The democratic premium expanded: What is the influence of presidential charisma on the
U.S. stock market?
The existence of a democratic premium in U.S. stock market returns and volatilities has been established before. This paper expands this by combining the democratic premium with a president’s personal character trait, namely his level of charisma. Specifically, the effects of solely the president’s political party as well as his charisma score on S&P 500 returns and volatilities are examined. Moreover, the democratic premium is tested when the effects are controlled for charisma. Finally, the interaction effect of the president’s political party and his charisma level on returns and volatilities are studied. Four control variables are included to control for business cycle fluctuations. The sample consists of 11 presidents who were active from 1960 until 2020, resulting in 5 Democrats and 6 Republicans. Three different charisma scores are being applied, one by Simonton (1988), one by Almahdi (2019) and one dummy variable to indicate whether a president was charismatic or not based on multiple authors.
Results show that there is indeed a democratic premium in the S&P 500 returns. Moreover, the charisma scores by Almahdi (2019) also show to have a positive significant effect on monthly returns. When the political effect on returns is controlled for charisma, the effect becomes stronger and is highly significant. No democratic effect is seen in the monthly volatility when the controls are added. The effect becomes significant when it is controlled for Simonton (1988) as well as the other controls. The interaction effect of the political variable and Simonton (1988) is significant. However, after including the controls, it is not anymore.
Keywords: democratic premium, presidential charisma, stock market returns, stock market volatility
Author: Julia de Koekkoek Student number: 12256404 Date of submission: 28/06/2021
Program: Bsc. Business Administration Track: Finance
Thesis supervisor: dr. R. de Bliek
Statement of Originality
This document is written by Student Julia de Koekkoek who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of Contents
1. Introduction ... 4
2. Theoretical framework ... 6
2.1 Financial impact following elections: political cycle effect ... 6
2.2 Presidential charisma ... 9
2.3 The political cycle effect and presidential charisma ... 10
3. Data and methodology ... 12
3.1 Sample ... 12
3.2 Dependent variables ... 15
3.3 Control variables ... 16
3.4 Analyses ... 17
4. Results and discussion ... 19
4.1 Effects on monthly S&P 500 returns... 19
4.1.1 Control variable testing ... 19
4.1.2 Hypothesis testing ... 23
4.2 Effects on monthly volatility ... 28
4.2.1 Control variable testing ... 28
4.2.2 Hypothesis testing ... 32
5. Conclusion ... 36
6. References ... 39
7. Appendix ... 43
1. Introduction
The closing prices of the S&P500 index increased with 14.3% since Joe Biden was elected as the president of the United States on November 3rd in 2020 until January 20th in 2021 (Ho, 2021). From the Second World War onwards, the economy of the Unites States has shown better performances under Democratic presidents than under Republican presidents (Egan, 2020). More specifically, the S&P 500 showed average annual gains of 11.2% and annual returns of 4.8% under Democratic presidency opposed to annual gains of 6.9% and annual returns of 2.7% under Republican presidency. However, the excess returns are not affected by a risk premium. Perhaps the president himself influences the stock market?
The presence of a democratic premium seems to be confirmed in the research conducted by Santa-Clara and Valkanov (2003). The study concludes that the excess return in the stock market is higher under Democratic than Republican presidencies which is the result of higher real stock returns and lower real interest rates. However, the excess returns are not affected by a risk premium since there are no differences in the riskiness of the stock market perceived across presidencies. Therefore, it remains a puzzle why there exists a premium for democratic presidencies. Besides the president’s political party, the president himself does have the ability to influence economic performances (Bohte & Heo, 2013) as well as the election outcomes via communication style (Ahmadian et al., 2017). A common characteristic observed by public figures is charisma. Being charismatic is beneficial as a president because it entails having a powerful personal appeal and the ability to captivate others (Bligh & Kohles, 2009).
Thus, recent research has proven the existence of the democratic premium in stock market returns.
Additionally, it is established that being charismatic is beneficial for presidents. Nevertheless, the effect of the political party together with the president’s level of charisma has not been studied yet. Therefore, this research will investigate the democratic premium, the effect of charisma and the effect of the interaction term between the president’s political party and his level of charisma on the U.S. stock
market returns and volatility. It will also be investigated how the effects of the president’s political party on returns and volatility change when it is controlled for charisma.
This study focuses on the presidents who were active from 1961 until 2020, resulting in 6 Republican presidents and 5 Democratic presidents. The level of charisma of the U.S. presidents is determined via three different scores on charisma. First of all, Simonton (1988) identified charisma via assessors who assigned presidential style items to descriptions of presidents. Secondly, Almahdi (2019) used transcripts of presidential speeches to measure the readability, subjectivity of text and polarity of text to identify charisma. Thirdly, a dummy variable is assigned whether a president is regarded as charismatic or not based on multiple authors. Regarding the effects on the U.S. stock market, this study will focus on the monthly returns and volatility reported for the S&P 500 index from the first of January 1961 until the 31st of December 2020. Following the research design of Santa-Clara and Valkanov (2003), this study will also control for variations in the business cycle. More specifically, the variables that will be controlled for are the term spread, dividend yields, the default spread, and the real interest rates.
After completing the regression analyses, it is expected to find a democratic premium in the monthly S&P 500 returns. Regarding the volatility, the opposite of a premium is expected to be found, namely lower volatilities under Democratic presidencies. Moreover, a positive interaction effect is expected to be found between the political and charisma variables on the stock returns. Subsequently, the coefficients of the interaction effect are expected to be negative when tested for the volatilities, resulting in lower volatilities when a president is charismatic and from the Democratic Party. Finally, it’s expected that when the charisma score is included as a control variable, the effect of the political party on the returns/volatility decreases.
2. Theoretical framework
2.1 Financial impact following elections: political cycle effect
Politicians govern a country according to their beliefs and values and aim to pursue their goals as long as possible. However, these decision-makers change over time due to re-election procedures. In order to win the re-elections and maintain their spot in the government, politicians have incentives to execute policies in favor of their chances of getting re-elected (Bove et al., 2017). Political cycle theories investigate the effect of government incentives on policy choices. A two-split is seen in these theories, namely the opportunistic (or electoralist) and the partisan theories. According to Bove et al. (2017), the opportunistic theories state that irrespective of political preferences all governments choose economic policies around elections to increase the possibility of getting re-elected. On the other side, the partisan theories state that left-wing governments follow income redistribution and expansionary policies during their active period as opposed to right-wing governments that appeal more to capital owners (Bove et al., 2017; Potrafke, 2012).
Rogoff (1990) identified that political cycles influence the composition of government spending, rather than the level of total spending. Citizens base their votes on the part of the government spending that directly applies to them. Therefore, the government shifts their spending to more visible public goods.
Drazen and Eslava (2010) concluded in their study that citizens value spending on some goods more than on other goods. Therefore, the government will spend on goods that are preferred by the people to increase their chances of getting re-elected. Regarding the tradeoff between social and military expenditures, Bove et al. (2017) found that decision-makers might bias their expenses towards social instead of military affairs.
