Trump voters and markets

Note: enjoyed by tens of thousands and >200 likes on social media.

A fantastic inquiry was brought my way from Colin Barr, an editor at the Wall Street Journal, and his journalist team looking at datasets of voters and stock holders.  There are many gaps in our otherwise rich understanding of which types of voters have much benefited from the recent stock market highs.  We generally find ourselves looking at univariate categories of voters (e.g., by gender, by age, etc.) and making broad guesses as to what happened to specific pockets of voters as we experienced this unprecedented Trump Bump (a consistent rally since Election Day nearly 10 months ago!)  This article will only discuss the details of the advanced mathematics involved in processing such a complicated set of data and drive it down to the most granular demographic information.  We leave it to the fine folks at the Wall Street Journal to consider any other germane investment and political context, as well as synopsis.

Exit polling data segments large demographic factors of voters and solicits information about who they voted for.  We can see univariate dimensions, such as by gender, education, etc.  It’s not readily assimilated in a format one can analyze, and we have the same dynamic with the layout of summary information for the population of stock holders.  The best we can hope for in this sort of analysis is to consider the broad joint factors, particularly when the dimensions come to us bi-variately distributed (e.g., after matrix algebra manipulations we’ll show intersections of race and age in this article).

Table 1 – U.S. voter population

Age	W	B	L
18-29	12%	3%	3%
30-49	16%	4%	4%
50-64	30%	5%	4%
65+	13%	1%	1%

The issue with such a table, and this is common in many government data sets, which then attempts to be secondarily re-applied to other analysis, is that we often need to invert the probability distribution matrix so that we can see the conditional probabilities we seek.  In this case, two layered iterations of Bayesian analysis were performed to drive the insights into the demographics of voters, and then -further- define the demographics of actual Trump voters.  The tremendous insight is worth stand-alone sharing here as it is not widely recognized by the public at large.  Having done deep election probability analysis shared widely by the Trump campaign, the results here are not surprising.  Voters that Hillary thought were assuredly in her camp, since those folks voted in high numbers for Obama, had instead could be seen slipping away within our statistical data (e.g., she handily lost the many White female voters who in early election polls stated they might vote for her).

Table 2 – All Trump voters

Age	W	B	L
18-29	13%	1%	2%
30-49	19%	1%	2%
50-64	41%	1%	3%
65+	17%	0%	1%

Clearly there is a bias in the distribution of voters who tended to vote for Trump.  Namely a disproportionate number of older Whites.  And while in general there are less younger minorities who voted, then older Whites, a disproportionally lower number of the former could be found among Trump voters. 

How does this match up to the beneficiaries of the strong record-high stock market run?  In order to analyze this information, consider the trials inherent in stock ownership data only being segmented univariately (e.g., by race, by age, etc.)  But not in the joint distribution that we need to enjoy as per above.  Further, consider that we would need to understand the actual bi-variate distribution, within race and age combinations. 

This requires the missing ingredient, or the introduction of multiple new datasets.  Those that cover this distribution of the population for the differing race categories.  Note that the U.S. Labor Department and the Census Bureau, each account for Blacks of Hispanic origin differently, and hence both datasets are needed to reconcile with each other. 

Through the final round of probability modeling, we see the Bayesian distribution below of the “stock holding” Trump voters, as opposed to Table 2 above which is our solution for the demographic distribution of all Trump voters (i.e., regardless of any stock ownership).

Table 3 – Stock owners, distributed among Trump voters

Age	W	B	L
18-29	11%	0%	1%
30-49	20%	1%	2%
50-64	43%	1%	2%
65+	17%	0%	0%

Whether looking at Trump voters, or looking at stock owners (irrespective of political segmentation), both tend to bias towards older Whites.  To some degree the opposite is true for young minorities and for low-stock proprietors, and hence this demographic of Trump voters didn't of course benefit from his namesake rally as much.  Table 3 above illustrates that the combination however does statistically amplify such a biased effect and leads to a highly curious outcome!  That the beneficiaries of the market rise are slightly weighted more greatly towards the subset of the demography that both own stock and voted for Trump, which we discover in the math here essentially one in the same. 

Last, we'll note that simply having the Dow as one’s “report card” has been an acute political calculus for a reason (see distribution below for Hillary voters), more so then when I ran the TARP analytics team for the prior Administration.

Table 4 – Stock owners, distributed among Hillary voters

Age	W	B	L
18-29	11%	3%	3%
30-49	15%	8%	6%
50-64	25%	9%	5%
65+	12%	2%	1%