1. Run create_sample.R.
This will create Garcia (2013) measures, lags, etc. and merge them with the market data. 
The output is sentiment_data.csv, which is the only file required in the next step. 
2. Run tables_123456_fig_1_predictability.R.
This will create Tables 1,2,3,4,5,6,7,8 (not in this order -- Tables 3 and 4 are at the end of the file) and Figure 1 in the paper. Generally, this code is responsible for the predictability sections of the paper. The only required input is sentiment_data.csv from the previous step. 
3. Run table_9_sffed.R. 
This will create Table 9 in the paper. It requires SF Fed News sentiment index data, Dow returns, ISRECD and sentiment_data.csv (to construct Garcia frequency measures). 
4. Run tables_10_11_12_fig23_properties.R,
This will create tables 10,11,12, Figures 2 and 3 in the paper. The only required input is sentiment_data.csv.

Note: file names and table numbering in the code annotation might not reflect the latest version  of the paper as it underwent multiple edits. However, the structure is the same: the first script creates the inputs, the second one does all predictability tests, and the third one computes statistical properties of the measures. 

I do not provide literature measures apart from the Garcia (2013) in this package; you can either obtain them from the original authors or request an exact copy used in this research from me (if and only if public versions result in material differences from the paper). 

Garcia (2013) data is available at https://leeds-faculty.colorado.edu/garcia/data.html
If reused, please attribute the original author (Diego Garcia) appropriately.