Congress and the Stock Market

Transcription

1 March 13, 2006 Congress and the Stock Market by Michael F. Ferguson* and H. ouglas Witte** JEL Codes: G1, G10, G14, G18 *epartment of Finance, College of Business, University of Cincinnati, P.O. Box , Cincinnati, Ohio ; (513) ; **College of Business, The University of Missouri, Columbia, ; (573) ; We thank Ed Kane, Richard Shockley, and Steve Wyatt for helpful comments on an earlier version of this paper.

2 Congress and the Stock Market We find a strong link between Congressional activity and stock market returns that persists even after controlling for known daily return anomalies. Stock returns are lower and volatility is higher when Congress is in session. This Congressional Effect can be quite large more than 90% of the capital gains over the life of the JIA have come on days when Congress is out of session. The Effect varies systematically with the public's opinion of Congress: returns are lower and volatility higher when a relatively unpopular Congress is active. Public opinion appears to play a fundamental role in market prices. This is consistent with a mood-based explanation that sees Congress as depressing the average investor. Alternatively, our results can also be reconciled with rational explanations that view Congressional activity as a proxy for regulatory uncertainty or rent-seeking behavior.

3 This country has come to feel the same when Congress is in session as we do when a baby gets hold of the hammer. It s just a question of how much damage he can do with it before we take it away from him. --Will Rogers 1 Throughout much of American history Congress has been unpopular. The epigraph is anecdotal evidence of this. 2 Empirically, over the past 40 years Congress has averaged a 41% approval rating versus a 48% disapproval rating. 3 Anecdotes and the polling data suggest that the public views Congressional activity skeptically; i.e., on average, an active Congress is considered bad news. Could such a reaction have a direct effect on stock prices? The burgeoning field of behavioral finance suggests that the answer is yes. Congressional activity may--for some heretofore unidentified and perhaps less than fully rational reason--affect stock investors moods and attitudes. Hirshleifer (2001) cites a wide range of research from the psychology literature suggesting that evaluation of information and attitudes toward risk are profoundly affected by mood. In particular, depression leads to greater risk aversion and more pessimistic forecasts than upbeat moods. Hirshleifer and Shumway (2003) argue that the Sunshine Effect in US equity markets (first identified by Saunders (1993)) is an example of the impact of mood on asset valuations. Similarly, Kamstra, Kramer, and Levi (2003) document a seasonal affective disorder (SA) effect in stock returns around the world related to the hours of daylight during the day. Thus, it is possible that Congress being in session depresses the market in a fashion similar to other mood dampening influences such as the SA and Sunshine Effects. 4 In this paper we seek to find out if there is in fact any connection between Congressional activity and stock market returns. Numerous studies have investigated the partisan effects of Presidential elections and/or 1 Congress Session, Rogers Says, Is Like Baby Getting a Hammer, (July 5, 1930), New York Times, pg Ridiculing Congress was a staple of the populist humor of both Will Rogers and, in an earlier era, Mark Twain. 3 Source: Harris and Gallup polls We will rely on these polls of the general public as a proxy for investors attitudes toward Congress. 4 In this analogy, Congress being in session is like a day without sunshine. 1

4 administrations on the stock market. 5 In addition, numerous studies have examined the impact of specific legislative actions (e.g., see Mitchell and Netter (1989)) on financial markets. In contrast, our interest is in the reaction of the broad market to Congressional activity. Given the strong negative reaction that frequently greets almost any political activity, we want to know if there is a Congressional Effect that is analogous to other, apparently mood-driven, effects like the Sunshine Effect or SA Effect. We find the following. There is a strong relationship between daily stock returns and a proxy for Congressional activity. Returns are significantly higher and volatility is significantly lower when Congress is out of session. This Congressional Effect persists after controlling for previously documented stock market seasonalities such as the January Effect and Pre-Holiday Effect. Our results give empirical credence to the (largely anecdotal) association of an active Congress with poor stock market returns. It is possible of course that these results are spurious. We conduct a number of robustness checks to ensure that this is not a mere statistical aberration or rediscovery of some other previously identified anomaly, but it is impossible to rule out entirely the possibility that this is a lucky draw from the data. 6 However, if a mood-based explanation is to carry any weight, at a minimum the Congressional Effect should be correlated with investors attitudes towards Congress. For example, investors presumably do not view all Congressional activity as bad news. Their assessment should depend upon the specific actions contemplated by Congress. It is possible that Congressional actions will be viewed sufficiently favorably so as to offset any mood- -induced increase in the market risk premium. uring such periods Congressional activity will increase the value of publicly traded firms. For example, as the financial fraud and accounting scandals of the late 1990s came to light, investor s confidence in the markets may have been enhanced by the Congressional inquiries and hearings that ultimately led to Sarbanes-Oxley. In general, we anticipate that investors will evaluate Congressional activity through time and adjust their expectations accordingly. Therefore, if the Congressional Effect we document is not simply a spurious correlation, there should be a systematic 5 Siegel (2002) summarizes the evidence. Santa-Clara and Volkanov (2003) is a good recent example. 6 We have a tiny bit of out of sample evidence. The main results in this paper all got stronger when we updated an 2