Political cycles and economic performance have been studied widely. Potrafke (2012) examined data on annual GDP growth in 21 OECD countries over the years 1951 until 2006 to investigate the influence of electoral motives and government ideology on short-term economic performance. He found that political cycles are more prevalent in countries two-party systems because citizens can penalize the current government directly for their performance. In these countries GDP growth is boosted before
elections, and under leftwing governments in the first two years of their incumbency. Besides election periods, partisan theories state that under leftwing governments the annual GDP growth is higher because of their responsory fiscal and monetary policies (Potrafke, 2012). This last statement is also confirmed in a working paper by Blinder and Watson (2014) in which they observed the time period of 1947:Q1 through 2013:Q1. They showed that GDP growth increases under Democrat leadership and decreases under Republican leadership. Moreover, they found that over a four-year Democrat presidency the U.S. economy grew by 18.6% opposed to 10.6% during a Republican presidency. The Democratic growth advantages are a result of more spending on consumer durables and private investment. Additionally, of the 49 quarters in the study by Blinder and Watson (2014) classified as in a recession, just 8 fell in Democratic reign and 41 in Republican reign. Regarding employment in the U.S., Blinder and Watson (2014) found that the annual growth rate of payroll employment is 1.42 percentage points higher under Democrat presidency and income inequality rises under Republicans but falls under Democrats. There is also a significant gap of -1.9 percentage points in the change in the unemployment rate: unemployment fell by 0.8 percentage points on average during Democratic periods of leadership opposed to an increase of 1.1 percentage points under Republican leadership. Nevertheless, inflation tends to rise under Democrats and fall under Republicans. However, as is also discussed by Blinder and Watson (2014), this might be caused by weaker GDP and employment growth. Alesina and Sachs (1988) confirm in their research that the Democratic party is inclined to combat unemployment more and is less averse to inflation than the Republican party. This is in line with the partisan theory of monetary policy.
Besides GDP growth and employment rates, the partisan gap is also visible in stock market returns.
Under Democratic leadership, annualized stock market returns on the S&P 500 are 5.4 percentage points higher than under Republican leadership (Blinder & Watson, 2014). Early research from Allvine and O’Neill (1980) correspond with these findings by confirming that changes in stock prices on the long run are not random but driven by the four-year election cycle. Their results show that in the period between 1948 and 1978 annual returns rose by 22.1% on average in the year beginning two years prior to the election, 9.2% in the year before the election, 0.6% in the year after the election and 0.7% in the
second year after the election. Additionally, Santa-Clara and Valkanov (2003) conclude that the difference in returns grows gradually over the presidential term. On the contrary, they found that differences in returns are not concentrated around election dates. By using data since 1927, they discovered that the average excess return of the value weighted CRSP index under Democratic leadership is 11% and just 2% under Republican leadership, estimated over the three-month Treasury bill rate. The difference is caused by the real market returns being 5% higher in combination with real interest rates being 4% lower when Democrats are in charge. To test whether the returns are not caused by variations in the business cycle, Santa-Clara and Valkanov (2003) controlled for macro variables that forecast the stock market: dividend-price ratio, the default and term spreads and the relative interest rate.
Their results show that the variations in returns are largely uncorrelated to business cycle fluctuations.
Moreover, by regressing the realized returns on the business cycle variables they examined whether the differences in returns are caused by expected or unexpected returns. After taking the fitted values of the regression as the expected returns and the residuals as the unexpected returns, the analysis shows that most of the differences in returns can be attributed to unexpected returns. Furthermore, the study investigates whether the higher realized returns are a compensation for the carried risk. Unexpectedly, the market volatility is higher under Republican presidency than under Democratic presidency. The article therefore concludes that it remains a puzzle why the returns differ since it is not a compensation for risk. Cahan et al. (2005) expand this conclusion by investigating real stock market returns in New Zealand. They conclude that real stock market returns are lower under the leadership of the Labour party than under the National party. Since this is contradictory to the observations in the U.S., it can be concluded that the presidential puzzle is not directly applicable to other countries with similar two party democracies. Based on the previously discussed findings, it is expected to also find a democratic premium in the examined data, implying that Democratic presidents lead to higher S&P 500 returns as opposed to Republican presidents.
Market uncertainty is caused by uncertain election results (Smales, 2015). Pantzalis et al. (2000) found that valuations rise two weeks before an election, assumed that the political uncertainty decreases during this period. This phenomenon is identified as the uncertain information hypothesis (Brown et al., 1998).
Nevertheless, when the outcome of the election is uncertain, the stock market volatility increases (Li &
Born, 2006). On the other hand, Gemmill (1992) found that the volatility of the FTSE 100 index increased in the last two weeks even though opinion polls were predicting a conservative outcome at the same time. Besides volatility around election periods, Santa-Clara and Valkanov (2003) showed in their study that the market volatility is higher under Republican presidents than under Democratic presidents.
Therefore, it is expected that the results of this study are in line with the before mentioned results, implying that the monthly volatility of S&P 500 returns are lower under the leadership of Democratic presidents than under Republican presidents.
2.2 Presidential charisma
Besides the political parties, voters are inclined to base their perceptions of the president on the state of the economy (Bohte & Heo, 2013). However, the amount of control presidents have on the economy is limited due to factors like existing federal laws and international influences according to presidency scholars. Since there was no systematic research conducted to investigate presidential management of the economy, Bohte and Heo (2013) performed this with the objective to assess the limitations of presidents in their goal to influence economic performances via fiscal policies. They find that fiscal policies significantly influence unemployment and economic growth in the U.S. For example, personal consumption decreases due to higher marginal tax rates and unemployment rates decrease as a result of deficit spending. Therefore, it can be concluded that the president together with the congress does have the ability to influence economic performance.
Ahmadian et al. (2017) solely focused on Donald Trump and how he won the elections from his more experienced competitors. They examined the former president’s communication style and found that he scored the highest on grandiosity ratings, use of first-person pronouns, greater pitch dynamics, and informal communication compared to the top nine Republican contenders since October 2015. Simonton (1988) examined presidential styles and found that charismatic presidents are person-orientated, “aiming their energies toward other human beings”. Secondly, it’s found that a charismatic leader is outward- active, meaning that he or she aims to implement a program and a vision. Charismatic leaders have a
powerful personal appeal and captivate others (Bligh & Kohles, 2009). Furthermore, the research by Mio et al. (2005) presented that charismatic presidents use nearly twice as many metaphors than non- charismatic presidents. The results of the same study also showed that using metaphors contributes to inspiring the audience. Bligh and Kohles (2009) explored charismatic leadership theory in the context of the 2008 U.S. presidential elections. Charismatic leadership theory has namely been identified as being important in situations where the social gap between leaders and followers is big, but also in times of crisis. Nevertheless, charismatic leaders should prevent that their phraseology will be classified as empty rhetoric. Thus, besides a leader’s delivery style, charismatic leaders should also be successful in forming ideals into approachable messages with widespread and emotional appeal. Finally, Deluga (1998) found that presidential proactivity is positively associated with charismatic leadership and rated performance.
