In February, I introduced you to my most simplistic forecast of the American presidential election, the only one of my forecasts that can run prior to the election year itself. Here is a link to the previous installment.
Prior to election year, the only predictor I can rely on is the 16-year cycle that has held steadily, with very few exceptions, in each presidential election starting in 1944 (arguably as early as 1904): 2 elections which Democrats win, then 2 elections which Republicans win. Since 1944, only 1980 and 2020 were exceptions; and one of my earlier articles provided statistical evidence to suggest very strongly that the 2020 election was not honest.
I mentioned in the first installment that the 16-year cycle is unforgiving, and cares not whether a party won or lost unexpectedly; the cycle grinds on. Thus, while 2020 was likely to have seen a Republican win the White House, 2024 is more likely to see a Democrat win. Both years had a forecast of a 13.2% margin of victory in the popular vote.
Recall that my models forecast only the national popular vote, not the electoral vote which is what determines the winner. The popular vote and the electoral vote are positively correlated, of course, with Republicans holding a slight electoral college advantage.
The first predictive model had an unacceptably large margin of error: 16.5%. We need more data to make a more precise prediction, and early in the election year itself, we get that data. It is an economic data point.
The best predictor I have found for the presidential election is not the unemployment rate or the inflation rate, but rather the annual percent change in real disposable personal income per capita. Personal income is, of course, the portion of GDP paid to individuals; disposable personal income adds transfer income and subtracts taxes; real means adjusted for inflation (I use the Consumer Price Index); per capita means adjusted for population size.
I found that the best time to use this statistic is its value in the last quarter of the year prior to the election year. That is, on average, voters are swayed by the value of this number in the 4th quarter of 2023 compared to the 4th quarter of 2022 (and before that, 2019Q4 compared to 2018Q4, and so on). This is when the average voter makes a commitment. Other voters will decide earlier; still other voters will decide later.
The most recent value of this statistic is 3.07%, larger than the average of 1.93% in the elections from 1952 to 2016. (Recall that I explained in the previous installment why I would not use 2020 as part of my “training” data set, due to the dishonesty of the outcome.)
This larger value for the economic statistic bodes well for the incumbent party in the White House (Democrats, in this case). Combining the economic statistic with the 16-year cycle, the Democrats’ predicted margin of victory in the national popular vote for president rises to 17.9%. Importantly, the margin of error falls substantially from the previous model, to plus or minus 9.5%, making this model more precise. R-squared, the predictive power of the model, increases substantially to 81.84%.
I have a third predictive variable that, in conjunction with the first two, enables me to lower the margin of error to less than 9%, raising R-squared over 85% as well. This variable is based on the performance of the nominees in the primary elections, comparing the share of the primary votes in the two major parties. We will not know the value of this variable until the end of primary season this summer. So far, it appears that Biden has taken a larger share of Democrat primary voters than Trump has among Republican primary voters. This is an ominous sign for Republicans.
There are multiple dynamics that could invalidate my predictive models. Never before in American history, as far as I know, have we had an administration deliberately erasing the country’s borders, flooding the country with millions of illegal aliens whose presence endangers, more than anyone, voters of the president’s own party. The resentment this has stirred among those voters could well undermine that president’s base, especially if the president grants amnesty to 32 million illegal aliens and gives them voting rights.
The other dynamic is one of the strongest third party challenges America has seen in a century. Kennedy’s support may well explode upward after the presidential debates. He should not be underestimated by either party.
We shall revisit the predictive models after the primary elections conclude.
Hi Surak, I must question your methodology. You admittedly seek an approach that bears itself out with historical data. Okay, sounds rational. However, the problem is the count of approaches possible. For example, if we want to look for approaches that properly predict the last 25 elections, knowing all too well that winners are going to go back and forth and have limited repeats, then we may only need to look at a hundred approaches before we find one that does a good job. As such, I am skeptical of your ability to use your approach to make valuable predictions. But it is fun!