This paper examines the impact on political risk of having a divided government in the United States. We consider almost 60 years of data and use stockmarket return volatility as a measure of risk. Results show a positive and statistically significant relationship between periods of divided government and higher volatility. Divided governments are associated with an increase of 2.8 percentage points in annual volatility. Divided branch governments are found to lead to an increase of 4.1 percentage points in volatility, whereas a divided legislative government is linked to an increase in volatility of 5.8 percentage points. The President’s party does not seem to be, in itself, a driver for market volatility. However, a Democrat President coinciding with a unified government leads to a significant decrease in volatility of 2.8 percentage points. Overall, our findings support the view that divided governments increase political risk. This result lends support to the balancing model and is difficult to reconcile with gridlock theory.
Este trabajo examina el impacto en el riesgo político de tener un gobierno dividido en los Estados Unidos. Consideramos casi sesenta años de datos y utilizamos la volatilidad en el precio de las acciones como medida del riesgo. Los resultados muestran una relación positiva y estadísticamente significativa entre los estados de gobiernos divididos y una mayor volatilidad. Los gobiernos divididos están asociados con un aumento de 2,8 puntos porcentuales en la volatilidad anual. Se concluye que los gobiernos con ramas divididas conducen a un aumento de 4,1 puntos porcentuales en la volatilidad, mientras que un gobierno legislativo dividido está vinculado a un aumento de 5,8 puntos porcentuales en esa misma variable. El partido del presidente no parece ser, en sí mismo, un motor de la volatilidad del mercado, pero un presidente demócrata bajo un gobierno unificado lleva a una disminución significativa de la volatilidad de 2,8 puntos porcentuales. En general, nuestros hallazgos respaldan la opinión de que los gobiernos divididos aumentan el riesgo político. Este resultado favorece el modelo de equilibrio politico y es difícil de conciliar con la teoría de bloqueo.
Governments shape the economic environment in which businesses operate: they may launch new taxes, grant subsidies, promote investments in infrastructure projects or in specific industries, and implement regulatory measures. Because these discontinuities in the business environment resulting from political change may be difficult to anticipate, firms face what might be called political risk.
Political risk is an important notion that is often brought up in institutional economics, finance and political economy. It may thus be defined as the “uncertainty about the impact of an administration’s future policies” (
While there has been increasing academic interest in the intersection of politics and economics, relatively little attention has been paid to the relationship between patterns of institutional control (divided government vs. unified government) and political risk. This omission in the case of the studies pertaining to the US political system is especially surprising given the extraordinary importance of that economy and the increasingly common occurrence of divided governments in that country since the 1980s.
In this paper we fill this gap by exploring the impact of government on political risk in the US. In particular, we empirically assess whether having a divided government (as opposed to a unified government) has a significant impact in the political risk to which the United States’ firms (and thus, stock market investors) are exposed. Furthermore, we analyse the combined effects of partisan effects (leftwing governments vs. rightwing governments) and government status on political risk.
Following the testing framework proposed by Füss and Bechtel (
Our results show a positive and statistically significant relationship between divided states of government and higher volatility. Divided governments are associated with an increase of 2.8 percentage points in annual volatility even after controlling for a set of economic and political variables. We also examined the importance of two different forms of divided government distinguishing between divided branch government and divided legislative government. Divided branch governments are found to lead to an increase of 4.1 percentage points in volatility whereas a divided legislative government is linked to an increase of 5.8 percentage points in that same variable. The President’s party does not seem to be, in itself, a driver for market volatility, but a Democrat President under a unified government leads to a significant decrease in volatility of 2.8 percentage points.
Overall, these results suggest that partisan conflict between the executive and legislative branches do not affect the possibility of economic policy change, thus lending support to the balancing model proposed by Fiorina (
The remainder of the paper is structured as follows. In the next section, we present the theoretical framework and discuss the relationship between government status and political risk. In Section 3, we describe our sample and explain the estimation strategy and the specification of the models. Section 4 reports the outcomes of the estimation of the models. Finally, section 5 discusses these results and concludes the paper proposing avenues for further research.
