My Final 2016 Poll Analysis
November 04, 2016
Since the 2000 election, I have been tracking how the presidential election poll numbers follow in the footsteps of the stock market, with the key insight that we can gauge how the popular vote is likely to turn out based on how the stock market was moving the week before the election. It is on that basis that I am now predicting that a Trump victory is likely next week.
Understand that this prediction is only for the popular vote, which is what the poll numbers can model. The math of electoral college totals is a whole separate question, and not factored into this analysis.
This election is much more complicated than past ones, and for a variety of reasons. When I started this effort back in 2000, Gallup and Zogby were the two big polling firms, and hardly anyone else was involved. Those two firms have since left that business, and now Real Clear Politics (RCP) calculates their polling averages using as many as 24 different polling firms.
In theory, having more polls in the average makes for a better outcome, because it means that the sample size overall is larger. In reality, though, there are a variety of methodologies employed by the varying firms, and so the RCP average can have a shifting mix of methodologies from one day to the next.
The chart above shows the RCP average in 2-way race polls, where the poll respondents are only given the choice of Clinton or Trump. It has been tracking the movements of the DJIA pretty well going all the way back to the beginning of RCP’s efforts to pull together this average. It is important to note that I have to use a 10 trading day lag to get the two plots to align and to show that correlation.
In years past, I used a 5 trading day lag. There is an inherent lag in poll reporting, because the surveyors have to contact the respondents, tabulate their results, do statistical adjustments in case the sample is overly weighted in one or more demographic ways, and then publish the results. All of those tasks take time. Some polls even use responses gathered over a rolling 3-day lookback period, adding to the lag. And then finally, RCP averages together polls published over the past 2 weeks, which adds still another lag in the poll response to changing public moods.
My hypothesis is that public mood changes much more in real time with changes in the stock market. If people are in a bad mood, they tend to dislike the party in power, and they also tend to sell stocks. So the movements of the DJIA can be a reflection of public mood which then shows up after a lag period in the poll numbers.
RCP also does a separate average involving 4-way race polls, with Gary Johnson (Libertarian) and Dr. Jill Stein (Green) added to the poll questions. Those two have seen their share of the poll responses declining as the election gets closer, and the Clinton-Trump differential is not appreciably different between the 2-way and 4-way averages:
You may notice in both of these charts that the time period around both major parties’ nominating conventions does not fit the DJIA’s pattern. This is normal, and the nonstop press coverage of each party during its convention tends to benefit that party for a short time in the poll numbers. Afterward, the poll numbers resume tracking the path of the DJIA, more or less. In the 2-way race, the spread is now down to just 1.3 percentage points.
There was an additional anomaly this year, starting in early October when a videotape was released showing Mr. Trump making offensive comments. The effect of that news appears to be decaying in its importance with the passage of time, and the poll differential is now more in his favor than it was before that release.
Also interesting this year is the way that the vacillations in the poll numbers have matched the pattern from 2000, albeit with the parties reversed:
The absolute values and the magnitudes do not match up, but the timing of the changes in direction do match up nicely. Remember that in 2000, George W. Bush was up by as much as 13 percentage points as late as Oct. 26, 2000, and was still up by 6 points on the Thursday before election day, and yet the numbers moved against him to a dead heat in the actual vote. We could see the same sort of movement this time. I was able to successfully predict that outcome in 2000, because I had noted how the stock market rallied strongly during the week before the election, something which had not yet shown up in the poll numbers. I wrote in our twice monthly newsletter then that the election was going to be “too close to call,” and indeed for 6 weeks it was.
With as many as 24 different polling firms in the 2016 RCP averages, there are understandably going to be some differences in the quality and accuracy of each firms’ results. Analyzing all of them is beyond my ability, but I have focused on 2 individual polling firms this year. Rasmussen has been around through several election cycles, and has developed a reputation for statistical rigor. They also offer daily tracking polls starting in late September, giving us more data to look at:
The lag time here is shorter than for the RCP average, since it does not look back as far into the past. An 8 trading day lag seems to give the best fit. The latest value from that poll shows a 3 point lead for Mr. Trump, and the continued decline by the DJIA this week suggests that this lead could widen over the remaining days until the election.
A new entrant to the polling game this year has been the LA Times/USC poll. It uses a different and unique methodology, weighting responses according to how people say that they voted in the 2012 election. That has brought it some criticism, and so has the fact that it has more consistently favored Mr. Trump than other polls have.
Its latest reading shows a 5-point lead for Mr. Trump, which is about the same as before the Oct. 7 videotape release. What I find interesting about this poll is that it seems to lag the movements of the DJIA by only about 3 trading days.
Regardless of which poll or average one prefers, the point remains that the poll numbers tend to follow the movements of the DJIA. And the meaningful decline in stock prices this week is bad news for Mrs. Clinton’s election prospects. My best guess, based on the totality of all the polling evidence, is a 3-4 point margin of victory for Mr. Trump in the popular vote. How the electoral college turns out is, of course, a much different matter.
Subscribe to Tom McClellan’s free weekly Chart In Focus email.
|Oct 20, 2016
The Post-Election Stock Market
|Sep 08, 2016
DJIA Leads Poll Numbers
|Oct 19, 2012
Market and Polling Not Following the Typical Pattern