Stock ownership and congressional elections: the political economy of the mutual fund revolution.

Duca, John V. ; Saving, Jason L.

I. INTRODUCTION

In recent years, political analysts have conjectured that voting has been affected by the rise of an investor class in the United States, where the stock ownership rate has doubled from under 25% in the late 1970s to over 50% by 2001. (1) This shift in portfolio behavior has accompanied an upward shift in the two-party share of the House popular vote for candidates from the more capital-friendly Republican Party (Figure 1).

While the notion that property interests affect voting is an old and intuitive one, empirical tests of the effect of stock ownership on voting have been hampered by the lack of continuous time series data on stock ownership. Fortunately, continuous data on equity mutual fund costs are available from Duca (2006) that are highly and negatively correlated with discontinuous stock ownership rates as shown by Duca (2001, 2005). In line with research on factors affecting stock ownership by Aiyagari and Gertler (1991) and Heaton and Lucas (2000), we argue that mutual fund costs are a good proxy for stock ownership rates and use them in short- and long-run models of the two-party share of the popular vote for Congressional Republican candidates that use data spanning 1954-2004.

We analyze the link with congressional voting because the shorter cycle of races allows for more degrees of freedom than presidential races, and we focus on House rather than Senate elections, the latter of which cover only one-third of that body and are not geographically balanced. Also, relative to presidential elections and Senate elections, the larger number of House races in national vote totals and the lower profile of individual representatives limit the impact of personality on elections, perhaps allowing for a cleaner test of the role of property interests. We focus on the national share of the popular vote rather than on the number of House seats because the latter is influenced by the design of House districts, which is subject to political gerrymandering.

[FIGURE 1 OMITTED]

We test whether the rise in the Republican share of the House popular vote from around 44% to 50% since the late 1980s is linked to wider stock ownership. Our findings accord with the view that the mutual fund revolution has contributed to a shift in voting. In particular, our results suggest the Republican share of the House popular vote will likely fluctuate around 50% until other factors trigger a political realignment.

Our study is organized as follows. Section II reviews the history of the debate over linking voting rights with property ownership in the United States and the literature on how stock ownership rates could affect voting. Section III empirically tests whether there is a long-run relationship between the House popular vote and stock ownership. Section IV addresses whether property interests, as reflected in this long-run relationship, help explain short-run changes in voting in the presence of more conventional short-run variables (e.g., midterm elections). To assess whether these findings are robust, Section V similarly analyzes the House vote in the South, where there has been a long-run shift toward the Republican Party. Section VI analyzes the popular vote shares in Senate elections. Although the latter is noisier than the House series and is not from a geographically balanced electorate unlike House races, applying the methodology to the Senate provides a robustness check on the House results. Section VII summarizes our findings and discusses their implications for future research.

II. VOTING AND PROPERTY OWNERSHIP IN U.S. HISTORY

The idea that asset ownership could affect voting is an old one in the United States and was much discussed at the 1787 constitutional convention. George Mason proposed that owning land be required for Senators because the chamber was intended "to secure the rights of property," which presumably could not be guaranteed if Senators did not own property (Madison 1987, 200). Pennsylvania delegate Gouverneur Morris argued that wealthy states ought to receive more House seats than poorer ones because citizens of wealthy states would be more likely to protect the assets of the propertied (Madison 1987, 244). This led South Carolina delegates John Rutledge and Pierce Butler to propose that House seats be apportioned by wealth and population. However, disputes over how to measure wealth (by federal taxes or land values), dedication to private property (ownership of land vs. general assets), and applying economic values to slaves convinced most delegates that apportioning the House by wealth was impractical.

The most important debate (from our perspective) was over suffrage. Gouverneur Morris proposed limiting the right to vote in federal elections to landowners (Madison 1987, 401). John Dickinson of Delaware supported this as "a necessary defense against the dangerous influence of those multitudes without property and without principle" (Madison 1987, 402). Madison (1987, 403) noted that landowners are the "safest depositories of Republican liberty," because only they can safeguard "the rights of property and the public liberty" but nevertheless argued against Morris, believing that limiting voting rights would lay the groundwork for dictatorship.

While many of the American republic's founders feared the prospect of rule by the general populace, Karl Marx welcomed it. Recognizing that workers with no hope of advancement and no prospect of wealth ownership had literally "nothing to lose but their chains," Marx (1977) theorized in his classic work, Capital, that the masses would inevitably confiscate the assets of the rich and establish an egalitarian society. What Marx never anticipated was that common individuals would gain the ability to become owners. The last decade of the 20th century featured the largest rise in stock ownership the United States has seen, and greater stock ownership has shaped many middle-class Americans into what many observers, such as Kudlow (1997), have called an "investor class." Kennickell, Starr-McCluer, and Surette (2000) found that Americans' net worth grew strongly during the 1990s with "a continued rise in the holding of stock equity ... account[ing] for a substantial part of the rise in net worth." This rise was "broadly shared by different demographic groups," with the biggest percentage point increases in ownership occurring in the middle quintiles of the income distribution.

Policies favoring capital formation and accumulation have gained popularity. "Thinking like the shareholders and business owners that they are," writes Glassman (2000), "members of the investor class want low taxes on capital, low taxes on individual and corporate income, light regulation of business and limits on litigation." And polls suggest that stock ownership induces middle-income Americans to support procapital policies like lower capital gains and estate taxes as Nadler (1999) and Gigot (1999) note.

In a basic two-party median voter framework such as Downs (1957), a change in the electorate would not affect the fortunes of either political party because the parties would adjust their platforms to match median voter preferences. Gross (2000) argued that the Democratic Party has adjusted by increasingly accommodating the views of "New Democrats," who combine social liberalism with moderate-to-conservative economic views. But for many reasons, including primary elections and ideological conviction, a complete convergence of party platforms to the views of the median voter may not occur as discussed in Mueller (2003). Simply put, neither a party nor its supporters are willing to entirely shed the principles that motivate them to seek power, whether those principles protect the poor or reward the rich--even at some added risk of losing elections. Thus, the traditional gap between the two major parties on wealth issues could plausibly persist, implying that owning stocks raises the likelihood of voting for the Republican Party that has traditionally pursued capital-friendly policies as Faucheux (1999) maintained.

III. STOCK OWNERSHIP AND HOUSE ELECTIONS IN THE LONG RUN

Although continuous stock ownership rates are unavailable, continuous data on equity mutual fund costs from Duca (2006) are a good proxy for stock ownership, having a strong, negative correlation (-0.96) with stock ownership rates (Figure 2), where equity fund costs (MFCOST, Appendix A) equal the annual expense ratio plus the annualized, average cost of front-end and back-end commission fees (loads) for mutual fund investments over 5 yr following Duca (2005, 2006). (2,3) We assess the time series relationship between stock ownership rates and voting behavior by using equity mutual fund costs as a proxy for unavailable continuous stock ownership rates. This section first discusses the link between the share of households owning stock and the mutual fund costs and then focuses on testing the long-run relationship between equity fund costs and voting.

[FIGURE 2 OMITTED]

A. Why Equity Mutual Fund Costs Are a Good Proxy for Stock Ownership Rates

There are good arguments for using equity mutual fund costs to proxy for stock ownership rates because higher equity participation has been associated with declines in the cost of investing in equity mutual funds, consistent with recent research on stock investing by Guiso, Haliassios, and Jappelli (2003), Heaton and Lucas (2000), and Siegel (1999) and on mutual fund costs by Duca (2000, 2001, 2005, 2006). In particular, Heaton and Lucas (2000) analyzed why stock ownership was lower than implied by conventional theory given a high equity premium. In their optimization model and that of Vissing-Jorgensen (2002), transaction costs and nondiversifiable labor market risk deter middle-income families from owning stocks despite a high equity premium. As a result, transaction costs can have larger portfolio effects than in conventional models, and higher mutual fund fees before the early 1980s may explain the lower stock ownership rates of that era. In this calibration model, lower transaction costs can induce higher equity participation, especially among low- to middle-income families, consistent with Dixit (1989), and a greater rise in stock ownership rates in the 1990s among these families than among high-income families, as noted by Duca (2006) and Kennickell, Starr-McCluer, and Surette (2000).

Owing to limited wealth, many families are more apt to acquire a diversified stock portfolio by buying mutual funds rather than directly buying stocks. For such families, the relevant transaction costs for investing in stocks are mutual fund fees, and if these lees fall, stock ownership rates should rise. Figure 2 shows that the rise in ownership mostly owes to greater indirect ownership, which is even more negatively correlated with annual equity fund costs (correlation of -0.98) than is overall ownership (correlation of -0.96) over commonly available years (1969-2004). Furthermore, other data show that the rise of indirect ownership primarily occurred through increased mutual fund ownership. The much higher fees of the 1970s and early 1980s may thus account for the low, pre-1990 stock ownership rates that were analyzed by Aiyagari and Gertler (1991) and others.

Using three decades of data, Duca (2005) found that lower mutual fund costs and greater confidence boosted the percent of stocks owned by families via mutual funds. He argued that lower fund fees could induce greater reliance on mutual funds by spurring some shareholders to shift assets from stock shares to mutual funds and by inducing some households to become shareholders. In addition, Duca (2005) found evidence linking mutual fund costs with technology. In particular, he showed that the measure of mutual fund costs used here is cointegrated with and negatively related to productivity in commercial banking, the only financial sector for which a long time series of productivity estimates exists. Together, recent calibration models and evidence on household portfolio behavior suggest that falling mutual fund costs have boosted equity ownership rates.

[FIGURE 3 OMITTED]

The positive correlation between stock ownership rates and the Republican share of the House vote and the negative one between stock ownership and equity fund costs imply a negative relationship between the Republican vote share and the equity fund costs. Indeed, a downtrend in equity fund costs has coincided with of slightly led an up-trend in the Republican House vote share since the late 1980s (Figure 3). Other data suggest a link between stock investors and voting. For example, much of the voter shift toward Republicans has occurred among the middle class, who have posted the largest rises in stock ownership rates since the late 1980s. Nadler (1999), Rasmussen Research (2000), and Zogby International (2000) noted polls showing a strong link between stock ownership and voting. (4)

Nevertheless, some changes in the vote share in Figure 4 likely stem from short-run factors such as midterm elections, income growth swings, and presidential approval. We analyze how such factors affect changes in the vote share along with information from the long-run link between equity fund costs and the House popular vote share.

[FIGURE 4 OMITTED]

B. Empirical Approach to Testing for Long-Run Relationships

Cointegration techniques are used given evidence of a unit root in key long-run variables. The Republican share of the House popular vote has a unit root according to augmented Dickey-Fuller (ADF) tests, whether or not the tests allow for a deterministic time trend in this variable, while stock mutual fund costs exhibit a unit root allowing for a time trend (Table 1). (5) We use the Johansen-Juselius approach rather than the dynamic ordinary least squares approach due to the upward shift in Republican share near the end of our 1954-2004 sample, which starts when revised expense ratios on mutual funds became available.

