It's a precarious time to be an analyst. Regulators want to make the whole business model irrelevant. The fell hand of cost-trimming has cut down numerous analysts in their prime. Now, thanks to natural language processing, aka The Shining, they have this to deal with:
By leveraging FactSet Symbology at the individual security level, we can explore patterns and trends at a more granular level comparing transcripts and estimates. For example, are companies more likely to call on analysts with favorable ratings after or before a down quarter? Are analysts who participate on company earnings calls more accurate than their peers who do not? This example highlights how symbology at the individual security level unlocks the ability to augment or enhance existing models.
Using this method, we would have noticed analyst Colin Langan (UBS) had an overall sentiment score that was below the average of the call (0.111 vs 0.189) and one of the most negative sentences (-0.423). Shortly after the call, he released an updated research report titled Tesla Motors Q3 Good News to Turn Bad in Q4.
That's courtesy of FactSet, which used a natural language processing script to analyze sentiment Tesla's Q3 2016 earnings call. The script uses a complex algorithm and what we can only assume to be ancient black magic to tease out the true feelings of those doing the talking. For instance, here's what FactSet found in the depths of Elon Musk's soul:
...We can diagram the call and identify how the sentiment of Elon Musk’s comments change between the Management Discussion Section (the prepared remarks) and the Question and Answer (off-the-cuff). [...] Here we can see the average sentiment of the call was positive, with a score of approximately 0.2, but the range of values varied quite dramatically from the Management Discussion to the Q&A, where the tone of the call took a turn for the worse upon mention of then-recent fatalities involving Tesla’s Autopilot.
It's good to know that, when confronted with deaths he has caused, Musk feels acceptably human emotions. But as we learn from the algos, so do the analysts. Here's a chart of their sentiment scores, according to the analysis:
Should analysts care that they're now the target of digital psychic wizardry? Probably not all that much. But it certainly doesn't make the job any easier.