Forecasting the American presidential election

Part one: before the presidential election year

We will construct our predictive model for the American presidential election gradually, step by step, corresponding to information that becomes available at various points prior to the election. We will only forecast the national popular vote, and we will not use polls.

In this first installment, we consider only information available before the presidential election year. I have discovered one pattern that has held in most presidential elections from 1904 to the present. That pattern is the 16-year cycle. Beginning in 1904, with very few exceptions, Republicans won the White House for 2 terms, followed by Democrats winning the White House for 2 terms.

Let’s examine the record. The following chart lists all the presidential elections from 1904 to the present, with the predicted result based on the 16-year cycle, and the actual election results.

Year Prediction Actual

1904 R R

1908 R R

1912 D D

1916 D D

1920 R R

1924 R R

1928 D R

1932 D D

1936 R D

1940 R D

1944 D D

1948 D D

1952 R R

1956 R R

1960 D D

1964 D D

1968 R R

1972 R R

1976 D D

1980 D R

1984 R R

1988 R R

1992 D D

1996 D D

2000 R R

2004 R R

2008 D D

2012 D D

2016 R R

2020 R ?

2024 D ?

The 16-year cycle accurately predicted the election results in each case except 1928, 1936, 1940, and 1980.

There is an important feature of the 16-year cycle; it is relentlessly unforgiving. A party losing an election unexpectedly does not get to make up its loss next time, or ever. The 16-year cycle grinds forward, uncaring. For example, when President Carter unexpectedly lost re-election in 1980, in violation of the 16-year cycle, it was followed by two more Republican victories.

For reasons I have stated previously, I believe that Donald Trump won the 2020 presidential election. You can see that column at the following link.

How I know the 2020 presidential election was stolen

My full predictive model for the national popular vote in the US presidential election is trained on data from 1952-2016 alone, because I don’t have access to the other components of my model prior to 1952. I also don’t use the 2020 results, which appear to be fraudulent, based on the Substack linked immediately above.

The goal of all versions of my predictive model is to determine the margin of victory in the national popular vote in favor of the party currently incumbent in the White House. If that margin is positive, that party is re-elected to the White House, and if the margin is negative, that party loses the White House. (My model overlooks the disparity between the popular vote and the electoral vote, a disparity that slightly favors the Republican party.)

I found that when victory for the party incumbent in the White House affirms the 16-year cycle, its average margin of victory is 13.2%, but when victory for the party incumbent in the White House is inconsistent with the 16-year cycle, its average margin of victory is -3.0% (thus it is expected to lose).

Based on this model alone, President Trump could have expected a 13.2% margin of victory in 2020; but for the same reason, Democrats can expect a 13.2% margin of victory in 2024.

The weak side of this model is its huge margin of error, plus or minus 16.5%. (For those who care, R-squared is 41.24%: better than nothing, but not great.) The enormous margin of error, larger than the expected margin of victory, makes the model’s forecasts very imprecise. We need more information, available during the election year, to build a better forecast. We will do that in the next installment.