Midterms 2018: Lessons for political risk
The US midterm elections yesterday are the biggest story of the week. Here are some quick thoughts about what they can teach us for political risk and its practice.
Don’t be afraid to outsource analysis
From an analytical perspective, there were a lot of winners last night. As I write this, FiveThirtyEight projects that Democrats will pick up 34 seats.
Pre-election forecasts were as follows:
- FiveThirtyEight deluxe model: 36 seats
- CNN: 32 seats
- Cook Political Report: 30–40 seats
- The Economist: 35 seats
- PredictIt: 35 seats (midpoint of the two highest performing markets)
Forecasting US elections down to the seat level is an area where the technical skills needed are relatively high (constructing models requires historical data, knowledge of statistics, access to current polls, and lots of time for testing) and where skilled people make their results publicly available.
Every political risk team faces resource constraints. A major part of analyzing one’s own political risk is knowing when not to reinvent the wheel. Sometimes using existing forecasts allows a team to redeploy energy to research with more value-add.
One of the most bedeviling challenges in political risk is forecasting one-off events. How do we know what a 60% chance of winning an election is if the results are either a win or a loss?
The midterm elections offered enough results to put probabilities to the test.
Called races (as of noon on Wednesday) and victories by favored party:
- Solid: 100%
- Likely: 95%
- Lean: 93%
- Tossup: 63% Dem
There are still 12 Democratic-favored and toss-up races still to be called, but the distribution shows how probabilities should work. There will be upsets in individual races and the sample size may be small, but, by categorizing by probability, we should not be overly surprised by the frequency of upsets.
In political risk, many events are not repeated. But political risk teams will be making numerous predictions. Using probability in predictions allows you to see if your 10% predictions come true 50% of the time. If so, you know that it’s time to recalibrate your analysis.
Beware the context-less trend piece
Some of the immediate reactions have fallen into the category of “not a blue wave, but more of a splash.” Democrats taking the House and losing seats in the Senate has been used to illustrate that the results of the election was a split decision. This is typical of too much political writing, focusing on results as indicative of larger phenomena without including crucial context.
Democrats faced one of the most difficult Senate maps in modern US political history. Because of the way the 6-year terms rotate, this is partly random, and partly because these seats were last elected in 2012 when Democrats performed well. The combined effect was that Democrats were defending 26 seats to Republicans’ 9. Ten of the Democratic seats were in states that Trump won in 2016. They were favored by most forecasters to lose seats to Republicans and, in another year, might have lost plenty.
Now, we can certainly discuss whether Democrats should have done better in some races and what this means for Democrats long-term that the Presidential and Senate maps appear to be merging. But stories based on which party outperforms expectations must always begin with a reiteration of those expectations. Otherwise, you’re extrapolating a trend that might be going in the opposite direction.
Prioritize what matters
There were a lot of races happening last night. 435 Representatives. 35 Senators. 36 governors. Hundreds of state legislators.
But from a political risk perspective, there was only one question.
Would Democrats win one house of Congress?
All other elections, while certainly important, paled in comparison to that one question. Control of the House decides whether the Republican agenda can pass in 2019 and 2020 (not legislatively, at least) and who will have subpoena power in Congress (those determined to investigate the Executive Branch).
Often, finding the right answer in political risk depends on asking the right question. Prioritizing which questions will matter most is the first and most important step to proper political risk analysis.