regime. To ensure that our main results reflect differences in optimism that derive from changes in the political climate rather than economic conditions, we repeat our analysis with explicit 17

Some households enter in the middle of the sample period and some leave the sample in the middle.

27

controls for macroeconomic conditions. As in Table 3, we construct an index motivated by Korniotis and Kumar (2013), which captures macroeconomic innovations with components reflecting income growth, relative unemployment, and the housing collateral ratio. We report results using a local (state-level) version of this index, but find similar results when we use a national-level macroeconomic index. We add this index as well as the interaction between this index and the household’s political affiliation to the existing regression specifications. These additional terms control both for the direct effects of economic conditions and for any differences between Democratic and Republican investors in their response to economic conditions. We find that the coefficient estimates of the DCONTROL × Political Affiliation interaction remains largely unchanged, and in some cases, the estimates are even stronger when we explicitly control for economic conditions. Another related concern is that, rather than simply being optimistic about the economy due to one’s preferred political party being in power, investors are expecting that the party in power will reward its districts with disproportionate federal spending. We examine this issue directly using data on political alignment and on the allocation of federal funds to states, and find that there is some positive relationship between state-level federal spending (per dollar of federal taxes paid) and state-level political alignment with the party of the president (measured using the political alignment index from Kim et al. (2012), which measures the fraction of state elected officials and members of Congress that belong to the same political party as the president). In a simple univariate regression with state fixed effects, we find that a one-standard-deviation increase in state political alignment is associated with federal spending that is higher by 0.04 standard deviations (t-statistic = 2.22). The relation is statistically significant but economically small. Further, using our index of state-level macroeconomic conditions (relative unemployment, housing collateral ratio, and state income growth), we find no relation between state economic conditions and current or lagged federal spending or political alignment. Thus, even though 28

there is a slight increase in federal spending for states that are more aligned with the party of the president, this does not translate into any real benefit that we can observe in state-level economic indicators. While investors may expect to benefit more from preferential federal spending when their preferred party is in power, any benefit appears to be more perceived than real, which is consistent with our optimism conjecture. Nevertheless, to ensure that our results are not driven by preferential federal spending, we also control for state-level federal spending (per dollar of taxes paid) directly in our regressions. We report these results in Table 9. We find that when the market beta is the dependent variable, DCONTROL × Democrat interaction has a coefficient estimate of 8.359 (t-statistic = 3.60) in the extended specification. In contrast, DCONTROL × Democrat has a coefficient estimate of 6.083 (t-statistic = 2.33) in column (4) of Table 8. Overall, these results indicate that the variation in portfolio choices we observe are induced by changes in the political climate and investors’ own political preferences, rather than simply a differential response to economic conditions or expectations of federal spending. A related concern in the interpretation of our results is that Republican- and Democratleaning areas may experience different economic conditions, and that our results may simply reflect regional variation in economic conditions rather than differences in optimism based on household political views with respect to the current political regime. When we include an analogous index of macroeconomic innovations defined at the state-level in our original specifications, we again find little change in the joint effects of the political regime and political affiliation. We report these results in Panel B of Table 9. For example, when the CAPM beta is the dependent variable (see Table 8), DCONTROL × Democrat interaction has coefficient estimates of 8.359 (t-statistic = 3.60) and 6.083 (t-statistic = 2.33) in the extended and original specifications, respectively. This evidence indicates that our findings do not merely reflect the geographical variation in economic climate across the U.S, nor variation in federal 29

allocations to states aligned with the political party in power.

4.9.

Political climate and active portfolio adjustment

In this subsection, we examine the degree to which the observed shifts in relative market exposures result from active portfolio changes, as opposed to passive changes in the betas of the stocks investors hold. We estimate the active component of the monthly changes in households’ portfolio betas using the following regression:

∆βit = γZ Zit + γt t + εit ,

(1)

where ∆βit = βit − βi,t−1 is the change in beta of the investor’s portfolio between months t − 1 and t. Zit is a set of two dummy variables which equal (i) one if the investor purchased stocks in month t and, zero otherwise, and (ii) one if the investor sold stocks in month t and, zero otherwise. t is a set of time fixed effects. The time fixed effects control for any systematic changes in beta that might occur passively due to the type of stocks being held. The demographic characteristics of investors are not included in equation (1) because they are constant over time and drop out of the difference equation specification. ? The quantity γZ Zit represents an estimate of the contribution of active trading to the change in households’ portfolio betas. A positive value of γZ Zit would indicate that investors’ trading is associated with higher portfolio betas, while a negative value would indicate that trading is associated with lower portfolio betas. We estimate γZ Zit separately for four subsamples: (i) Democratic investors during the DCONTROL period, (ii) Democratic investors during the split government periods, (iii) Republican investors during the DCONTROL period, and (iv) Republican investors during the split government periods. Note that by estimating the active changes separately for each subsample, the time fixed effects control for passive beta changes that could arise from any tendencies across political affiliation groups or time periods in the types of stocks held. 30

Table 10 reports the γZ Zit estimates for each subsample as well as the differences across subsamples. γZ Zit is positive (= 0.0061) for Democratic investors during the DCONTROL period, when we expect them to be more willing to take on systematic risk, and negative (= −0.037) during the split government periods. Thus, trading by Democratic investors is associated with increases in portfolio betas when their preferred party is in control, and with reductions in portfolio betas during the split government periods. Similarly, the changes in portfolio betas load more positively on the trading indicators for Republicans during the split government periods than during the period of unified Democratic control. Most importantly, we find that the difference-in-difference estimate between Democrats and Republicans across political regimes is positive and significant. Specifically, the difference in the active component of portfolio beta changes between Democratic and Republican is higher during the DCONTROL period by 0.0116 (t-statistic = 4.67). We find qualitatively similar results for RMRF, SMB, HML, and UMD betas from a four-factor model, although the results are weaker and statistically insignificant (t-statistic = 1.39) for the HML beta. It should be noted that our estimates of γZ Zit simply represent the component of portfolio beta changes that are associated with active trading, and cannot be interpreted as causal. A more direct way of addressing whether the investors are actively adjusting portfolio risk across political regimes is to examine the betas of stocks purchased and sold. In unreported results, we find that the difference in the betas of purchased stocks between Democrats and Republicans is significantly higher during the DCONTROL period. However, the differencein-difference estimate for the betas of stocks sold is also significantly positive. Thus, we do not find direct evidence that investors are actively increasing their portfolio betas by selling low-beta stocks and buying high-beta stocks when their preferred party is in power. Rather, the betas of both stocks purchased and stocks sold tends to be higher, so that the direct net impact that we can observe from trades is neutral. Taken together, a conservative interpretation of our findings is that investors appear to 31

both hold and trade stocks that exhibit higher betas when their preferred party is in power. While the shifts in portfolio betas that we observe across political regimes may be partly passive, investors’ trading activity at least maintains, if not amplifies, the changes in market exposure.

4.10.

Local stock preference and flight to familiarity

In this subsection, we examine whether shifts in relative optimism across political regimes affect investors’ bias toward local stocks. This test is partly motivated by the evidence in Kumar (2009a), who shows that investors exhibit a stronger preference for domestic stocks as well as local stocks when the level of economic uncertainty is high. We estimate the same set of fixed effects regressions as in the previous section, but use a measure of local bias as the dependent variable. We construct a monthly, investor-level measure of local bias by computing the average distance between an investor’s location and the firms she chooses to hold in her portfolio. This distance is then scaled by the expected distance to a portfolio with similar style in terms of size, book-to-market, momentum, and industry composition. A more detailed definition of the local bias measure is provided in the Appendix. Table 11 reports the local bias regression estimates. The negative coefficient estimates for DCONTROL × Democrat and DCONTROL × High Democrat interaction terms indicate that investors in highly Democratic areas have 0.0139 lower average local bias relative to other investors when their own party is in power (see column (2)). Relative to the mean local bias of 0.195 for the full sample, this represents a 7.12% differential. In contrast, investors in highly Republican neighborhoods on average have 0.0082 higher local bias during the period of Democratic control (see column (3)), which reflects a 4.21% differential relative to the mean local bias. Overall, our local bias results are consistent with the conjecture that investors exhibit a stronger preference for local stocks when they become relatively pessimistic about the econ-

32

omy due to a change in the political climate. This behavior may reflect investors simply taking comfort in more familiar stocks when they have a poor economic outlook. Alternatively, it may reflect stronger attempts by investors to pursue perceived informational advantages by investing in local stocks when they are relatively less optimistic about the performance of the broader market.

4.11.

