green data lines on screen
November 3, 2016
Tyler Durden
  • Polls have tightened in the US presidential election, in Goldman Sachs', view due to a combination of recent campaign developments and mean reversion following the presidential debates and other events over the last month. However, even after adjusting Sec. Clinton’s lead in state polls for her diminished margin nationally, she still appears likely to win more than the 270 electoral votes needed to clinch the White House.
  • Early voting patterns in Colorado, Florida, Nevada, and North Carolina suggest that turnout among independent voters is likely to exceed the share of likely voters assumed in many polls; the Republican and Democratic shares both appear on track to be slightly lower than expected, with a slightly greater deficit for Republicans than Democrats overall.
  • Prediction markets now view the election to be more competitive as well, and the moves seen over the last few days appear to be roughly proportional with the change in public opinion polls. Nevertheless, Sec. Clinton is still seen as twice as likely to win the White House next week as Mr. Trump.

Goldman Sachs' Alec Phillips answers the four biggest questions they are getting from clients:

Q: What have we seen in the polls?

Sec. Clinton’s lead in national tracking polls has declined by about 2pp since late last week. We follow five tracking polls, conducted by the Washington Post/ABC, the LA Times, Investor’s Business Daily, Rasmussen, and Lucid/Times-Picayune. In three of these polls, Sec. Clinton’s lead declined by 2pp between polls released Friday, October 28 (with a sample period ending prior to October 28) and polls released Tuesday November 1 (with a sample period covering mostly or entirely the period after October 28). The effect in key states appears similar; comparing swing-state polls by pollsters that conducted polls last week and again over the last couple of days, the median decline in Sec. Clinton’s margin was 1pp.

Sec. Clinton’s diminished lead appears to be due more to an increase in support for Mr. Trump than a decline in support for her (Exhibit 1). This trend appears to have started more than a week ago, predating the most recent campaign developments—namely, the release on October 28 of FBI Director Comey’s letter to Congress—and is likely due to the fading effect of prior developments that had been negative for Mr. Trump as well as declining support for Libertarian candidate Gary Johnson. Put differently, Sec. Clinton has led by an average margin of 4pp in national four-way polling (including the two most popular third-party candidates) since May; this has ranged from a deficit of 1pp shortly after the Republican convention in late July to a lead of 9pp in August shortly after the Democratic convention. Some of the tightening in the race over the last week or so is probably also simply reversion to the mean.

Q: So where does this put the state of the presidential race?

Sec. Clinton’s national lead in the one-week average of polling as of last Friday stood at around 4.5pp, suggesting that, all else equal, her margin over Mr. Trump in national polls might have diminished to around 2pp to 3pp as a result of developments over the last few days. Of course, what matters more is the outcome in the key swing states that will determine which candidate reaches 270 votes in the Electoral College. The left panel of Exhibit 2 shows the Democratic margin in each state, organized by electoral votes (horizontal axis) and current polling margin in the state (vertical axis). The right panel shows the one-week average of national polls and the one-week average of polls in whichever state would, based on current polling, provide the 270th marginal electoral vote. For states where no polling has been released, we extrapolate the state trend using the most recent state polling and the change in national polls since state polling was conducted. The upshot is that the race has tightened somewhat on a national basis and at least as much in the marginal states, but that Sec. Clinton retains an advantage in the Electoral College.
That said, this tightening has reopened a potential path to victory for Mr. Trump, albeit a narrow one. As a starting point, note that the reliably Democratic states that Sec. Clinton currently leads by at least mid-single digits are worth 263 electoral votes (EVs; this includes all of the “safe” Democratic states as well as the states to the left of and including New Hampshire in the left panel of Exhibit 2). To reach 270, she must add to this North Carolina (15 EVs, Clinton leads by around 2pp), Colorado (9 EVs, Clinton leads by 1-2pp); and/or Nevada and the 2nd Congressional District of Maine (6 EVs plus 1 EV, Clinton leads by around 1pp in both).