These tests reflect two considerations. First, we focus on results not allowing for deterministic time trends in the variables because allowing for a trend in the vote share implicitly allows the variance of this variable to increase without bound. Such an assumption would be at odds with the 0% and 100% share range of this variable and the implications of median voter theory that the parties be competitive with one another. Nevertheless, qualitatively and quantitatively similar results for the House popular vote are obtained, allowing for possible deterministic trends in the variables (Appendix Table A1). A second consideration is that since the sample has 26 elections, vote fluctuations from short-run factors may affect long-run coefficient estimates. Accordingly, we use vector error correction models (VECMs) to jointly estimate the long-run relationship in a cointegrating vector and short-run effects in first-difference equations, respectively:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where the lags of first-difference endogenous variables minimize the Akaike information criterion (AIC), X is a vector of exogenous factors, [[epsilon].sub.it] are residuals, and the [[lambda].sub.i], [[gamma].sub.i] and [[delta].sub.i] are row vectors of coefficients.

The sets of exogenous, short-run factors include variables tracking the impact of midterm elections, income growth, the Watergate scandal, presidential approval, and the Gingrich revolution of 1994. To control for the short-run impact of midterm election shifts away from the party in the White House, MIDTERM equals 0 in presidential elections but in midterm elections, equals 1 under Republican presidents and -1 under Democratic presidents. According to conventional wisdom and past studies, MIDTERM should have a negative coefficient. As Tufte (1975) noted, the president's party almost always loses House seats. Campbell (1966) discussed the "surge and decline" explanation for midterm election effects, under which victorious presidents boost the popularity of their party in on-year elections but not in off-year elections when they are not on the ballot. Erikson (1988) argued that there also is a "presidential penalty" effect under which the electorate punishes the president's party regardless of his performance, perhaps because citizens are less likely to vote in off-year elections unless they wish to voice disapproval of the president as noted by Kernell (1977) and Lau (1985).

We also consider short-term economic factors, whose impact on voting has been shown by Peltzman (1990). In a seminal paper, Kramer (1971) found that real disposable income growth, when adjusted for the party of the president, was significantly correlated with the Republican share of the House vote. However, Kramer's disposable income and vote share variables have unit roots. When we tested whether they are cointegrated, cointegration was rejected in most cases, with the exceptions indicating a counterintuitive relationship between real income growth and the House vote. Instead, we include a stationary term for the first difference of a variable interacting a variable equaling 1 for Republican presidents and -1 for Democrats with real disposable, per capita income growth in an election year on a third-quarter basis (REALINCOME). (6)

Although earlier studies, such as Conway and Wyckoff (1980) and McLeod, Brown, and Becker (1977), usually do not find a Watergate effect, we test WATERGATE (which equals 1 in 1974 and 0 otherwise) in case an effect emerges in a longer sample. Other short-run variables include controls for the "Gingrich revolution" (GINGRICH equals 1 in 1994 and 0 otherwise) and presidential approval (PRESAPPROVAL). Based on Campbell (1997), the latter equals the log of the percent of Americans approving of a Republican president in the last Gallup Poll before each House election or the log of 100 minus the percent approving of a Democratic president. (7) Objections to these variables are that they are defined after the fact and may introduce simultaneity bias because they are not purely exogenous. In particular, they may reflect that Republicans ran a better national campaign in 1994 or when their presidential candidate ran a better campaign. Some models omit these variables partly out of this concern and partly as a robustness check. (8)

C. Are Stock Fund Costs and House Elections Related in the Long Run?

Table 1 reports cointegration tests using different sets of short-run factors in VECMs with data over 1954-2004. Since researchers may differ over which short-run variables to include, we present results from VECMs that include three plausible sets of short-run factors as well as one omitting them. (Table 2 presents results for the first difference of vote share from VECMs and conventional models.) One set includes MIDTERM and REALINCOME, to which a second adds WATERGATE and a third also adds GINGRICH and PRESAPPROVAL. As only MIDTERM, PRESAPPROVAL, and REALINCOME are the only significant short-run factors in that set, another set includes just these three extra variables. In three cases, a lag length of 3 on first-difference Y terms minimizes the AIC, while a lag length of 1 does so for third model.

Test statistics indicate the existence of only one significant cointegrating vector in each case, as the eigenvalue and trace statistics reject the null hypothesis of no significant long-run relationship exists between the stock fund costs and the Republican vote share. In line with unit root test results, there is no evidence that two or more cointegrating vectors exist (which would imply that each variable is stationary). In each case, stock fund costs have significant and negative coefficients on equity fund costs ranging from -0.24 to -0.34. Ordinary least square (OLS) regressions of the log-level vote share on log-level mutual fund costs yielded similar and statistically significant coefficients using similar sets except for using the non-first differenced version of REALINCOME and a dummy for the advent of 401(k) plans (table available upon request). Also encouraging is that the vectors imply similar estimates if the VECMs are estimated allowing for possible deterministic time trends in the long-run variables. For example, in the presence of MIDTERM and REALINCOME, the baseline and deterministic trend equilibrium levels are, respectively,

ln(REPVOTE) = 4.023977 - 0.275105 x ln(MFCOST), (Vector 5, Table 1) and

ln(REPVOTE) = 3.998250 - 0.266393 x ln(MFCOST), (Vector 5, Table A1).

As shown in Figure 4, the equilibrium values from both vectors track the long-run movements in the Republican share of the House popular vote (in logs) over 1954-2004.

IV. STOCK OWNERSHIP AND HOUSE ELECTIONS IN THE SHORT RUN

To see if long-run relationships help explain short-run movements in the House popular vote, we examine short-run changes in the Republican popular vote share from using the same set of VECMs used to estimate long-run relationships and compare these findings with results from OLS "conventional" models of the first difference of the log of Republican vote share using the same corresponding sets of short-run variables.

D. A Conventional Model of the Change in the House Popular Vote

Table 2 reports eight models of the percent change in the Republican share of the House popular vote. Models 1-4 are conventional models containing conventional short-run variables but omit an error correction term for shifts in stock ownership rates via mutual fund costs. These models are estimated using OLS since all the variables in them are stationary according to unit root tests (not shown). However, since the log levels of the Republican vote share and the stock ownership proxy are cointegrated, omitting an error correction term renders the conventional models misspecified.

Each conventional model includes a constant and three lags of the change in the Republican vote share to mirror the inclusion of these lags in the VECMs, which limits the sample to 1962-2004. To the baseline model 1--corresponding to Model 5 in Table 2--Model 2 adds MIDTERM and REALINCOME, corresponding to Model 6. In addition to the Model 2 variables, Model 3 includes the WATERGATE, GINGRICH, and PRESAPPROVAL. Finally, Model 4 adds just MIDTERM, REALINCOME, and PRESAPPROVAL to the baseline to correspond to the last VECM (Model 8) in Table 2.

E. Results from Estimating Models of the Change in the House Popular Vote

Several patterns emerge across the conventional models. First, the midterm effect is significant and negative, consistent with Campbell (1997). Second, presidential approval is significant with the expected positive sign. Third, the Gingrich variable is positive but statistically insignificant. Fourth, the Watergate coefficient is negative and is only marginally significant, which may reflect two difficulties in detecting a Watergate effect. First, the economy weakened between 1972 and 1974 during the first OPEC oil shock and perhaps partly because the scandal weakened confidence. Indeed, as Kiewiet (1983) noted, real per capita income fell by an unusually large 2% and the share of voters concerned about the economy (especially inflation) doubled. Consistent with this view, REALINCOME loses statistical significance when WATERGATE is present (Models 2 vs. 3). Another difficulty in identifying a Watergate effect is that the first congressional election (1974) after the scandal was a midterm election with a Republican president. (9)

A fifth notable pattern across the models is that the lagged change in Republican vote share is negative but insignificant. This finding may reflect a weak tendency for the unusual success or failure in a prior election to unwind in the next election. Finally, these models explain between 20% and 83% of the variance in vote share changes, with the range rising to between 64% and 83% when some short-run variables are present.

In the presence of cointegration, proper specification requires including an error correction term. Accordingly, Models 5-8 include error correction terms testing for the impact of stock ownership on short-run voting. In each case, [EC.sub.t-1] equals the gap between the actual House vote and the equilibrium share implied by equity fund costs from the corresponding VECM in Table 1. When the actual share exceeds the equilibrium share in the prior election, long-run equilibrium implies a tendency for the vote share to fall in the current election and thus a negatively signed coefficient on [EC.sub.t-1].

Models 5-8 contain sets of short-run variables corresponding to Models l-4. In each model, the error correction term is negative and highly statistically significant, implying that the Republican vote share tends to fall when actual exceeds the equilibrium share in the prior election, where the latter is decreasing in equity fund costs. In the absence of conventional short-run variables, the error correction term exceeds 1 (177%), which likely indirectly reflects a tendency for midterm election and other conventional short-run effects to reverse in subsequent elections. Indeed, the error correction coefficients are more reasonable in Models 6-8, implying that 73%-103% of the prior election's gap between the actual and the equilibrium vote shares is unwound. In these VECMs, the lagged changes in equity fund costs, which are included to pick up any short-run dynamic effects, are insignificant, while the lagged dependent variable is significant in only one model. Results are similar when the VECMs are estimated under an assumption allowing for time trends (see Appendix Table A2).

There are several notable patterns across the error correction models pertaining to the short-run, conventional variables. First, presidential approval, PRESAPPROVAL, is very significant and positive. Second, coefficients on MIDTERM are negative and statistically significant, ranging from -0.0565 to -0.0646. Because the Republican share is near 50%, this implies a midterm effect of about 3 percentage points. (10) Third, GINGRICH is positive but insignificant as before. Fourth, the Watergate coefficient is negative but insignificant, whereas it is marginally significant in the conventional model. Another small difference is that the coefficients on REALINCOME and PRESAPPROVAL are estimated with smaller standard errors in VECMs, which may reflect that they are easier to identify in a better specified model. After all, omitting an error correction term omits valuable information and is a form of misspecification. Finally, estimates indicate that the Republican vote share rises (falls) by about 0.09% for each log percentage point in the approval rating of a Republican (Democratic) president, which is qualitatively consistent with Campbell's evidence of a presidential election effect on House races. (11) Nevertheless, because of the potential for simultaneity bias, results regarding PRESAPPROVAL should be cautiously interpreted.