Political climate and portfolio performance

Finally, we test whether the relative portfolio performance of Democrat and Republican investors differs across political regimes. In Table 12, we present coefficient estimates from a set of performance regressions. In columns (1)-(3), the dependent variable is the marketadjusted monthly return on the investor’s portfolio, net of transaction costs. In columns (4)-(6), the dependent variable is the Daniel et al. (1997) characteristic-adjusted net return. The independent variables are identical to those used in the previous regressions. We find that the market-adjusted returns of Democrats increase when the political regime switches from being Republican to Democratic. During the same period, investors in strongly Republican areas experience a decline in the market-adjusted performance. For example, the coefficient estimate of DCONTROL × High Republican interaction dummy (estimate = −0.080, t-statistic = −2.35) indicates that investors located in strongly Republican regions earn 0.080 × 12 = 0.96% lower incremental return on an annualized basis when Democrats are in power. This evidence indicates that investors who decrease the riskiness of their portfolios earn lower market-adjusted returns and is consistent with our previous evidence on the beta shifts across regimes.18 Our results are also consistent with the evidence obtained using self-reported performance measures in the Gallup data. In both instances, investors earn higher raw or market-adjusted returns when their own party is in power. 18

We obtain very similar results when we do not account for market risk and estimate the performance regressions using raw returns.

33

The performance differences become weaker when we use characteristic-adjusted returns to measure performance. For example, in economic terms, the coefficient estimate of DCONTROL × High Republican interaction dummy (estimate = −0.050, t-statistic = −1.47) indicates that investors located in highly Republican regions earn 0.050 × 12 = 0.60% lower incremental return on an annualized basis when Democrats are in power. However, the coefficient estimate is statistically very weak. These performance regression estimates indicate that the risk-shifting behavior of investors is less likely to be strategic and does not have an economically significant effect on their overall portfolio performance once we properly account for differences in portfolio risk. Alternatively, it is likely that investors earn consistently better risk-adjusted return when their own party is in power but our voting-based procedure for identifying political affiliation is noisy and lacks strong statistical power.

5.

Conclusion

We find evidence that the prevailing political climate combined with investors’ political affiliations jointly influence their optimism towards financial markets and the macroeconomy. To our knowledge, this is the first study to document the dynamic and differential impact of political environment on optimism and investment decisions. We find that investors become more optimistic and perceive the markets to be less risky and more undervalued when their own party is in power. These shifts in perceptions of risk and reward affect investors’ portfolio decisions. Specifically, when the political climate is aligned with their political identity, investors increase allocations to riskier assets and exhibit a stronger preference for high market beta, small-cap, and value stocks. While some of the observed shifts in portfolio betas we observe may be passive in nature, we find that investors’ actively maintains their higher risk exposures through their trading during periods when their preferred party is in power. We also demonstrate that when investors’ 34

preferred party is not in power, they exhibit a stronger preference for familiar local stocks. Due to these portfolio reallocations, investors improve their raw portfolio performance when their own political party is in power, but the improvement in risk-adjusted performance is economically small. In future work, it would be interesting to examine whether the changing political climate also affects the portfolio decisions of institutional investors. In addition, it would be interesting to examine whether the influence of changing political climate extends beyond the investment decisions of retail and institutional investors to the aggregate market. For example, if Democrats and Republicans have distinct preferences for certain types of stocks, those stocks could exhibit different behaviors across political regimes. Specifically, stocks favored by Democrats may become overpriced when Democrats are in power due to an increase in the optimism of Democratic investors. The turnover and liquidity of those stocks could also be affected by shifts in optimism. More broadly, our evidence of a connection between political affiliation and portfolio choice may have implications for the real business cycle. Specifically, a shift in the political regime may cause certain subsets of investors to be systematically less optimistic about the U.S. economy and grow relatively more optimistic about foreign economies. This optimism shift may subsequently generate capital outflows from certain sectors or industries of the U.S. economy into foreign countries. It would be useful to incorporate political climate-induced time-varying optimism into real business cycle models.

35

References Amromin, G., and S. A. Sharpe. 2008. Expectations of risk and return among household investors: Are their Sharpe ratios countercyclical? Finance and Economics Discussion Series 2008-17, Board of Governors of the Federal Reserve System (U.S.). URL http://ideas.repec.org/p/fip/fedgfe/2008-17.html. Barber, B. M., and T. Odean. 2000. Trading is Hazardous to Your Wealth: The Common Stock Investment Performance of Individual Investors. Journal of Finance 55:773–806. Barber, B. M., and T. Odean. 2001. Boys will be Boys: Gender, Overconfidence, and Common Stock Investment. Quarterly Journal of Economics 116:261–292. Bertrand, M., E. Duflo, and S. Mullainathan. 2004. How Much Should We Trust DifferencesIn-Differences Estimates? The Quarterly Journal of Economics 119:249–275. URL http://qje.oxfordjournals.org/content/119/1/249.abstract. Carroll, C. D. 2003. Macroeconomic Expectations Of Households And Professional Forecasters. The Quarterly Journal of Economics 118:269–298. URL http://ideas.repec.org/a/tpr/qjecon/v118y2003i1p269-298.html. Daniel, K. D., M. Grinblatt, S. Titman, and R. Wermers. 1997. Measuring Mutual Fund Performance with Characteristic-Based Benchmarks. Journal of Finance 52:1035–1058. Dhar, R., and N. Zhu. 2006. Up, Close and Personal: An Individual Level Analysis of the Disposition Effect. Management Science 52:726–740. Driscoll, J. C., and A. C. Kraay. 1998. Consistent Covariance Matrix Estimation with Spatially Dependent Panel Data. Review of Economics and Statistics 80:549–560. Fama, E. F., and K. R. French. 1997. Industry Costs of Equity. Journal of Financial Economics 43:153–193. Graham, J. R., C. R. Harvey, and H. Huang. 2009. Investor Competence, Trading Frequency, and Home Bias. Management Science 55:1094–1106. Graham, J. R., and A. Kumar. 2006. Do Dividend Clienteles Exist? Evidence on Dividend Preferences of Retail Investors. Journal of Finance 61:1305–1336.

36

Hilary, G., and K. W. Hui. 2009. Does Religion Matter in Corporate Decision Making in America? Journal of Financial Economics 93:455–473. Hong, H. G., and L. Kostovetsky. 2012. Red and Blue Investing: Values and Finance. Journal of Financial Economics 103:1–19. Jacobsen, B., J. B. Lee, and W. Marquering. 2008. Are Men More Optimistic? Working Paper (February), Erasmus University; Available at SSRN: http://ssrn.com/abstract=1030478. Jones, J. M. 2009. In U.S., Cautious Optimism About Economy in Year Ahead. Gallup.com September 28. Available at http://www.gallup.com/poll/123209/cautious-optimismabout-economy-year-ahead.aspx. Kaustia, M., and S. Torstila. 2011. Stock Market Aversion? Political Preferences and Stock Market Participation. Journal of Financial Economics 100:98–112. Kim, C. F., C. Pantzalis, and J. C. Park. 2012. Political geography and stock returns: The value and risk implications of proximity to political power. Journal of Financial Economics 106:196–228. URL http://www.sciencedirect.com/science/article/pii/S0304405X12000785. Korniotis, G. M., and A. Kumar. 2011. Do Behavioral Biases Adversely Affect the MacroEconomy? Review of Financial Studies 24:1513–1559. Korniotis, G. M., and A. Kumar. 2013. State-Level Business Cycles and Local Return Predictability. Journal of Finance 68:1037–1096. Kumar, A. 2009a. Hard-To-Value Stocks, Behavioral Biases, and Informed Trading. Journal of Financial and Quantitative Analysis 44:1375–1401. Kumar, A. 2009b. Who Gambles in the Stock Market? Journal of Finance 64:1889–1933. Newey, W. K., and K. D. West. 1987. A Simple, Positive semi-Definite Heteroskedasticity and Auto-Correlation Consistent Variance-Covariance Matrix. Econometrica 55:703–708. Puri, M., and D. Robinson. 2007. Optimism and Economic Choice. Journal of Financial Economics 86:71–99.

37

Strong, N., and X. Xu. 2003. Understanding the Equity Home Bias: Evidence from Survey Data. Review of Economics and Statistics 85:307–312. Vissing-Jorgensen, A. 2003. Perspectives on Behavioral Finance: Does “Irrationality” Disappear with Wealth? Evidence from Expectations and Actions. NBER Macroeconomics Annual 18:139–194. Zuckerman, G. 2009. Pessimism Exacts a Price on the Skeptics. The Wall Street Journal September 28:C1.