Based on abilities like engaging the audience, being outward-active and having a powerful personal appeal that are connected to being charismatic, it is expected that charismatic leadership results in higher stock market returns as opposed to a situation in which a non-charismatic president is in charge.
Consequently, the importance of charisma in times of crisis and the ability to captivate others (and thus to comfort others) leads to the expectation that a charismatic president results in lower stock market volatility.
2.3 The political cycle effect and presidential charisma
Based on the previously discussed literature, it can be concluded that a democratic premium exists.
However, it remains a puzzle why this premium exists. Moreover, it has been identified that the president can individually influence the economic state of the U.S. and charismatic character traits contribute to inspiring people and close the gap between a leader and his followers. Finally, charismatic leadership is positively associated with proactivity and rated performance. Therefore, it is expected that a positive interaction effect exists between the political variable and charisma scores on the S&P 500 returns.
Secondly, a negative interaction effect between the political variable and charisma scores on the volatility of the S&P 500 returns is expected, implying that a more positive interaction leads to lower
volatility. Finally, it is anticipated that the political effect on the returns and volatility decreases when these effects are controlled for charisma.
3. Data and methodology
3.1 Sample
This study will focus on the presidents who were active from the first of January 1961 until the 31st of December 2020 in the United States. This starting point is chosen because of the aftermath of the Second World War which might influence the outcomes of this study. Officially, Kennedy took office on 20/01/1961, but since he was elected 2 months before, it is assumed that he was active as a president since the first of January. For the other presidents, it is assumed that they entered the White House on the first day of the month in which they officially took office. For example, Obama entered the White House officially on the 20th of January 2009, but in this study it is assumed that he was active as a president since the first of January 2009. The sample of presidents ranges from Kennedy up until Trump.
The sample contains 11 presidents in total, of which 6 Republicans and 5 Democrats.
The identified presidential charisma scores differ across authors. To start with, Simonton (1988) identified presidential charisma scores up to and including Reagan. In the study, seven undergraduate psychology majors independently assessed presidential descriptions based on 82 presidential style items constructed by the Historical Figures Assessment Collaborative at the Institute for Personality Assessment and Research. After internal consistency reliability calculations, 49 items had alpha coefficients of 0.6 or higher and therefore proved to be reliable to assess the presidents. A principal- components analysis resulted in 5 presidential styles. With the loadings in parentheses, the charismatic presidential style entails a president who ‘“finds dealing with the press challenging and enjoyable” (.93),
“enjoys the ceremonial aspects of the office” (.91), “is charismatic” (.82), “consciously refines his own public image” (.81), “has a flair for the dramatic” (.81), “conveys clear-cut, highly visible personality”
(.80), is a “skilled and self-confident negotiator” (.76), “uses rhetoric effectively” (.69), is a “dynamo of energy and determination” (.69), is “characterized by others as a world figure ….” (.61), “keeps in contact with the American public and its moods” (.60), “has ability to maintain popularity” (.60),
“exhibits artistry in manipulation” (.60), and “views the presidency as a vehicle for self-expression”
(.54), but rarely “is shy, awkward in public” (−.87).’ The individual presidential scores were then
identified via exclusive factor scores, meaning that an item is solely assigned to a factor with which it has the highest loading.
Simonton (1988) identified charisma in a public setting: seven assessors who have not met any of the presidents based their views on descriptions of these presidents. Therefore, the identified charisma scores reflect charisma seen from a distance. Data scientist Almahdi (2019) also reviewed public charisma by using transcripts of all the presidential speeches. Charisma in the speeches was measured by readability, subjectivity of text, and polarity of text. The score on charisma is then calculated as the mean of Flesch reading ease, polarity and subjectivity scores. Trump was excluded from the final scores since only presidents with completed presidential terms were examined. However, it is identified that Trump has the highest reading ease. Moreover, he scored the highest on the polarity measure (after Truman). Since this implies being perceived as charismatic, Trump should be assigned a high charisma score relative to the other scores identified by Almahdi (2019). Regarding reading ease solely, Trump scored 11.88% higher as Truman. Regarding polarity, Truman scored 3.47% higher than Trump.
Therefore, Trump should get a final score which is 8.41% higher than the final score of Truman (which is 0.7251), resulting in 0.7861.
Contrary, House et al. (1991) examined charismatic relationships in the context of a leader’s actual behavior to a leader’s subordinates. Therefore, the perceived charisma reflects private charisma as opposed to public charisma. House et al. (1991) obtained charismatic behavioral scores by gathering biographies of two or more cabinet members reporting to each president up until Reagan. Subsequently, one-volume scholarly biography of each cabinet member was selected based on the best references. The documents were screened to identify remarks about the president’s behavior. Theories on charismatic leadership were used to code the extracts. The final charismatic scores are based on all occurrences of charismatic behavior found in all biographical extracts. Since Ford was not elected, he is not included in this study. Reagan was also not included because no complete data could be obtained. For Nixon and Carter, no sufficient reference materials could be obtained. Due to the circumstance that only two presidents in this study’s sample are assigned a score, the scores by House et al. (1991) cannot be used
as an independent variable. However, we do see that the scores approximately correspond with the scores identified by Simonton (1988) and Almahdi (2019): Kennedy received the highest score by House et al. (1991), the second highest by Simonton (1988) but is in the middle according to Almahdi (2019).
Eisenhower received the second highest score by House et al. (1991) and is in the middle according to Simonton (1988) and Almahdi (2019).
Another way of assessing charisma, is to solely identify whether a president was charismatic or not instead of assigning a score. Concerning the three previously discussed scores, Simonton’s (1988) scores are the only ones identifying negative scores on charisma. From this research, Ford and Carter have been identified as non-charismatic. Since Simonton’s (1988) study was published before George H.W.
Bush became president, the last 5 presidents should be assigned differently. As being perceived as inspiring and famous for his ‘thank you’ notes, George H.W. Bush was regarded as a charismatic president (Clifford, 2018). Moreover, most political observers identified Clinton as a charismatic president (Wen, 2017; Klein, 2015) which might come at a cost (Bloland, 2000). Bligh et al. (2004) examined George W. Bush’s rhetoric before and after the crisis of 9/11. They concluded that the former president’s rhetoric language became more charismatic after the crisis which was in line with the media’s portrayal of the president as a charismatic leader. Moreover, Obama’s positive score on charisma is caused by his ability to captivate others, his emotional appeals, and his delivery style (Bligh
& Kohles, 2009). Finally, Trump positioned himself as an outsider to build up his charisma and to create an us-versus-them narrative (Khazan, 2016). Furthermore, Trump’s campaign speeches show that he used hyperbolic crisis rhetoric and emphasized a shared social identity and common goal (Ghazal Aswad, 2019). Therefore, he is also perceived as charismatic. To sum up, Ford and Reagan are the only two presidents being perceived as non-charismatic. The negative/positive scores on charisma have been identified in Table 1 under the column “multiple authors” as a dummy variable (0 being perceived as non-charismatic and 1 being perceived as charismatic). Table 1 also shows the other two charisma scores plus the years in the White House and political party of all the presidents examined in this study.