According to Menefee-Libey (
The impact of government status on the volatility of stock returns is based on the premise that the value of a firm is equal to the present value of its expected cash flows, whereas the discount rate represents investors’ required rate of return. If there is uncertainty regarding the future implementation of economic policies, the range of realizations for expected cash flows and discount rates for individual firms and for the market as a whole should be wider and the variance of returns should increase accordingly.
The theoretical literature regarding the expected impact of the government status on the market’s volatility provides no consensus. In fact, the balancing model and the gridlock theory predict opposing effects. On one hand, the balancing model proposed by Fiorina (
In order to understand how the causality goes from government status (divided government vs. unified government) to the financial market, it is important to appreciate the main drivers that may lead to a different level of political risk as a consequence of having a different pattern of institutional control. Those drivers, according to the way through which they may have an impact, may be divided in two main types: economic and political.
Regarding the economic consequences, it is fair to say that there is a consensus that the government status matter to the economy. For example, Lohmann and O’Halloran (
In what concerns the second set of factors (political factors) through which divided government may result in a different level of political risk, there is a lower level of consensus. While some authors such as Mayhew (
Other authors sustain that the analysis of successful legislation may lead to biased inferences. For instance, Edwards
An additional factor that may strengthen the relationship between government status and stock market’s volatility is the presidential attitude, which is plausible to be different under a divided or a unified government. According to Nicholson
When it comes to the impact of the government status on the political risk, reflected in the stock market volatility, there are to our best knowledge only two published studies. The first one was conducted by Bechtel and Füss (
The second study was carried out by Kim
Partisan differences between Democrats and Republicans are found to be important in explaining political risk (
Overall, these studies highlight the importance of understanding the relationship between the political power in general and the government status in particular, and the uncertainty felt in the economy. Moreover, the divergent findings of scholars suggests that a more systematic examination of divided government is necessary. However, to the best of our knowledge, there are no studies looking at the effect of government status on political risk in the US. With the current paper we intend to fill this gap in the literature.
Our sample covers a period of 58 years, from 1950 until 2007, which compares favorably with similar studies, such as the already mentioned Füss and Bechtel (2008), which only included 35 years of data. We decided not to include in the sample the period after 2008 since from that year on, with the onset of the crisis in the subprime segment, the US stock market was marked by several episodes of high volatility which are hardly attributable to government status.
Data on stock market returns was retrieved from Thomson Reuters DataStream and from Quandl. The stock market index used to perform the analysis was the S&P 500 since it is the main indicator for the overall performance of the US stock market.
We define stock market volatility as the 20-day standard deviation of returns. This is a common way to measure volatility in the finance literature. We compute it by firstly converting close prices into a logarithmic return series. From this return series we then computed the 20-day standard deviation of returns and annualized the values obtained by multiplying them by the square root of the number of trading days in a year.
By looking at a scatter plot of volatility through time, one can point out that the periods of high stock market volatility seem to coincide with years in which the US witnessed financial crashes. In fact, the peaks of volatility associated with the first oil shock in 1973/1974 as well as those resulting from the stock market crashes in October 19, 1987 (“the Black Monday”) and in 2002 (“tech bubble crash”) are well visible in the figure.
The volatility histogram as well as the most relevant descriptive statistics pertaining the volatility data series are presented in
The volatility data set has a mean of 0.1329 and a standard deviation of 0.0511. The maximum shown in the figure below, 0.3373, and the minimum, 0.0526, concern to the years of 1987 and 1964, respectively.
As explained before, the main explanatory variable in this study is a dummy variable, which is set to 1 in the case of a divided government (cases in which the political party of the President does not control both the Senate and the House of Representatives), and 0 otherwise. It is also important to point out that, from 1950 until 2007 divided government is the predominant status of government (35 years, against 23 years of unified governments).
From the analysis of
Finally, the configuration of the box-plot and the positive values for the skewness statistic indicate that in periods of divided government the distribution of the volatility is skewed to the right, i.e., there is a significant number of large volatility events. This suggests, once again, that it is important to control for the presence of outliers in the multivariate analysis.
