Fundamental Model Predicts Electoral College Tie

On February 16, 2012 Patrick Hummel and I released our fundamental model for the 2012 presidential election. A fundamental model does not include polling or market-data, but focuses on core indicators: past voting, presidential approval, incumbency, economic growth, and senatorial voting record. The full model is detailed in this academic paper (and outline in more depth below). The 2012 forecast was excellent: 50 of 51 states binary correct (President Obama was 35% to win Florida, which he won with 50.01% of the vote!) with a median absolute error of 2.1 percentage points (mean of 2.8).

Screenshots from 2012 Fundamental Model Write-up

On February 28, 2016, we released the same fundamental model and it also did very well. While it had just 46 of 51 states binary correct, it had 292 electoral votes going to President Trump (he got 304). The error was even smaller than 2012: just 1.8 median absolute percentage points off (2.7 mean). The model had Pennsylvania flipping to Trump but missed Wisconsin and Michigan. Of course, we did not put too much stock in this model: fundamental models are about the generic Democratic and Republican candidate and we assumed that 2016 would veer further from that generic match-up than normal. Of course, we were wrong, the fundamental model held!

2020 Fundamental Model

Today, February 27, 2020, we release our fundamental Electoral College model for 2020. I want to emphasize that I ran the data without knowing what would happen: it comes out as a 269 to 269 tie. I know, horrifying. There are just seven serious swing states. Three states leaning Democratic: Nevada, Michigan, and Pennsylvania. And there are four states leaning Republican: Wisconsin, Florida, North Carolina, and Arizona. Easy prediction: it is going to be a long 8 months and 7 days.

More details on the model broken up by variables that are national, vary by both state and national, and state-specific:

National Variables that do not vary by state.

1) Presidential Approval: is not very good for an incumbent, the Republican, with a decent economy. I averaged the RealClearPolitics and FiveThiryEight averages. RealClearPolitics is a simple average, but opaque on what it includes and tends to favor right-wing polls, and FiveThirtyEight includes everything but over engineers it’s smoothing. So neither is ideal, but they are the publicly available choices.

2) Incumbency: is a binary term that is turned on if a party has held the office for eight or more years, so it is 0 this cycle.

Mixed Variables that vary on both the national and state level.

3) Income: is both national and state-by-state. Technically the variable is changes in state-level income over the last few quarters. But, this term has a lot of national-level variation that drives most of the movement (i.e., state income in any given year is highly correlated with national movement in state income). This variable is relatively very strong for the Republicans this year.

As I write this the stock market is plunging, which could be a bad sign for President Trump. We note in the paper that the ideal economic indicators is a shift in personal income from the 1st quarter of 2019 through the 1st quarter of 2020 (i.e., the 9th to 13th quarter of the cycle). People remember movement, not levels, and fix their idea of the economy in the late spring of the election year.

State-level Variables.

4) Past Election: includes the last two presidential elections. Obviously, President Trump in 2016 reversed much of President Obama’s strength in 2012.

5) Changes in State Legislature: is the change in Democratic representation in the lower house of the state legislature in the last election and was relatively good for the Democrats this cycle. This picks up what happened since the last election.