A Fundamental Model of US Elections
How Bad Is It?
Liberals stand for higher taxes and a more generous welfare state; conservatives for fiscal discipline and lower taxes. Whatever the truth behind these ideas, they are strongly baked into public perceptions — with important implications for any fundamental model of electoral competition in the United States. Specifically, because Dems are perceived to be less tight-fisted, macroeconomic variables have a clear political valence. In general, we should find periods of stable inflation and strong growth to favor Democratic fortunes; while periods of macroeconomic instability that seem to call for fiscal retrenchment should favor the GOP.
G. Elliott Morris, at the Economist, shared an interesting result in the Checks and Balances newsletter recently. He showed that accounting for political polarization, presidential tenure and economic growth, Biden’s awful poll numbers are exactly where you’d expect them to be. This is especially surprising given that the model seemingly does not include inflation.
Rapid growth in American incomes and plentiful employment opportunities pull Democrat fortunes in one direction; high inflation, and perhaps house price inflation, pull in another. What is the net effect? Put another way, before conditioning on political variables, we should first consider a purely fundamental model of Democratic fortunes. A fundamental model cannot predict electoral outcomes with good confidence. It is blind to things like the class-political confrontations over schools and covid. What it can tell us is merely how macroeconomic fluctuations condition electoral fortunes. In what follows, we make a first pass at building such a model.
We construct a panel dataset as follows. We obtain electoral data on presidential, senatorial and congressional outcomes by state and year from MIT election lab. We obtain macroeconomic data from the St Louis Fed. The features we consider include state-level unemployment rates, state-level house price inflation, headline CPI inflation, expected inflation from the Michigan Survey of Consumers, and real GDP growth. We also obtain stock market excess returns from Kenneth French’s website. We stochastically detrend all variables except the last by first-differencing. We lag all features by a year, so that, in principle at least, voters can observe the macroeconomic variables before casting their ballot. We run panel regressions with change in Democrat vote share as the response. We import PanelOLS from linearmodels. Introducing state fixed-effects was found to leave the results unchanged. In what follows, we report estimates without state fixed-effects.
We first run single-factor regressions for each of our features separately in order to understand the relative significance of the features for Democrat electoral fortunes. We measure relative significance of the features by within R-squared — the explained percentage of variation in change in Democrat vote share within each state. We find that expected inflation is the strongest conditioner of Democrat electoral fortunes in presidential elections, followed by house price inflation and unemployment. We find very small effects of our macroeconomic features on senate elections; while unemployment, inflation and real growth are moderately strong conditioners of Democrat fortunes in congressional elections. In what follows, we restrict attention to the presidential elections since these are more strongly conditioned by macroeconomic fluctuations than Congressional races.
Expected inflation and headline inflation are highly correlated (r=0.86), with the result that if we include both in the model, headline inflation bears the wrong sign (while expected inflation continues to bear the right sign). This suggests that expected inflation matters more than headline inflation in conditioning electoral outcomes.
Our selected presidential model includes state-level unemployment rate, state-level house price inflation, market excess returns and expected inflation. We also test an alternative specification with headline inflation instead of expected inflation. The results are qualitatively similar.
The selected model explains 18 percent of the variation in Democrat vote share in presidential contests within states.
The estimated parameters of the presidential model are of considerable interest. We find that an increase in state-level unemployment by one percentage point reduces Democrat vote share by 2 percent. Most significantly, a percentage point increase in expected inflation reduces Democrat vote share by 1.89 percentage points. Surprisingly, we find that a one percentage point increase in state-level house price inflation reduces Democrat vote share by 0.18 percentage points. This is the opposite of what we would expect if we paid attention to the wealth effect of rising home values. But the evidence shows that house price inflation is treated very much like inflation in general — Democrats are punished for it at the ballot box. This is not true of the stock market — higher stock market returns improve Democrat fortunes. Each time the stock market doubles in value, Democrat vote share increases by 5 percentage points.
Given that the unemployment rate is falling and the stock market doing well on the one hand, and expected inflation and house price inflation rising on the other, what is the net effect on Democrat political fortunes? We find that if a presidential election were to be held today, Democrat vote share would be down 8 percent on average across the fifty states. The average conceals a lot of variation. But the big story is clear: Democrats face very significant macropolitical headwinds.
A map of the predicted decline in Democrat vote share shows that Nevada and Hawaii are especially at risk from macroeconomic headwinds.
Note that this is not a prediction for 2022 or 2024. In fact, the way the model has been built, we need to wait for the macroeconomic observations in 2021 and 2023, for the mid-term reckoning in 2022 and the presidential contest in 2024, respectively. What this exercise amounts to instead is a quantitative estimate of the headwinds faced by Democrats at the time of writing.
We have argued that macroeconomic fluctuations have clear political effects that can in principle be estimated from panel data. We have shown that Democrats face significant headwinds when expected inflation is accelerating and when house prices are rising more rapidly. Meanwhile, Democrats face macropolitical tailwinds when unemployment is falling and stocks are rising. These two sets of forces are pulling in opposite directions as of writing: unemployment is falling but inflation is accelerating; stocks are sky-high, but so are house prices. The net effect for Democrats is negative because the headwinds from rising inflation — both consumer and house price — trump the tailwinds from falling unemployment and rapidly rising stocks.
Again, this fundamental macropolitical model is just the point of departure. Much more important than macroeconomic conditions are specific class-political confrontations. They will be the topic of a future dispatch where we consider the crisis of the Biden presidency.
Postscript. Published the panel dataset on my GitHub.