5 relationship between investors perceptions of Congressional activity and the Congressional Effect. The more strongly the public disapproves of Congressional activity, the lower returns should be when Congress is active. We investigate the consistency of investors opinions and the Congressional Effect by utilizing public polling data as a proxy for investor s attitudes toward the general direction of Congressional activity. The evidence indicates that investors attitudes toward Congress play a significant role in the Congressional Effect. The Congressional Effect varies systematically with our proxy for stock market investors disapproval of Congress. Stock returns are lowest and volatility highest when Congress is in session and the public has a relatively low opinion of Congress. However, our regression results show that the Congressional Effect disappears when the public has an unusually high opinion of Congress. Thus, our results suggest that an active Congress, per se, is not enough to induce poor stock market returns. Public opinion plays a fundamental role. Finally, we employ two methodologies to assess the economic significance of the Congressional Effect. We first look at Kandel and Stambaugh s (1996) model of how an observed statistical relationship affects an investor s portfolio allocation between cash and the market index. We also consider the Britten-Jones (1999) model of the impact a strategy has on an investor s optimal portfolio. The result of each analysis is that strategies based on the Congressional Effect can have a significant impact on an investor's asset allocation decision. Our findings are consistent with a mood-based explanation of the Congressional Effect. On the other hand, we can think of two potential explanations that fit the data that do not rely on the mood of the market or investor psychology. These can be characterized as the regulatory uncertainty and rent seeking explanations. First, while Congress is in session uncertainty exists about whether the tax and regulatory constraints that publicly traded firms confront will be altered. Information about potential changes in the tax and earlier version of the paper with data. 3

6 regulatory environment is generated during committee meetings, floor speeches, etc. In a competitive equity market this information will impact stock prices as it is learned. 7 One implication of this is that Congressional activity will be positively correlated with stock market volatility. At a minimum, when Congress is active it is more likely to alter existing economic arrangements. This increases uncertainty and, hence, the volatility of returns. Malkiel (1979) in his presidential address to the American Finance Association attributes sluggish economic returns in the 1970s to an increase in the risk-premium that investors required as compensation for greater exposure to a variety of risks. In particular, he cites the regulatory risk inherent in escalating business regulation as an important source of economic uncertainty. Significantly, Malkiel contends it is not so much the direct cost of regulation that has inhibited investment but rather the unpredictability of future regulatory change (emphasis added, p. 297). This suggests that the stock market might decline (i.e., the risk-premium might increase) during periods of greater regulatory uncertainty. The hypothesized negative ex post relationship between risk and realized return arises precisely because there is a positive ex ante relationship between risk and expected return. This volatility-feedback effect on market prices has been formally modeled and empirically estimated recently by Kim, Morley and Nelson (2004) and Mayfield (2004). 8 Our tests can be viewed as an investigation of Malkiel s (1979) conjecture that regulatory uncertainty is a negative influence on the stock market if we treat Congressional activity as a simple proxy for regulatory uncertainty. 9 In this interpretation we are comparing returns and volatility across periods of varying degrees of regulatory uncertainty. If Congress is in session we deem it to be active and regulatory uncertainty to be high, if Congress is out of session we deem it to be inactive and regulatory uncertainty to be low. 10 The 7 Mitchell and Netter (1989) provide a dramatic example. They argue that a tax bill containing anti-takeover provisions proposed in the U.S. House of Representatives triggered the 1987 stock market crash. 8 In general, the empirical relationship between volatility and realized market returns is ambiguous. The evidence is summarized in Scruggs (1998). The volatility-feedback hypothesis advanced by Malkiel (1979) was initially studied in Pindyck (1984), Poterba and Summers (1986), and French, Schwert, and Stambaugh (1987). 9 Regulatory uncertainty should be interpreted broadly. When Congress is in session uncertainty exists with respect to many factors (taxes, trade policy, spending, etc.) that may affect firm values. 10 An obvious objection to this explanation is that volatility feedback models require long-lasting regime changes to 4