Another difference across the two groups of models is the better fit of corresponding mutual fund models, reflecting the statistical significance of the error correction terms. Indeed, the corrected [R.sup.2]s from the mutual fund models are higher by .31 comparing Models 1 and 5,. 16 comparing Models 2 and 6, .11 comparing Models 3 and 7, and .12 comparing Models 4 and 8. Of models including some conventional variables (Models 2-4 and Models 6-8), information from mutual fund costs helps explain 11%-16% more variation in how the vote share changed from one election to the next, implying that our proxy for stock ownership is both statistically and economically significant. Another sign that the mutual fund models are better specified is that they have well-behaved residuals according to LM and Q statistics, unlike the conventional models which omit the statistically significant and substantive error correction terms.

To illustrate the information added by the stock ownership proxy, estimates from two sets of corresponding models are plotted with actual outcomes. Of the models containing midterm and income variables (Models 2 and 6), the error correction model (Model 6) tracks voting changes better in the 1990s (Figure 5)--when stock ownership rates shifted up as mutual fund costs fell--and outperforms the conventional model in seven of the last eight elections since 1990. Similar results arise in models that include presidential approval with income growth and midterm election effects (Models 4 and 8).

[FIGURE 5 OMITTED]

An alternative interpretation is that the upward shift in the popular vote share owes to an incumbency effect, in that the 1994 jump in the number of Republican House seats (the Gingrich revolution) had a hysteresis-like effect on the popular vote share. However, the Republican share of House seats was in a flat-to-down range between 1986 and 1992, whereas that party's share of the popular vote drifted upward (Figure 6) since the mid-1980s. In fact, the 1994 jump in the popular vote share of the popular vote appears to be a temporary positive outlier along an upward trend that is more consistent with the stock ownership hypothesis. Indeed, Model 8 notably outperforms Model 4 in the 1990 and 1992 elections and Model 5 performs better than Model 1 in these elections as well, which predated the Gingrich revolution and the upward shift in the Republican share of House seats in 1994. Furthermore, improved fit and better estimates of coefficients from accounting for stock ownership effects are not limited to recent decades of rising stock ownership rates and Republican gains in House elections. Indeed, both equity fund models also outperform their conventional counterparts in the 1964, 1970, and 1974 House elections when the two-party vote for the Republican Party plunged.

V. ANALYZING THE HOUSE POPULAR VOTE IN THE SOUTH

One concern with the national results is that they might reflect southern cultural shifts caused by the civil rights revolution and not investor effects. To address this, we analyze the impact of economic variables (regional income and mutual fund costs) on the Republican share (REPVOTESO) of the House vote in the South (defined by Congressional Quarterly as states from the Confederacy plus Kentucky and Oklahoma), which exhibit a unit root but not so for other states without allowing for a time trend. (12)

[FIGURE 6 OMITTED]

This analysis is limited because voting could have been affected by migration and the extension of voting rights to African Americans in the South, which altered the voting pool and induced whites to become Republicans in the early 1960s. But the lifting of voting barriers enabled many poor whites to vote, and Husted and Kenny (1997) found that their votes and those of new African American voters raised southern support for income maintenance programs after the early 1960s. So the cultural shift story is not as simple as it is sometimes portrayed. With these caveats in mind, analyzing the South provides a robustness check because if the findings were qualitatively similar to U.S. results, they would imply that investor class effects operate amid regional political realignments.

A. Long-Run Patterns in Southern Popular Vote Share

The Republican vote share has unevenly risen in the South, jumping in the early 1960s (Figure 7) when the civil rights views of Presidents Kennedy and Johnson were unpopular. The rise of Republican vote share in the South in the 1960s and 1970s implies that other factors were at play. One plausible factor is that the South became less poor in the 1960s and 1970s when the ratio of median disposable income per capita in the South to the United States (SINCSHARE, Figure 7) rose. (13) As this ratio rose, the South may have had less incentive to vote for Democrats because the South has less to gain from redistributive policies associated with the Democratic Party, consistent with the notion that relative income shapes political preferences, perhaps as inequality raises the potential gains to the poor from redistribution, as Boskin and Sheshinski (1978), Frank (1985), and Easterlin (1995) discussed. Lungqvist and Uhlig (2000) even found that people prefer more progressive economic policies when they are relatively poor than when they are absolutely poor.

For these reasons, we use a relative income variable to analyze long-run voting behavior in the South. The larger, earlier income gap between the United States and South arguably induced southerners to vote for the Democratic Party that favored more progressive economic policies, and as that gap eroded, southerners should shift toward the Republicans. Since lower income voters favor Democrats, regional income convergence over the 1960s-1970s could be linked to a shift toward the Republicans in the South, with later shifts since the early 1980s plausibly stemming from stock ownership effects.

[FIGURE 7 OMITTED]

To see if stock ownership and income differentials could account for southern voting patterns, cointegration tests were run using VECMs that include similar sets of exogenous variables as in the national models with two differences. First, the log of SINCSHARE is included as a long-run variable. Second, to control the permanent shift toward the Republican Party associated with the civil rights movement in the early 1960s, VECMs include CIVILRIGHTS, which equals 0 before 1962, 1/2 in 1962, and 1 since 1964. Since this variable is not really endogenous and would not be reasonable to model among the long-run variables in a VECM, it is included as an exogenous variable. By its construction, CIVILRIGHTS has a long-lasting effect on short-run changes, in Republican vote share which effectively corrects for the tendency of SINCSHARE and MFCOST by themselves to underpredict the Republican vote share since the early 1960s.

As with the national results, eigenvalue and trace statistics indicate the existence of only one significant cointegrating vector using each set of exogenous variables and assuming no time trends in the variables or vector (Table 3). In each case, a lag length of 1 minimized the AIC and CIVILRIGHTS is included. The sets of exogenous variables differ from the national sets in excluding the insignificant REALINCOME variable in the fourth set. In each case, equity fund costs are significantly and negatively related to Republican voting share (Table 3), and relative southern income is significant, with the hypothesized positive sign. The implied equilibrium relationships also line up well with log-level vote shares when adjusted for estimated impact of the civil rights variable, which is statistically significant as an exogenous variable (Models 5-8, Table 4). For example, the implied equilibrium relationship from the VECM including MIDTERM and REALINCOME (Model 6 in Tables 3 and 4) adjusted for estimated civil rights effects tracks the Republican popular vote share in the South (Figure 8), and similar results are obtained using the VECM including MIDTERM and PRESAPPROVAL (not shown).

B. Short-Run Movements in Southern Popular Vote Share

Following the national House analysis, we examine short-run changes in the Republican Southern vote share using the same set of VECMs used to estimate long-run relationships and compare these models with results from OLS conventional models. The latter contain similar sets of short-run variables as their VECM counterparts along with the first difference of the vote share to mirror the lag length of 1 on first-differences long-run variables in the VECMs. This lag length implies a sample of 1958-2004. One difference between the corresponding groups of models is that the conventional models include the first difference of the civil rights variable, which is significant in conventional models that focus on tracking just changes in vote share. In contrast, the level version of CIVILRIGHTS, which was designed to control for the long-run influence of civil rights effects in the long-run portion of the mutual fund VECMs, was insignificant but did not yield any other qualitatively different result (a table of results is available upon request).

[FIGURE 8 OMITTED]

Several patterns arise in Table 4 that are similar to the national results. First, MIDTERM is statistically significant in all models. Second, REALINCOME is insignificant but is included in Models 5-7 for making comparisons with the national results. Third, the error correction terms are significant with the expected negative sign. Reflecting this is the fourth notable pattern that the mutual fund models outperform their conventional models, with corrected [R.sup.2]s higher by .23-.28 in corresponding VECMs. Figure 9 illustrates how mutual fund Model 5 outperforms its conventional counterpart. Overall, the Southern results are consistent with the national House findings.

VI. STOCK OWNERSHIP AND SENATE ELECTIONS

This section evaluates whether the stock ownership proxy adds information to models of the two-party split of the total popular vote in Senate elections (SREPVOTE). Because Senate elections in any particular cycle are not representative of the United States, since there are so few Senate races in each election cycle, and because Senators tend to enjoy much greater name recognition than House members, one should not expect that the results will be as strong of the models as tightly fitting as those for the House. Still, qualitatively similar results would bolster House findings that stock ownership matters.

As before, we first review long-run relationships. ADF tests (Table 5) indicate that there is a unit root in the aggregate Senate vote. Accordingly, the methodology for the Senate analysis is similar, with one notable difference. In contrast to House elections, Senate vote totals are not from a nationwide electorate since one-third of the Senate is chosen when House elections are held. As a result, the two-party split in the Senate popular vote fluctuates more than the House vote, first because there are three different subsamples of Senate electorates and second because there are fewer Senate races and a smaller Senate electorate. To control for the former, models include two 0-1 variables (SENATEGROUP1 and SENATEGROUP2) for two groupings (with the omitted group implicitly controlled for by default) as exogenous variables in the VECMs and as right hand side variables in the conventional specifications. (14)

[FIGURE 9 OMITTED]

A. Stock Ownership and Senate Elections in the Long Run

As with the House vote, several VECMs are estimated with similar sets of short-run exogenous variables to test for robustness in a limited sample. Each set includes the two senate group 0-1 variables. One set includes no other exogenous variables; another adds MIDTERM and REALINCOME; and one included MIDTERM, REALINCOME, WATERGATE, and PRESAPPROVAL. Since only MIDTERM and PRESAPPROVAL are significant in the last set and in a series of regressions that drop the least significant of the insignificant variables, a last set just adds these two variables.

Results in Table 5 indicate that the two-party Senate vote split is cointegrated with mutual fund costs. As with the House findings, the long-run mutual fund cost variable (log-level of MFCOST) is statistically significant with an implied and hypothesized negative effect on the Republican share of the Senate vote. Although the size of the long-run MFCOST coefficient is smaller than that in House elections, this may reflect modeling problems stemming from different subsample sets of Senate races--interestingly, one-senate group variable (SENATEGROUP1) is statistically significant.

B. Stock Ownership and Senate Elections in the Short Run

In important ways, the Senate results are similar to those for the House. First, the error correction term is highly significant and negative. Recall that mutual fund costs are negatively related to stock ownership rates and to the Republican share of the Senate vote. Since the error correction term equals actual minus the estimated equilibrium level, the negative EC coefficients imply that long-lasting decreases in mutual fund costs are associated with increases in the Republican popular vote share in the short run as well as the long run. A second, related similarity is that mutual fund models have much higher adjusted [R.sup.2]s than corresponding conventional models, with model fits rising by .15 comparing Models 1 and 5 and by .12 comparing Models 4 and 8. Third, residuals are well behaved in the mutual fund models but not in most conventional models. Fourth, the midterm effect is negative and significant although smaller in size than in the House models. Fifth, presidential approval is significantly positive, implying a "presidential pulse effect" for having a more popular president from one's party. Finally, the Watergate variable is statistically insignificant, as in the models of the House vote.

