############################################################
## README file for the paper "The colour of finance words,"
## by Garcia, Hu and Rohrer (2021).
## Dated 20210426.
############################################################

############################################################
## FinWordColour01.zip
############################################################
## If you are interested in just getting dictionaries, 
## the file in FinWordColour01.zip are small and give you 
## some flexibility as to how to tweak the choices we make
## in our paper.
##
## The csv files are the dictionaries discussed in Section
## 5.1 of the paper.
##
## The RData object, together with the metadata file and the
## R code allows the reader to adapt the choices we make
## when constructing the dictionaries (default settings
## yield the output in the csv files discussed above).
############################################################

Dictionary files:
- ML_positive_bigram.csv: ML positive bigram dictionaries, 12130 tokens.
- ML_negative_bigram.csv: ML negative bigram dictionaries, 13330 tokens.
- ML_positive_unigram.csv: ML positive unigram dictionaries, 617 tokens.
- ML_negative_unigram.csv: ML negative uniigram dictionaries, 727 tokens.
- ML_dictionaries.RData: RData object with tons on colour (see *.R file for details).
- meta_public.rds: metadata for all 61,041 earnings calls in our dataset.
- dictionary_construction.R shows how we construct uni/bigram dictionaries as presented in Section 5.1. 
- replication_kaggle.R: replicates  Table 1 column 2 and Table 2 column 1 & 4 in the paper, using the dtms provided above.

############################################################
## FinWordColour02.zip
############################################################
## If you are interested in testing the training algorithm
## and seeing how it performs out of sample, the files in
## FinWordColour02.zip should provide some colour.
##
## We matched our database to public sources (from Kaggle,
## see https://www.kaggle.com/borismarjanovic/price-volume-data-for-all-us-stocks-etfs).
##
## The data from this match, in particular the abnormal returns,
## are from the meta_public.rds file included in FinWordColour01.zip
##
## We note that the data from Kaggle stops in 2017
## (slightly smaller sample than in our paper).
##
## We provide the document-term-matrix with 65K terms (2^16),
## both with tf and tf-idf weights.
##
## The code replication_kaggle.R (see above) runs through
## the training given a cutoff date, then looks at out-of-sample
## performance.
############################################################

Document term matrices/metadata:
- dtm_bigram.rds: document-term-matrix for all 61,041 earnings calls in our dataset.
- dtm_bigram_tfidf.rds: document-term-matrix using tf-idf weights for all 61,041 earnings calls in our dataset.