38

0.4

Optimism differential (smoothed) Optimism differential (raw)

Democrat − Republican optimism differential

0.3

0.2

0.1

0

−0.1 −0.2 −0.3 −0.4 −0.5 −0.6

Clinton Pesidency

Bush Pesidency

−0.7 −0.8

Oct96 Feb97 Mar98 Jan00 Mar00 May00 Jul00 Sep00 Dec00 Feb01 Apr01 Jun01 Aug01 Oct01 Dec01 Feb02 Apr02 Jun02 Aug02 Oct02 Dec02

Calendar time (October 1996 to December 2002) Figure1. 1. Optimism Optimism shifts shifts around around change change in in political political regime Figure regime This figure figure shows shows the the difference difference in in reported reported optimism optimism regarding This regarding the the economy economy between between Democratic Democratic and and Republican survey respondents in the UBS/Gallup Investor Optimism Index. Optimism is defined as Republican survey respondents in the UBS/Gallup Investor Optimism Index. Optimism is defined asaa dummy variable variable that that equals equals one one ifif the the respondent respondent was was “somewhat” dummy “somewhat” or or “very” “very” optimistic optimistic with with respect respecttoto stockmarket marketperformance, performance, economic economic growth, growth, income, income, employment, stock employment, investment investment goals, goals, or orinflation inflationduring duringthe the subsequent 12 months. The dark line indicates the smoothed (five-month moving average) difference in subsequent 12 months. The dark line indicates the smoothed (five-month moving average) difference in average optimism between Democrats and Republicans over the sample period. The solid vertical line marks average optimism between Democrats and Republicans over the sample period. The solid vertical line marks the start of George W. Bush’s presidency. Please note that there is a gap in our data for 1999, when Gallup the start of George W. Bush’s presidency. Please note that there is a gap in our data for 1999, when Gallup did not record the political affiliation of survey respondents. We have compressed this gap in the plot, so did not record the political affiliation of survey respondents. We have compressed this gap in the plot, so that the data point for December 1998 is followed directly by the data point for January 2000. This does that the data point for December 1998 is followed directly by the data point for January 2000. This does not affect the interpretation of the plot, as the optimism differential is similar before and after the gap in not affect the interpretation of the plot, as the optimism differential is similar before and after the gap in the data. the data.

39 39

Market beta

SMB beta 2

Beta difference

Beta difference

2 1 0 −1 −2

0 −1 −2

REP D2 D3 D4 D5 D6 D7 D8 D9 DEM

REP D2 D3 D4 D5 D6 D7 D8 D9 DEM

HML beta

UMD beta 2

Beta difference

2

Beta difference

1

1 0 −1 −2

1 0 −1 −2

REP D2 D3 D4 D5 D6 D7 D8 D9 DEM

REP D2 D3 D4 D5 D6 D7 D8 D9 DEM

Political identity decile

Political identity decile

Figure 2. Political identity and beta differential across political regimes This figure shows the mean factor exposure differentials for political identity sorted household groups. We use the four-factor model to estimate the factor exposures and the proxy for political identity of an investor is the proportion of county-level population that voted for the Democratic party. We compute the portfolio betas for each household portfolio when Democrats are in control (DCONTROL = 1) and when there is a split government (DCONTROL = 0). The average values of the portfolio factor exposure differences are shown. To facilitate meaningful comparisons across the various risk measures, we standardize the portfolio beta measures separately for the two periods (mean is set to zero and the standard deviation is one).

40

Panel beta Panel A: Market beta Panel A: A: Market Market beta 0.05 0.05

Portfolio Portfoliomarket marketbeta beta(Democrat (Democrat−−Republican) Republican)

DCONTROL DCONTROL Market beta differential Market beta differential Market beta differential Market beta differential

0.04 0.04

(smoothed) (smoothed) (raw) (raw)

0.03 0.03

0.02 0.02

0.01 0.01

0 0Jan91 Jan91

Jul91 Jul91

Jan92 Jan92

Jul92 Jul92

Jan93 Jan93

Jul93 Jul93

Jan94 Jan94

Jul94 Jul94

Jan95 Jan95

Jul95 Jul95

Jan96 Jan96

Jul96 Jul96

Jan95 Jan95

Jul95 Jul95

Jan96 Jan96

Jul96 Jul96

Portfolio PortfolioHM HMLLbeta beta(Democrat (Democrat−−Republican) Republican)

Panel L beta Panel B: B: HM HM L Panel B: HM L beta beta 0.02 0.02 0.01 0.01

DCONTROL DCONTROL HM L beta differential HM L beta differential HM L beta differential HM L beta differential

(smoothed) (smoothed) (raw) (raw)

0 0

-0.01 -0.01 -0.02 -0.02 -0.03 -0.03 -0.04 -0.04 -0.05 -0.05 Jan91 Jan91

Jul91 Jul91

Jan92 Jan92

Jul92 Jul92

Jan93 Jan93

Jul93 Jul93

Jan94 Jan94

Jul94 Jul94

Figure 3. series of Democrat − Republican beta differential Figure 3. Time Time series of of Democrat betadifferential differential Figure 3. Time series Democrat−−Republican Republican beta

This figure shows the time series of the mean difference in the portfolio betas between Democratic This This figure shows thethe time series in the theportfolio portfoliobetas betas between Democratic figure shows time seriesofofthe themean mean difference difference in between Democratic and Republican investors. The period of Democratic control (DCONTROL = 1) is shaded. The dashed Republican investors. The periodofofDemocratic Democratic control control (DCONTROL is shaded. TheThe dashed line Republican investors. The period (DCONTROL= =1) 1) is shaded. dashed shows the time series using the raw data and the solid line shows the five-month moving average shows the time series using the raw data and the solid line shows the five-month moving average of this shows the time series using the raw data and the solid line shows the five-month moving average of of series. The market betas and the HM L betas are estimated using the four-factor model. series. The market betas and the HM L betas are estimated using the four-factor model. series. The market betas and the HM L betas are estimated using the four-factor model.

41

41 41

and and line line this this

Table 1 Summary statistics This table presents the main summary statistics for the UBS/Gallup and discount brokerage data sets. Panel A reports statistics for variables from the UBS/Gallup Investor Optimism Survey, while Panel B reports statistics for a sample of investors from a large U.S. discount brokerage. The sample period for the Gallup survey data is from 1996 to 2002, except for the following measures: (i) perceived risk, which is available for the March to December 2002 period and (ii) perception of undervaluation, which is available only in March, June, and September of 2002. The sample period for the brokerage account data is from January 1991 to November 1996. The Appendix provides brief definitions of all variables.

42

Table 1 (Continued)

Summary statistics

Panel A: Gallup survey sample

Percentile Variable

Mean

Optimism measures (scaled response) Optimism index (OPTIDX ) 3.61 Stock market (MKTOPT ) 3.36 Economic growth (GROPT ) 3.49 Employment (EMPOPT ) 3.41 Income (INCOPT ) 3.96 Inflation (INFLOPT ) 3.35 Short-term investment (SINVOPT ) 3.69 Long-term investment (LINVOPT ) 4.01 Optimism measures (dummy) Optimism index (OPTIDXD) 0.607 Stock market (MKTOPTD) 0.545 Economic growth (GROPTD) 0.595 Employment (EMPOPTD) 0.548 Income (INCOPTD) 0.769 Inflation (INFLOPTD) 0.505 Short-term investment (SINVOPTD) 0.667 Long-term investment (LINVOPTD) 0.805 Political affiliation and regime Democrat 0.201 Republican 0.292 D in Power 0.385 Investor characteristics Age 48.95 Education 14.51 White 0.840 Male 0.560 Income ($K) 87.66 Assets ($K) 224.67 Perceived risk and mispricing Market risk (MKTRISK ) 6.25 Market undervalued (UNDERVAL) 0.299

Std Dev

10th

25th

50th

75th

90th

N

0.71 1.09 1.04 1.14 1.06 1.08 1.11 0.96

2.57 2 2 2 2 2 2 2

3.14 2 2 2 3 2 3 4

3.71 3 4 3 4 3 4 4

4.14 4 4 4 5 4 4 5

4.43 4 5 5 5 5 5 5

57, 428 55, 992 57, 428 56, 094 56, 869 56, 066 56, 813 56, 721

0.489 0.498 0.491 0.498 0.421 0.500 0.471 0.396

0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 1

1 0 1 1 1 0 1 1

1 1 1 1 1 1 1 1

1 1 1 1 1 1 1 1

57, 428 55, 992 57, 428 56, 094 56, 869 56, 066 56, 813 56, 721

0.401 0.455 0.487

0 0 0

0 0 0

0 0 0

0 1 1

1 1 1

57, 428 57, 428 57, 428

13.84 2.23 0.360 0 44.79 340.81

32 9 0 0 35 55

39 14 1 0 55 55

48 15 1 1 67.50 55

59 17 1 1 150 150

70 17 1 1 150 750

56, 712 57, 169 57, 428 57, 428 53, 801 46, 928

2.02 0.458

4 0

5 0

7 0

8 1

10 1

9, 913 2, 192

43

Table 1 (Continued)

Summary statistics

Panel B: Brokerage sample

Percentile Variable

Mean

Portfolio choice measures Market beta (CAPM) 0.985 Market beta (4F) 1.066 SMB beta 0.418 HML beta −0.045 UMD beta −0.141 Portfolio performance measures Market-adjusted return −0.002 Characteristic-adjusted return −0.002 Political affiliation proxies % Democrat 56.35 High Democrat 0.200 High Republican 0.200 Investor characteristics Age 50.63 Education 23.97 % White 78.86 Male 0.881 Income ($K) 88.97 Portfolio Size ($K) 57.75