By contrast, Mr. Trump has a solid lead in states worth 186 EVs (this includes the states to the right of and including Iowa in the left panel of Exhibit 2); to reach 270 he would need to add to this Arizona (11 EVs, Trump leads by around 1pp), Florida (29 EVs, Trump leads by around 1pp), Ohio (18 EVs, Trump leads by around 2pp), plus Colorado, Nevada, and North Carolina, or some other combination that scores electoral votes even deeper into Democratic territory.

This is still a significant polling deficit for Mr. Trump to overcome in a large number of states, particularly in contrast to Sec. Clinton, who simply needs to win one of the very competitive states she currently leads (Colorado, Nevada, or North Carolina) and maintain her sizable lead in the others to win.

Q: How does early voting compare to what the polls have been expecting?

The accuracy of the polls will hinge on the estimates of likely voters that those polls are based on. Most pollsters use 2012 turnout patterns as a starting point, adjusting their estimate based on demographic factors and the answers from poll respondents to certain questions about likelihood of voting and voting history. However, now that early voting has started in most states, we have another cross-check. To gauge whether the early vote matches what pollsters are expecting, we first gather early voting and mail ballot data for each state from 2016 and 2012 to find the difference in the share of early voting by party. Most states make this available at least by party registration, and some provide other demographic detail (e.g. race, gender, age, etc). We then calculate the share of likely voters by party in current polls minus the share in 2012 exit polls. Exhibit 3 shows the difference between these two figures, with positive numbers suggesting a group has seen greater participation (based on early voting) than what polls assume (based on the difference from 2012 exit poll shares) and negative numbers suggesting underperformance.
Source: Colorado Sec. of State, Florida Sec. of State, North Carolina State Board of Elections, Nevada Secretary of State, New York Times, Real Clear Politics, NBC-WSJ-Marist Poll, Quinnipiac Poll, Emerson Poll, Bloomberg, CBS-YouGov Poll, Goldman Sachs Global Investment Research

The most important takeaway from this analysis is that there has been a surprising increase in turnout among voters not affiliated with either major political party. This has mixed implications. Unaffiliated voters are younger and are more often Hispanic or Asian than the broader electorate, suggesting they might lean more Democratic as a group. That said, polling of unaffiliated voters in these states suggests that they lean slightly toward Mr. Trump on net.

The second implication is that Republican voters appear to be trailing expectations slightly more than Democratic voters in the states we analyzed. We note that states report early voting based on party registration, whereas polling is typically conducted based on self-reported party identification, and statewide shares based on these two measures often differ by several percentage points. However, since we are comparing the change rather than the level, we do not believe this is a critical distinction.

Q: What are prediction markets saying?

The probability implied by online prediction markets that Sec. Clinton wins the White House has declined over the last several days, from between 80% and 90% in most markets over the last couple of weeks, to between 64% and 75% as of November 1. Exhibit 4 puts this move in historical context, using the Iowa Electronic Markets (IEM), which have been operating since 1992. The left panel shows the implied probability that Mr. Trump or Sec. Clinton wins the White House against the range of previous election winners; the right panel compares where Sec. Clinton stands today to the perceived outlook for other candidates in recent elections. For most of the year, Sec. Clinton has been seen as more likely to win the election than any recent candidate at the same point in the race, but the election is now seen to be roughly as uncertain as the average election since 1992.
On its face, the change in probabilities implied by the prediction markets might seem to be an overreaction to the most recent campaign developments—namely, the letter regarding from FBI Director Comey to Congress on October 28—but it is likely that recent events simply refocused attention on the already-tightening polls. If so, the change in perception regarding the likely outcome appears roughly warranted by the change in the polls, as shown in Exhibit 5. That said, Sec. Clinton is still seen by prediction markets to be twice as likely to win the election as Mr. Trump. Based on the relationship between polling and implied probabilities we have seen to date, Mr. Trump would need to lead the average of national polls by a couple of points over the next few days before markets would view him to be as likely to win the White House as Sec. Clinton.