Nevertheless, there are a few differences across the House and Senate results in Tables 2 and 6. First, the size of the error correction coefficients in the Senate VECMs exceeds 100%. Second, the disposable income variable (REALINCOME) is insignificant in Senate models, whereas it is significant in modeling House popular vote patterns. Third, the model fits are somewhat higher for the House models. In this regard, the more prominent role played by personality in Senate races suggests that economic variables may not explain as much of the variation in the aggregate Senate as in nationwide House popular vote. Furthermore, the smaller aggregate Senate electorate in any given election may also add volatility to Senate voting fluctuations that is difficult to model.

Some of these differences are plausibly linked to the difficulty posed by modeling voting across three different groups of Senate races, which may not be fully captured by including 0-1 variables. Although the large error correction coefficients for the Senate suggest that overshooting behavior may arise in Senate elections, they may instead reflect that short-run variables may not fully capture short-run shifts arising from shifts in the composition of Senate races. Because the estimated equilibrium values used to construct the EC terms effectively average across the three groups of Senate races, any missed shifting effects may be picked up by a larger-than-normal estimated coefficient on the EC term which would inadvertently pick up a tendency to return toward the mean. Nevertheless, the basic qualitative results for the Senate are consistent with the more credible House findings of an investor class effect of higher stock ownership rates.

VII. CONCLUSIONS

Since the late 1980s, there has been a large rise in stock ownership rates through mutual funds and an upward shift in the Republican share of the House popular vote, leading some analysts, such as Glassman (1999, 2000, 2001), to argue that an investor class effect has affected voting. However, testing this hypothesis has been hampered by the lack of continuous time series data on stock ownership rates.

Fortunately, new data on equity mutual fund costs line up well with discontinuous stock ownership rates as Duca (2005, 2006) showed. Drawing on this relationship and new studies of equity ownership by Duca (2001, 2005, 2006) and Heaton and Lucas (2000), we use mutual fund cost data to proxy for stock ownership and find evidence supporting the hypothesis that an investor class effect has increased the Republican share of the House popular vote. In particular, results show that the long-run decline in mutual fund costs not only is strongly correlated with the Republican vote share in the long run, but also is helpful in tracking short-run changes. In addition, these qualitative results were also obtained when analyzing voting behavior in the South--controlling for shifts in relative regional income--and even in nationally unbalanced Senate vote totals.

Although the notion that property interests affect voting is old and intuitive, our study contributes to the empirical literature and suggests new research topics. Changes in household portfolios could, via their impact on voting, affect the coalitions and agendas of the major parties, spurring both to compete more for actual or potential shareholders. Indeed, the presidential candidates from the two major parties in 2000 each had proposals regarding mutual fund investing, and there are many congressional proposals as well. (15) Future voting may also be influenced by property interests affected by pension or Social Security legislation or by new financial products, such as exchange traded funds.

APPENDIX A: HOW EQUITY MUTUAL FUND COSTS ARE MEASURED

Following Duca (2005, 2006), mutual fund costs are based on a sample of large equity mutual funds. Mutual fund fees are measured using the weighted-average front-end and back-end load as a percent of assets at a 5-yr horizon. The back-end load is for withdrawals 5 yr after assets are invested, front- and back-end loads are annualized by dividing by 5, and data are weighted by a fund's relative asset size. Because mutual funds may substitute annual expenses for loads, expense ratios are added to loads to track total costs, MFCOST. It is reassuring that MFCOST moves with post-1980 overall equity fund estimates by Rea and Reid (1998) from the Investment Company Institute, which fell in the late 1990s as lower in loads outweighed slightly higher expense ratios.

Because pre-1980 data are sketchy, mutual fund costs were based on a sample of large mutual funds. Funds were included if assets were at least: $1 billion at year-end 1991 if the fund began before the mid-1980s, $2 billion at year-end 1994 if the fund began after 1983, $5 billion at year-end 2003 of 2000, or $250 million at year-end 1975. The first criterion reflects whether a fund was large during the early 1990s, when bond and equity mutual fund loads fell and households shifted assets out of M2 into mutual funds. The second criterion follows Duca (2000) and reflects whether a growing but new fund was large near the end of these portfolio shifts. The third criterion reflects whether a fund was large in the early 2000s. Given the stock appreciation of the 1990s, the hurdles for newer funds were higher for the 1994 and early 2000s to keep data gathering costs from exploding. The fourth criterion includes funds that were relatively large in 1975 when few funds existed. As in Duca (2005), funds are omitted if they were closed-end or only open to employees of a specific firm with a few reasonable exceptions.

[FIGURE 10 OMITTED]

APPENDIX B: WHY THE HOMEOWNERSHIP RATE IS UNLIKELY TO ACCOUNT FOR THE LEVEL SHIFT IN THE TWO-PARTY SPLIT IN THE HOUSE POPULAR VOTE

For three empirical reasons, there does not appear to be a long-run relationship between the homeownership rate and the House popular vote. First, U.S. homeownership data start in 1966, which shortens the sample, which may in turn account for why homeownership rates (in logs or levels) do not appear to be I(1), in contrast to mutual fund costs and the Republican share of the House vote which are I(1) over the longer 1954-2004 period. Second, U.S. homeownership rates were flat over 1966-1996, before rising in the late 1990s. This increase in the late 1990s lags behind the uptrend in the Republican share of the House popular (Figure 10), in contrast to stock ownership rates (mutual fund costs), which rose (fell) slightly ahead of voting shares. It is more plausible that a change in property ownership status would lead or coincide with a change in voting behavior. Third, the size of the shift in the U.S. popular vote (about 5 percentage points) exceeds that of the 3-percentage point rise of homeownership rates in the late 1990s, which apparently was concentrated among minority households, whose voting behavior did not change much in the last 10 yr. Given some preexisting division in party loyalties, the 5-percentage point shift in voting patterns appears more plausibly linked to the larger 26-percentage point rise in the stock ownership rate. On these empirical grounds, a rise in stock ownership rates, rather than homeownership rates, appears to be a more relevant factor behind an apparent realignment toward the Republican Party in House elections. Intuitively, homeownership may create more of an incentive for voters to seek economic stability, whereas stock ownership may give voters more of an incentive to back candidates with more business-friendly policy positions.

TABLE A1
Cointegration Results for Republican Share of Two-Party House
Popular Vote

 Cointegrating
 Equilibrium
 Relationship
 (Assumption: No
 Trend in
 Vector, but
Model and No. of Significant Trend in Log Likelihood
Vector Vectors Variables (AIC)

No short-run variables present
5 1 ln(REPVOTE) = 4.039361
 - 0.340668 ln (MFCOST) ** 94.16094 (-6.923722)
 (-6.00)
With MIDTERM and disposable income growth present
6 1 ln(REPVOTE) = 3.998250
 - 0.266393 ln(MFCOST) ** 104.5917 (-7.871969)
 (-3.68)
With MIDTERM, disposable income growth, WATERGATE, GINGRICH, and
presidential approval present
7 1 ln(REPVOTE) = 3.993854
 - 0.258452 ln(MFCOST) ** 117.6296 (-9.057240)
 (-3.19)
With MIDTERM, disposable income growth, and presidential approval
present
8 1 ln(REPVOTE) = 3.995695
 - 0.261777 ln(MFCOST) ** 112.7119 (-8.610172)
 (-4.03)
Unit root test statistics, ADF statistics

 Eigenvalue (Trace Statistics,
 Max-Eigen Statistics

Model and No. of Significant
Vector Vectors 0 Vector 1 Vector

No short-run variables present
5 1 0.631112 0.001801
 (21.97945 **, (0.039667,
 21.93979 **) 0.039667)
With MIDTERM and disposable income growth present
6 1 0.510962 0.068391
 (17.29549 *, (1.558534,
 15.73695 *) 1.558534)
With MIDTERM, disposable income growth, WATERGATE, GINGRICH, and
presidential approval present
7 1 0.488966 0.105707
 (17.22689 *, (2.457892,
 14.76900 *) 2.457892)
With MIDTERM, disposable income growth, and presidential approval
present
8 1 0.587261 0.066148
 (20.97430 **, (1.505610,
 19.46869 **) 1.505610)
Unit root test statistics, ADF statistics

 Level 5% Critical Level
 (AIC Lag) for Lag

ln(REPVOTE) 0.449068 (4) -3.012363
ln(REPVOTE) 0.166639 (6) -3.673616
ln(REPVOTE) 0.174860 (3) -3.004861
ln(REPVOTE) -0.729981 (2) -3.622033

 1% Critical Level First Difference
 for Lag (AIC Lag)

ln(REPVOTE) -3.788030 -7.959103 ** (0)
ln(REPVOTE) -4.532598 -7.976779 ** (0)
ln(REPVOTE) -3.769597 -1.146193 (3)
ln(REPVOTE) -4.416345 -5.308305 ** (0)

 5 Critical level 1 Critical level
 for Lag for Lag

ln(REPVOTE) -2.991878 -3.737853
ln(REPVOTE) -3.612199 -4.394309
ln(REPVOTE) -3.012363 -3.788030
ln(REPVOTE) -3.612199 -4.394309

 Assumptions

ln(REPVOTE) Constant/no trend
ln(REPVOTE) Constant/no trend
ln(REPVOTE) Constant/no trend
ln(REPVOTE) Constant/no trend

Notes: t Statistics are in parentheses except when the AIC
statistic is reported. For each of the vectors listed above,
a lag length of 3 minimized the AIC in the cointegration tests
and yielded significant and unique vectors. The modified AIC
was used to determine the lag length (up to 10) in the ADF unit
root tests. Data used span 1954-2004.

* and ** denotes significance at the 95%, and 99%, levels,
respectively.

