Examining the Relationship between Expected Market Returns and Political Variables Using Generalized Method of Moments Procedures

by

Print Friendly, PDF & Email

Ian Wright and Dr. Keith Vorkink, Finance

Investors, economists and financial analysts are constantly seeking to develop models that may help them learn what the stock market will do and how it functions. They do this so that they may take advantage of any market anomalies as soon as they appear and try to make “easy money.” Various pricing models (see Sharpe (1964), Litner (1965) and Fama and French (1993) for example1) have been and continue to be constructed that try to predict returns, and variables suspected to be correlated with returns have been included in these models in an attempt to determine specific predictors of returns.

Due to the large role the government can play in the state of the economy, it is quite natural to investigate the potential predictability that political variables, such as the partisan composition of the United States Presidency, give to returns in the stock market. Various studies have attempted to uncover whether any statistically significant correlations between political variables and returns exist (for one example see Santa-Clara and Valkanov (2003)), with varying results. Some researchers claim no such significant relationships exist, while others claim that returns are higher during gridlock, others during Democratic administrations, and so forth. There has been no answer that has been generally accepted by the financial world.

Conflicting results and lack of a generally accepted relationship between, in particular, the partisan composition of the U.S. Presidency and returns in the U.S. stock market, led me to take a new approach in investigating this relationship in 2006-2007, with hopes of shedding additional light on the puzzle. I built on my work then with further extensions this year (2008-2009). Of the studies done previously, most have employed a method of estimation called Ordinary Least Squares (OLS), or variations of it. In hopes of finding firm and robust results, I decided to use a method of estimation known as Generalized Method of Moments (GMM), which encompasses OLS as a special case, but, in general, is broader in its capabilities and possibilities for use in estimation. In 2006-2007 I used a given set of moment restrictions and only generated GMM point estimates, without performing any significance tests of those point estimates. This past year I completed the previous GMM estimation by performing significance tests, and revamped some of the theory behind my estimation.

I obtained data for the project from the CRSP (Center for Research in Security Prices) database using the years 1946 to 2005, and chose models of estimation based on Newey (1988), Fama and French (1993), Santa-Clara and Valkanov (2003) and the suggestions of my mentor. These details are fully discussed in the completed work. Due to the particular GMM estimation I performed, I found it to my advantage to code using the programming language Matlab, where I could freely code my own estimators. In terms of my learning and growth, coding in Matlab was far more productive than simply using a canned statistical package, although it was more difficult along the way. Initially I had no a priori hypothesis as to whether or not returns would be significantly affected by political variables, due to the conflicting voices presented in the recent literature.

Results showed that GMM and OLS estimates of various coefficients were quite close to one another in both sign and magnitude, which is not surprising, given that the GMM moment restrictions I employed were simply an extension of the OLS first order conditions. Specifically, I found that statistically significant differences between returns in Republican and Democrat presidential administrations exist in some particular models where various macroeconomic factors are controlled for, but less so in models including the Fama and French factors. In models where such differences are significant, returns are estimated to be anywhere from 5 to over 20 percent higher (on an annualized basis) when a president is a Democrat relative to a Republican. I also found that a president serving an additional term in office has no significant effect upon returns. Additionally, preliminary investigation of the effect of the partisan demographic of Congress on returns revealed that the partisanship of Congress has no statistically significant effect upon returns, neither in conjunction with, nor independent of presidential partisanship.

Some ideas for further work include obtaining the heteroskedastic auto-correlated (HAC) corrected standard errors of the GMM parameter estimates and then using those to make inference, as well as varying the moment restrictions used in GMM estimation. Additionally, a question of reverse causality could also be looked at by analyzing returns near political elections: do returns drive the political variable even if the political variables do not drive returns?

The complete 70-page work, entitled “Examining the Relationship between Expected Market Returns and Political Variables Using Generalized Method of Moments Procedures,” is available from the author upon request.

References

  • Fama, Eugene F. and French, Kenneth R., (1993), “Common risk factors in the returns on stocks and bonds,” Journal of Financial Economics, Vol. 33, No. 1, 3-56.
  • Litner, John, (1965), “The valuation of risk assets and the selection of risky investments in stock portfolios and capital budgets,” Review of Economics and Statistics, Vol. 47, 13-37.
  • Newey, Whitney K., (1988), “Adaptive Estimation of Regression Models Via Moment Restrictions,” Journal of Econometrics 38, 301-339.
  • Santa-Clara, Pedro, and Valkanov, Rossen, (2003), “The Presidential Puzzle: Political Cycles and the Stock Market,” The Journal of Finance, Vol. 58, No. 5, 1841-1872.
  • Sharpe, William F., (1964), “Capital asset prices: a theory of market equilibrium under conditions of risk,” The Journal of Finance, Vol. 19, 425-442.