Table 1. Sample of presidents with their corresponding years in the White House, political party and charisma scores.
Charisma Scores
Presidency President Political Party Simonton (1988)
Almahdi (2019)
Multiple authors
01/01/1961* – 22/11/1963 John F. Kennedy Democratic 1.3 0.6557 Positive 22/11/1963 – 20/01/1969 Lyndon B. Johnson Democratic 1.5 0.6786 Positive 20/01/1969 – 09/08/1974 Richard Nixon Republican 0.3 0.6798 Positive 09/08/1974 – 20/01/1977 Gerald Ford Republican -0.1 0.6374 Negative 20/01/1977 – 20/01/1981 Jimmy Carter Democratic -0.4 0.6516 Positive 20/01/1981 – 20/01/1989 Ronald Reagan Republican 1.2 0.6533 Negative 20/01/1989 – 20/01/1993 George H.W. Bush Republican n/a 0.6706 Positive 20/01/1993 – 20/01/2001 Bill Clinton Democratic n/a 0.6870 Positive 20/01/2001 – 20/01/2009 George W. Bush Republican n/a 0.6697 Positive 20/01/2009 – 20/01/2017 Barack Obama Democratic n/a 0.6714 Positive 20/01/2017 – 31/12/2020** Donald Trump Republican n/a 0.7861 Positive
*Kennedy took office on 20/01/1961, however the time period examined in this study starts on 01/01/1961.
** Trump left the office on 20/01/2021, however the time period examined in this study ends on 31/12/2020.
3.2 Dependent variables
U.S. Stock market performance In order to measure the effect of a president’s political party and level
of charisma on the U.S. stock market, the monthly returns on the S&P 500 are used in the analysis. The S&P 500 is a market-capitalization-weighted index of the 500 largest companies listed on the stock exchange in the U.S. The data is collected via the CRSP Index File on the S&P 500. The range chosen is 1961/01 – 2020/12 with a monthly data frequency. Value-weighted returns (excluding dividends) are selected, as is also used by Santa-Clara and Valkanov (2003).
To investigate the heteroskedasticity that might be viewed in the dependent variable, the Breusch-Pagan test is used. Four regressions are run in total, every time using another independent variable. The Chi- square test and corresponding p-values after using the estat hettest command in Stata are shown in Appendix table 1. Even though the p-value for Almahdi (2019) is not significant, the significant p-values for the political variable, Simonton’s score (1988) and the general score show that there is heteroskedasticity in the monthly S&P 500 returns. To check for stationarity, Appendix figure 1 is created that shows a scatterplot of t (time in months) and the monthly S&P 500 returns. The plot does not show any trends nor seasonal effects and therefore the time series can be regarded as stationary.
Moreover, the Dickey-Fuller test is performed with the command dfuller. The test is also performed
with a trend term and drift term. The Dickey-Fuller tests show that the null hypothesis of monthly returns exhibiting a unit root can be rejected (see Appendix table 2).
Volatility The volatility of the monthly S&P 500 returns is included in this analysis to examine whether
a president’s political party and/or charisma level have an effect on the stock market volatility. By using the CRSP Index File on the S&P 500, monthly volatility is calculated from within-monthly daily return data by using the Egen SD command in Stata. In line with the previous dependent variable, the monthly volatility is also checked for heteroskedasticity and stationarity. The Breusch-Pagan test shows that all the models (expect for the general charisma score) are significant and therefore heteroskedasticity is present in the monthly volatility (see Appendix table 3). Regarding stationarity, the scatterplot of t (time in months) and the monthly volatility does show some outliers (see Appendix figure 2). However, Appendix table 5 shows that the null hypothesis of monthly volatility exhibiting a unit root can be rejected after performing the Dickey-Fuller test.
3.3 Control variables
Regarding the control variables, the study by Santa-Clara and Valkanov (2003) is followed. To test whether the returns and/or volatility are not caused by variations in the business cycle, the study will control for macro variables that forecast the stock market. Other studies also included similar (or almost similar) control variables, like Keim and Stambaugh (1986), Campbell and Shiller (1988), Fama and French (1988;1989), and Fama (1991).
Term spread The difference between 10-year treasury constant maturity rates and federal fund rates
are retrieved from the online database Federal Reserve Economic Data.
Dividend yield The data on dividends that are paid out relative to the stock prices is gathered via
quandl.com; S&P 500 Ratios; S&P 500 Dividend Yield by Month.
Default spread By including default spreads, a comparison between corporate bonds and risk-free alternatives is possible. ‘Moody’s seasoned Baa corporate bond minus federal funds rate’ data is used, retrieved from the Federal Reserve Economic Data database.
Real interest rate The real interest rate is added as a control variable to control for the effects of
inflation. The yearly data up to and including 2019 is provided by the World Bank; ‘Real interest rate (%) – United States’. The monthly real interest rates are calculated by dividing the annual rates by 12.
The monthly real interest rates of 2020 are based on https://www.longtermtrends.net/real-interest-rate/.
Table 2. Descriptive statistics of the dependent and control variables.
Variable Obs Mean Std. Dev. Min Max
Monthly returns 720 .0067724 .0428515 -.2179442 .1648143 Monthly volatility 720 .0086826 .0054771 .0019637 .0585351
Dividend yield 720 2.903514 1.132374 1.11 6.24
Term spread 720 1.048833 1.628819 -6.51 3.85
Default spread 720 3.087986 2.036405 -4.05 8.82
Real interest rate 720 .0029164 .0024386 -.0126 .0071622
3.4 Analyses
The regressions discussed in the following part will be performed with Newey-West standard errors (1987) to correct for heteroskedasticity and serial-correlation. The number of lags is determined by the number of time periods to the power of 0.25, resulting in lag (5). The R-squared and Adjusted R-squared values are obtained by simply running the regressions with the code ‘reg’. The analysis is divided in two main parts: in the first part, the effects of the independent variables will be tested for monthly returns and in the second part they will be tested for monthly volatility. Subsequently, every main part starts with a stepwise approach to test the control variables: 5 models are tested, and in each model an extra control variable is added. The control variables are tested with the political variable as the independent variable but also with the charisma scores as the independent variable. Hereafter, regressions are run that test for the effects of the political and charisma variables on the monthly returns and volatility.