TABLE A2
Models of the Percent Change in the Republican Share of the Two-Party
Popular Vote in House Elections (VECM Results Allowing for Time Trends
in Variables, 1962-2004)

 No Error Correction Term

Variable Model 1 Model 2

Constant 0.0121 (0.83) 0.0122 (1.23)
[EC.sub.t-1]
[MIDTERM.sub.t] 0.0711 ** (-4.59)
[REALINCOME.sub.t] 0.0098 ** (2.77)
[WATERGATE.sub.t]
[GINGRICH.sub.t]
[PRESAPPROVAL.sub.t]
[DELTAln -0.653 * (-2.79) -0.3045 ([dagger])
 [(REPVOTE).sub.t-1] (-1.72)
[DELTA]ln
 [(MFCOST).sub.t-1]]
Adjusted [R.sup.2] .204 .630
LM(l) 0.15 0.02
LM(2) 5.00 ([dagger]) 2.51
Q(12) 12.58 10.40

 No Error Correction Term

Variable Model 3 Model 4

Constant -0.5887 ** (-4.27) -0.5026 ** (-3.17)
[EC.sub.t-1]
[MIDTERM.sub.t] -0.0683 ** (-4.94) -0.0819 ** (-6.45)
[REALINCOME.sub.t] 0.0013 (0.47) 0.0057 * (1.86)
[WATERGATE.sub.t] -0.07731 (-1.86)
[GINGRICH.sub.t] 0.0415 (1.07)
[PRESAPPROVAL.sub.t] 0.1547 ** (434) 0.1325 ** (3.25)
[DELTAln -0.2589 * (-2.08) -0.2627 ([dagger])
 [(REPVOTE).sub.t-1] (-1.87)
[DELTA]ln
 [(MFCOST).sub.t-1]]
Adjusted [R.sup.2] .796 .768
LM(l) 0.92 3.361
LM(2) 6.71 * 3.73
Q(12) 10.21 17.81

 Error Correction Term

Variable Model 5 Model 6

Constant 0.0453 * (2.18) 0.0107 (0.90)
[EC.sub.t-1] -1.9081 ** (-3.40) -1.1038 * (-2.60)
[MIDTERM.sub.t] -0.0554 ** (-4.02)
[REALINCOME.sub.t] 0.0089 ** (3.12)
[WATERGATE.sub.t]
[GINGRICH.sub.t]
[PRESAPPROVAL.sub.t]
[DELTAln 0.8016 (1.67) 0.4243 (1.27)
 [(REPVOTE).sub.t-1]
[DELTA]ln 0.7370 (1.59) 0.1889 (0.64)
 [(MFCOST).sub.t-1]]
Adjusted [R.sup.2] .508 .779
LM(l) 1.63 4.24 *
LM(2) 1.64 4.47
Q(12) 13.04 10.94

 Error Correction Term

Variable Model 7 Model 8

Constant -0.4000 ** (-3.03) -0.3970 ** (-3.26)
[EC.sub.t-1] -0.73551 (-2.06) -0.9284 * (-2.97)
[MIDTERM.sub.t] -0.0564 ** (-4.95) -0.0668 ** (-6.31)
[REALINCOME.sub.t] 0.0056 ([dagger]) 0.0059 * (2.62)
 (2.10)
[WATERGATE.sub.t] -0.0398 (-0.96)
[GINGRICH.sub.t] 0.0386 (1.01)
[PRESAPPROVAL.sub.t] 0.1050 ** (3.09) 0.1041 ** (3.35)
[DELTAln 0.1372 (0.47) 0.3239 (1.32)
 [(REPVOTE).sub.t-1]
[DELTA]ln 0.0393 (0.17) 0.1281 (0.60)
 [(MFCOST).sub.t-1]]
Adjusted [R.sup.2] .876 .881
LM(l) 2.47 0.16
LM(2) 3.18 0.16
Q(12) 14.27 11.00

Notes: t Statistics are in parentheses. EC terms are from VECMs that
jointly estimate long- and short-run relationships. t - 2 and t - 3
lags of [DELTA]ln(REPVOTE) and [DELTA]ln(MFCOST) are not reported
to conserve space.

*, **, and ([dagger]) are significant at the 95%, 99%, and 90%
levels, respectively.

ABBREVIATIONS

ADF: Augmented Dickey-Fuller

AIC: Akaike Information Criterion

GNP: Gross National Product

LM: Lagrange Multiplier

OLS: Ordinary Least Square

VECM: Vector Error Correction Models

REFERENCES

Aiyagari, S. R., and M. Gertler. "Asset Returns with Transactions Costs and Uninsured Individual Risk." Journal of Monetary Economics, 27, 1991, 311-31.

Aizcorbe, A. M., A. B. Kennickell, and K. B. Moore. "Recent Changes in U.S. Family Finances: Evidence from the 1998 and 2001 Survey of Consumer Finances." Federal Reserve Bulletin, 89, 2003, 1-32.

Barro, R. J., and X. Sala-i-Martin. "Convergence across States and Regions." Brookings Papers on Economic Activity, 1, 1991, 107-82.

Boskin, M. J., and E. Sheshinski. "Individual Welfare Depends Upon Relative Income." Quarterly Journal of Economics, 92, 1978, 589-601.

Bucks, B. K., A. B. Kennickell, and K. B. Moore. "Recent Changes in U.S. Family Finances: Evidence from the 2001 and 2004 Survey of Consumer Finances." Federal Reserve Bulletin, 92, 2006, A1-38.

Campbell, A. "Surge and Decline: A Study of Electoral Change," in Elections and the Political Order, edited by A. Campbell, P. E. Converse, W. E. Miller, and D. E. Stokes. New York: Wiley, 1966, 40-62.

Campbell, J. E. The Presidential Pulse of Congressional Elections. 2nd ed. Lexington, KY: The University of Kentucky Press, 1997.

Cochrane, S. "Regional Income Convergence." Dismal Economist, 9 April 1999, 1-4.

Conway, M. M., and M. L. Wyckoff. "Voter Choice in the Congressional Elections." American Politics Quarterly, 8, 1980, 3-14.

Dixit, A. K. "Entry and Exit Decisions Under Uncertainty." Journal of' Political Economy, 97, 1989, 620-37.

Downs, A. An Economic Theory of Democracy. New York: Harper, 1957.

Duca, J. V. "Financial Technology Shocks and the Case of the Missing M2." Journal of Money, Credit, and Banking, 32, 2000, 820-39.

--. "The Rise of Stock Mutual Funds." Southwest Economy, Federal Reserve Bank of Dallas, January/February 2001, 1, 6-9.

--. "Why Have Households Increasingly Used Mutual Funds to Own Equity?" Review of Income and Wealth, 51, 2005, 375-96.

--. "Mutual Funds and the Evolving Long-Run Effects of Stock Wealth on U.S. Consumption." Journal of Economies and Business, 58, 2006, 202-21.

Durkin, T. A., and G. E. Elliehausen 1977 Consumer Credit Survey. Washington, DC: Board of Governors of the Federal Reserve System, 1978.

Easterlin, R. A. "Will Raising the Incomes of All Enhance the Happiness of All?" Journal of Economic Behavior and Organization, 27, 1995, 35-47.

Erikson, R. S. "The Puzzle of Midterm Loss." Journal of Politics, 50, 1988, 1011-29.

Faucheux, R. "Politics of the New Stockholder Majority." Campaigns and Elections, July 1999, 7.

Fiorina, M. P. Retrospective Voting in American Elections. New Haven, CT: Yale University Press, 1981.

Frank, R. H. "The Demand for Unobservable and Other Nonpositional Goods." American Economic Review, 75, 1985, 101-16.

Gigot, P. "This Isn't What Marx Meant by Das Kapital." Wall Street Journal, 19 March 1999, A19.

Glassman, J. K. "The Rise of the Investor Class." AEI on the Issues, The American Enterprise Institute, May 1999, 1-4. http://www.aei.org/publications/ pubID.102811pub_detail.asp [accessed Jan 2001].

--. "Election 2000: Year of the Investor Class?" Reason Magazine, 10, July 2000. http://www.reason.com/ news/show/36063.html [accessed Jan 2001].

--. "Shareholders of the World Unite." The American Enterprise, January/February 2001, 17.

Gross, D. Bull Run: Wall Street, the Democrats, and the New Polities of Personal Finance. New York: Public Affairs, 2000.

Guiso, L., M. Haliassios, and T. Jappelli. "Household Stockholding in Europe: Where Do We Stand and Where Do We Go?" Economic Policy, 18, 2003, 123-70.

Heaton, J., and D. Lucas. "Portfolio Choice in the Presence of Background Risk." Economic Journal, 110, 2000, 1-26.

Husted, T. A., and L. W. Kenny. "The Effect of the Expansion of the Voting Franchise on the Size of Government." Journal of Political Economy, 105, 1997, 54-82.

Katona, G., C. A. Lininger, and E. Mueller. 1967 Survey of Consumer Finances. Ann Arbor, Michigan: University of Michigan, 1968.

Katona, G., W. Dunkleberg, G. Hendricks, and J. Schmiedeskamp. 1969 Survey of Consumer Finances. Ann Arbor, Michigan: University of Michigan, 1970.

Katona, G., L. Mandell, and J. Schmiedeskamp. 1970 Survey of Consumer Finances. Ann Arbor, Michigan: University of Michigan, 1971.

Kennickell, A. B., M. Starr-McCluer, and B. J. Surette. "Recent Changes in U.S. Family Finances: Results from the 1998 Survey of Consumer Finances." Federal Reserve Bulletin, 86, 2000, 1-29.

Kernell, S. "Presidentail Popularity and Negative Voting: An Alternative Explanation of the Midterm Electoral Decline of the President's Party." American Political Science Review, 71, 1977, 44-66.

Kiewiet, D. R. Microeconomics and Macropolitics: The Electoral Effects of Economic Issues. Chicago, IL: University of Chicago Press, 1983.

Kramer, G. H. "Short-Term Fluctuations in U.S. Voting Behavior, 1896-1964." American Political Science Review, 65, 1971, 131-43.

Kudlow, L. A. American Abundance: The New Economic and Moral Prosperity. New York: Forbes Publishing Group, 1997.

Ladd, E. C. "The 1994 Congressional Elections: The Postindustrial Realignment Continues." Political Science Quarterly, 110, 1995, 1-23.

Lau, R. R. "Two Explanations of Negativity Effects in Political Behavior." American Journal of Political Science, 29, 1985, 119-38.

Lungqvist, L., and H. Uhlig. "Tax Policy and Aggregate Demand Management Under Catching Up with the Joneses. American Economic Review, 90, 2000, 356-66.

Madison, J. Notes of Debates in the Federal Convention of 1787 Reported by James Madison. New York: W.W. Norton and Company Inc., 1987.

Marx, K. Capital." A Critique of Political Economy. New York: Vintage Books, 1977.

McLeod, J. M., J. D. Brown, and L. B. Becker. "Watergate and the 1974 Congressional Elections." Public Opinion Quarterly, 41, 1977, 181-95.

Mueller, D. C. Public Choice III. New York: Cambridge University Press, 2003.

Nadler, R. "The Rise of Worker Capitalism," Cato Policy Analysis 359, Washington, DC: Cato Institute, 1999.

Peltzman, S. "How Efficient Is the Voting Market?" Journal of Law and Economics, 33, 1990, 27-63.

Rasmussen Research. "Exit Survey Data." Posted on 7 November 2000. www.portraitofamerica.com\html\ poll-1528.html [accessed Nov 2000].

Rea, J.D., and B.K. Reid. "Trends in Ownership Cost of Equity Mutual Funds." Perspective, 4, 1998, 1-15.

Rey, S. "Spatial Dependence in the Evolution of Regional Income Distributions." Working Paper REAL 01-T03, University of Illinois and Federal Reserve Bank of Chicago, 2001.

Sherwood-Call, C. "The 1980s Divergence in State Per Capita Incomes: What Does It Tell Us?" Economic Review, Federal Reserve Bank of San Francisco, January 1996, 14-25.

