| July 06, 2020
RavenPack data is used to analyze sentiment, media attention, and historical voting patterns to predict the U.S. presidential election in 2020.
is said to be able to help forecast the outcome of the November 2020 presidential election, but how does it achieve such a feat?
Behind the monitor is a sophisticated predictive model developed in-house by RavenPack’s Data Scientists.
If the model had been applied to past elections it would have correctly predicted the winning candidate in four out of the five elections since 2000, with an average confidence of 75%.
The one election where it would have failed was the 2000 race between George W Bush and Albert Arnold Gore, but it could be argued that this was an anomaly, since it was one of the closest elections in American history. Ultimately the matter had to be decided in court, therefore making it almost impossible to forecast successfully.
The RavenPack platform derives election insights using three main analytics - one that measures the volume and share of news about different candidates and two that reflect sentiment.
The Media Attention Ratio is based on the relative share of media exposure for different candidates. This has been found to correlate closely with successfully predicting the winner since 2000, as reflected in the graphic below.
The next indicator, the Candidate Sentiment Score (CSS), measures the relative sentiment of both main candidates at a state level. This is based on news stories about non-business events such as those relating to politics, candidate policies, and even their private lives.
“While media attention is important, it does not tell us what the overall sentiment of the
electorate towards the candidate is. It is expected that certain voters, at least the indecisive ones, tend to choose candidates based on their attitude and affection towards them,” says Peter Hafez, the Chief Data Scientist at RavenPack.
The Incumbent Sentiment Score (ISS) provides a proxy for the approval rating for the president currently sitting. It is based on an analysis and aggregation of sentiment derived from socio-economic news events at a state-level. This is based on the insight that the economy is seen as a key variable for determining voter satisfaction with the incumbent.
The RavenPack presidential election forecasting model is based on the principle that a future election result in a given state is going to be similar to the previous election result plus or minus a margin depending on candidate news exposure and sentiment.
For each state, the model runs thousands of simulations based on different news analytics scenarios. It then plots the results as distribution curves with the central tendency providing the forecast estimate. The white paper shows distributions from 300 to 1 days before the election.
“Our model attempts to capture these complex dynamics and translate them into
insights that provide a statistical assessment of the electoral dynamics, and ultimately point to the likely winning candidate,” says Peter Hafez.
The model was overall fairly accurate in predicting future election results in most U.S. states. For a more in-depth exploration of the science behind the monitor´s predictive model, you can access the white paper on the subject here.
Please use your business email. If you don't have one, please email us at email@example.com.
We will process your personal data with the purpose of managing your personal account on
RavenPack and offering our services. You can exercise your rights of access, rectification,
erasure, restriction of processing, data portability and objection by emailing us at firstname.lastname@example.org. For more information, you can
Your request has been recorded and a team member will be in touch soon.
We consider incorporating sentiment signals from news, earnings call transcripts, and insider transactions to
boost the risk-adjusted returns, and revive factor performance.
We find stronger, more predictable market reactions when the words of company executives agree with their actions.
We have gathered 12 insights from 2021 research that can be leveraged in 2022.