7 conclusion in this case is the stock market reacts negatively to greater regulatory uncertainty. A second explanation that appears to be consistent with the data is to note that it may be perfectly rational for investors to, on average, dislike Congressional activity. There is a long literature that argues that in the political marketplace concentrated economic interests have the incentive and ability to overcome the broader public interest. 11 This rent-seeking activity may take many forms, but the bottom line, as Rajan and Zingales (2003) argue, is that there are powerful financial and industrial incumbents that actively seek to make markets work less efficiently, limiting access to a privileged few. Our tests can be viewed as an investigation of Rajan and Zingales (2003) thesis that powerful economic interests act as a negative influence on the stock market if we treat Congressional activity as a simple proxy for the level of rent seeking. In this interpretation, the public opinion polls reflect investor s attitudes toward Congressional activity. Moreover, these attitudes frequently reflect an accurate assessment of the wealth-destroying activities of the political class. So, under this interpretation the Congressional Effect is not a psychological reaction, but rather a rational interpretation of Congressional activities. The conclusion in this case is that the stock market reacts negatively to increased rent-seeking. The plan of the rest of this paper is as follows. In Section I we compare returns and volatility across days Congress is in and out of session. The main finding is that returns are higher and volatility is lower when Congress is out of session. In Section II we document that the Congressional Effect persists after controlling for known daily return anomalies. In Section III we control for public opinion and find that the Congressional Effect is driven by the public s attitude toward Congress. Section IV confirms that a similar Congressional Effect exists in volatility. Section V assesses the economic impact of the Congressional Effect. Section VI concludes. affect prices significantly. But, potential legislative changes are long-lasting. One way to think about this is that whenever Congress is in session there is a small chance that they will do something dramatic that has nearly permanent consequences. Thus, the markets are switching back and forth between a normal and a peso-risk regime as Congress goes out and back into session. 5

8 I. Stock Returns and the Congressional Calendar We compare daily returns when Congress is in and out of session for four major stock indexes the ow Jones Industrial Average (JIA), the S&P 500 index (S&P 500), the CRSP value-weighted index (CRSP VW), and the CRSP equal-weighted index (CRSP EW). The indexes we study encompass a wide range of firm types and time periods. The JIA consists primarily of large firms with return data spanning the entire 20 th century, having begun in The S&P 500 is more representative of the overall stock market and dates to 1957 in its current form. The CRSP VW index is the most broad based of the indexes, with coverage beginning in The CRSP EW index is similar to the CRSP VW; however, it places greater weight on small firms. We determined whether Congress was in session or not by consulting the Congressional Record. In Table I we report the mean daily returns when Congress is in and out of session as well as the number of days Congress is in and out of session. We also report the standard deviations of the daily return series. For every index except the JIA, daily volatility is lower when Congress is out of session. The differences range from 5-6 basis points per day and are statistically significant at well below the one percent level for each index. Volatility for the JIA is nearly 6 basis points higher when Congress is out of session over its entire sample period. However, over the same time period as the CRSP indexes, JIA volatility is actually 5.7 basis points lower (90.0 basis points vs basis points) when Congress is out of session, a difference significant at the one percent level. These results are generally consistent with the conjecture that relatively more information is generated on days Congress is in session. epending on the index, daily returns when Congress is in session range from 1 to 4 basis points per day. When Congress is out of session returns range from 5 to 15 basis points per day. The differences in returns range from 4 to 11 basis points per day and are statistically significant at the one percent level for each index. These daily return differences are smaller than those reported for daily 11 See, for example, Olson (1971), Stigler (1971), and Becker (1983). 6

9 anomalies such as the Pre-Holiday Effect, the Turn-of-the-Month Effect, or the January Effect. 12 However, Congress is out of session many more days per year than are affected by these anomalies. Congress has been out of session about 84 trading days per year for the 43 years covered by the CRSP indexes and Congress has average 117 trading days out of session over the 108 years of the JIA. Therefore, these somewhat smaller daily differences may cumulate into larger annual differences than those implied by the January Effect, etc. [Table I: Mean aily Returns and Standard eviations: Congress In Session vs. of Session] In Table II we report the annualized returns across days Congress is in and out of session. As expected, annualized returns are substantially higher when Congress is out of session for each index. Returns have ranged from 3.3% to 6.5% higher per year when Congress is out of session. In fact, more than 90% of the capital gains on the JIA have occurred on days Congress is out of session. For the shorter CRSP and S&P 500 time series, more than two-thirds of the annual returns have been earned when Congress is not in session. This is as true for the small-firm dominated CRSP EW index as for the CRSP VW or the S&P 500 index. Note that the proportion of the annualized returns to each index attributable to days Congress is out of session is far in excess of the proportion of days Congress is out of session. For example, approximately 93% of the returns to the ow have occurred on the 43% of the days Congress has been out of session. 13 The evidence in Tables I and II indicates there have been large, statistically significant differences in the properties of the daily return to the major stock indexes when Congress is in and out of session. These daily differences imply large annualized differences in returns across the Congressional calendar that are accompanied by lower volatility. The combination of higher 12 There are 8 Pre-Holiday trading days (Ariel (1990) reports an increase of 36 basis points for CRSP VW and 53 basis points for CRSP EW); the January Effect is concentrated in the first 5 trading days of the year; the Turn-of-the- Month Effect covers the last trading day of the month and the first 3 days of the next month for a total of 48 trading days per year (Ariel (1987) reports an increase of 12 basis points per day). 13 Note that Congress has been in session more frequently in the past 48 years than it was in the previous 60 years. For the entire time span of the JIA Congress was out session about 43% of the time; however, since 1957 Congress has been out of session only about a third of the time. 7