Std Dev

10th

25th

50th

75th

90th

N

0.643 0.516 0.990 1.039 0.608

0.184 0.522 −0.521 −1.21 −0.852

0.629 0.781 −0.242 −0.579 −0.423

1.01 1.03 0.194 −0.004 −0.091

1.34 1.32 0.801 0.469 0.182

1.71 1.69 1.64 1.02 0.471

1,892,994 1,978,230 1,978,230 1,978,230 1,978,230

0.096

−0.108

−0.054

−0.005

0.046

0.104

1,886,262

0.076

−0.079

−0.033

0

0.027

0.073

1,966,820

11.61 0.396 0.399

41.86 0 0

47.63 0 0

56.21 0 0

64.48 0 0

69.39 1 1

55,433 55,433 55,433

12.76 12.50 18.30 0.319 64.83 166.87

36 8.80 58.13 0 25 5.42

42 13.91 71.46 1 45 10.15

48 22.46 79.42 1 62.50 21.50

58 32.31 87.63 1 112.50 48.24

70 41.24 97.99 1 250 118.94

44,760 52,387 52,387 48,030 48,168 51,924

44

Table 2 Political affiliation, political climate, and optimism: sorting results This table reports optimism scores for investor groups identified by political affiliation and political regime. In columns (1) to (3), we report the fraction of investors who are optimistic about the economy. In columns (5) to (8), we report the mean optimism scores of those investor groups, where a score of 5 means very optimistic and a score of 1 means very pessimistic. We consider multiple optimism measures, including measures of optimism about the stock market and the overall economy. There are two distinct political regimes: (i) 1996 to 1998, when President Clinton (a Democrat) held office, and (ii) 2001 to 2002, during which President Bush (a Republican) was in office. The difference reported in column (4) is the fraction of Republicans who gave an optimistic response minus the fraction of Democrats who gave an optimistic response. Responses are considered optimistic if respondents answered 4 or 5 on a 5-point scale (“somewhat” or “very” optimistic). For the composite optimism index, investors are considered optimistic if the average response to the optimism questions is greater than 3.5 on the 5-point scale. *, **, and *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively. The Appendix provides brief definitions of all variables. Proportion Optimistic

Optimism Measure Clinton Presidency (1996-98) Optimism index Stock market optimism Economic growth optimism Employment optimism Income optimism Inflation optimism Short-term investment optimism Long-term investment optimism Bush Presidency (2001-02) Optimism index Stock market optimism Economic growth optimism Employment optimism Income optimism Inflation optimism Short-term investment optimism Long-term investment optimism

Repub (1)

Indep (2)

Dem (3)

0.696 0.663 0.633 0.609 0.746 0.528 0.755 0.806

0.710 0.642 0.668 0.642 0.746 0.561 0.732 0.798

0.742 0.683 0.726 0.676 0.745 0.596 0.763 0.824

0.571 0.522 0.570 0.453 0.746 0.511 0.633 0.837

0.421 0.407 0.429 0.355 0.673 0.432 0.510 0.748

0.358 0.362 0.378 0.317 0.651 0.392 0.485 0.715

45

R−D (4)

Average Optimism Repub (5)

Indep (6)

Dem (7)

R−D (8)

−0.046*** −0.021* −0.093*** −0.067*** 0.001 −0.068*** −0.007*** −0.018***

3.772 3.687 3.601 3.595 4.008 3.411 3.942 4.045

3.787 3.650 3.682 3.676 3.966 3.509 3.894 4.020

3.847 3.732 3.818 3.728 3.984 3.561 3.957 4.061

−0.075*** −0.045*** −0.217*** −0.133*** 0.023* −0.150*** −0.015* −0.016*

0.213*** 0.161*** 0.192*** 0.137*** 0.095*** 0.119*** 0.148*** 0.122***

3.558 3.279 3.426 3.171 3.983 3.389 3.584 4.101

3.296 2.992 3.086 2.898 3.792 3.186 3.292 3.848

3.167 2.859 2.922 2.755 3.695 3.048 3.176 3.728

0.390*** 0.420*** 0.504*** 0.416*** 0.288*** 0.341*** 0.408*** 0.373***

Table 3 Optimism Regression Estimates This table reports optimism regression estimates from the UBS/Gallup data. The following optimism measures are used as dependent variables: (1) continuous optimism index (OPTIDX), (2) optimism index dummy (OPTIDXD), (3) stock market optimism (MKTOPT ), (4) economic growth optimism (GROPT ), (5) employment optimism (EMPOPT ), (6) income optimism (INCOPT ), (7) inflation optimism (INFLOPT ), and (8) short-term investment optimism (SINVOPT ). We present the OLS estimates in all columns except column (2), which reports the marginal effects from a probit estimation. In Panel B, we add interaction terms between the political affiliation indicators and an index of US macroeconomic conditions (defined in the Appendix). Income and Assets are expressed in millions to give the coefficient estimates a readable magnitude. t- or z-statistics are reported in parentheses below the coefficient estimates. The sample period is from October 1996 to December 2002, excluding 1999 when political affiliation data are unavailable and 2000, when the horizon of the optimism questions overlapped presidential regimes. The Appendix provides brief definitions of all variables.

46

Table 3 (Continued)

Optimism Regression Estimates

Panel A: Optimism regression estimates (1) (2) (3) (4) (5) (6) (7) (8) Independent Variable OPTIDX OPTIDXD MKTOPT GROPT EMPOPT INCOPT INFLOPT SINVOPT D × D in Power R × D in Power Ind × D in Power D in Power Democrat Republican Age Education White Male Income Assets Constant Time Dummies Observations Adjusted R2

0.119 (4.84) −0.195 (−8.93) −0.027 (−1.29) 0.357 (33.70) −0.079 (−7.46) 0.156 (16.17) −0.002 (−7.36) −0.002 (−1.06) 0.013 (1.32) 0.096 (14.15) 1.049 (12.10) 0.160 (13.83) 3.392 (108.38) Yes 42,197 0.082

0.030 (1.41) −0.165 (−8.16) −0.034 (−1.82) 0.136 (4.91) −0.025 (−3.23) 0.110 (15.89) −0.001 (−3.05) −0.001 (−0.926) 0.004 (0.54) 0.063 (12.48) 0.560 (8.77) 0.084 (9.55)

Yes 45,287 0.057

0.162 (4.33) −0.120 (−3.61) 0.014 (0.46) 0.443 (27.48) −0.086 (−5.35) 0.177 (12.03) 0.001 (1.50) −0.012 (−4.96) −0.022 (−1.48) 0.088 (8.50) −0.148 (−1.12) 0.129 (7.29) 3.249 (68.43) Yes 44,329 0.049

47

0.188 (5.29) −0.296 (−9.35) −0.076 (−2.57) 0.428 (27.89) −0.120 (−7.84) 0.190 (13.55) 0.001 (1.51) −0.006 (−2.39) −0.022 (−1.55) 0.098 (9.92) 0.170 (1.35) 0.073 (4.36) 3.336 (71.03) Yes 44,815 0.049

0.153 (3.95) −0.218 (−6.33) −0.049 (−1.51) 0.559 (33.38) −0.129 (−7.73) 0.109 (7.14) −0.001 (−1.85) −0.003 (−1.13) 0.101 (6.47) 0.095 (8.76) 0.499 (3.62) 0.080 (4.36) 3.106 (60.68) Yes 44,387 0.061

−0.003 0.169 (−0.09) (4.58) −0.166 −0.176 (−5.15) (−5.39) −0.096 0.066 (−3.18) (2.14) 0.152 0.291 (9.70) (18.27) −0.039 −0.031 (−2.51) (−1.96) 0.131 0.128 (9.13) (8.80) −0.010 0.003 (−25.03) (6.37) 0.003 0.010 (1.09) (3.98) −0.014 0.061 (−0.93) (4.09) 0.058 0.259 (5.75) (25.24) 2.739 1.252 (21.22) (9.54) 0.144 0.202 (8.35) (11.54) 4.021 2.600 (87.36) (53.33) Yes Yes 44,957 44,389 0.044 0.044

0.107 (2.82) −0.216 (−6.39) −0.060 (−1.88) 0.414 (25.14) −0.099 (−6.01) 0.181 (12.03) −0.001 (−1.69) −0.011 (−4.57) −0.024 (−1.53) 0.023 (2.22) 0.956 (7.06) 0.202 (11.20) 3.677 (73.16) Yes 44,874 0.041

Table 3 (Continued)

Optimism Regression Estimates

Panel B: Control for sensitivity to macroeconomic conditions (1) (2) (3) (4) (5) (6) (7) (8) Independent Variable OPTIDX OPTIDXD MKTOPT GROPT EMPOPT INCOPT INFLOPT SINVOPT D × D in Power R × D in Power Ind × D in Power D in Power Democrat Republican D × Macro Index R × Macro Index Ind × Macro Index US Macro Index Age Education White Male Income Asset Constant Time Dummies Observations R2

0.061 (2.35) −0.175 (−7.47) −0.037 (−1.69) 0.280 (21.46) −0.054 (−4.84) 0.167 (16.86) 0.258 (6.04) −0.062 (−1.54) 0.077 (1.80) 0.314 (8.71) −0.002 (−6.22) −0.002 (−1.31) −0.011 (−1.05) 0.094 (13.81) 1.102 (12.74) 0.150 (13.25) 3.510 (104.67) Yes 41, 395 0.110

