Trading the US Election – Profiting from “Known Unknowns”
You’ve probably noticed that there’s a US election on the horizon. This is an event of known uncertainty: a “known unknown” in the now immortal language of Donald Rumsfeld.
In trading, we sometimes observe marginal pricing inefficiencies around these “known unknowns”.
For example, ahead of stock earnings announcements or significant economic or policy announcements, we tend to find:
- more significant trend effects (auto-correlations in asset returns)
- enhanced risk premia (the returns to holding risk positions tend to be higher, on average, perhaps as a premium for taking extra risk)
- implied volatility tends to become expensive (post hoc vs subsequent realized volatility.)
What does this mean for stocks ahead of the coming US Election?
In our new Robot Wealth Research Lab – one of our members, Ben, has analyzed stock index return patterns ahead of an election.
With the limited data available, he finds evidence of significant excess returns to holding the SPX index for the 5 days before the US Election Day. These excess returns tend to reverse in the 10 days following the election day.
- US election dates scraped from Wikipedia
- SPX index data pulled from Bloomberg
And some wrangling in Pandas, Ben creates the dataset below:
returnsis daily simple returns calculated on the close column
T2Lis the number of days to the closest election day. –9 means 9 days before an election, 0 means day of the election, and 9 means 9 days after the election.
typeindicates whether the closest election is a midterm (M) or general election (G)
winnerindicates whether Democrats (D) or Republicans (R) ultimately won the election.
In doing analysis, we like to start with The Simplest Thing.
- takes the whole data set
- filters for -9 < Days to Election < +9
- groups by the Days to Election (T2L) column
- calculates the mean (arithmetic) returns for that day, over the whole data set.
We see mean returns being higher in the days before the election than after.
This hints at a conditional risk premium effect, in which the holder of stocks receives higher rewards, on average, when uncertainty is highest. (Uncertainty is resolved by the announcement of the winner of the election.)
Next, we might ask the following questions:
- How consistent has this effect been over time? (We’d be less interested in this if all the outperformance was prior to 2000)
- To what extent is the excess returns ahead of election accompanied by excess variance that would reduce our compounded returns?
To start answering those questions, we show the cumulative returns of a simple strategy that “buys” the index a number of days before each election date and sells it on or after the election date.
D-5 to D-0
We simulate buying the index 5 days ahead of the election and selling at the close on election day.
D-5 to D+1
This simulates buying the index 5 days ahead of the election and selling at the close the day after election day.
D-5 to D+3
This simulates buying the index 5 days ahead of the election and selling at the close 3 days after election day.
We find the effect to be positive and robust to how we choose the days to trade.
On average we see positive expected returns of over 0.2% per day – which is significantly higher than the equity risk premium.
Is the Effect Similar for Midterm Elections and General Elections?
Seems to be.
Does it Look Different if we condition (post hoc) on the election result?
What’s the Trade Here?
We have limited data here (elections just don’t come along that often) but the limited data we have points to a significant edge. This may be understood as a conditional risk premium for holding risky stocks under conditions of significant outcome uncertainty.
The trade expression is simple:
- From 5 days before the election (i.e. now), overweight your exposure to the SPX index (via ES, MES, SPY etc.)
- Remove the overweight a day after the election.
These trades are not things you can build a business around because they don’t come along very often. But they are useful trades to add to your portfolio. Small edges add up.
But It’s Different This Time Because…
Yeah, maybe. But the nature of statistical trading is that we grind out small edges, on average. The law of large numbers is our friend. We trade multiple blunt edges and we allow our edge to play out over time. There is huge uncertainty associated with any one bet. Trading edges can only be analyzed in aggregate over large data samples.
Other People Have Looked At This, Right?
In Stock Market Volatility around National Elections, Jedrzej Bialkowski , Katrin Gottschalk and Tomasz Piotr Wisniewsk show evidence of a similar drift in the 5 days ahead of US election days.
Trading uncertainty around “known unknowns” can be a good place to go hunting for edges. The lead up to an election is one such situation, where one can make the argument for elevated risk premia as compensation for the heightened uncertainty.
The limited data that we do have suggests a significant, if infrequent edge. This is backed up by the work of others who’ve looked at the same effect.