10 volatility and lower realized returns is consistent with Malkiels (1979) conjecture that regulatory uncertainty has a negative impact on market returns. It is also consistent with prior empirical studies by Campbell (1987), French, Schwert, and Stambaugh (1987), Glosten, Jagannathan, and Runkle (1993) and Whitelaw (1994) that failed to find a positive ex post relationship between market volatility and realized returns. [Table II: Annualized Returns In Session vs. of Session] In Figure 1 we present cumulative returns for strategies that invest $1 in the JIA when either Congress is in session or when it is out of session and invest in cash otherwise. These strategies starkly illustrate the findings presented in Tables I and II. The -of-session strategy invests in the market index on days Congress is not in session and in cash (earning, we assume, 1 basis point per day) when Congress is in session. Conversely, the In-Session strategy invests in the market index on days Congress is in session and in cash (earning, again, 1 basis point per day) on days Congress is out of session. Figure 1 indicates that for the JIA the cumulative returns since 1897 are more than 100 times greater ($216 vs. $2) for the strategy. Obviously, these strategies are simplistic and do not account for transaction costs; nonetheless, they indicate that the return differences documented in Tables I and II may be of significant economic interest. 14 We will return to the question of economic significance in Section VI. [Figure 1: JIA portfolio returns (In vs. strategy)] II. Is the Congressional Effect Merely a Proxy for Previously Identified Calendar Anomalies? An obvious question raised by the evidence in Tables I and II is whether or not there is a correlation between the Congressional calendar and previously identified calendar anomalies such as the ay-of-the- 14 The results are less dramatic for the CRSP indexes due, in part, to the shorter compounding period. Nonetheless, the strategy would have accumulated from 3 (for VW) to 9 (for EW) times as much wealth as the In strategy. 8

11 Week Effect, January Effect, and the Pre-Holiday Effect. 15 Table III reports the distribution of days Congress is in and out of session by day of the week and in relation to three other well-known calendar anomalies. For all four indexes Congress is out of session more frequently on Fridays, the day before a holiday, around the turn of the month, and during the first five days of January than it is out of session in the entire time series. Since these are all periods that have been identified as having above average returns, it is possible that the higher returns we observe when Congress is not in session are simply proxies for these previously identified regularities. [Table III: istribution of Congressional Sessions by ay of the Week] To investigate this possibility we run the following regression for each index r t = β Hol Mon Hol Mon Turn Tue Tue Turn Wed Jan Wed Jan Thu Thu + ε t Fri Fri Sat Sat (1) where r t is the daily return to the index, Mon, Tue, Wed, Thu, Fri, and Sat are dummy variables for the days of the week; Hol is equal to one if it is the trading day before a holiday for which the exchange is closed; 16 Turn is equal to one if it is the last trading day or one of the first three trading days of the month; Jan is equal to one if it is one of the first five trading days of January; and, is equal to one if Congress is not in session. The results are reported in Table IV. Recall from Table II that the unconditional Congressional Effect ranged from 4-11 basis points. After controlling for known seasonalities, Table IV indicates that there is still an observable Congressional Effect of about 3-6 basis points per day. Thus, previously identified anomalies account for no more than half of the Effect identified in Table I. The effect is statistically significant for all four indexes including the longest time series (JIA--which is also primarily made up of very large firms) and the index most sensitive to small firms (CRSP EW). [Table IV: Congressional Effect Regressions Controlling for Known Seasonalities] 15 Previous research has established that returns are lower on Mondays, and relatively higher on Fridays, during the first 5 days of January, on the day before a holiday, and over the last trading day of the month and the first three days of the month. 16 These are New Year s ay, President s ay (formerly Washington s Birthday), Good Friday, Memorial ay, 9

12 The results for β reported in Table IV are statistically significant and represent a range of indexes and time periods. However, the result for each index does not really represent an independent piece of evidence. The returns are drawn from overlapping time periods and it is well known that the series are highly correlated. An additional concern about our tests is the possibility that the results are driven by a handful of outliers that have, coincidentally, occurred while Congress is out of session. Further, we cannot dismiss the possibility that our findings may represent a spurious statistical relationship where none really exists. In order to address these concerns and to determine an overall (joint) as well as individual significance level for the four related indexes, we employ a bootstrap-randomization procedure that formally addresses the question of how likely we would be to observe the ummy coefficients and corresponding t-statistics in Table IV if there were in fact no relationship between returns and the Congressional calendar. We begin by holding the daily returns, days of the week, etc. fixed and randomly reassigning (without replacement) the Congressional regressor. This reassignment procedure makes the Congressional variable truly independent of returns while keeping the total number of days in and out-of-session fixed. 17 This allows the general power of our tests to be held relatively constant. Also, once has been reassigned, its value (in or out) for a particular day (e.g. May 4, 1977) is the same for each index. This allows us to account for the fact that the test results across the four indexes are not independent. Regression (1) is then rerun for each index yielding a new estimate of j βˆ and a new t j. 18 This is repeated 10,000 times for j = 1,, j 10,000. The bootstrapped distributions of β and t are the 10,000 draws of βˆ and t j. We compute twosided significance levels α for the reported β as ([the number of times β reported in Table IV] + j βˆ j [the number of times βˆ -β reported in Table IV])/10,000. We do a similar computation for the Fourth of July, Labor ay, Thanksgiving, and Christmas. 17 We actually reassign the days from 1897 through April 1957, reassign the days from May 1957 through June 1962, and reassign the days from July 1962 through Once a day is reassigned, it has the same value (in or out) across each data set. In this way we keep the total number of days in and out of session fixed for each of the indexes. 10