Siegel, J. "The Shrinking Equity Premium: Historical Facts and Future Forecasts." Journal of Portfolio Management, 26, 1999, 10-17.

Tufte, E. R. "Determinants of the Outcomes of Midterm Congressional Elections." American Political Science Review, 69, 1975, 812-26.

Vissing-Jorgensen, A. "Towards an Explanation of Household Portfolio Choice Heterogeneity: Nonfinancial Income and Participation Cost Structures." NBER Working Paper 8884, April 2002.

Zogby International. "Investors Think, Act, and Vote Differently Than Non-Investors." 23 October 2000. http:/zogby.com/news/ReadNews. dbm?ID=290

(1.) See Faucheux (1999), Gigot (1999), Glassman (2000), Gross (2000), Kudlow (1997), and Nadler (1999).

(2.) Appendix B reviews why homeownership is unlikely to have affected long-run House voting patterns.

(3.) Rates include direct and indirect ownership (mainly mutual funds), with both components--especially the latter, including retirement assets. The rates plotted are not fully consistent for minor differences (Duca 2005) and are from Aizcorbe, Kennickell, and Moore (2003), Bucks, Kennickell, and Moore (2006), Durkin and Elliehausen (1978), Katona et al. (1968, 1970, 1971), and Kennickell, Starr-McCluer, and Surette (2000). 1983 Data are computed using Survey of Consumer Finances sample weights and definitions from earlier surveys. Equity fund costs are from Duca (2005, 2006).

(4.) In Zogby International's October 2000 poll, investors were evenly split between intentions to vote for Democrats or Republicans, with noninvestors favoring Democrats 43%-32%. Investors were split between self-described Democrats and Republicans (38% vs. 37%), with noninvestors more likely to be Democrats (45% vs. 27%). In 2000 exit polls, shareholders favored Bush to Gore by 51%-45% and nonshareholders favored Gore by 51%-43% (Rasmussen Research 2000). Using Rasmussen Research's January 1999 poll, Nadler analyzed party affiliation by income and whether people had portfolios of at least $5,000. In four groups, portfolio owners favored Republicans over nonowners by 12-20 percentage points. For the 831 out of 6,400 respondents earning above $75,000, the gap was only 2 percentage points.

(5.) Data are in logs because it helps to better estimate separate effects from other variables in modeling short-run changes in voting and to more accurately identify long-run relationships (cointegrating vectors).

(6.) Kramer (1971) found that incumbency-adjusted unemployment rate, nominal GNP growth, and inflation were insignificant in the presence of significant income growth. Finding parallel results with first differences, we only show results using REALINCOME as the only short-run economic variable to conserve space.

(7.) Campbell (1997) used Gallup data but focused on midterm elections.

(8.) To limit simultaneity bias, PRESAPPROVAL was replaced in other models with PRESPOPWIN, equal to 1 if the Republican won the popular vote, -1 if the Democrat won, and 0 otherwise. As qualitative results were similar but model fits were better with PRESAPPROVAL, we report these runs to conserve space.

(9.) Fiorina (1981) argued that Watergate's effect on party affiliation was permanent not transitory, while Ladd (1995) saw the shift as insufficient to curtail broader trends favoring a Republican realignment.

(10.) Campbell (1997) found that each percentage point margin of victory in the presidential popular vote cuts a party's House popular vote share by 0.2 percentage points in the next midterm election. Campbell's results and the average 9-point victory margin in our sample imply a midterm effect of 2 points, near our estimates.

(11.) Campbell (1997) found that a party's popular vote share rises 0.4 percentage points per percentage point margin of victory in the presidential popular vote, for an average win effect of 4 percentage points.

(12.) Allowing for a time trend yields qualitatively similar findings for the non-South as mutual fund costs are cointegrated with Republican vote share with the expected significant sign; EC terms are significant in modeling first differences; and VECMs outperform corresponding conventional models (results available).

(13.) As non-South median income is unavailable, we use the ratio of southern to U.S. median income, which is similar to state income dispersion data that converged in most of the 1900s (Barro and Sala-i-Martin 1991) but not much since the early 1980s (Cochrane 1999; Rey 2001, 4; Sherwood-Call 1996).

(14.) SENATEGROUP1 (SENATEGROUP2) equals 1 every sixth year since 1954 (1956) and 0 otherwise. Perhaps reflecting three groups of Senate elections, a greater role of personalities in Senate elections, and a smaller number of Southern Senate elections, the Southern two-party split of the Senate popular vote does not have a unit root.

(15.) George Bush proposed raising the annual limits on tax-deferred contributions to thrift plans and IRA contributions and converting some portion of Social Security into a defined contribution-like plan. Al Gore proposed giving many families access to a national thrift plan, with federal matching of contributions based on income. In June 2001, legislation was enacted that increased IRA and thrift plan limits starting in 2002.

JOHN V. DUCA and JASON L. SAVING, We would like to thank Alan Viard, two anonymous referees, and session participants at the American Economic Association, Western Economic Association International, and American Political Science Association meetings for suggestions. We also thank Daniel Wolk, Jamie Lee, and Christine Rowlette for excellent research assistance. The views expressed are those of the authors and do not necessarily reflect those of the Federal Reserve Bank of Dallas or the Federal Reserve System. Our study analyzes shifting political trends and in no way should be interpreted as endorsing or criticizing any political party or figure. Any remaining errors are ours alone.

Duca: Vice President and Senior Economist, Research Department, Federal Reserve Bank of Dallas, P.O. Box 655906-5906, Dallas, TX 75265. Phone 1-214-922-5154, Fax 1-214-922-5194, E-mail john.v.duca@dal. frb.org

Saving: Senior Economist, Research Department, Federal Reserve Bank of Dallas, P.O. Box 655906-5906, Dallas, TX 75265. Phone 1-214-922-5167, Fax 1-214-922-5194, E-mail jason.saving@dal.frb.org

TABLE 1
Cointegration Results for Republican Share of U.S. Two-Party House
Popular Vote, 1954-2004

 Cointegrating Equilibrium
 No. of Relationship
Model and Significant (Assumption: No Trend in
 Vector Vectors Vector or in Variables)

No short-run variables present
5 1 ln(REPVOTE) = 4.071488 ** - 0.341267 ln(MFCOST) **
 (102.04) (-6.21)
With MIDTERM and disposable income growth present
6 1 ln(REPVOTE) = 4.023977 ** - 0.275105 ln(MFCOST) **
 (80.12) (-3.98)
With MIDTERM, disposable income growth, WATERGATE, GINGRICH, and
presidential approval present
7 1 ln(REPVOTE) = 3.363681 ** - 0.271182 ln(MFCOST) **
 (26.53) (-6.27)
With MIDTERM, disposable income growth, and presidential approval
present
8 1 ln(REPVOTE) = 3.652723 ** - 0.246681 ln(MFCOST) **
 (35.40) (-3.77)
Unit root test statistics, ADF statistics

 Eigenvalue (Trace Statistic, Max-
Model and Log Likelihood Eigen Statistic)
 Vector (AIC)
 0 Vector

No short-run variables present
5 93.10325 (-6.918477) 0.647815 (25.11423 **,
 22.95918 **)
With MIDTERM and disposable income growth present
6 103.2153 (-7.837754) 0.530662 (20.95276 *,
 16.64149 *)
With MIDTERM, disposable income growth, WATERGATE, GINGRICH, and
presidential approval present
7 104.3688 (-8.669888) 0.813277 (41.95596 **,
 36.91885 **)
With MIDTERM, disposable income growth, and presidential approval
present
8 114.4425 (-8.585678) 0.738777 (33.57684 **,
 31.12530 **)
Unit root test statistics, ADF statistics

 Eigenvalue (Trace Statistic, Max-
Model and Eigen Statistic)
 Vector
 1 Vector

No short-run variables present
5 0.093312 (2.155057,
 2.155057)
With MIDTERM and disposable income growth present
6 0.177960 (4.311263,
 4.311263)
With MIDTERM, disposable income growth, WATERGATE, GINGRICH, and
presidential approval present
7 0.204639 (5.037109,
 5.037109)
With MIDTERM, disposable income growth, and presidential approval
present
8 0.167931 (4.044482,
 4.044482)
Unit root test statistics, ADF statistics

 Level 5% Critical Level 1% Critical Level
 (AIC Lag) for Lag for Lag

ln(REPVOTE) 0.449068 (4) -3.012363 -3.788030
ln(REPVOTE) 0.166639 (6) -3.673616 -4.532598
ln(MFCOST) 0.174860 (3) -3.004861 -3.769597
ln(MFCOST) -0.729981 (2) -3.622033 -4.416345

 First Difference 5 Critical Level 1 Critical Level
 (AIC Lag) for Lag for Lag

ln(REPVOTE) -7.959103 ** (0) -2.991878 -3.737853
ln(REPVOTE) -7.976779 ** (0) -3.612199 -4.394309
ln(MFCOST) -1.146193 -3.012363 -3.788030
ln(MFCOST) -5.308305 ** (0) -3.612199 -4.394309

 Assumptions

ln(REPVOTE) Constant/no trend
ln(REPVOTE) Constant and trend
ln(MFCOST) Constant/no trend
ln(MFCOST) Constant and trend

Notes: t Statistics are in parentheses except when the AIC
statistic is reported. For Vectors 5, 6, and 8, a lag length
of 3 minimized the AIC in the cointegration tests and yielded
significant and unique vectors. For Vector 7, the AIC implied
a lag length of 1. The modified AIC was used to determine the
lag length (up to 10) in the ADF unit root tests. Data span
1954-2004.

* and ** denote significance at the 95%, and 99% levels,
respectively.