0.011 (1.41) −0.044 (−4.45) −0.023 (−2.92) 0.027 (9.45) −0.002 (−1.09) 0.021 (9.67) 0.001 (0.13) −0.012 (−1.15) −0.006 (−0.60) 0.028 (3.14) −0.001 (−11.49) 0.000 (1.22) 0.001 (0.58) 0.009 (5.58) 0.176 (7.63) 0.019 (5.43)

Yes 44, 386 0.0637

0.036 (0.87) −0.149 (−4.10) −0.069 (−1.96) 0.424 (20.85) −0.067 (−4.00) 0.179 (11.91) 0.519 (7.96) 0.134 (2.15) 0.328 (5.01) 0.063 (1.14) 0.001 (2.35) −0.012 (−4.89) −0.041 (−2.51) 0.087 (8.31) −0.120 (−0.90) 0.123 (6.87) 3.372 (65.20) Yes 43, 475 0.059

48

0.099 (2.64) −0.300 (−8.82) −0.145 (−4.40) 0.351 (18.55) −0.089 (−5.54) 0.203 (14.27) 0.530 (8.67) 0.050 (0.86) 0.283 (4.62) 0.315 (6.13) 0.001 (2.92) −0.006 (−2.59) −0.059 (−3.74) 0.096 (9.72) 0.210 (1.66) 0.058 (3.45) 3.435 (69.63) Yes 43, 922 0.077

0.064 (1.54) −0.204 (−5.45) −0.084 (−2.30) 0.387 (18.70) −0.080 (−4.70) 0.146 (9.65) 0.384 (5.82) 0.002 (0.03) 0.205 (3.10) 0.715 (12.80) 0.000 (−0.01) −0.004 (−1.49) 0.043 (2.55) 0.092 (8.67) 0.625 (4.63) 0.058 (3.28) 3.045 (59.63) Yes 43, 516 0.115

−0.006 (−0.14) −0.143 (−3.96) −0.067 (−1.95) 0.102 (5.14) −0.024 (−1.43) 0.134 (9.18) 0.097 (1.53) −0.103 (−1.71) −0.066 (−1.04) 0.218 (4.09) −0.009 (−23.93) 0.002 (0.97) −0.027 (−1.74) 0.058 (5.64) 2.780 (21.57) 0.135 (8.27) 4.166 (82.99) Yes 44, 064 0.049

0.204 (4.99) −0.091 (−2.42) 0.139 (3.82) 0.214 (10.42) −0.014 (−0.83) 0.133 (8.99) −0.061 (−0.95) −0.359 (−5.83) −0.228 (−3.51) 0.322 (5.91) 0.003 (6.52) 0.009 (3.60) 0.048 (2.88) 0.258 (24.99) 1.285 (9.73) 0.198 (11.41) 2.594 (49.77) Yes 43, 510 0.048

0.017 (0.42) −0.223 (−6.06) −0.087 (−2.61) 0.346 (17.14) −0.075 (−4.29) 0.195 (12.53) 0.329 (5.05) 0.061 (0.98) 0.186 (2.85) 0.283 (5.20) 0.000 (−0.72) −0.012 (−4.64) −0.051 (−3.06) 0.019 (1.81) 1.004 (7.33) 0.191 (10.83) 3.535 (70.40) Yes 43, 983 0.056

Table 4 Perceived risk and under-valuation regression estimates This table reports estimates from risk and under-valuation regressions. The independent variables are optimism (OPTIDX), Democrat, Republican, and other demographic characteristics. In specifications (1) and (2), the dependent variable is the investor’s assessment of the riskiness of the stock market, on a scale of 1 (no risk) to 10 (very high risk). The dependent variable in specifications (3) and (4) is a dummy variable that equals one if the investor felt the market was currently undervalued. To prevent the coefficient estimates on Income and Assets from becoming very small, we divide these two variables by 106 and make more readable presentation. z- or t-statistics are reported in parentheses. The perceived risk is available for the March to December 2002 period, while the perception of market under-valuation measure is available only in March, June and September of 2002. The Appendix provides brief definitions of all variables. Dependent Variable Market Risk Independent Variable

(1)

(2)

0.067 (0.71) −0.261 (−3.18) −0.003 (−1.13) 0.001 (0.05) −0.313 (−2.40) −0.215 (−3.02) 0.855 (0.95) −0.150 (−1.72) 6.535 (18.52) Yes 3,606 0.028

−0.766 (−15.00) 0.089 (0.95) −0.055 (−0.66) −0.007 (−2.30) 0.002 (0.11) −0.362 (−2.72) −0.129 (−1.80) 0.549 (0.60) −0.033 (−0.37) 9.193 (22.39) Yes 3,360 0.104

Optimism (OPTIDX) Democrat Republican Age Education White Male Income Assets Constant Time Dummies Number of Households Adjusted/Pseudo R2

Under-Valuation

49

(3)

(4)

0.020 (0.69) 0.081 (3.20) −0.001 (−1.64) −0.001 (−0.25) −0.013 (−0.34) 0.033 (1.50) 0.387 (1.46) −0.012 (−0.35)

0.070 (4.44) 0.013 (0.42) 0.075 (2.83) −0.001 (−1.29) −0.003 (−0.62) −0.022 (−0.56) 0.026 (1.12) 0.275 (1.01) −0.029 (−0.86)

Yes 1,928 0.010

Yes 1,813 0.019

Table 5 Political Climate and Portfolio Allocation Decisions This table reports difference-in-difference estimates for portfolio allocations, measured as the value of holdings in stocks, bonds, and mutual funds divided by net worth. The data are from the National Longitudinal Survey of Youth (NLSY). We report the mean allocation for groups defined by political preference and political regime, and report the differences between Democrat and Republican respondents and across political regimes, as well as the difference-in-difference estimate, along with their corresponding t-statistics. The portfolio allocations are regionally adjusted within each year to remove any systematic differences across Census regions. In Panel A, we define the political regimes as Split Government in 1991-1992 and 1995-1996 (where Congress and the Presidency were controlled by different parties), or DCONTROL in 1993-1994 when the Democratic party controlled both Congress and the Presidency. In Panel B, we adopt an alternate definition of political regime defined as the number of branches among the Presidency, the Senate, and the House of Representatives that are controlled by the Democratic Party. In both panels, the sample is restricted to respondents who indicated that their political preference was “strong”. The data are annual, from 1988 to 2000.

Panel A: Main Political Regime Definition Investor Type Political Regime

Democrat

Republican

D−R

Split (1988-92, 1995-2000)

0.0147

0.0365

DCONTROL (1993-94)

0.0154

0.0263

DCONTROL−Split

0.0007 (0.29)

−0.0102 (−1.01)

−0.0218 (−3.91) −0.0109 (−9.24) 0.0109 (1.06)

Panel B: Alternate Political Regime Definition Investor Type Political Regime

Democrat

Republican

D−R

(1): D President and R Congress; 1995-2000

0.0178

0.0412

(2): R President and D Congress; 1988-92

0.0132

0.0342

(3): D President and D Congress; 1993-94

0.0154

0.0263

−0.0024 (−1.19)

−0.0148 (−10.20)

−0.0233 (−10.53) −0.0210 (−2.41) −0.0109 (−9.24) 0.0124 (4.93)

(3)−(1)

50

Table 6 Political affiliation and stock preferences: panel regression estimates This table reports estimates from fixed effects panel regressions for Republican and Democrat investor groups, where the excess weight assigned to a stock in the aggregate group portfolio is the dependent variable. We use county-level voting data from the 1992 and 1996 presidential elections to obtain a proxy for the political affiliation of each investor in the brokerage sample. Investors in the lowest (highest) quintile of % Democrat variable are included in Republican (Democrat) categories. % Democrat is the fraction of votes cast for the Democratic presidential candidate in the county where the investor resides. The excess portfolio weight w −w allocated to stock i in month t is given by: EW ipt = iptwimtimt × 100, where, wipt is the actual weight assigned to stock i in group portfolio p in month t and wimt is the weight of stock i in the aggregate market portfolio in month t. The aggregate group portfolio is constructed by combining the portfolios of all investors in the group. The main independent variable is the politically sensitive stock dummy, which is set to one for stocks that are likely to benefit from the policies of the Republican party (see Subection 4.2). Other independent variables include market beta, firm size, book-to-market ratio, past one-month and twelve-month stock returns, stock price, idiosyncratic volatility, firm age, an S& P 500 dummy, a dividendpaying stock dummy, turnover, and return skewness. All independent variables are measured at the end of month t − 1. We follow the Driscoll and Kraay (1998) method to correct the standard errors for serial and cross-sectional correlations. We winsorize all variables at their 0.5 and 99.5 percentile levels, and standardize the independent variables such that each has a mean of zero and a standard deviation of one. The t-statistics for the coefficient estimates are shown in parentheses.