13 reported t-statistic in Table IV. These randomized α s are reported at the bottom of Table IV. The significance levels are similar to the p-values of the reported t-statistics for each of the four indexes. As noted above, due to the overlapping time periods covered by the indexes, these four randomized α s are not four independent pieces of evidence. So we also compute the number of times that all four j βˆ or t j are greater than the minimum (hence, least significant) β or t reported across the four regressions. 19 We also tally the number of times that all four j βˆ or t j are less than the negative of the minimum β or t reported across the four regressions. Table IV reports this two-sided α is.0010 (=10/10,000) for β and.0017 (=17/10,000) for t. Thus, the randomization indicates that is very unlikely that we would see the least significant relationship yield an absolute coefficient estimate of 3.30 or an absolute t-statistic of 2.19 if there were in fact no relationship between returns and the Congressional calendar. As a final robustness check we re-run regression (1) for each day of the week; i.e, the independent variable is Monday returns, Tuesday returns, etc. The ummy coefficients and t-statistics are reported in Table V. Of the 21 coefficient estimates, 20 are positive; and of those, 8 are significantly positive. Of course we lose power in partitioning the sample by day of the week. To get a different sense of how robust the Congressional Effect is to splitting the sample in this way we perform another randomization bootstrap. This bootstrap addresses the following question: How likely would it be to find 20 out of 21 coefficient estimates having the same sign if there were truly no relationship between Congress and returns? We randomly reassign the regressor over the sample period as before but now we also require for each day of the week that the total number of in and out of session days equal the actual sample numbers. The two-sided randomization α is 0.64% (= 64/10,000). This indicates that is unlikely that we would find 20 of 21 coefficients with the same sign if there were no relationship between returns and the Congressional calendar. 18 The bootstrapping literature generally favors focusing on the t-statistic. We report bootstrapping results for both the coefficient and the t-statistic. 19 This is a relatively conservative standard. For example, it is more conservative than asking, say, how often the average t-statistic in the randomization experiment is greater than the average reported t-statistic. 11

14 [Table V: Congressional Effect by ay of the Week] III. Is the Congressional Effect Sensitive to the Public s Attitude toward Congress? In the previous section we confirmed that the Congressional Effect exists independently of previously identified anomalies. Moreover, the randomization results suggest that this effect is a robust feature of the data. If we are in fact capturing a real economic phenomenon, what accounts for it? In the Introduction we hypothesized that the Congressional Effect reflects investors' rational assessments of the impact of Congressional activity on the stock market. Therefore, a systematic connection should exist between the public s attitude toward Congress and the Congressional Effect. We will now test this hypothesis. We use public polling data to proxy for investors attitude toward Congress. There have been occasional public polls of attitudes toward Congress dating back as far as However, regular polling with consistent questions did not take place until the 1960s. We identified 17 national polls taken prior to 1963 and another 145 for the period from These are heavily concentrated in the later years with 112 having been conducted during Figure 2 presents 43 years of disapproval ratings for Congress based on public polling data. The numbers in the figure are the percent of people who say that they disapprove of the job Congress is doing in polls conducted by Harris ( ) and Gallup ( and ). 20 A notable feature of Figure 2 is that Congress disapproval rating has fluctuated substantially over the years, ranging from 10% (October 2001) to 78% (March 1992). [Figure 2: Public Opinion of Congress] This variation in the Congressional approval rating provides us with an opportunity to examine the hypothesis that the Congressional Effect varies systematically with Congressional disapproval. If the public s attitude toward Congress is really the source of the Congressional Effect we would expect the Effect to be greater when public disapproval of Congress is greater than average. If the Congressional Effect is merely a 20 In the polls of the 1960s and early 1970s the question was How would you rate the job Congress is doing? We combined the excellent and good categories (for approval) and fair and poor (for disapproval). Gallup began asking, 12