More specifically, the following models will be tested for both the monthly returns as well as the monthly volatility as the dependent variable:
Model (1): effects of the political variable on the monthly returns/volatility
Model (2): effects of the charisma scores by Simonton (1988) on monthly returns/volatility Model (3): effects of the charisma scores by Almahdi (2019) on monthly returns/volatility Model (4): effects of the general charisma scores on monthly returns/volatility
Model (5): effects of the political variable on the monthly returns/volatility while controlling for the term spread, dividend yield, default spread, and real interest rate
Model (6): effects of the charisma scores by Simonton (1988) on the monthly returns/volatility while controlling for the term spread, dividend yield, default spread, and real interest rate
Model (7): effects of the charisma scores by Almahdi (2019) on the monthly returns/volatility while controlling for the term spread, dividend yield, default spread, and real interest rate
Model (8): effects of the general charisma scores on the monthly returns/volatility while controlling for the term spread, dividend yield, default spread, and real interest rate
Model (9): effects of the political variable on the monthly returns/volatility while controlling for the charisma scores by Simonton (1988)
Model (10): effects of the political variable on the monthly returns/volatility while controlling for the charisma scores by Almahdi (1988)
Model (11): effects of the political variable on the monthly returns/volatility while controlling for the general charisma scores
Model (12): effects of the political variable on the monthly returns/volatility while controlling for the charisma scores by Simonton (1988), term spread, dividend yield, default spread, and real interest rate Model (13): effects of the political variable on the monthly returns/volatility while controlling for the charisma scores by Almahdi (1988), term spread, dividend yield, default spread, and real interest rate Model (14): effects of the political variable on the monthly returns/volatility while controlling for the general charisma scores, term spread, dividend yield, default spread, and real interest rate
Model (15): effects of the interaction term of the political variable and the charisma scores by Simonton (1988) on the monthly returns/volatility
Model (16): effects of the interaction term of the political variable and the charisma scores by Almahdi (2019) on the monthly returns/volatility
Model (17): effects of the interaction term of the political variable and the charisma scores by Simonton (1988) on the monthly returns/volatility while controlling for the term spread, dividend yield, default spread, and real interest rate
Model (18): effects of the interaction term of the political variable and the charisma scores by Almahdi (2019) on the monthly returns/volatility while controlling for the term spread, dividend yield, default spread, and real interest rate
The general charisma score showed too little variation and perfect collinearity exists between the political variable and the interaction. Therefore, this interaction term is omitted.
4. Results and discussion
4.1 Effects on monthly S&P 500 returns 4.1.1 Control variable testing
In order to check the model, a stepwise estimation is used implying that an extra control variable is added in every regression. The results for the political variable on monthly returns are shown in Table 3.
Table 3. Model testing for the effects of the political variable on monthly S&P 500 returns, with the t-statistics in parentheses.
Regressor Model (1) Model (2) Model (3) Model (4) Model (5) Political .0046
(1.52)
.0043 (1.42)
.0044 (1.45)
.0051 (1.69)
.0050 (1.69)
Term spread .0020
(1.86)
.0020 (1.86)
-.0000 (-0.00)
.0001 (0.02) Dividend yield
.0004 (0.25)
.0005 (0.34)
.0003 (0.17) Default spread
.0018 (0.47)
.0018 (0.48) Real interest rate
.8297 (0.86) Intercept
.0046 (1.89)
.0027 (1.07)
.0015 (0.29)
-.0026 (-0.26)
-.0045 (-0.43) R2
.0029 .0085 .0086 .0095 .0117
Adjusted R2 .0015 .0058 .0045 .0040 .0047
As can be concluded from the regression results shown in Table 3, monthly S&P returns are 0.0046 percentage points higher when the president is a Democrat instead of a Republican when not including any control variables. However, this effect is not significant since the t-statistic has a value of 1.52. After including the term spread and the dividend yield as controls, the effect remains insignificant. However, the term spread itself is significant at the 10% level in Model (2) and (3). After including the default spread and the real interest rate, the political variable shows to have a significant effect on a 10%
significance level on the monthly returns in Model (4) and (5). The coefficient of the political variable is the highest in Model (4), but this is only a difference of 0.0001 percentage points. The R2 value does
increase every time an additional control is added. However, model (5) still has a low overall explanatory power with an R2 value of 0.0117. The Adjusted R2 increases to 0.0058 as opposed to the starting value of 0.0015 when adding the term spread as a control. It decreases in Model (3) and (4) and increases a little again to 0.0047 in Model (5). Since the coefficient of the dependent variable changes every time a control is added, the coefficient is significant in the last two models, the R2 value is the highest for Model (5) and the Adjusted R2 value for Model (5) is the second highest for all models, all the examined control variables will be used in the following analyses. The change in coefficients after a control is added is caused by the control being responsible for a part of the values of the monthly returns.
Subsequently, a stepwise estimation is used to examine the model including the charisma scores, monthly returns and control variables. Table 4 shows the model including the scores identified by Simonton (1988), Table 5 shows the model for the scores by Almahdi (2019) and Table 6 shows the model for the general scores.
Table 4. Model testing for the effects of Simonton’s (1988) charisma scores on monthly S&P 500 returns with the t-statistics in parentheses.
Regressor Model (6) Model (7) Model (8) Model (9) Model (10) Simonton (1988) 0.0015
(0.47)
-.0004 (-0.11)
.0008 (0.24)
.0016 (0.45)
.0040 (0.66)
Term spread .0039
(3.05)
.0041 (3.13)
-.0086 (-2.10)
-.0095 (-2.03)
Dividend yield .0024
(0.86)
-.0017 (-0.55)
-.0001 (-0.03)
Default spread .0114
(3.18)
.0122 (3.10)
Real interest rate -1.0978
(-0.51) Intercept .0045
(1.25)
.0038 (1.05)
-.0064 (-0.55)
-.0101 (-0.87)
-.0156 (-0.95)
R2 .0006 .0264 .0283 .0558 .0569
Adjusted R2 -.0024 .0206 .0195 .0444 .0426
The results show that the charisma scores identified by Simonton (1988) do not have a significant effect on monthly returns, regardless of the inclusion of control variables. The signs of the effects of the charisma scores ranges between -.0004 and .0040. It changes in Model (7) to a negative effect but changes to positive again from Model (8) onwards. The negative coefficient indicates that a higher charisma score results in lower market returns. The control variable term spread is significant at the 1%
level in Models (7) and (8) and significant at the 5% level in Models (9) and (10). The default spread is significant at the 1% level for Models (9) and (10). However, the dividend yield and real interest rate are insignificant for every model. The R2 value increases every time a control variable is added. The Adjusted R2 value increases for every model compared to Model (6) and is the highest for Model (9).
Since the R2 is the highest for Model (10) and the Adjusted R2 is the second highest in Model (10), every control variable will be added in the regression.
Table 5. Model testing for the effects of Almahdi’s (2019) charisma scores on monthly S&P 500 returns with the t-statistics in parentheses.