51

Table 6 (Continued)

Political affiliation and stock preferences: panel regression estimates

Independent Variable Pol. Sensitive Stock Dummy

Republican Democrat (1) (2)

R−D (3)

Republican Democrat (4) (5) 0.090 (4.54) 0.150 (3.98) −0.700 (−4.37) −0.209 (−6.66) 0.016 (0.40) 0.033 (0.90) −0.122 (−6.88) 0.410 (6.89) −0.045 (−4.60) 0.111 (5.14) −0.144 (−8.49) 0.224 (6.55) 0.006 (1.00) Yes 411,516 0.066

0.085 (6.01)

−0.044 (−4.55)

0.129 (8.30)

Yes 430,686 0.012

Yes 430,686 0.011

Yes 430,686 0.011

Market Beta Firm Size Book-to-Market Ratio Past 1-Month Stock Return Past 12-Month Stock Return Stock Price Idiosyncratic Volatility Firm Age S&P 500 Dummy Dividend Paying Stock Dummy Monthly Turnover Skewness Month Fixed Effect N Adjusted R2

52

−0.032 (−2.93) 0.124 (5.48) −0.485 (−6.56) −0.162 (−5.78) 0.015 (0.49) −0.044 (−1.66) −0.120 (−6.40) 0.180 (4.47) 0.020 (1.98) 0.120 (6.33) −0.328 (−6.62) 0.180 (6.90) −0.012 (−0.80) Yes 411,516 0.054

R−D (6) 0.122 (4.81) 0.026 (2.11) −0.215 (−6.55) −0.047 (−2.57) 0.001 (0.13) 0.077 (1.90) −0.002 (−0.40) 0.230 (4.70) −0.065 (−3.18) −0.009 (−0.80) 0.184 (5.50) 0.044 (2.64) 0.018 (1.60) Yes 411,516 0.032

Table 7 Systematic risk exposures across political regimes This table reports difference-in-difference estimates for measures of the systematic risk of aggregate Democrat and Republican investor portfolios across political regimes. For each month, we form aggregate Democrat and Republican portfolios by aggregating the holdings of households in top and bottom quintiles of the Democrat measure, defined as the fraction of votes cast for the Democrat presidential candidate in the county where the household resides. We consider five measures of systematic risk: the portfolio beta from a CAPM regression, and RMRF, SMB, HML, and UMD betas estimated from a four-factor model. We report the average systematic risk for the Democrat and Republican investor portfolios in each of two political regimes, which we define as Split Government in 1991-1992 and 1995-1996 (where Congress and the Presidency were controlled by different parties), and DCONTROL in 1993-1994 when the Democratic party controlled both Congress and the Presidency. In addition, we report the average difference in systematic risk between the Democrat and Republican portfolios in each political regime as well as the difference in difference across regimes. We also report a t-statistic, shown in parentheses, for the difference-in-difference estimate. The tstatistic is computed using the Newey and West (1987) correction where we allow for all lags to be potentially important. The data are from January 1991 to November 1996. The Appendix provides brief definitions of all variables. Risk Measure

Democrat

Republican

D−R

Beta (CAPM) DCONTROL (1993–1994) Split (1991–1992, 1995–1996)

0.982 0.944

0.957 0.930

0.026 0.014

RMRF Beta (4F) DCONTROL (1993–1994) Split (1991–1992, 1995–1996)

1.060 1.111

1.041 1.097

0.020 0.014

SMB Beta DCONTROL (1993–1994) Split (1991–1992, 1995–1996)

0.660 0.550

0.635 0.532

0.025 0.018

HML Beta DCONTROL (1993–1994) Split (1991–1992, 1995–1996)

0.104 0.091

0.104 0.112

0.000 −0.021

−0.102 −0.167

−0.093 −0.162

−0.009 −0.005

UMD Beta DCONTROL (1993–1994) Split (1991–1992, 1995–1996)

53

Diff-in-Diff

0.012 (4.08)

0.006 (3.80)

0.007 (3.42)

0.021 (6.89)

−0.005 (−0.70)

Table 8 Systematic risk exposure regression estimates This table reports estimates from fixed effects panel regressions of systematic risk exposure on measures of political affiliation and other controls. Panel A reports results from household fixed effect regressions where CAPM beta is the dependent variable. Panels B reports estimates using RMRF beta, SMB beta, HML beta, and UMD beta as one of the dependent variables. We use the four-factor model to estimate the factor exposures. Panel B also reports abbreviated results from alternative estimation methods, including a specification with time (year-month) fixed effects and a cross-sectional difference-in-difference regression. The primary independent variables are measures of political affiliation (Democrat, High Democrat dummy, and High Republican dummy) interacted with a DCONTROL dummy variable that equals one for months when the Democrat party controlled both the Presidency and Congress (1993-1994). Additional investor demographic characteristics are included as controls in specifications (4)-(6), each of which are also interacted with DCONTROL. All coefficient estimates have been scaled by 100 to enhance readability. Robust tstatistics, clustered by household and party-month, are reported in parentheses. The Appendix provides brief definitions of all variables.

Panel A: Portfolio beta (CAPM) regression estimates Dependent Variable: Ex-ante Portfolio Beta (CAPM) Independent Variable DCONTROL × Democrat DCONTROL × High Dem.

(1)

(2)

(3)

6.039 (2.56)

(4)

(5)

(6)

6.083 (2.33) 0.616 (0.79)

DCONTROL × High Repub.

0.220 (0.27) −1.291 (−1.91)

−1.558 (−2.12) DCONTROL × HH Age −0.068 −0.067 −0.068 (−3.35) (−3.33) (−3.36) DCONTROL × Education −0.005 0.013 0.004 (−0.18) (0.50) (0.16) DCONTROL × Male −0.626 −0.732 −0.673 (−0.82) (−0.96) (−0.88) DCONTROL × White −0.037 −0.047 −0.040 (−1.03) (−1.32) (−1.12) DCONTROL × HH Income 0.003 0.003 0.003 (0.75) (0.69) (0.71) DCONTROL × Portfolio Size 0.847 0.877 0.852 (0.50) (0.52) (0.51) DCONTROL −1.952 1.332 1.714 2.874 6.310 6.591 (−1.29) (1.32) (1.65) (1.18) (3.07) (3.19) Fixed Effects Household Household Household Household Household Household N 1,822,880 1,822,880 1,822,880 1,405,133 1,405,133 1,405,133 Number of Households 43,181 43,181 43,181 33,136 33,136 33,136 2 Adjusted R 0.047 0.047 0.047 0.047 0.047 0.047

54

Table 8 (Continued)

Systematic risk exposure regression estimates

Panel B: Estimates from additional tests Democrat× High D× High R× Democrat× High D× High R× DCONTROL DCONTROL DCONTROL DCONTROL DCONTROL DCONTROL Independent Variable (1) (2) (3) (4) (5) (6) Household Fixed Effects RMRF Beta (4F) 5.826 (1.74) SMB Beta 5.368 (1.38) HML Beta 19.765 (3.76) UMD Beta 0.904 (0.27) Time Effects CAPM Beta 7.493 (2.35) RMRF Beta (4F) 7.380 (2.92) SMB Beta 7.835 (1.70) HML Beta 23.619 (4.22) UMD Beta 3.235 (0.96) Cross-Sectional Diff-in-Diff Market Beta (CAPM) 4.684 (2.42) RMRF Beta (4F) 6.356 (3.94) SMB Beta 7.448 (2.62) HML Beta 19.728 (6.00) UMD Beta −1.650 (−0.87)

0.239 (0.20) 0.164 (0.13) 3.076 (2.02) −0.359 (−0.34)

−1.231 (−1.10) −1.350 (−1.17) −3.344 (−2.18) 0.520 (0.52)

2.746 (0.79) 4.094 (0.95) 13.151 (2.47) 2.378 (0.69)

−0.305 (−0.25) −0.546 (−0.41) 1.472 (0.93) 0.409 (0.37)

−0.720 (−0.64) −1.300 (−1.04) −2.253 (−1.40) 0.184 (0.18)

0.899 (1.42) 0.619 (1.05) 1.004 (0.92) 4.323 (3.71) 0.097 (0.12)

−1.619 (−2.59) −1.684 (−2.81) −1.197 (−1.20) −3.374 (−2.93) 0.253 (0.34)

7.493 (2.44) 6.924 (2.74) 5.751 (1.13) 22.611 (4.01) 4.516 (1.22)

0.646 (0.92) 0.329 (0.51) −0.006 (0.00) 3.607 (2.85) 1.172 (1.32)

−1.926 (−2.86) −1.792 (−2.70) −1.166 (−1.02) −3.791 (−2.94) 0.207 (0.25)

0.262 (0.46) 0.063 (0.13) 0.215 (0.26) 2.360 (2.42) −1.721 (−3.07)

−1.239 (−2.22) −1.380 (−3.04) −1.449 (−1.76) −3.693 (−3.99) 0.666 (1.24)

4.980 (2.13) 2.431 (1.26) 8.336 (2.46) 12.723 (3.24) −0.238 (−0.11)

0.077 (0.12) −0.612 (−1.11) 0.021 (0.02) 0.937 (0.84) −1.255 (−1.99)

−1.589 (−2.45) −0.660 (−1.25) −2.092 (−2.22) −2.431 (−2.29) 0.029 (0.05)

55

Table 9 Political regime and macroeconomic conditions This table reports estimates from fixed effects panel regressions of systematic risk exposure on measures of political affiliation, macro-economic conditions, state-level federal spending, and other controls. The reported regressions have the CAPM beta of the household’s portfolio, or the RMRF beta, SMB beta, HML beta, or UMD beta from a four-factor regression, as the dependent variable, and include household fixed effects. In addition to our main independent variable of interest, the DCONTROL × Affiliation interaction (where affiliation is measured using Democrat, a High Democrat dummy, or a High Republican dummy), we also include a State-Level Macro Index × Affiliation variable to control for sensitivity to local macro-economic conditions and a measure of state-level federal spending. For brevity, we only report the coefficients for DCONTROL × Affiliation, the Macro Index × Affiliation, and federal spending. The DCONTROL dummy variable equals one during the period when the Democratic party controlled both the Presidency and Congress (19931994). Additional investor demographic characteristics are included as controls in specifications (4)-(6), each of which are also interacted with the Democrat control dummy. All coefficient estimates have been scaled by 100 to enhance readability. t-statistics are reported in parentheses. The Appendix provides brief definitions of all variables.