15 'lucky find' (data mining) we would expect there to be no relationship between the public's attitude toward Congress and the Congressional Effect. To examine this we run the following regression for each index over the time period July 1962 through ecember 2004 r t = β Hol Mon Hol Mon Turn Tue Tue Turn Wed Jan Wed Jan Thu Thu Fri * IS Fri IS% + ε t (2) This is regression (1) without the Saturday dummy since trading never occurred on Saturdays during this time period and with the addition of an interaction term for Congress being out of session with the disapproval percentage, IS%, reported in the most recent public polling data. The coefficient of interest here is the interaction coefficient β *IS. This coefficient measures the incremental Congressional Effect per unit of Congressional disapproval. The results of regression (2) are reported in Table VI. First, note that the Congressional activity effect by itself is now negative, as indicated by a negative estimate for β, although only significantly so for one of the indexes. However, β would only represent the complete Congressional Effect if the disapproval percentage ( IS% ) were zero and, hence, the coefficient on the interaction term IS% were zero. This is because the net (i.e. total) Congressional effect is given by β * is IS% for the model in (2). The lowest disapproval number in our dataset is 10%. Hence, the negative β can more easily be interpreted as a prediction of what the Congressional effect would be if public disapproval for Congress were 0%. Given this interpretation of β and the positive estimates for β * is, the implication is that the higher returns observed on days Congress is out of session are entirely due to the public s attitude toward Congress. The interaction coefficient β* is is positive and significant for each index. For the ow, the 0.24 estimate indicates that daily returns are higher on days Congress is out of session by.24 basis points per percentage point of disapproval. Thus, if we observe an increase of 10% o you approve of the job Congress is doing? in

16 in the Congressional disapproval rating we would expect to see daily returns increase by about 2.4 basis points more when Congress is out of session. The CRSP EW index is the most sensitive to an increase in disapproval; a 10% increase in the disapproval rating will increase daily returns by 3.5 more basis points. Since the range of disapproval ratings is about 68% over the 40 years we have regular estimates, these estimates imply a range of basis points per day in the Congressional Effect. These results are remarkable. It seems clear that the Congressional Effect is driven by the public's attitude toward Congress. 21 [Table VI: Congressional Effect controlling for known seasonalities and public opinion] Given the model specification in (2) we can compute a breakeven Congressional disapproval rating, IS% B/E, where the net Congressional effect is zero. That is, β *IS IS% B/E = 0. Using the coefficients from Table VI for JIA, *IS% B/E = 0 IS% B/E = 31%. For disapproval ratings below 31%, the net Congressional Effect is negative returns are lower when Congress is out of session. When public opinion is more negative towards Congress disapproval greater than 31%--returns are higher when Congress is out of session. The breakeven disapproval ratings for the other indexes are: 34% for S&P 500; 32% for CRSP EW; and 32% for CRSP VW. These are strikingly similar estimates. The Congressional disapproval rating has averaged 48% over Thus, the average net Congressional Effect is about 4-5 basis points per day for each index. At the bottom of Table VI we report randomized αs computed in an analogous way to those reported in Table IV. All regressors are held fixed while the disapproval % is randomly reassigned over the sample. This creates a randomized interaction regressor IS% and makes returns truly independent of Congressional disapproval. This allows us to compute a bootstrapped distribution for 21 It is important to note that even if stock returns and the public attitude toward congress are endogenous, the 14

17 the coefficient of interest β *IS. The individual randomization αs for each index is similar to the regression p-values for each index. The overall randomization αs for the interaction coefficients and t- statistics are.0248 and.0364, respectively. These αs assesses the overall likelihood of finding, by chance, all four interaction coefficients or t-statistics either 1) greater than the minimum coefficient (.24 basis points) or t-statistic (1.66) or 2) less than the negative of the minimum coefficient (i.e basis points) or t-statistic (i.e ). These randomization results imply that the impact of the disapproval rating on the Congressional Effect is not likely spurious. IV. Is There a Congressional Effect in Volatility? In Table I we documented an unconditional Congressional Effect in mean daily returns and in volatility. Table IV documents that after controlling for known daily return seasonalities, the Congressional Effect in mean returns persists. Table VI demonstrates that this effect is driven by public disapproval of Congress. In this section we will document an analogous phenomenon in volatility. We begin by estimating the Congressional Effect in volatility after controlling for seasonalities as in equation (1). Maximum likelihood estimation is performed for each index on the following mean and volatility equations: r t = α X t + ε t (3) σ t = δ + πx t 2 where ε t ~ N(0, σ t ), α and δ are intercepts, and β and π are coefficient vectors. Xcontains all the regressors used in mean equation (1) except the Saturday dummy. The estimations are over the July 1962 through ecember 2004 for better comparison of results controlling for public opinion of Congress. The π OUT coefficient from the volatility equation for each index is reported in Table VII in the column labeled Model Congressional Effect we find is not necessarily implied. 15

18 3a. For each index volatility is significantly lower when Congress is out of Session. This Congressional Effect in volatility ranges from 4 to 6 basis points of return standard deviation. This is consistent with the idea that information relevant for stock valuation is more frequently generated (e.g., regulatory uncertainty is greater) when Congress is in session. Following the logic we used in examining mean returns in equation (2), we next augment X by incorporating the public opinion variable IS%. The π OUT and π OUT*IS% coefficients from the standard deviation equation for each index are reported in Table VIII in the column labeled Model 3b. The coefficient estimates for π are positive and significant at the 1% level for three of the four indexes. However, analogous to the case for mean returns, the net Congressional Effect in volatility is given by π + π * isis%. The positive estimates for π in Model 3b can be interpreted as only a prediction of the Congressional effect in volatility if, hypothetically, public disapproval for Congress were 0%. More importantly, for all four indexes we find negative estimates for the coefficient on the interaction term IS%. The t-statistics are significant at the one percent level for each index. This implies that the Congressional Effect in volatility, like the effect in mean returns, is driven in large part by public disapproval of Congress. Specifically, Congressional activity affects daily return volatility in proportion to the public s disapproval of Congress. Taking the JIA as an example, the coefficient estimate of -.96 for π *is indicates that if the public disapproval percentage rose 10%, the difference in daily return volatility vis-à-vis whether or not Congress is in session would change by 9.6 basis points of return standard deviation. As noted earlier, there has been a range of 68% between the highest and lowest disapproval ratings; this translates into a daily volatility range of 65 basis points. Given the positive coefficient estimates for π and negative estimates for π * IS, the implication is that the lower volatility of returns observed on days Congress is out of session are entirely due to the public s attitude toward Congress. The significantly lower volatility associated with Congress being out of 16

