WRDSMerger.jl

This package is designed to perform common tasks using the Wharton Research Database Services (WRDS). In particular, there is a focus on data from CRSP (stock market data), Compustat (firm balance sheet and income data) and the links between these datasets. It also implements an abnormal return calculation.

Installation

This package is registered, so in the Julia REPL:

]add WRDSMerger

or

julia> using Pkg; Pkg.add("WRDSMerger")

General Usage

This package requires a subscription to WRDS and can only access datasets that are included in your subscription. There are several ways to connect to the database. The simplest and most reliable is to use LibPQ.jl (requires at least v1.16 of LibPQ.jl), to initiate a connection run:

conn = LibPQ.Connection(
    """
        host = wrds-pgdata.wharton.upenn.edu 
        port = 9737
        user='username' 
        password='password'
        sslmode = 'require' dbname = wrds
    """
)

Note, running the above too many times may cause WRDS to temporarily block your connections for having too many. Run the connection at the start of your script and only rerun that part when necessary. I have found that LibPQ is the easiest way to connect to WRDS since there are no restrictions on length of query and the data has a consistent format.

Alternatively, you can connect to WRDS through an ODBC driver using ODBC.jl. I recommend following the setup steps listed under WRDS support for connecting with Stata (since that also uses ODBC). You can find that information here.

The third method is if you download the data to your own database, such as a SQLite database using SQLite.jl (This is the method this package uses for testing). SQLite requires slightly different names for tables, so you will need to change the table defaults:

conn = SQLite.DB("db.sqlite")
WRDSMerger.default_tables["comp_funda"] = "compa_funda"
WRDSMerger.default_tables["comp_fundq"] = "compa_fundq"
...

ODBC vs LibPQ

The two largest packages I am aware of for connecting to a Postgres database in Lulia are ODBC.jl and LibPQ.jl. Both of these have various advantages.

Starting with LibPQ, adding LibPQ to your project is the full installation process. To use ODBC, an extra driver, with extra setup, needs to occur before use. In addition, as far as I can tell, LibPQ does not have a limit on length of query. Some functions in this package (such as crsp_data) create exceptionally long queries to reduce the total amount of data downloaded, which LibPQ handles easily.

For ODBC, it is considerably faster at converting data to a DataFrame. For example, downloading the full CRSP Stockfile (crsp.dsf, which includes returns for every stock for each day and is about 100 million rows), takes about 4 minutes to download and make into a DataFrame with ODBC on a gigabit connection. LibPQ takes about 24 minutes. Most of this difference appears to be type instability while converting the LibPQ result to a DataFrame, since the initial LibPQ result only takes a minute and @time reports 80% garbage collection time. ODBC also stores your password separately (in the driver settings) making it a little easier to share a project without compromising your password.

Calculating Abnormal Returns and Other Return Statistics

This functionality is now part of the package AbnormalReturns.jl

Disclaimer

This package is still early in development and many things could change. WRDSMerger.jl also has no association with WRDS or Wharton.