56

Table 9 (Continued)

Interaction Terms Market Beta (CAPM) DCONTROL × Affiliation State Macro Index State-Level Fed Spending RMRF Beta (4F) DCONTROL × Affiliation State Macro Index State-Level Fed Spending SMB Beta DCONTROL × Affiliation State Macro Index State-Level Fed Spending HML Beta DCONTROL × Affiliation State Macro Index State-Level Fed Spending UMD Beta DCONTROL × Affiliation State Macro Index State-Level Fed Spending

Political regime and macroeconomic conditions Democrat (1)

High D (2)

High R (3)

Democrat (4)

High D (5)

High R (6)

8.537 (4.43) 2.357 (8.92) 0.950 (0.24)

0.824 (1.46) 2.195 (8.37) 0.849 (0.21)

−1.680 (−3.04) 2.245 (8.54) 0.830 (0.21)

8.359 (3.60) 2.209 (7.36) −5.363 (−1.15)

0.383 (0.59) 2.071 (6.93) −5.441 (−1.17)

−1.868 (−2.91) 2.123 (7.10) −5.413 (−1.16)

4.641 (2.68) −0.658 (−2.70) 14.839 (4.24)

0.121 (0.24) −0.753 (−3.10) 14.797 (4.22)

−1.024 (−2.09) −0.715 (−2.94) 14.791 (4.22)

1.750 (0.85) −0.610 (−2.21) 13.937 (3.39)

−0.390 (−0.66) −0.647 (−2.35) 13.932 (3.38)

−0.582 (−1.02) −0.623 (−2.26) 13.936 (3.39)

−1.738 (−0.58) −6.341 (−14.59) 6.437 (1.06)

−0.487 (−0.55) −6.316 (−14.57) 6.448 (1.06)

−0.210 (−0.24) −6.297 (−14.52) 6.453 (1.07)

−1.949 (−0.54) −6.192 (−12.49) −1.098 (−0.16)

−1.008 (−0.99) −6.174 (−12.48) −1.074 (−0.15)

−0.462 (−0.46) −6.144 (−12.41) −1.080 (−0.15)

12.265 (3.59) −5.896 (−11.13) 34.320 (4.62)

2.367 (2.31) −6.100 (−11.50) 34.232 (4.61)

−2.088 (−2.13) −6.070 (−11.45) 34.198 (4.60)

6.629 (1.62) −5.736 (−9.52) 31.415 (3.62)

0.951 (0.81) −5.834 (−9.67) 31.366 (3.61)

−1.336 (−1.17) −5.808 (−9.63) 31.400 (3.61)

−2.788 (−1.24) −3.345 (−9.76) 1.619 (0.32)

−0.694 (−1.02) −3.302 (−9.72) 1.637 (0.32)

1.113 (1.74) −3.331 (−9.79) 1.649 (0.33)

−0.683 (−0.26) −3.036 (−7.78) 2.901 (0.50)

0.174 (0.23) −3.021 (−7.79) 2.903 (0.50)

0.613 (0.83) −3.043 (−7.84) 2.895 (0.50)

57

Table 10 Active change in systematic risk across political regimes This table reports difference-in-difference estimates between Democratic and Republican investors across political regimes of active changes in the market risk of their portfolios. The active change in portfolio market risk is the component of the change in portfolio market beta due to trading, estimated as the quantity γZ Zit from the regression ∆βit = γZ Zit + γt t + εit , where ∆βit = βit − βi,t−1 is the change in the beta of the investor’s portfolio between months t − 1 and t, Zit is a set of two dummy variables to capture trading activities, and t is a set of time fixed effects. We estimate the above regression for four subsamples based on household political preference (Democrat or Republican) and political regime (DCONTROL or split government). We report the average of γZ Zit for each group, as well as the differences and t-statistics for the differences across groups (in parentheses). We consider five measures of systematic risk: the portfolio beta from a CAPM regression, and RMRF, SMB, HML, and UMD betas estimated from a four factor model. Political regimes are defined as Split Government in 1991-1992 and 1995-1996 where Congress and the Presidency were controlled by different parties, and DCONTROL in 1993-1994 when the Democratic party controlled both Congress and the Presidency. The data are from January 1991 to November 1996. The Appendix provides brief definitions of all variables. Investor Type Political Regime

Democrat

Beta (CAPM) Split (1991-1992, 1995-1996) DCONTROL (1993-1994)

−0.0037 0.0061

0.0051 0.0033

−0.0088 0.0028

Market Beta (4F) Split (1991-1992, 1995-1996) DCONTROL (1993-1994)

−0.0005 0.0008

0.0007 0.0000

−0.0012 0.0008

SMB Beta Split (1991-1992, 1995-1996) DCONTROL (1993-1994)

−0.0006 0.0014

0.0018 −0.0001

−0.0024 0.0015

−0.0020 −0.0016

−0.0028 −0.0023

0.0008 0.0007

0.0011 0.0021

0.0023 0.0015

−0.0012 0.0006

HML Beta Split (1991-1992, 1995-1996) DCONTROL (1993-1994) UMD Beta Split (1991-1992, 1995-1996) DCONTROL (1993-1994)

58

Republican

D−R

Diff-in-Diff

0.0116 (4.67)

0.0020 (2.83)

0.0039 (6.01)

0.0002 (1.39)

0.0019 (6.23)

Table 11 Local stock preference regression estimates This table reports estimates from fixed effects panel regressions of local stock preference on measures of political affiliation and other controls. The dependent variable is a measure of local stock preference, measured as the average distance to stocks in the investors portfolio, relative to the expected average distance of a portfolio of comparable stocks (the Appendix provides a more detailed definition of the local bias measure). The primary independent variables are measures of political affiliation (Democrat, High Democrat dummy, and High Republican dummy) interacted with a DCONTROL dummy variable that equals one during the period when the Democratic party controlled both the Presidency and Congress (1993-1994). Additional investor demographic characteristics are included as controls in specifications (4)-(6), each of which are also interacted with the DCONTROL dummy. Further, all coefficient estimates have been scaled by 100 to enhance readability. Robust t-statistics are reported in parentheses. The Appendix provides brief definitions of all variables.

Dependent Variable: Local Bias Independent Variable DCONTROL × Democrat DCONTROL × High Dem.

(1)

(2)

(3)

−3.664 (−2.70)

(4)

(5)

(6)

−2.456 (−1.51) −1.122 (−2.73)

DCONTROL × High Repub.

−0.749 (−1.57) 0.880 (2.36)

0.696 (1.62) DCONTROL × HH Age −0.029 −0.029 −0.029 (−2.12) (−2.12) (−2.12) DCONTROL × Education −0.026 −0.028 −0.029 (−1.50) (−1.70) (−1.72) DCONTROL × Male 0.661 0.669 0.678 (1.22) (1.23) (1.25) DCONTROL × White 0.010 0.013 0.011 (0.42) (0.52) (0.46) DCONTROL × HH Income −0.001 −0.001 −0.001 (−0.44) (−0.45) (−0.41) DCONTROL × Portfolio Size 1.214 1.211 1.212 (1.34) (1.34) (1.34) DCONTROL 2.212 0.366 −0.032 2.735 1.474 1.224 (2.88) (2.13) (−0.18) (1.83) (1.23) (1.02) Fixed effects Household Household Household Household Household Household N 1,790,049 1,790,049 1,790,049 1,379,540 1,379,540 1,379,540 Number of Households 42,772 42,772 42,772 32,812 32,812 32,812 Adjusted R2 0.663 0.663 0.663 0.664 0.664 0.664

59

Table 12 Portfolio performance regression estimates This table reports estimates from fixed-effect panel regressions of portfolio performance on measures of political affiliation and other controls. In columns (1)-(3), we report results from regressions in which the dependent variable is the market-adjusted monthly return on the investor’s portfolio, net of transaction costs. In columns (4)-(6), we report regression results in which the dependent variable is the Daniel et al. (1997) characteristic-adjusted net return. The primary independent variables are measures of political affiliation (Democrat, High Democrat dummy, and High Republican dummy) interacted with a DCONTROL dummy variable that equals one during the period when the Democratic party controlled both the Presidency and Congress (1993-1994). Additional investor demographic characteristics are included as controls in specifications (4)-(6), each of which are also interacted with the DCONTROL dummy. Robust t-statistics are reported in parentheses. The Appendix provides brief definitions of all variables. Dependent Variable Market-Adjusted Return Characteristic-Adjusted Return Independent Variable DCONTROL × Democrat DCONTROL × High Dem.