Regressor Model (11) Model (12) Model (13) Model (14) Model (15) Almahdi (2019) .0291
(0.71)
.0363 (0.89)
.0518 (1.14)
.0523 (1.16)
.0846 (1.73)
Term spread .0021
(1.95)
.0023 (2.02)
.0012 (0.29)
.0014 (0.34)
Dividend yield .0009
(0.54)
.0010 (0.57)
.0011 (0.63)
Default spread .0010
(0.26)
.0011 (0.29)
Real interest rate 1.1565
(1.22) Intercept -.0129
(-0.46)
-.0120 (-0.71)
-.0333 (-0.99)
-.0357 (-1.02)
-.0618 (-1.65)
R2 .0005 .0067 .0071 .0074 .0111
Adjusted R2 -.0009 .0040 .0030 .0019 .0042
As is seen with Simonton’s (1988) charisma scores, the scores identified by Almahdi (2019) are also not significant in the first 4 models (see Table 5). However, the effect on monthly returns is significant at the 10% significance level in Model (15). Moreover, the term spread is significant at the 10%
significance level in Model (12) and at the 5% level in Model (13). After adding default spread as a control variable, the term spread is not significant anymore. The dividend yield, default spread and real interest rate are all not significant as a control variable in all the regressions. However, R2 increases every time a control variable is added and the Adjusted R2 is the highest when all controls are included in the regression.
Table 6. Model testing for the effects of general charisma scores on monthly S&P 500 returns, with the t- statistics in parentheses.
Regressor Model (16) Model (17) Model (18) Model (19) Model (20)
General -.0042
(-0.94)
-.0037 (-0.84)
-.0051 (-0.93)
-.0048 (-0.73)
-.0031 (-0.43)
Term spread .0020
(1.83)
.0018 (1.58)
.0014 (0.33)
.0013 (0.31)
Dividend yield -.0008
(-0.45)
-.0007 (-0.34)
-.0006 (-0.29)
Default spread .0004
(0.10)
.0006 (0.16)
Real interest rate .7204
(0.71) Intercept .0103
(2.46)
.0078 (1.82)
.0115 (1.25)
.0102 (0.61)
.0057 (0.30)
R2 .0014 .0071 .0074 .0074 .0089
Adjusted R2 .0000 .0043 .0032 .0019 .0020
In line with Simonton’s (1988) scores, the general charisma estimates do not have a significant effect on the monthly S&P 500 returns in all the regression (see Table 6). The term spread is significant at the 10% level in Model (17) but is just not significant in Model (18) and has a low t-statistic in Models (19) and (20). The dividend yield, default spread and real interest rate are all not significant as a control variable. The R2 value increases every time a control is added. Nevertheless, the Adjusted R2 decreases after adding the dividend yield and is slightly higher for regression (20) than for regression (19).
From these findings it can be concluded that the term and default spreads are the only controls that are significant in some of the regressions. However, the Adjusted R2 increases in some regressions as more controls are added and the effect of the independent variables on the monthly returns does change every time an additional control is included. The R2 is the highest for the models including all the control variables and the last model is significant for the political variable and Almahdi’s (2019) score.
Therefore, the subsequent analysis will be performed with every control variable in the regressions.
4.1.2 Hypothesis testing
Table 7 shows the regression results with the political variable and the three charisma scores separately as the independent variable with the monthly S&P 500 returns being the dependent variable. From Model (5) onwards, the control variables are added. Without the control variables, nor the political variable nor the charisma scores have a significant effect on the monthly S&P 500 returns. Looking at the explanatory power of the models, the R2 and Adjusted R2 values are quite low. However, the political variable seems to have the biggest explanatory power compared to the charisma scores. After adding the control variables, the political variables as well as the scores by Almahdi (2019) show to have a significant effect on the monthly returns at the 10% significance level. This implies that a Democratic president is responsible for returns which are .0050 percentage points higher than when the president is a Republican. Secondly, for every additional point received by Almahdi (2019), the stock returns increase by .0846 percentage points. The R2 values as well as the Adjusted R2 values increase due to the addition of the control variables. Model (5) predicts .47% of the variance in the monthly S&P 500 returns and the scores by Almahdi predict .42%. Even though the effects in Models (5) and (7) are significant, the explanatory power of the models is quite low.
Table 7. Effects of political variable and charisma scores on monthly S&P 500 returns, with the t-statistics in parentheses.
Regressor Model (1)
Model (2)
Model (3)
Model (4)
Model (5)
Model (6)
Model (7)
Model (8)
Political .0046 (1.52)
.0050 (1.69)
Simonton (1988) .0015
(0.47)
.0040 (0.66)
Almahdi (2019) .0291
(0.71)
.0846 (1.73)
General score -.0042
(-0.94)
-.0031 (-0.43)
Term spread .0001
(0.02)
-.0095 (-2.03)
.0014 (0.34)
.0013 (0.31)
Dividend yield .0003
(0.17)
-.0001 (-0.03)
.0011 (0.63)
-.0006 (-0.29)
Default spread .0018
(0.48)
.0122 (3.10)
.0011 (0.29)
.0006 (0.16)
Real interest rate .8297
(0.86)
-1.0978 (-0.51)
1.1565 (1.22)
.7204 (0.71) Intercept .0046
(1.89)
.0045 (1.25)
-.0129 (-0.46)
.0103 (2.46)
-.0045 (-0.43)
-.0156 (-0.95)
-.0618 (-1.65)
.0057 (0.30)
R2 .0029 .0006 .0005 .0014 .0117 .0569 .0111 .0089
Adjusted R2 .0015 -.0024 -.0009 .0000 .0047 .0426 .0042 .0020
Table 8 shows the regression results with the political variable as the independent variable and controlling for the charisma scores. In Models (12) till (14), the other control variables are also added.
When controlling for the general charisma score, the political variable becomes significant at the 5%
significance level. Moreover, the coefficient of the political variable becomes higher when controlling for Almahdi (2019) and the general score as opposed to the coefficients in Model (1). After adding the other controls, the political effect in all three models show to be significant: Model (12) at the 1% level and Models (13) and (14) at the 5% level. Furthermore, the coefficient of the political variable is higher
for Model (12) and (13) as opposed to Models (9) and (10), respectively. The explanatory power of the models is again quite low. However, the R2 and Adjusted R2 values are higher when including the standard controls as opposed to the models without the standard controls. The Adjusted R2 value is the highest for Model (12) in which the effect is controlled for Simonton’s (1988) score and the four standard controls. The model explains 8.21% of the variance in the monthly S&P 500 returns.
Table 8. Effects of political variable on monthly S&P 500 returns while controlling for the charisma scores, with the t-statistics in parentheses.