TABLE 2
Models of the Percent Change in the Republican Share of the Two-Party
Popular Vote in House Elections (VECMs, Results over 1962-2004
Using Data over 1954-2004)

 No Error
 Correction Term

Variable Model 1 Model 2

Constant 0.0121 (0.83) 0.0122 (1.23)
[EC.sub.t - 1]
[MIDTERM.sub.t] 0.0711 ** (-4.59)
[REALINCOME.sub.t] 0.0098 * (2.77)

[WATERGATE.sub.t]
[GINGRICH.sub.t]
[PRESAPPROVAL.sub.t]
[DELTA]ln -0.653 * (-2.79) -0.3045 (-1.72)
 [(REPVOTE).sub.t-1]
[DELTA]ln
 [(MFCOST).sub.t-1]
Adjusted [R.sup.2] .204 .630
LM(1) 0.15 0.02
LM(2) 5.00 ([dagger]) 2.51
Q(12) 12.58 10.40

 No Error
 Correction Term

Variable Model 3 Model 4

Constant -0.5887 ** (-4.27) -0.5026 ** (-3.17)
[EC.sub.t - 1]
[MIDTERM.sub.t] -0.0683 ** (-4.94) -0.0819 ** (-6.45)
[REALINCOME.sub.t] 0.0013 (0.47) 0.0057 ([dagger])
 (1.86)
[WATERGATE.sub.t] -0.07731 (-1.86)
[GINGRICH.sub.t] 0.0415 (0.07)
[PRESAPPROVAL.sub.t] 0.1547 ** (4.34) 0.1325 ** (3.25)
[DELTA]ln -0.2589 ([dagger]) -0.2627 ([dagger])
 [(REPVOTE).sub.t-1] (-2.08) (-1.87)
[DELTA]ln
 [(MFCOST).sub.t-1]
Adjusted [R.sup.2] .796 .768
LM(1) 0.92 3.36 ([dagger])
LM(2) 6.71 * 3.73
Q(12) 10.21 17.81

 No Error Error
 Correction Term Correction Term

Variable Model 5 Model 6

Constant
[EC.sub.t - 1] -1.7693 ** (-3.30) -0.9875 * (-2.40)
[MIDTERM.sub.t] -0.0565 ** (-4.10)
[REALINCOME.sub.t] 0.0088 * (3.08)

[WATERGATE.sub.t]
[GINGRICH.sub.t]
[PRESAPPROVAL.sub.t]
[DELTA]ln 0.7118 (1.53) 0.353 (1.07)
 [(REPVOTE).sub.t-1]
[DELTA]ln 0.7833 (1.71) 0.2456 (0.83)
 [(MFCOST).sub.t-1]
Adjusted [R.sup.2] .514 .778
LM(1) 0.44 2.00
LM(2) 0.44 2.01
Q(12) 12.97 9.93

 Error Correction Term

Variable Model 7 Model 8

Constant
[EC.sub.t - 1] -0.7258 ** (-7.68) -1.0307 ** (-5.07)
[MIDTERM.sub.t] -0.0613 ** (-6.76) -0.0646 ** (-7.47)
[REALINCOME.sub.t] 0.0039 * (2.23) 0.0063 ** (3.05)

[WATERGATE.sub.t] -0.0454 (-1.63)
[GINGRICH.sub.t] 0.0301 (1.14)
[PRESAPPROVAL.sub.t] 0.1216 ** (7.79) 0.0904 ** (5.04)
[DELTA]ln 0.1225 (1.25) 0.3919 * (2.19)
 [(REPVOTE).sub.t-1]
[DELTA]ln 0.0856 (0.54) 0.1355 (0.70)
 [(MFCOST).sub.t-1]
Adjusted [R.sup.2] .908 .890
LM(1) 0.33 0.50
LM(2) 3.23 0.58
Q(12) 15.79 8.79

Notes: t Statistics are in parentheses. EC terms are from VECMs
that jointly estimate long- and short-run relationships. t - 2 and
t - 3 lags of [DELTA]ln(REPVOTE) and [DELTA]ln(MFCOST) are
not reported to conserve space.

*, **, and ([dagger]) are significant at the 95%, 99%, and
90%, levels, respectively.

TABLE 3
Cointegration Results for Republican Shares of Southern Two-Party
House Popular Vote, 1954~2004

 Cointegrating
 Equilibrium
 Relationship
 (Assumption: No
 Trend in
Model and No. of Significant Vector or
Vector Vectors in Variables

Only CIVILRIGHTS present
5 1 ln(REPVOTESO) = -4.989362 **
 (-5.65)
 - 0.799579 ln(MFCOST) **
 (-15.82)
 + 1.921269 ln(SINCSHARE) **
 (9.67)
With CIVILRIGHTS, MIDTERM, and disposable income growth present
6 1 ln(REPVOTESO) = 5.746194 **
 (-6.82)
 - 0.776692 ln(MFCOST) **
 (-17.44)
 + 2.099087 ln(SINCSHARE) **
 (11.03)
With CIVILRIGHTS, MIDTERM, disposable income growth, WATERGATE,
GINGRICH, and presidential approval present
7 1 ln(REPVOTESO) - 5.525293 **
 (-6.23)
 - 0.789051 ln(MFCOST) **
 (-16.61)
 + 2.000153 ln(SINCSHARE) **
 (9.92)
With CIVILRIGHTS, MIDTERM. and presidential approval present
8 1 ln(REPVOTESO) = -5.416989 **
 (-6.41)
 - .802878 * ln(MFCOST) **
 (-17.66)
 + 1.980887 * ln(SINCSHARE) **
 (10.24)

Unit root test statistics, ADF statistics

Model and No. of Significant Log Likelihood
Vector Vectors (AIC)

Only CIVILRIGHTS present
5 1 167.9728
 (-12.66440)
With CIVILRIGHTS, MIDTERM, and disposable income growth present
6 1 179.2281
 (-14.43567)
With CIVILRIGHTS, MIDTERM, disposable income growth, WATERGATE,
GINGRICH, and presidential approval present
7 1 196.8619
 (-15.07182)
With CIVILRIGHTS, MIDTERM. and presidential approval present
8 1 189.3963
 (-14.44969)
Unit root test statistics, ADF statistics

 Eigenvalue (Trace Statistics,
 Max-Eigen Statistics

Model and No. of Significant
Vector Vectors 0 Vectors 1 Vector

Only CIVILRIGHTS present
5 1 0.0893983 0.375633
 (66.53954) **, (12.67970,
 53.859883 **) 11.30440)
With CIVILRIGHTS, MIDTERM, and disposable income growth present
6 1 0.915862 0.353355
 (72.02558 **, (12.61843.
 59.40715 **) 10.46298)
With CIVILRIGHTS, MIDTERM, disposable income growth, WATERGATE,
GINGRICH, and presidential approval present
7 1 0.926991 0.450651
 (77.55257 **, (14.74058,
 62.81198 **) 14.37651)
With CIVILRIGHTS, MIDTERM. and presidential approval present
8 1 0.917013 0.404728
 (72.98215 **, (13.24446,
 59.73769 **) 12.44968)
Unit root test statistics, ADF statistics

 5% Critical
 Level (AIC Lag) Level for Lag

ln(REPVOTESO) -1.591496 (1) -2.991878
ln(REPVOTESO) -2.624879 (0) -3.612199
ln(REPVOTENO) (a) -3.499933 * (1) -2.991878
ln(REPVOTENO) -3.473299 (1) -3.612199

 1% Critical First Difference
 Level for Lag (AIC Lag)

ln(REPVOTESO) -3.737853 -5.662519 ** (0)
ln(REPVOTESO) -4.394309 -5.633538 ** (0)
ln(REPVOTENO) (a) -3.737853 -5.436498 ** (0)
ln(REPVOTENO) -4.394309 -5.272608 ** (0)

 5 Critical 1 Critical
 Level for Lag Level for Lag

ln(REPVOTESO) -2.991878 -3.737853
ln(REPVOTESO) -3.612199 -4.394309
ln(REPVOTENO) (a) -2.998064 -3.752946
ln(REPVOTENO) -3.622033 -4.416345

 Assumptions

ln(REPVOTESO) Constant/no trend
ln(REPVOTESO) Constant and trend
ln(REPVOTENO) (a) Constant/no trend
ln(REPVOTENO) Constant and trend

Notes: t Statistics are in parentheses except when AIC statistic
is reported. For each vector, a lag length of 1 minimized the AIC
in the cointegration tests and yielded significant and unique vectors.
The modified AIC was used to determine the lag length (up to 10)
in the ADF unit root tests. Data used span 1954-2004.

(a) REPVOTENO denotes Republican share of non-South vote.

* and ** denote significance at the 95% and 99% levels, respectively.

TABLE 4
Models of the Percent Change in the Republican Share of the Two-Party
Popular Vote in Southern House Elections 1958-2004

 No Error Correction Term

Variable Model 1 Model 2

Constant 0.0248 (1.17) 0.0353 * (2.10)
[EC.sub.t-1]
[MIDTERM.sub.t] 0.0871 ** (-3.94) -0.0813 ** (-2.82)
[REALINCOME.sub.t] -0.0006 (-0.12) -0.0001 (-0.00)
[WATERGATE.sub.t]
[GINGRICH.sub.t]
[PRESAPPROVAL.sub.t]
[DELTA]ln -0.1465 (-0.98) -0.1170 (-0.87)
 [(REPVOTESO).sub.t-1]
[DELTA]ln
 [(MFCOST).sub.t-1]
[DELTA]ln
 [(SINCSHARE).sub.t-1]
[DELTA][CIVILRIGHTS 0.6301 ** (4.58) 0.5186 ** (4.74)
 .sub.t] or
 [CIVILRIGHTS.sub.t]
Adjusted [R.sup.2] .469 .687
LM(1) 2.27 0.04
LM(2) 2.52 2.10
Q(12) 9.58 6.40

 No Error Correction Term

Variable Model 3 Model 4

Constant -0.0593 (-0.17) -0.1590 (-0.51)
[EC.sub.t-1]
[MIDTERM.sub.t] -0.0936 ** (-4.02)
[REALINCOME.sub.t]
[WATERGATE.sub.t] -0.0070 (-0.08)
[GINGRICH.sub.t] 0.0816 (0.92)
[PRESAPPROVAL.sub.t] 0.0230 (0.26) 0.0494 (0.63)
[DELTA]ln -0.1037 (-0.64) -0.0677 (-0.51)
 [(REPVOTESO).sub.t-1]
[DELTA]ln
 [(MFCOST).sub.t-1]
[DELTA]ln
 [(SINCSHARE).sub.t-1]
[DELTA][CIVILRIGHTS 0.5367 ** (4.58) 0.5177 ** (4.79)
 .sub.t] or
 [CIVILRIGHTS.sub.t]
Adjusted [R.sup.2] .654 .693
LM(1) 0.35 0.00
LM(2) 1.56 1.87
Q(12) 5.00 6.35

 Error Correction Term

Variable Model 5 Model 6

Constant
[EC.sub.t-1] -1.0816 ** (-6.35) -0.8850 ** (-9.24)
[MIDTERM.sub.t] -0.0892 ** (-7.86)
[REALINCOME.sub.t] 0.0028 (1.12)
[WATERGATE.sub.t]
[GINGRICH.sub.t]
[PRESAPPROVAL.sub.t]
[DELTA]ln -0.3086 ([dagger]) 0.2707 * (2.83)
 [(REPVOTESO).sub.t-1] (-1.92)
[DELTA]ln 0.6083 (1.25) 0.7832 ** (3.11)
 [(MFCOST).sub.t-1]
[DELTA]ln -0.2761 (-0.26) -1.2977 * (-2.30)
 [(SINCSHARE).sub.t-1]
 [DELTA][CIVILRIGHTS 0.6202 ** (7.44) 0.4861 ** (11.79)
 .sub.t] or
 [CIVILRIGHTS.sub.t]
Adjusted [R.sup.2] .697 .921
LM(1) 1.19 1.86
LM(2) 1.79 4.22
Q(12) 16.13 8.07