Regarding the data related to the US government status in the period under scrutiny, we used a database developed by Baumgartner
Obviously, in order to understand whether the differences in the stock market volatility can be attributed to the government status, one needs to consider a set of control variables.
In order to examine if there is a systematic influence of the government status on the stock market volatility, one needs to develop a model that includes a set of control variables. In this regard, the set of control variables to be included may be divided in two groups: economic/financial variables and political variables.
Firstly, in what concerns the set of economic/financial variables, we control for the United States’ GDP growth rate, inflation, federal deficit (in percentage of GDP), and years of stock market crashes. Data on the GDP growth rate were retrieved from the Bureau of Economic Analysis - Department of Commerce. Data on inflation and federal deficit were retrieved from the Federal Reserve Bank of St. Louis. Given the impact of these variables on financial markets’ volatility, they are commonly used as control variables in the fields of political and financial economics (
When it comes to political variables, we considered a total of three variables. The first two control variables intend to capture the depth of the union or division of the government, from two different angles. The first variable is “Distance” which refers to the ideological distance between the majority and the opposition during divided government in the House of Representatives. It assesses how far, in terms of ideology, the government and the opposition are (
In order to empirically test whether the government status does ultimately affect stock market volatility, we follow the testing framework suggested by Bechtel and Füss (
Annual Volatility =
Annual Volatility =
Annual Volatility =
We also develop a different specification of divided and unified government. Thus, three possibilities of government status will be considered: unified government (President plus both chambers belonging to the same party), weak divided (President plus one chamber belonging to the same party) and strong divided (President belonging to a party different from the one that controls both chambers). In these models, we will exclude the dummy variable that accounts for the existence of crash years since it raises concerns of multicollinearity. This new specification will be tested using the following two models:
Annual Volatility =
Annual Volatility =
Finally, to enhance the quality of the analysis, we will develop three additional models. With these models, we will examine whether the feature “political party” is of relevance in the issue under study. It could be the case that times of higher (or lower) volatility are intrinsically related to a specific President’s party, and not necessarily related to the government status. In the first of those two models the government status will not considered as an explanatory variable since we only intend to understand if there is any relationship between annual volatility and the president’s party. In the second model, we will analyze if there is any significant difference between the cases of a democratic and a republican unified governments. So, we take both aspects into account: presidential party and government status. The last model will include, in addition, a set of control variables. The models are as follows:
Annual Volatility =
Annual Volatility =
Annual Volatility =
|
|
|
|
---|---|---|---|
Constant | 0.1068 |
0.0925 |
0.0981 |
|
|
|
|
Inflation | 0.1940 |
0.2538 |
|
|
|
|
|
Deficit | 0.2632 |
0.1123 |
|
|
|
|
|
Presidential Election | -0.0227 |
-0.0251 |
|
|
|
|
|
Distance | -0.0002 |
||
|
|
||
N | 58 | 58 | 50 |
R-squared | 0.1750 | 0.6381 | 0.5790 |
Adjusted R-squared | 0.1603 | 0.5874 | 0.4843 |
F-statistic | 11.883 | 12.5954 | 6.1146 |
Prob. (F-statistic) | 0.0010 | 0.0000 | 0.0000 |
Prob. (Wald F-statistic) | 0.0006 | 0.0000 | 0.0000 |
Notes: The dependent variable is the annual volatility measured by the annualized standard deviation of the S&P500’s daily returns from 1950 to 2007 in the case of models (1) and (2) and from 1950 to 1999 in the case of model (3). Estimates in the case of models (1) and (2) follow the OLS approach with a HAC Newey-West estimator (p-values in parenthesis) and in the case of model (3) with a White- heteroskedasticity-consistent estimator given that with the HAC estimator there were signs of multicollinearity. Divided Government, the explanatory variable, is a dummy variable that takes the value of 1 if the presidential party does not control both congressional chambers, and 0 otherwise. Inflation, GDP growth, Deficit, Crash Year, Distance and Cohesiveness are variables to control for the economic environment. Presidential Election and Congress Election are dummy variables to control for election years.
