Trump Mood Index (TMI) : Anticipate bold declarations that could move the market

Published by newspill Team | 2019-07-24

        For the past few months at Sysmo, we’ve been seeking validation of our technology from the Market Finance Ecosystem: Hedge Funds, Banks, Asset Managers, Private Bankers, Market Makers, Brokers etc.
As we went, we started noticing a recurrent topic of conversation that was both source of anxiety and inspiration for our interlocutors: Donald Jr. Trump, sitting president of the United States. His declarations are known to make a lot of noise in the Media and most traders are fearing or expecting his next moves.

His weapon: Twitter.
His secret ingredient: Unpredictability.

One might say that every human is in some way predictable once you’ve collected enough data. The thing is, we have a lot of data on Trump thanks to his heavy tweeting habit.
What if we could predict Trump’s mood and therefore anticipate bold declarations that could move the market?

Method :

Building a Trump Mood Predictor is a two-step process:

  • Step n°1 : Defining a relevant scale to describe the mood of Donald Trump

       We decided to build a Trump Mood Index (TMI from 0 to 100) standardising his level of Cheerfulness vs. Rage (100 being the highest level of Rage).

Any quantitative measure of an emotional state involves defining and measuring a proxy of the emotional state. Because any such definition is arbitrary, this approach inevitably has limitations. However, we combined various proxy measures in order to reduce this bias:
- Polarity of the sentiment (Positive vs. negative)
- Analytical vs. emotional discourse (Complexity of expression)
- Context of the sentiment (Average negative vs. positive mood over a period)
- Custom Human ranking of tweets (we asked a sample of individuals to rank 3000 tweets on a cheerfulness-rage scale)

NB : Using generic sentiment scores allows us to effectively correct for the potential bias in our custom measure of cheerfullness-rage. Including context and complexity of expression measures allows us to quantify the intensity of the emotional state.

  • Step n°2 : Building a model to predict the evolution of his mood in the near future
       Once our Index defined, we used Machine Learning to build a predictive model of his mood. The index itself was showing some promising trends that could be exploited through basic technical analysis as if it was a financial product (e.g. Momentum, MACD, RSI, etc.).
We decided to complete the model with extra features that showed decent predictive power (e.g. Time of the tweets, Volume of tweets, Number of retweets, etc.).

Building the TMI backward with weighted combinations of these measures, here are the results we found for the past 2 years :


Results :

        What we’ve published here is an easily accessible packaging of our predictions that can help investors of every level to anticipate a change in Trump’s mood.
We also think that it can be used as a “weather forecast” of the White House for journalists & entertainers.

Note that the details of the methodology used (i.e. ML Workflow) could be part of a dedicated data-science article if we realize that there is a demand for it from readers.

Check out our live Trump Mood Predictor :