Oct 20, 2020

Our probably final updates on U.S. election probabilities

Looking good for Democrats – don't count out the underdogs, though.

  • For the White House, we currently see a 75% probability of a Democratic win, with 25% for the Republican nominee winning.
  • As for combined Congressional and Presidential results, we now see 53% probability for a Democratic sweep.
  • Voter engagement is unusually high this year, introducing plenty of uncertainty – in both directions.
to read

Only two more weeks to go until Election Day.

Indeed, almost 28 million votes have already been cast nationally, corresponding to 20% of the total turnout in 2016.[1] Time for one more update of our U.S. election probabilities. It is probably going to be the final one.

For the White House, we currently see a 75% probability of a Democratic win, with 25% for the Republican nominee winning. As for combined Congressional and Presidential results, the probabilities for a Democratic sweep, a Republican sweep and various forms of divided government are 53%, 7% and 40% respectively.  Of the 40% divided government, roughly 14% corresponds to the status quo, i.e. Trump winning re-election, Democrats holding on to the House and Republicans to the Senate. The remaining 26% largely correspond to a new Biden administration facing at least partial Republican control in Congress, notably a 20% probability of Republicans holding on to the Senate despite a Biden win, while failing to win back the House. (All figures are rounded to the next full percentage point – see below table for details.)

Long-time followers of our forecasts will notice that we are finally taking out the 5% for “Other” scenarios, as well as putting the names of the two contenders in below table in quotation marks. That 5% was initially designed to capture three things. Namely, the shrinking chance of a third-party candidate winning; the risk of one of the two nominees falling sick; or the process descending into chaos, with the next President being picked by some other means than the Electoral College.

Following President Donald Trump’s recent bout with Covid-19, we feel that now is the time to formally remove this set of scenarios from our table. During a global pandemic, further health developments can certainly not be ruled out for either ticket. However, these are difficult to quantify. With voting already under way, it also seems increasingly clear how either party would react if one of the two Septuagenarians at the top of the ticket had to withdraw: most probably, they would be replaced by their running mate.

Taking out the “Other” scenarios also has the advantage of making our probabilities more easily comparable with those of other forecasters, such as FiveThirtyEight and The Economist. Avid readers will notice that our forecasts are showing a more Republican-leaning picture throughout – not just for the Presidency, but also for the Senate and House, where separate forecasts are available. That has one practical and several more philosophical reasons.

The practical reason is that we initially put together our basic forecasting approach more than 1 year ahead of November’s vote, so that its results could feed into our 12-month investment strategy. As a result of this and other methodological choices described in previous publications,[2] we try to mimic what is happening in all 538 Congressional and all 56 Electoral-College contests and showing net results for the overall, national picture. The alternative would be having formal models for each event, which is a lot more labor-intensive and largely meaningless until a few months before Election Day (before that, it is often not clear who the candidates are or how the public might respond to specific candidates). In this, we partly rely more than other forecasters on somewhat heuristic inputs such as the Generic Congressional ballot (see an excellent quality-weighted version) which tend to be lagging indicators in the final stages of U.S. election races. In other words, our forecasts implicitly assume that come Election Day, recent polling trends favorable for Democrats (both nationally and in most swing states) are more likely to partially reverse, rather than accelerate further. How confident are we in this assumption? Well, that takes us to our – slight – philosophical differences with some U.S. forecasters.

As we described in previous publications, there are good reasons to be pretty confident in U.S. polling. The question is how confident. By historical standards, Joe Biden’s lead looks very solid indeed. Perhaps most significantly for forecasting purposes, there are far fewer Electoral-College (EC) contests where independent or third-party candidates play a meaningful role.  That means most of them are likely to be decided by the candidate winning more than 50% of the vote. (That was the case for all 56 EC contests in 2012, as we noted in our January “Roadmap,” where we went on to note: “In 2016, by contrast, 15 contests went to a "winner" of less than 50% of votes within those states or districts. Those contests accounted for 157 electoral votes out of 538, i.e. easily enough for an Electoral-College "landslide" of either Clinton or Trump.”) Given polling for these states for both head-to-head match ups and presidential approval, we agree with other forecasters that Biden looks well placed to be that candidate in more than enough of those swings states.[3]

U.S. elections 2020: Everything will be obvious, once you know the answers

The upcoming U.S. elections could be remembered as the most important since 1980.
Open the special