10 %
5 %
1 %
, represent significance at the 10 %, 5 % and 1 % levels respectively.
|
|
|
---|---|---|
Constant | 0.1068 |
0.1012 |
|
|
|
Weak Divided | 0.0557 |
0.0586 |
|
|
|
GDP growth | -0.0429 |
|
|
|
|
Presidential Election | -0.0340 |
|
|
|
|
N | 58 | 58 |
R-squared | 0.1859 | 0.2842 |
Adjusted R-squared | 0.1563 | 0.1840 |
F-statistic | 6.2822 | 2.8363 |
Prob (F-statistic) | 0.0034 | 0.0143 |
Prob (Wald F-statistic) | 0.0020 | 0. 0001 |
Notes: The dependent variable is the annual volatility measured by the annualized standard deviation of the S&P500’s daily returns from 1950 to 2007. Estimates follow the OLS approach with a HAC Newey-West estimator (p-values in parenthesis). Strong Divided is a dummy variable that takes the value 1 if the presidential party does not control any of the congressional chambers, and 0 otherwise. Weak Divided is a dummy variable that takes the value 1 if the presidential party controls only one of the congressional chambers, and 0 otherwise. Inflation, GDP growth, and Deficit are variables to control for the economic environment. Presidential Election and Congress Election are dummy variables to control for election years.
10 %
5 %
1 %
, represent significance at the 10 %, 5 % and 1 % levels respectively.
The results are congruent to the previous findings. Once again, the models are globally significant at the conventional levels. Both the explanatory variables are positive and significant at a 95 % confidence level in models (4a) and (4b). We can observe that a “weak divided” government is expected to increase the annual volatility by 5.5 percentage points and 5.8 percentage points in models (4a) and (4b), respectively. On the other hand, a “strong divided” government is shown to provoke an increase in the annual volatility by 4 percentage points and 4.1 percentage points in models (4a) and (4b), respectively. Note that it would probably be expected that, in line with the previous findings, a strong division would yield an higher uncertainty than a weak division. However, the classification of “strong” and “weak” is rather ambiguous. In this case the classification is assumed from the standpoint of the President (President against the chambers). If a distinct perspective was assumed the results would, naturally, be switched.
Models (5a), (5b), and (5c) are developed to understand if the party of the president is, by itself, responsible for a different level of volatility.
|
|
---|---|
Constant | 0.1134 |
|
|
Inflation | 0.2140 |
|
|
Deficit | 0.2056 |
|
|
Presidential Election | -0.0183 |
|
|
N | 58 |
R-squared | 0.607602 |
Adjusted R-squared | 0. 552666 |
F-statistic | 11.06022 |
Prob (F-statistic) | 0.000000 |
Prob (Wald F-statistic) | 0. 000000 |
Notes: The dependent variable is the annual volatility measured by the annualized standard deviation of the S&P500’s daily returns from 1950 to 2007. Estimates follow the OLS approach with a HAC Newey-West estimator (p-values in parenthesis). Democratic President is a dummy variable that takes the value 1 if the president belongs to the Democratic Party, and 0 otherwise. Inflation, GDP growth, Deficit, and Crash Year are variables to control for the economic environment. Presidential Election and Congress Election are dummy variables to control for election years.
10 %
5 %
1 %
, represent significance at the 10 %, 5 % and 1 % levels respectively.
|
|
|
---|---|---|
Constant | 0.1500 |
0.1148 |
|
|
|
Unified Republican | -0.0377 |
-0.0169 |
|
|
|
GDP growth | 0.1310 |
|
|
|
|
Crash Year | 0.1127 |
|
|
|
|
Congress Election | 0.0092 |
|
N | 58 | 58 |
R-squared | 0.1767 | 0.6414 |
Adjusted R-squared | 0.1468 | 0.5829 |
F-statistic | 5.9050 | 10.958 |
Prob (F-statistic) | 0.0047 | 0.0000 |
Prob (Wald F-statistic) | 0.0030 | 0.0000 |
Notes: The dependent variable is the annual volatility measured by the annualized standard deviation of the S&P500’s daily returns from 1950 to 2007. Estimates follow the OLS approach with a HAC Newey-West estimator (p-values in parenthesis). Democratic President is a dummy variable that takes the value 1 if the president belongs to the Democratic Party, and 0 otherwise. Inflation, GDP growth, Deficit, and Crash Year are variables to control for the economic environment. Presidential Election and Congress Election are dummy variables to control for election years.