(1)

(2)

0.342 (2.37)

(3)

(4)

(5)

(6)

0.212 (1.74) 0.029 (0.71)

DCONTROL × High Repub.

−0.013 (−0.39)

−0.080 −0.050 (−2.01) (−1.47) DCONTROL × HH Age 0.175 0.177 0.174 0.046 0.048 0.045 (1.46) (1.47) (1.45) (0.45) (0.47) (0.44) DCONTROL × Education 0.379 0.464 0.433 0.189 0.261 0.222 (2.56) (3.22) (2.99) (1.51) (2.15) (1.82) DCONTROL × Male 0.048 0.043 0.045 0.050 0.046 0.049 (1.01) (0.91) (0.95) (1.25) (1.14) (1.21) DCONTROL × White −0.061 −0.116 −0.084 0.124 0.081 0.109 (−0.27) (−0.52) (−0.37) (0.65) (0.43) (0.57) DCONTROL × HH Income 0.303 0.294 0.293 0.208 0.196 0.202 (1.19) (1.15) (1.15) (0.96) (0.91) (0.93) DCONTROL × Portfolio Size −0.071 −0.068 −0.069 −0.052 −0.051 −0.052 (−0.93) (−0.91) (−0.92) (−0.83) (−0.80) (−0.82) DCONTROL −0.603 −0.413 −0.396 −0.327 −0.203 −0.198 (−4.55) (−3.89) (−3.73) (−2.92) (−2.27) (−2.21) Fixed Effects Household Household Household Household Household Household N 1,521,226 1,462,993 1,462,993 1,521,157 1,521,157 1,521,157 Number of Households 34,799 34,799 34,799 34,799 34,799 34,799 2 Adjusted R 0.028 0.028 0.028 0.025 0.025 0.025

60

Appendix Brief Definitions and Sources of Main Variables In this table we briefly define the main variables used in the empirical analysis. The data sources are: (i) Gallup: UBS/Gallup Investor Optimism Index; (ii) Brokerage: Large U.S. discount brokerage house; (iii) Election: U.S. Election Atlas (www.uslectionatlas.org); (iv) Census: U.S. Census County Files; and (v) Created: constructed by authors using data from above sources.

Variable Name

Description

Source

Optimism Measures Responses to each of the optimism questions from the Gallup survey are answered using the following scale: (1) very pessimistic; (2) somewhat pessimistic; (3) neither; (4) somewhat optimistic; (5) very optimistic. Stock Market Optimism (MKTOPT )

As far as the general condition of the economy is concerned, how would you rate performance of the stock market, over the next twelve months?

Gallup

Economic Growth Optimism (GROPT )

As far as the general condition of the economy is concerned, how would you rate Economic growth, over the next twelve months?

Gallup

Employment Optimism (EMPOPT )

As far as the general condition of the economy is concerned, how would you rate the unemployment rate, over the next twelve months?

Gallup

Income Optimism (INCOPT )

Thinking now about your own household, and the things that impact on your ability to invest over the next twelve months, how would you rate your ability to maintain or increase your current income over the next twelve months?

Gallup

Inflation Optimism (INFLOPT )

As far as the general condition of the economy is concerned, how would you rate Inflation, over the next twelve months

Gallup

Short-Term Investment Optimism (SINVOPT )

Overall, how optimistic or pessimistic are you that you will be able to achieve your investment targets over the next twelve months?

Gallup

Long-Term Investment Optimism (LINVOPT )

Overall, how optimistic or pessimistic are you that you will be able to achieve your investment targets over the next five years?

Gallup

Composite Optimism Index (OPTIDX )

(MKTOPT + GROPT + EMPOPT + INCOPT + INFLOPT + SINVOPT + LINVOPT )/7 .

Created

61

Political Affiliation Variables Democrat

A binary variable that takes the value of one if the respondent answers “Democrat” (zero otherwise) to the following question: In politics as of today, do you consider yourself a Republican, a Democrat, or an Independent?.

Gallup

Republican

A binary variable that takes the value of one if the respondent answers “Republican,” zero otherwise.

Gallup

Independent

A binary variable that takes the value of one if the respondent answers “Independent,” zero otherwise.

Gallup

D in Power

A binary variable that takes the value of one if the survey was conducted when the Democratic party was in power (i.e., before February 2000), zero otherwise.

Created

Perceptions of Risk and Reward Perceived Market Risk (MKTRISK )

Using a ten-point scale, where 1 means no risk, and 10 means very high risk, how would you rate the CURRENT level of risk for investing in the stock market?

Gallup

Perceived Market Under-Valuation (UNDERVAL)

Do you think the stock market is: (1) Overvalued; (2) Valued about right; (3) Undervalued; (4) Unsure.

Gallup

Age

Age of the investor, in years.

Gallup

Education

Equals 9 if the respondent was a high school graduate or less. We assign a value of 14 if the respondent had attended a college or had receiving any educational training. We assign a value of 15 for those who graduated from college, and assign a value of 17 for those who had postgraduate degrees.

Created

White

One if investor is white, zero otherwise.

Gallup

Male

One if investor is male, zero otherwise.

Gallup

Income

Total household total annual income before taxes for the previous year. This is coded as the midpoint of income ranges given in the survey, and takes one of the following values: $10,000, $25,000, $35,000, $45,000, $55,000, $67,500, $87,500, or $150,000.

Created

Assets

Total size of all savings and investments in the household, coded as the midpoint of asset value ranges given in the survey. It takes one of the following values: $50,000, $150,000, $350,000, $750,000, or $1,500,000.

Created

Investor Characteristics

62

Portfolio Measures Market Beta (CAPM)

Value-weighted CAPM beta of stocks in the portfolio, where the stock-level beta estimates are obtained using monthly data over the past four years.

Brokerage

RMRF Beta, SMB Beta, HML Beta, and UMD Beta

Value-weighted factor betas of stocks in the portfolio, where the stock-level beta estimates are obtained using a four-factor regression on the RMRF, SMB, HML, and UMD factors using monthly data over the past four years.

Brokerage

Local Bias

LBIAS = 1 − Dact /Dportf . Dact is the average distance between an investor’s location and stocks in her portfolio, while Dportf is the average distance between an investor’s location and other characteristic-matched portfolios not held by the investor. The distance between an investor’s location and a PNi portfolio p is computed as D(i, p) = k=1 wk d(i, k), where wk is the weight of stock k in investor’s portfolio, d(i, k) is the distance between the zip code of the residence of investor i and the headquarter of stock k, and Ni is the number of stocks in the investor portfolio. The matching stock is in the same size, book-to-market, and momentum deciles of the original stock and, furthermore, it belongs to the same Fama and French (1997) industry as the original stock. In several instances, we are unable to find a stock that matches on all dimensions, but we match stocks on at least the size and the B/M dimensions.

Brokerage

Market-Adjusted Portfolio Return

Monthly return on the investor’s portfolio, net of transaction costs [which are computed as in Barber and Odean (2000)], minus the return on the market in that month.

Brokerage

Characteristic-Adjusted Portfolio Return

Value-weighted average of the return (net of transaction costs) on each stock in the investor’s portfolio, minus the corresponding Daniel et al. (1997) benchmark return.

Brokerage

Political Affiliation Proxies (Brokerage Data) DCONTROL

An indicator variable that takes the value of one during the 1993-1994 period of full Democratic control, zero otherwise.

Created

Democrat

(%Dem92 + %Dem96 )/2, where %Demyear is (Num. Democrat voters)/(Num. Democrat voters + Num. Republican voters) in that election year in the county where the investor resides.

Election

High Democrat

One if % Democrat is in the top quintile, zero otherwise.

Election

High Republican

One if % Democrat is in the bottom quintile, zero otherwise.

Election

63

Investor Characteristics (Brokerage Data) Age

Age of the investor, in years.

Brokerage

Education

Percentage of Zip Code residents above age 25 with bachelor’s degree or higher in the Zip Code where the investor resides.

Census

White

Percentage of Zip Code residents who are White in the Zip Code where the investor resides.

Census

Male

One if the investor is male, zero otherwise.

Brokerage

Income

Investor’s household income, in thousands of dollars.

Brokerage

Portfolio Size

Total market value of stocks in the investor’s portfolio, averaged over the months in which the investor was in the sample.

Brokerage

64