Set against this, one characteristic perhaps underappreciated by many U.S. observers is that since 1970, turn-out has typically been in the low to mid 50%. As one astute observer put it, almost 40 years ago: “By 1980 election, a mere half of the electorate voted in the presidential election and Ronald Reagan’s “landslide” was effected by little more than a quarter of the voting-age population.”[4] The same could be said of most subsequent winners, and not just those who owed victory to the vagaries of the Electoral College (like George W. Bush in 2000 and Trump in 2016). Take Barack Obama’s dramatic 2008 win in both the popular vote and the Electoral College in the highest turnout election since 1968, when Republican Richard M. Nixon defeated Democrat Hubert Humphrey. That still corresponded to only 1 in 3 of potentially eligible voters actually casting a ballot for the Obama-Biden ticket.[5]

Such low turn-out rates are not necessarily problematic for forecasting purposes, as long as the participation patterns of different segments of the electorate are fairly stable over time. For 2020, however, that is not necessarily a safe assumption. 2020 is likely to see a sharp revival in turn-out, with various measures of voter engagement long suggesting an unusually high level of interest.[6] Figuring out who these new or previously disengaged voters might opt for is methodologically hard. It is also a fairly unfamiliar challenge to U.S. pollsters, who typically rely on past voting behavior in constructing their likely-voter models.

When facing voters for the first time after their initial electoral breakthrough, the presence of new populist movements on the ballot tends to boost voter engagement.

That so many U.S. voters are engaged is in itself a consequence of the Trump presidency familiar from populist election victories in other countries. When facing voters for the first time after their initial electoral breakthrough, the presence of new populist movements on the ballot tends to boost voter engagement among both those in favor and those against the new movement. In our experience, this dynamic typically gives rise to familiar patterns, when comparing polling to actual results. In one, which is especially common in countries where populists surprisingly won power with mixed results, turnout goes up with previously disengaged voters wanting to register their opposition to the populist government and often by more than pollsters forecasted. Such polling errors can result from a number of factors. For example, youth turn-out tends can be quite low over long periods of time but suddenly spike if a candidate or issue moves center-stage that young voters care about. (The 2019 elections to the European Parliament, with its focus on climate change, in many countries provided a nice illustration.) How much of this effect is or should be captured in polls is notoriously hard to say in advance. To take the extreme case of young voters only just reaching voting age, pollsters tend to rely on the voting behavior of previous age cohorts to forecast participation. That will tend to understate turn-out among young voters, until pollsters have been wrong on a sufficient number of occasions to adjust their assumptions (and at some point probably begin to overstate turnout for the next cohort).

Another related dynamic, though, is that disengaged voters similar to a populist initial base, only come out in force in subsequent cycles, and it usually takes pollsters a while to catch on, just as it did in above example of youth voter turn-out. This would correspond to the Trump campaign discovering and mobilizing new pockets of disengaged, anti-establishment Trump fans, perhaps in segments of the population that for other reasons tend to be underweighted or under-sampled by enough pollsters to meaningfully distort polling averages.  Other sources of bias can also play a role, such as some groups being unusually reluctant to admit voting intentions to pollsters, one's neighbor or even oneself. The size of any such effect is obviously difficult to forecast, though relatively easy to assess retrospectively by assessing in which geographic areas a candidate outperformed. For example, it is pretty clear that Trump did not owe his 2016 victory to “shy” Trump voters.[7] However, it is quite possible that you could see shy Biden voters in rural and small-town America in 2020.    

All of which is to say that there is plenty of uncertainty in both directions, and probably more so in 2020 than models relying on past patterns of the last 50 years or so would suggest. Given the campaigns both sides have run so far, and for example, the seeming inclination of Republicans to deliberately make voting hard for low-propensity voters, we have leaned towards scenarios of previously disengaged voters wanting to register their opposition to the Trump presidency since the start of the campaign.[8] And perhaps, they will do so by even more than pollsters are forecasting. Set against this, however, Trump’s team has had four years to prepare for 2020 and try to change the electorate in ways that is conducive to its re-election efforts. Moreover, many things campaigns do only become clear with the benefit of hindsight, when campaign insiders explain what they were up to. And given their 2016 success, a 1 in 4 chance that Team Trump actually have some sort of a strategy not quite obvious from the outside seems about right. Keep in mind, after all, that the base rate of either party’s ticket winning a U.S. presidential election is about 50%.

Election-outcome probabilities (DWS Expectations)



Source: DWS Investment GmbH as of 10/16/20

Our 2020 U.S. elections CIO Special

Download the full 2020 U.S. elections CIO Special here

Get started

CIO View

You are now seeing the Sweden version of the page despite being located in USA. You can change the country below.

Other country

Other country