 Error Correction Term

Variable Model 7 Model 8

Constant
[EC.sub.t-1] -0.8670 ** (-9.91) -0.8518 ** (-10.06)
[MIDTERM.sub.t] -0.0932 ** (-7.30) -0.0931 ** (-8.78)
[REALINCOME.sub.t] 0.0029 (1.14)
[WATERGATE.sub.t] -0.0113 (-0.28)
[GINGRICH.sub.t] 0.0299 (0.79)
[PRESAPPROVAL.sub.t] 0.0546 ** (6.27) 0.0509 ** (6.21)
[DELTA]ln 0.2821 ** (3.14) 0.2360 ** (2.99)
 [(REPVOTESO).sub.t-1]
[DELTA]ln 0.7955 ** (3.31) 0.8208 ** (3.60)
 [(MFCOST).sub.t-1]
[DELTA]ln -1.4190 * (-2.67) -1.3508 * (-2.65)
 [(SINCSHARE).sub.t-1]
[DELTA][CIVILRIGHTS 0.4630 ** (8.90) 0.4579 ** (9.10)
 .sub.t] or
 [CIVILRIGHTS.sub.t]
Adjusted [R.sup.2] .931 .934
LM(1) 6.84 ** 1.71
LM(2) 7.82 * 2.18
Q(12) 7.75 10.71

Notes: t Statistics are in parentheses. EC terms from VECMs
jointly estimate long- and short-run relationships.

(a) ACIVILRIGHTS, is in Models 1-4, and CIVILRIGHTS,
is in Models 5-8.

*, **, and t denote significance at the 95%, 99%, and 90%
levels, respectively.

TABLE 5
Cointegration Results for Republican Share of U.S. Two-Party
Senate Popular Vote, 1954-2004

 Cointegrating
 Equilibrium
 Relationship
 (Assumption:
 No Trend in
Model and No. of Significant Vector or in
Vector Vectors Variables

No short-run variables present
5 1 ln(SREPVOTE) = 3.906146 **
 (123.96)
 - 0.122310 ln(MFCOST) **
 (2.85)
With MIDTERM and disposable income growth present
6 1 ln(SREPVOTE) = 3.890057 **
 (112.10)
 - 0.091274 ln(MFCOST) ([dagger])
 (-1.95)
With MIDTERM, disposable income growth, WATERGATE, and presidential
approval present
7 1 ln(SREPVOTE) = 3.604044 **
 (23.35)
 - 0.097294 ln (MFCOST) ([dagger])
 (-1.88)
With MIDTERM and presidential approval present
8 1 ln(SREPVOTE) = 3.572279 **
 (24.28)
 - 0.113408 ln(MFCOST) *
 (-2.29)
Unit root test statistics, ADE statistics

 Eigenvalue
 (Trace Statistics,
 Max-Eigen Statistics)

Model and No. of Significant Log Likelihood
Vector Vectors (AIC)

No short-run variables present
5 1 89.76563 (-6.730469)
With MIDTERM and disposable income growth present
6 1 93.77452 (-7.064543)
With MIDETERM, disposable income growth, WATERGATE, and
presidential approval present
7 1 100.3001 (-7.608341)
With MIDTERM and presidential approval present
8 1 95.15140 (-7.179283)
Unit root test statistics, ADF statistics

 Eigenvalue
 (Trace Statistics,
 Max-Eigen Statistics)

Model and No. of Significant
Vector Vectors 0 Vector

No short-run variables present
5 1 0.616312 (26.55036 **,
 22.99023 **)
With MIDTERM and disposable income growth present
6 1 0.528997 (21.90128 *,
 18.06940 *)
With MIDETERM, disposable income growth, WATERGATE, and
presidential approval present
7 1 0.585096 (23.26358 *,
 21.11301 *)
With MIDTERM and presidential approval present
8 1 0.572945 (23.35607 *,
 20.42022 **)
Unit root test statistics, ADF statistics

 Eigenvalue
 (Trace Statistics,
 Max-Eigen Statistics)

Model and No. of Significant
Vector Vectors 1 Vector

No short-run variables present
5 1 0.137861 (3.560122,
 3.560122)
With MIDTERM and disposable income growth present
6 1 (0.147568 (3.831880,
 3.831880)
With MIDETERM, disposable income growth, WATERGATE, and
presidential approval present
7 1 0.085710 (2.150573,
 2.150573)
With MIDTERM and presidential approval present
8 1 0.115141 (2.935848,
 2.935848)
Unit root test statistics, ADF statistics

 Level 5% Critical Level
 (AIC Lag) for Lag

ln(SREPVOTE) -0.526579 (6) -3.02997
ln(SREPVOTE) 7.322867 ** (0) -3.603202
ln(MFCOST) 0.248677 (3) -2.998064
In(MFCOST) -0.729981 (2) -3.622033

 1% Critical Level First Difference
 for Lag (AIC Lag)

ln(SREPVOTE) -3.831511 -9.489163 ** (0)
ln(SREPVOTE) -4.374307 -9.271089 ** (0)
ln(MFCOST) -3.752946 -1.146193 (3)
In(MFCOST) -4.416345 -5.308305 ** (0)

 5 Critical Level 1 Critical Level
 for Lag for Lag

ln(SREPVOTE) -2.991878 -3.737853
ln(SREPVOTE) -3.612199 -4.394309
ln(MFCOST) -3.012363 -3.78803
In(MFCOST) -3.612199 -4.394309

 Assumptions

ln(SREPVOTE) Constant/no trend
ln(SREPVOTE) Constant and trend
ln(MFCOST) Constant/no trend
ln(MFCOST) Constant and trend

Notes: t Statistics are in parentheses except when AIC statistic
is reported. For each vector, a lag length of 1 minimized the AIC
in the cointegration tests and yielded significant and unique vectors.
The modified AIC was used to determine the lag length (up to 10)
in the ADF unit root tests. Data used span 1954-2004.

(a) REPVOTENO denotes Republican share of non-South vote.

* and ** denote significance at the 95% and 99% levels, respectively.

TABLE 6
VECMs of the Percent Change of the Two-Party Popular Senate Vote,
1956-1904

 No Error Correction Term

Variable Model 1 Model 2

Constant -0.0517 ([dagger]) -0.0333 (-1.32)
 (-1.88)
[EC.sub.t-1]
[MIDTERM.sub.t] 0.0601 * (-2.61)
[REALINCOME.sub.t] 0.0004 (0.09)
[WATERGATE.sub.t]
[PRESAPPROVAL..sub.t]
[DELTA]ln -0.5663 ** (-3.26) -0.3979 ([dagger])
 [(SREPVOTE).sub.t-1] (-2.07)
[DELTA]ln
 [(MFCOST).sub.t-1]
[SENATEGROUP.sub.1] 0.1122 ** (2.91) 0.1001 ** (2.91)
[SENATEGROUP2.sub.t] 0.0519 (1.22) 0.0201 (0.51)
Adjusted [R.sup.2] .490 .604
LM(1) 7.37 ** 5.65 *
LM(2) 7.61 * 5.66 ([dagger])
Q(12) 17.47 11.75

 No Error Correction Term

Variable Model 3 Model 4

Constant -0.5537 * (-2.20) -0.5023 ([dagger])
 (-1.98)
[EC.sub.t-1]
[MIDTERM.sub.t] -0.0614 * (-2.76) -0.0746 ** (-3.52)
[REALINCOME.sub.t] -0.0034 (-0.78)
[WATERGATE.sub.t] -0.1283 ([dagger])
 (-1.79)
[PRESAPPROVAL..sub.t] 0.1365 ([dagger]) 0.1237 ([dagger])
 (2.07) (1.86)
[DELTA]ln -0.4157 * (-2.44) -0.4013 * (-2.68)
 [(SREPVOTE).sub.t-1]
[DELTA]ln
 [(MFCOST).sub.t-1]
[SENATEGROUP.sub.1] 0.0749 * (2.23) 0.0727 ([dagger])
 (2.09)
[SENATEGROUP2.sub.t] 0.0258 (0.72) 0.0114 (0.32)
Adjusted [R.sup.2] .690 .667
LM(1) 2.09 4.25 *
LM(2) 6.50 * 4.72 ([dagger])
Q(12) 15.87 15.95

 Error Correction Term

Variable Model 5 Model 6

Constant

[EC.sub.t-1] 1.6481 ** (-5.40) -1.3565 ** (-3.82
[MIDTERM.sub.t] -0.03441 (-1.76)
[REALINCOME.sub.t] 0.0031 (0.83)
[WATERGATE.sub.t]
[PRESAPPROVAL..sub.t]
[DELTA]ln 0.1872 (1.05) 0.1962 (0.97)
 [(SREPVOTE).sub.t-1]
[DELTA]ln -0.5310 (-1.51) -0.5713 (-1.41)
 [(MFCOST).sub.t-1]
[SENATEGROUP.sub.1] 0.0693 ** (3.51) 0.0679 ** (3.44)

[SENATEGROUP2.sub.t] 0.0373 (1.56) 0.0198 (0.80)
Adjusted [R.sup.2] .752 .753
LM(1) 0.09 1.78
LM(2) 0.32 4.27
Q(12) 9.96 11.51

 Error Correction Term

Variable Model 7 Model 8

Constant

[EC.sub.t-1] -1.1521 ** (-3.94) -1.1982 ** (-4.17)
[MIDTERM.sub.t] -0.0374 * (-2.06) -0.0441 * (-2.72)
[REALINCOME.sub.t] 0.0001 (0.02)
[WATERGATE.sub.t] -0.0877 (-1.47)
[PRESAPPROVAL..sub.t] 0.0853 ** (3.47) 0.1023 ** (3.75)
[DELTA]ln 0.0904 (0.49) 0.0689 (0.41)
 [(SREPVOTE).sub.t-1]
[DELTA]ln -0.4124 (-1.15) -0.3999 (-1.13)
 [(MFCOST).sub.t-1]
[SENATEGROUP.sub.1] 0.0604 ([dagger]) 0.0533 ([dagger])
 (2.07) (1.82)
[SENATEGROUP2.sub.t] 0.0283 (0.95) 0.0188 (0.65)
Adjusted [R.sup.2] .789 .790
LM(1) 0.02 0.56
LM(2) 0.38 0.91
Q(12) 7.58 10.55

Notes: t Statistics are in parentheses. EC terms are based on VECMs
that jointly estimate long-run and short-run relationships, where
the long-run relationship involves ln(SREPVOTE) and ln(MFCOST).

*, **, and ([dagger]) denote significance at the 95%, 99%, and 90%
levels, respectively.