19 session is driven entirely by periods in which Congress is relatively unpopular. We conclude that there is a Congressional Effect in volatility and that it also systematically varies with the public's attitude toward Congress. [Table VII: Congressional Effect in Volatility Controlling for Known Seasonalities and Public Opinion] V. Is the Congressional Effect Economically Significant? Regressions (1) and (2), reported in Tables IV and VI respectively, demonstrate that the Congressional calendar has had a statistically significant impact on stock market returns. In addition, Tables I and II and Figure 1 suggest that this Congressional Effect is economically important. In this section we further explore the economic significance of the Congressional Effect. We employ two frameworks that provide us with separate perspectives on the importance of the Congressional Effect for asset allocation decisions. First, we employ Kandel and Stambaugh s (1996) setup that considers the economic significance of the regression evidence from the perspective of a hypothetical risk-averse Bayesian investor who must allocate funds between a stock index and risk-free cash. Second, we consider the Britten-Jones (1999) regression framework for determining the allocation weights necessary to maximize the Sharpe ratio of an investment portfolio. Although typically applied to the returns of various assets (e.g., returns to several international indexes) we follow the lead of Hirshleifer and Shumway (2003) and use the returns to various investment strategies and ask whether any of these strategies can help an investor achieve an investment portfolio with a higher Sharpe ratio. Our study examines various investment strategies based on different aspects of the influence Congress has on stock returns. A. Allocating Between Cash and the Market Index Suppose that an investor knows whether or not Congress will be in session and knows the 17

20 Congressional disapproval rating. Given the regression evidence for the specification in (2) and presented in Table VI, how does this evidence influence the investor s asset allocation decision? Kandel and Stambaugh (1996) develop a framework for assessing the economic importance of predictive variables on investor decisions. Economic significance is judged by the sensitivity of an investor s allocation in an index to changes in the predictive variables. In this section we determine the optimal proportion of the investor s portfolio that should be allocated to the market versus left in cash and how sensitive this allocation is to the disapproval rating of Congress. We assume that our hypothetical investor has prior beliefs that are biased against a relationship between stock returns and any of the daily return explanatory variables (January, Monday, Pre-Holidays, Congressional disapproval, etc.) we have examined in this paper. The priors for all variables are equivalent to the posteriors that would result if an investor had started with diffuse (non-informative) priors and had observed a hypothetical pre-sample of data equivalent in all respects to the actual data set except with a sample R 2 equal to zero of an equivalent length to the actual sample. 22 Essentially, the investor is Bayesian and his beliefs are biased against any relationship between returns and the explanatory variables to the same degree as if the investor had started out agnostic and had observed over 40 years worth of returns that were completely uncorrelated with any of the regressors. The investor updates these prior beliefs by taking the following model (our regression (2)) to the actual data r t = β Hol Mon Hol Mon Turn Tue Tue Turn Wed Jan Wed Jan Thu Thu Fri * IS Fri IS% + ε t (4) ε ~ N(0, 2 σ ε ) Given the investor s prior beliefs about the data, the prior for β is multivariate t with T k degrees of freedom, where T is the number of observations and k is the number of regressors. The mean for β1 is r (the sample average return) and all other β means are equal to 0. The variance-covariance matrix is 22 See Kandel and Stambaugh (1996) for details. 18