10 %
5 %
1 %
, represent significance at the 10 %, 5 % and 1 % levels respectively.
The results remain, in general, in line with the previous findings. Unified governments are clearly associated with times of lower volatility in the stock market. Moreover, in this case we can see the difference between having a unified government with a democrat president and a unified government with a republican president. In model (5b), both the explanatory variables are statistically significant at the 1 % level. Democratic unified governments tend to be associated with the largest decrease in the stock market volatility: 4.5 percentage points against a decrease of about 3.8 percentage points in the case of a republican unified government. However, in model (5c), when all the control variables are considered, a republican unified government ceases to have any statistically significant effect in the stock market volatility. This may be due to the fact that the sample only included 6 years of such situation, which naturally makes the statistical test to become more demanding. Notwithstanding, the dummy “democratic unified government” is still significant at the 1 % level, signaling a decrease of 2.87 percentage points in the market volatility, even after considering the complete set of control variables.
Overall, the abovementioned set of models show, on a consistent basis, that divided governments in the US are associated with a higher stock market volatility. In the main models, the effect ranges from 2.5 percentage points to 4.8 percentage points. Moreover, the later models show that we can exclude the possibility of such result being driven by the difference in the presidential party.
We subject the models to a series of robustness tests to verify the results. Robustness checks include testing for omitted variables bias, testing for autocorrelation and testing for multicollinearity between explanatory variables.
Regarding the presence of omitted variables, we applied the standard Ramsey Regression Equation Specification Error Test (RESET) which indicated that the models do not suffer from functional misspecification, with a high degree of confidence.
Even though all the models were estimated with a HAC Newey-West which enables the statistical inference even in the presence of heteroskedasticity and autocorrelation, we conducted some tests to understand whether there is a high degree of serial correlation in the data. We used the standard Durbin-Watson test and the Ljung-Box test, which led to the conclusion that from all the models of the study only model (1) suffered from autocorrelation issues for both the lag 1 and lag 2. In the models (4b) and (5b) the problem is limited to the first lag only.
Although the potential multicollinearity problems have been dealt with in the course of the models’ estimation phase, we conducted a test based on Variance Inflation Factors (VIF). Those factors test the magnitude to which the variance of estimated coefficients is inflated because of multicollinearity issues. We concluded that there were no signs of extreme collinearity between variables that would decrease the meaning of the coefficients that were previously estimated.
Finally, a word should be said about the possibility of reverse causality in the correlation we are examining. It is not likely that our results could be affected by endogeneity problems since it is highly doubtful that the volatility of stock markets could play any relevant role in the definition of the US government status. This conjecture is confirmed by the results obtained with the Granger causality test. In fact, this test indicates that the direction of the (Granger) causality at the horizon of one year is from the divided government to the US stock market volatility and that there is no signs of reverse causation.
Overall, we found our results to be fairly robust so we are able to draw strong conclusions regarding the topic at hand.
Government plays a paramount role in modern economies. In this paper we used almost 60 years of data from the US to analyse the impact of the pattern of institutional control (divided government vs. unified government) on political risk. Following a recent trend in the political science literature, the stock market volatility was used as a measure of risk. Our empirical evidence strongly suggests that political environments in the US characterized by a divided government tend to increase stock market volatility. Divided governments are found to increase the annual volatility by 2.8 percentage points even after controlling for a set of economic and political variables. We also examined the impact of two different forms of divided government distinguishing between weak divided government (the presidential party controls only one congressional chamber) and strong divided government (the presidential party does not control any of the congressional chambers). Strong divided government is found to lead to an increase of 4.1 percentage points in volatility whereas a weak divided government is linked to an increase of 5.8 percentage points in that same variable. The President’s party does not seem to be, in itself, a driver for market volatility, but a Democrat President under a unified government leads to a significant decrease in volatility of 2.8 percentage points.