Regressor Model
(9)
Model (10)
Model (11)
Model (12)
Model (13)
Model (14) Political .0011
(0.23)
.0049 (1.57)
.0073 (2.17)
.0263 (4.59)
.0062 (2.01)
.0072 (2.16) Simonton (1988) .0015
(0.45)
-.0011 (-0.22)
Almahdi (2019) .0375
(0.88)
.1076 (2.10)
General score -.0084
(-1.66)
-.0084 (-1.04)
Term spread -.0226
(-5.25)
.0002 (0.05)
.0003 (0.07)
Dividend yield -.0068
(-1.53)
.0018 (1.00)
-.0012 (-0.53)
Default spread .0249
(6.19)
.0021 (0.56)
.0013 (0.32)
Real interest rate .9294
(0.46)
1.2329 (1.29)
.5097 (0.50) Intercept .0041
(0.93)
-.0208 (-0.70)
.0103 (2.45)
-.0248 (-1.72)
-.0845 (-2.08)
.0082 (0.43)
R2 .0007 .0037 .0074 .0985 .0158 .0142
Adjusted R2 -.0053 .0009 .0046 .0821 .0075 .0059
In Table 9, the interaction effect of the political variable and the charisma scores are shown. Without the control variables, both interaction terms show to be insignificant. The explanatory power of the models is also low: it is nihil for Model (15) and only .01% for Model (16). The interaction term with Simonton (1988) shows to be negative, meaning that a charismatic Democratic president results in lower returns. After including the control variables, the interaction term between the political variable and Simonton (1988) is almost significant and Model (18) is still insignificant. However, the R2 and Adjusted R2 values did increase. The interaction term with Simonton (1988) explains now 8.56% of the variance in the monthly S&P 500 returns and the interaction term with Almahdi (2019) explains .62%.
Model (16), (17), and (18) have positive interaction terms, meaning that a charismatic Democratic president results in higher S&P 500 returns.
Table 9. Interaction effects of the political variable and the charisma scores on monthly S&P 500 returns, with the t-statistics in parentheses.
Regressor Model
(15)
Model (16)
Model (17)
Model (18) Political*Simonton (1988) -.0031
(-0.39)
.0213 (1.58)
Political*Almahdi (2019) .1207
(0.71)
.0382 (0.16)
Term spread -.0225
(-5.18)
.0002 (0.05)
Dividend yield -.0071
(-1.58)
.0019 (0.93)
Default spread .0254
(6.41)
.0021 (0.56)
Real interest rate 4.3714
(1.64)
1.1834 (1.09)
Intercept .0026
(0.41)
-.0148 (-0.48)
-.0239 (-1.69)
-.0835 (-2.04)
R2 .0013 .0042 .1047 .0159
Adjusted R2 -.0077 .0001 .0856 .0062
Based on the literature, a democratic premium is expected to exist which shows higher S&P 500 returns when a Democrat is in charge (Blinder & Watson, 2014; Santa-Clara & Valkanov, 2003). The multiple positive coefficients of the political variable is therefore in line with the literature. After including the control variables, a significant democratic premium exists. The significant effect of the charisma scores by Almahdi (2019) on the monthly returns, after including the control variables, is in line with the presidential charisma theory. Presidents can individually affect the economic growth in the U.S. and are therefore also expected to be able to influence the stock market (Bohte & Heo, 2013). Moreover, a president’s ability to captivate others and especially in times of crisis results in the expectation of a positive and significant effect of charisma on S&P 500 returns. The positive coefficient of Simonton’s (1988) scores is also in line with the literature. Remarkably, the general charisma score shows a negative coefficient which is contradictory to the literature. However, these effects are not shown to be significant. After including the control variables, the political variable becomes significant and the coefficient increases when also controlling for either Simonton (1988), Almahdi (2019) or the general charisma score. Finally, the interaction terms show to be insignificant, regardless of adding control variables. This is not in line with previous studies, since a Democratic president who is also charismatic is expected to influence the stock market returns positively.
4.2 Effects on monthly volatility 4.2.1 Control variable testing
Similar to the dependent variable monthly returns, the control variables are also tested in combination with the dependent variable monthly volatility. The same stepwise estimation is used implying that an extra control variable is added in every regression. The results for the political variable on the monthly volatility of the S&P 500 returns is shown in Table 10 with t-statistics in parentheses.
Table 10. Model testing for the effects of the political variable on monthly S&P 500 volatility, with the t-statistics in parentheses.
Regressor Model (1) Model (2) Model (3) Model (4) Model (5) Political -.0015
(-2.21)
-.0015 (-2.19)
-.0016 (-2.35)
-.0001 (-0.12)
-.0001 (-0.11)
Term spread .0000
(0.02)
-.0001 (-0.31)
-.0046 (-6.63)
-.0046 (-6.60) Dividend yield
-.0004 (-1.56)
-.0001 (-0.27)
-.0000 (-0.21) Default spread
.0040 (6.24)
.0040 (6.28) Real interest rate
-.0397 (-0.21) Intercept
.0094 (17.60)
.0094 (19.08)
.0107 (10.08)
.0015 (0.99)
.0016 (1.03) R2
.0177 .0177 .0247 .3022 .3025
Adjusted R2 .0164 .0150 .0207 .2983 .2976
From the estimates in Table 10 it can be concluded that the political variable has a significant effect on the monthly S&P 500 volatility in Models (1), (2) and (3) at a 5% significance level. After the default spread is added in Model (4), the effect stays negative but becomes insignificant. The term spread and default spread are both highly significant, namely at the 1% significance level, in Models (4) and (5).
The dividend yield as well as the real interest rate are both not significant in any model. The R2 remains the same in the first two models but increases up to .3025 in Model (5). The Adjusted R2 starts at .0164, decreases in Model (2) and is the highest in Model (4) with a value of .2983. However, the Adjusted R2 in Model (5) is just slightly lower, namely .2976. Since the effect of the political variable becomes
insignificant in Model (4) and the term spread and default spread are both significant in Model (4) and (5), all the controls will be used in the subsequent analyses.
Subsequently, a stepwise estimation is used to examine the model including the charisma scores, monthly volatility and control variables. Table 11 shows the model including the scores identified by Simonton (1988), Table 12 shows the model for the scores by Almahdi (2019) and Table 13 shows the model for the general scores.
Table 11. Model testing for the effects of Simonton’s (1988) charisma scores on monthly S&P 500 volatility with the t-statistics in parentheses.
Regressor Model (6) Model (7) Model (8) Model (9) Model (10)
Simonton (1988) -.0007 (-1.39)
-.0006 (-1.27)
.0001 (0.15)
.0002 (0.50)
-.0000 (-0.03)
Term spread -.0003
(-1.31)
-.0002 (-0.92)
-.0024 (-4.78)
-.0024 (-4.45)
Dividend yield .0013
(3.78)
.0006 (1.57)
.0005 (0.70)
Default spread .0020
(4.31)
.0020 (4.48)
Real interest rate .0943
(0.34)
Intercept .0081
(16.83)
.0082 (18.17)
.0026 (1.84)
.0020 (1.44)
.0025 (1.11)
R2 .0131 .0240 .0817 .1723 .1730
Adjusted R2 .0102 .0182 .0734 .1623 .1605
The models in Table 11 show that Simonton’s (1988) charisma scores do not have a significant effect on the monthly volatility, regardless of adding control variables. Nevertheless, the dividend yield is significant in Model (8) and the term spread and default spread are both significant in Models (9) and (10), all on a 1% significance level. The R2 value starts with .0131 and increases up to .1730 in Model (10). The Adjusted R2 is the highest in Model (9) but is only slightly lower in Model (10) with a value of .1605. Because of the term spread and default spread being both significant in Model (10), the highest R2 value and the second-highest Adjusted R2 value in Model (10), all the controls will be added.
