@wtrumanrose let's just edit this list as we brainstorm all of the things we might do:
Use word embeddings along with firm performance lags to model quarterly firm performance.
Can we use sub-dictionaries or clusters from the word embeddings as features?
What is the place of "Granger Causality"?
Is there a place to use SUE?
Do we have data on marketing actions in addition to marketing term use for a give quarter?
How do we incorporate heterogeneity? Seems like a good place to use a hierarchical linear model with the form depending on which outcome we use.
Investigate other uses of collocation (e.g., when firms exceed their projected performance with surprise > 0).
Allow for an investigation of the collocation network graphs by putting them into a dashboard/Shiny environment.
Determine if there are any interesting prediction questions to answer with methods outlined in SMLTAR.
Look at differences in sentiment between questions from analysts and answers from corporate officers within Q/A session.
Explore different window sizes.
Use {clock} if we need to look at fiscal quarters again? Or {modeltime} for time series?
“Can you show whether talking about marketing in today’s earnings call is associated with higher firm performance in the future? What is the predictive value to investors when the earnings call uses a lot of marketing terms – is it a leading indicator of future firm performance?”
@wtrumanrose let's just edit this list as we brainstorm all of the things we might do:
surprise> 0).“Can you show whether talking about marketing in today’s earnings call is associated with higher firm performance in the future? What is the predictive value to investors when the earnings call uses a lot of marketing terms – is it a leading indicator of future firm performance?”