Overall, the present study challenges the gridlock theory and contradicts the conclusions reached by Bechtel and Füss (
One way of explaining our results relates to investor expectations and the effect such expectations may have on stock market prices. Stock market returns are more volatile in times of uncertainty. The gridlock theory posits that such uncertainty is abated when a government is divided because policies shaping business environments are harder to approve and implement. However, one could argue in alternative that in times of unified government, expectations regarding the policies to be implemented are clearer which decreases investor uncertainty and reduces stock market volatility. Moreover, under divided governments investors may be less capable of understanding the overall political agenda, since there are two different parties trying to set the tone in political terms. This possibility is especially relevant in that the literature suggests that Democrats and Republicans have a significantly different historical record in terms of the fiscal and regulatory policies they try to implement. Indeed, political parties can exert a significant effect on policy outcomes. For example, Alt and Lowry (
Another way of explaining these results is based on the research of Nicholson
Finally, our findings can also be understood at the light of the recent model developed by Azzimonti (
Low risk is crucial to any well-functioning economy, as it stimulates capital investment, facilitates growth, and enhances overall economic performance. An increase in stock market volatility may lead to the deterioration of capital investment by risk-averse investors. Considering such effects, our analysis suggests that states of divided government might be a key explanatory variable on the attraction of capital stock thus indirectly affecting consumption, investment and economic growth.
Our findings are also relevant to the stream of literature on the electoral causes of divided government. There is a growing body of evidence suggesting that moderate voters intentionally bring about divided government by voting for parties whose ideal points may strongly differ from their own preferred policies (
The veto players’ theory developed by Tsebelis (
We believe our findings carry serious implications for future research on political economics and financial markets since they suggest that government status is a variable that should be considered in the study of economic uncertainty.
There is much more to investigate regarding the impact of divided government. First, given that conflicting results are reached for different countries, it would be important to discern the underlying drivers of the market volatility that can be attributed to the government status. Is it a matter of investors’ perception and country’s national culture, or is it a matter of the specific political framework in which the country operates on? Whereas the impact of the patterns of institutional control is not dependent upon whether the risk is objective or subjective, the nature and extent of their contribution to risk certainly is. If uncertainty is objective, the contribution of political decisions to risk is a function of only the decisions themselves. If uncertainty is subjective, the contribution of political risk will be a function of both the decisions and the economic agents’ cognitive processes. Second, there is issue of the direction of causality between the government status and the stock market volatility. In this paper, we have implicitly assumed that political variables are exogenous events, that is, were not influenced by the volatility of capital markets. This assumption was given some support when we established that government status does not seem to statistically precede the evolution of the financial variable. However, there are contributions in the literature that suggest that the evolution of stock markets can significantly influence variables of political nature (
We subject the models to a series of robustness tests to verify the results. Robustness checks include testing for omitted variables bias, testing for autocorrelation and testing for multicollinearity between explanatory variables in all of the models that were applied to the full sample and that considered the full set of control variables. Finally, the robustness checks also include testing for endogeneity issues recurring to Granger causality tests.
Regarding the tests for omitted variables, we started by applying a standard Ramsey Regression Equation Specification Error Test (RESET) to models (2), (4b) and (5c). In the remaining models, the Ramsey’s RESET test cannot be performed since the only explanatory variables are dummy variables. The results are as follows:
|
|
|
---|---|---|
Model (2) | 0.6935 | Cannot reject |
Model (4b) | 0.5427 | Cannot reject |
Model (5b) | 0.5556 | Cannot reject |
Note:
From these results, we can say that none of the abovementioned models suffers from functional misspecification, with a statistically significant degree of confidence.
Even though all the models were estimated with a HAC Newey-West that enables the statistical inference in the presence of heteroskedasticity and autocorrelation, it is important to understand whether we have a high degree of serial correlation in the main models, as it sometimes happens with time series data.