21 s ( X X ), where s is the sample variance of returns ( r ) multiplied by σ ( T 1) /( T k) and X is a 2 9,694 by 10 matrix of regressor values through time. The prior for σ ε is given by 2 ( T k) σ 2 σ 2 ε r ~ χt k. Given these prior beliefs and the likelihood function implied by (4), the investor can form posterior densities for all the parameters and predictive densities for returns. These predictive return densities account for the influence of the explanatory variables, uncertainty with respect to the extent of this influence (estimation error) and the standard deviation of returns. Assuming that the investor has initial wealth equal to $1 and has utility-of-wealth given by 23 u( W ) 10W = 1 exp, we compute this risk-averse Bayesian investor s optimal (one-day) allocation to an index versus a riskfree asset (earning 2 basis points per day) under three scenarios. Short-selling the index is precluded, but buying on margin (2 basis points per day interest) is not. 24 In the first (benchmark) scenario, Congress is in session. In the second scenario, Congress is out of session and public disapproval of Congress is 37%. In the final scenario, Congress is out of session and Congressional disapproval is 59%. The alternative scenarios reflect disapproval ratings one standard deviation above and below the average disapproval rating (48%) for the period. In determining an optimal allocation, the investor integrates his utility function with respect to the predictive return density. The allocation that maximizes this integral is the utility-maximizing allocation 25. Table VIII reports the results of this calculation. [Table VIII: Changes in Portfolio Allocation induced by Congressional Effect by day of week.] 23 Relatively higher risk-aversion coefficients (e.g., 10) are more conservative than lower coefficients (e.g., 2). Combined with the priors we use, our experiment is biased away from finding dramatic evidence of portfolio reallocations based on the Congressional Effect. 24 For technical reasons (to ensure the expected utility integral is defined), negative allocations are prohibited. Positive allocations must be less than infinity. 25 See Kandel and Stambaugh (1996) for more details on the computations in this type of analysis and its 19

22 We can assess the economic significance of Congress being out of session conditional upon the Congressional disapproval rating by comparing the allocations to the index under the low and high disapproval scenarios relative to the allocation when Congress is in session. Table VIII reports the allocation when Congress is in session for each day of the week. As we would expect given the ay-of-the-week Effect, Mondays have no allocation to the index. For each day, the table reports the change in the allocation to the index under the high and low disapproval scenarios relative to the allocation to the index when Congress is in session. For Wednesday, the optimal allocation to the JIA is 15% when Congress is in session. The utility-maximizing allocation rises by 11% when Congress is out of session and the disapproval percentage is 37%. However, if disapproval were to rise to 59%, the optimal allocation would rise 24% (from 11% to 35%) over the benchmark. This increased allocation reflects the influence of disapproval percentages on the investor s beliefs about returns. For each and every index we document a positive change in allocation when Congress is out of session under both scenarios for each index. When not subject to the no short-sale constraint, the investor allocates up to 30 percent (CRSP EW) more to the index when Congress is out of session and disapproval is low. In contrast, the allocations increase up to 94 percent (CRSP EW) when Congress is out of session and disapproval is high. These results suggest that even a very risk averse investor with prior beliefs biased against any relationship between stock returns, the Congressional calendar, and Congressional disapproval would increase their allocation to the market index based on the evidence in Table VI. The changes are fairly significant when Congressional disapproval is low. However, when Congressional disapproval is high the changes are even more dramatic. The economic significance of the Congressional disapproval rating is reflected in the magnitude of the increases from the low to the high disapproval scenarios. When the short-sale restriction is not binding, these increases range from up to 65%. These rather dramatic asset reallocations suggest that the regression results in Table VI are economically significant. interpretation. 20

23 B. Allocating Across Many Strategies Hirshleifer and Shumway (2003) utilize the framework developed in Brtten-Jones (1999) to show that a portfolio based on the Sunshine Effect has about equal weight in the investor s optimal portfolio as the global market portfolio. 26 We also employ the Britten-Jones methodology to determine the extent to which trading strategies based on the Congressional Effect could be an important component in the tangency (maximum Sharpe ratio) portfolio for daily returns. 27 Britten-Jones shows that the scaled coefficients (sum to 100%) in the following type of regression are the weights in the investor s optimal risky portfolio 1 = β R R R R R R R (5) Mkt Mkt Mon Mon Fri Fri Hol Hol Turn Turn Jan Jan Cong Cong The 1s on the left hand side of regression (5) represent an ideal portfolio (with an infinite Sharpe ratio). The right hand side variables are the daily returns to various potential trading strategies. R Mkt is the return to the equally weighted portfolio of index returns ((JIA + S&P CRSP EW + CRSP VW)/4); we can think of R Mkt as the return to a strategy of investing in the market every day. R Mon is the return to a strategy that shorts the market on Monday and R Fri is long the market on Friday; R Hol is the return to a strategy that is long the market the day prior to a holiday; R Turn is the return to a strategy that is long the market the first three trading days and the last trading day of the month; and, R Jan is the return to a strategy that is long the market the first five trading days of January. We consider three potential trading strategies related to the Congressional Effect. In Strategy 1, R Cong is the return to being long the market on days Congress is out of session and short the market when Congress is in session. In Strategy 2, R Cong is the return to being long the market on days Congress is out of session and the disapproval rating is greater than 59%; the strategy is short the market when Congress is out of session and the disapproval rating is below 37%. At all other times the investor 26 Hirshliefer and Shumway consider a strategy that is long the market index when there is below average (seasonally adjusted) cloud cover in the early morning in the city where the major stock exchange is located for 26 countries. The also look at a strategy that is short the index when cloud cover is above average in the early morning and a strategy that is both long and short the index depending upon the degree of sunshine. 27 Strictly speaking, the framework of Britten-Jones calls for the use of excess returns. We do not have daily riskfree returns and so we use the actual returns to the index. Thus, our results produce the tangency portfolio relative to 21