Table 12. Model testing for the effects of Almahdi’s (2019) charisma scores on monthly S&P 500 volatility with the t-statistics in parentheses.
Regressor Model (11) Model (12) Model (13) Model (14) Model (15) Almahdi (2019) .0022
(0.17)
.0022 (0.16)
-.0049 (-0.31)
-.0030 (-0.23)
-.0046 (-0.45)
Term spread -.0000
(-0.07)
-.0001 (-0.44)
-.0047 (-6.83)
-.0047 (-6.78)
Dividend yield -.0004
(-1.29)
-.0001 (-0.37)
-.0001 (-0.40)
Default spread .0040
(6.28)
.0040 (6.33)
Real interest rate -.0572
(-0.32)
Intercept .0072
(0.79)
.0072 (0.80)
.0133 (1.17)
.0036 (0.38)
.0049 (0.65)
R2 .0002 .0002 .0055 .3024 .3030
Adjusted R2 -.0012 -.0026 .0013 .2985 .2981
Based on the results in Table 12, the charisma scores by Almahdi (2019) do not have a significant effect on the monthly S&P 500 volatility, regardless of adding control variables. However, after adding the dividend yield as a control, the coefficient becomes negative. As was also seen in Table 11, the term spread and default spread are significant on a 1% level in the last two models. The dividend yield and real interest rate are not significant in any model. The R2 starts with .0002 and increases up to .3030 in Model (15). The Adjusted R2 starts at nihil and increases up to .2985 in Model (14) and ends at .2981 in Model (15). Since the coefficient of Almahdi (2019) changes every time a control is added, the term spread and default spread are significant in the last two models and the Adjusted R2 is the second highest in Model (15), all the controls will be used in the further analyses.
Table 13. Model testing for the effects of the general charisma scores on monthly S&P 500 volatility with the t- statistics in parentheses.
Regressor Model (16) Model (17) Model (18) Model (19) Model (20) General score -.0009
(-1.10)
-.0009 (-1.10)
-.0022 (-1.99)
.0004 (0.33)
.0003 (0.27)
Term spread -.0000
(-0.14)
-.0002 (-0.75)
-.0047 (-6.41)
-.0047 (-6.44)
Dividend yield -.0008
(-2.23)
.0000 (0.08)
.0000 (0.07)
Default spread .0040
(5.79)
.0040 (5.83)
Real interest rate -.0273
(-0.14)
Intercept .0094
(13.33)
.0095 (12.79)
.0129 (7.00)
.0007 (0.26)
.0009 (0.30)
R2 .0039 .0040 .0199 .3026 .3028
Adjusted R2 .0025 .0012 .0158 .2987 .2979
The general charisma score only has a significant effect in Model (18) in Table 13 after the significant control variable ‘dividend yield’ is added. Both the general score and the dividend yield are significant on a 5% significance level. The coefficient of the general charisma score is negative in the first 3 models but becomes positive after the significant control variable ‘default spread’ is added. The default spread as well as the term spread are significant in Models (19) and (20) on a 1% significance level. The R2 value starts with .0039 and increases up to .3028. The Adjusted R2 decreases in Model (17) but rises again and is the highest for Model (19). However, the Adjusted R2 is only slightly lower for Model (20).
Therefore, all the controls will be used in the following analyses.
4.2.2 Hypothesis testing
Table 14 shows the regression results with the political variable and the three charisma scores separately as the independent variable with the monthly S&P 500 volatility being the dependent variable. From Model (5) onwards, the control variables are added.
Table 14. Effects of political variable and charisma scores on monthly S&P 500 volatility, with the t-statistics in parentheses.
Regressor Model (1)
Model (2)
Model (3)
Model (4)
Model (5)
Model (6)
Model (7)
Model (8) Political -.0015
(-2.21)
-.0001 (-0.11)
Simonton (1988) -.0007
(-1.39)
-.0000 (-0.03)
Almahdi (2019) .0022
(0.17)
-.0046 (-0.45)
General score -.0009
(-1.10)
.0003 (0.27)
Term spread -.0046
(-6.60)
-.0024 (-4.45)
-.0047 (-6.78)
-.0047 (-6.44)
Dividend yield -.0000
(-0.21)
.0005 (0.70)
-.0001 (-0.40)
.0000 (0.07)
Default spread .0040
(6.28)
.0020 (4.48)
.0040 (6.33)
.0040 (5.83)
Real interest rate -.0397
(-0.21)
.0943 (0.34)
-.0572 (-0.32)
-.0273 (-0.14) Intercept .0094
(17.60)
.0081 (16.83)
.0072 (0.79)
.0094 (13.33)
.0016 (1.03)
.0025 (1.11)
.0049 (0.65)
.0009 (0.30)
R2 .0177 .0131 .0002 .0039 .3025 .1730 .3030 .3028
Adjusted R2 .0164 .0102 -.0012 .0025 .2976 .1605 .2981 .2979
Before adding any control variables, Table 14 shows that the political variable has a significant effect on S&P 500 volatility on a 5% significance level. The coefficient implies that when a president is Democratic, the monthly volatility is .0015 percentage points lower than when the president is a Republican. The three charisma scores do not have a significant effect on the monthly volatility. After adding the control variables, the effect of the political variable becomes insignificant and the effects of the charisma scores remain insignificant. The R2 and adjusted R2 values do increase substantially due to
the addition of the control variables. The model with the political variable and the control variables explains 29.76% of the total variance in the monthly S&P 500 returns. For Simonton (1988), Almahdi (2019) and the general charisma score it is 16.05%, 29.81% and 29.79%, respectively.
Table 15 shows the regression results with the political variable as the independent variable and controlling for the charisma scores. In Models (12), (13) and (14), the other control variables are also added.
Table 15. Effects of political variable on monthly S&P 500 volatility while controlling for the charisma scores, with the t-statistics in parentheses.
Regressor Model
(9)
Model (10)
Model (11)
Model (12)
Model (13)
Model (14) Political -.0026
(-3.78)
-.0015 (-2.23)
-.0014 (-1.80)
-.0015 (-2.44)
-.0001 (-0.23)
-.0002 (-0.31) Simonton (1988) -.0005
(-1.17)
.0003 (0.42)
Almahdi (2019) -.0002
(-0.02)
-.0050 (-0.49)
General score -.0001
(-0.10)
.0005 (0.33)
Term spread -.0016
(-3.40)
-.0047 (-6.60)
-.0047 (-6.56)
Dividend yield .0008
(1.24)
-.0001 (-0.46)
.0000 (0.11)
Default spread .0013
(2.83)
.0040 (6.18)
.0040 (5.92)
Real interest rate -.0222
(-0.08)
-.0586 (-0.33)
-.0222 (-0.11) Intercept .0091
(14.81)
.0095 (1.06)
.0094 (13.32)
.0030 (1.32)
.0053 (0.70)
.0009 (0.27)
R2 .1030 .0177 .0178 .1871 .3030 .3030
Adjusted R2 .0976 .0150 .0150 .1723 .2972 .2971
