Respond to Libor's Email from 2/18

[ijyliu]

• time period 2010-2016 is fine, but if we can extend a little bit, would be good. E.g. 2008-2018. Note, there was a lot of rating action in 2009 (financial crisis), and in 2016 (oil supply shock from fracking).

• you can use S&P ratings for now only, to keep it simple

• change in rating - do you mean you have the rating info before change and after the change? Or you know how many notches the rating changed? Clarify

  • IL: Our data is of ratings issued (changed or affirmed) on certain days. Therefore, we know the rating after every change, and the rating before the changes that have a prior point in the dataset (assuming ratings are fixed between changes). When we have ratings before and after, we know how many notches the rating changed. Limiting the data to S&P ratings 2010-2016, we have 3,171 observations/issuances, of which 1,992 are affirmations (same rating before/after), 637 are ambiguous (no info about rating before), 292 are upgrades, and 250 are downgrades.

• for Altman Z - you can use it, but you can also try to throw the coeffs for A-E into the regression/ML to estimate them

• clarify the scope of the analysis: only U.S. public firms; how many UNIQUE FIRMS do you have that had ratings & earnings call data between 2010-2016 (or 2008-2018)? The number of unique firms is the most important sample size info. Then, how many rating changes per firm did we observe over the time frame (histogram). Do we have quarterly/annual earnings calls for all these firms.

  • IL: After limiting to S&P ratings from 2010-2016 (but without limiting to public firms or earnings call firms), we have 638 unique firms

 - _Of these firms, 319 have earnings call data that matches up clearly to credit ratings for an appropriate timeframe. Here's a histogram of the count of earnings calls by company (stock ticker)_

• Plot other info - bar chart of #of firms by sector, etc.

  • IL: It's easiest to use earnings call sectors (there are several credit rating sector variables that are not always available), so that's what I did for the below.

[ijyliu]

completed, ready to send this email

[ijyliu]

subsumed by all data eda except for this part

[ijyliu]

fully subsumed by

https://github.com/current12/Stat-222-Project/blob/39804d98067eac1aef180b6e553c2bdd74b10a73/Code/Exploratory%20Data%20Analysis/All%20Data%20EDA.ipynb