We first recurred to a standard Durbin-Watson test, to assess if there is autocorrelation of first order (following AR1 processes), and the results were as follows:
|
|
|
||
---|---|---|---|---|
|
|
|||
Model (1) | 1.2534 | 1.356 | 1.428 | Reject |
Model (2) | 1.8766 | 1.134 | 1.685 | Cannot Reject |
Model (4a) | 1.3262 | 1.320 | 1.466 | Inconclusive |
Model (4b) | 1.3950 | 1.134 | 1.685 | Inconclusive |
Model (5b) | 1.2713 | 1.320 | 1.466 | Reject |
Model (5c) | 1.9238 | 1.095 | 1.734 | Cannot reject |
Note:
Firstly, it is important to reinforce that all the models were estimated using a Newey-West HAC estimator that allows statistical inference even in cases of serial correlation. Notwithstanding, it is also important to point out that only in the simplest models we reject the null hypothesis. Those models have only one dummy or a pair of dummies as explanatory variables. In the core models of our analysis, e.g., in models (2), (4b) and (5c), that include all the control variables, the possibility of autocorrelation problems is not excluded by the Durbin-Watson test. Thus, to better assess the issue, we performed a more sophisticated test. Since the Breusch-Godfrey test only has asymptotical validity (its results are not valid for relatively small samples, as it is the case), we run a Ljung-Box test, with two lags, for all the six abovementioned models. The results were as follows:
|
|
|
|
|
---|---|---|---|---|
Model (1) | 0.004 | Reject |
0.017 | Reject |
Model (2) | 0.770 | Accept |
0.924 | Accept |
Model (4a) | 0.010 | Accept |
0.04 | Accept |
Model (4b) | 0.007 | Reject |
0.022 | Accept |
Model (5b) | 0.005 | Reject |
0.021 | Accept |
Model (5c) | 0.920 | Accept |
0.94 | Accept |
Note:
So, as we can see, most of the models do not suffer from autocorrelation issues. Only model (1) presents some autocorrelation issues for both the lag 1 and lag 2. In the remaining models, models (4b) and (5b), the problem concerns the first lag only.
Regarding the issue of multicollinearity, we resorted to the computation of Variance Inflation Factors (VIF) for those models were this problem can be of importance, i.e., we excluded the models with only one or two dummy explanatory variables. The results of the multicollinearity tests were as follows:
|
|
|
|
|
|
|
|
|
|
---|---|---|---|---|---|---|---|---|---|
Model (2) | 1.65 | 1.95 | 3.21 | 3.38 | 4.24 | 4.27 | 1.66 | – | Cannot reject |
Model (4b) | 1.33 | 2.56 | 1.66 | 3.06 | 2.29 | 2.31 | 1.87 | – | Cannot reject |
Model (5c) | 2.18 | 3.54 | 1.81 | 2.14 | 1.35 | 3.35 | 4.69 | 4.50 | Cannot reject |
Note: The threshold used in this test is the standard value in the literature: as it is usually considered, multicollinearity issues are raised when VIF stats are above 5 for one or more variables.
As it can be seen in
Regarding the issue of endogeneity, we recurred to the estimation of Granger causality between the variable that captures the existence of a divided government and the variable that reflects the volatility in US stock markets. The results of the Granger causality tests are as follows:
|
|
|
|
|
---|---|---|---|---|
Stock Market Volatility does not cause Divided Government | 1 | 0.10 | 0.75 | Do not reject |
Divided Government does not cause Stock Market Volatility | 1 | 4.58 | 0.04 | Reject |
Stock Market Volatility does not cause Divided Government | 2 | 0.87 | 0.42 | Do not reject |
Divided Government does not cause Stock Market Volatility | 2 | 2.17 | 0.12 | Do not reject |
Stock Market Volatility does not cause Divided Government | 4 | 0.40 | 0.81 | Do not reject |
Divided Government does not cause Stock Market Volatility | 4 | 1.26 | 0.30 | Do not reject |
Note: results are drawn upon a 5 % significance level.
Considering one lag in the estimation, the results show that the direction of the (Granger) causality seems to be from the divided government to the US stock market volatility at the horizon of one year. On the other hand, there are no signs of reverse causation from the stock market volatility to the government status since the computed F value was not statistically significant.
However, at two or more lags, there is no statistically discernible relationship between the two variables.
Overall, there seems to be a unidirectional causality of Granger from the divided government to the US stock market volatility.