reg-tables

Description

A simple linearmodels extension to run panel regressions with different specifications and export the results in a professional-looking latex table

Installation

pip install reg_tables

Sample Usage

from reg_tables import *

# Generate Random panel
N = 10**3
df = pd.DataFrame({
    'x1': np.random.randn(N),
    'x2': np.random.randn(N),
})
df['entity'] = np.random.randint(0,10,N)
df['time'  ] = np.random.randint(0,50,N)

# Generate the `y` variable 
df['y'     ] = 2 * df['x1'] - 0.5 * df['x2'] + np.random.randn(N)

# Generate the `y2`, with some fixed effects 
df['y2'    ] = df['y'] + (df['entity'] % 3)*10 + np.where(df['time']>10, -50, 0)

# Set the panel's double-index
df = df.set_index(['entity', 'time'])

# Define the baseline specification
baseline = Spec( df, 'y2', ['x1', 'x2'], double_cluster=True )

# The renaming dictionary
rename   = {
    'y2' : 'Salary',
    'x1' : 'Education',
    'x2' : 'Age',
}

# Create the model
model = Model(baseline, rename_dict=rename)

# Define some other regression specifications
model.add_spec(y='y2', entity_effects=True)
model.add_spec(y='y2', time_effects=True)
model.add_spec(y='y2', entity_effects=True, time_effects=True)

# Run all the specifications
res = model.run()
res

Classes and Methods

This package consists of two classes: Spec and Model.

Spec defines the regression specifications, including the panel dataset, the independent variable, and the independent variables. Optional arguments for this class include specifying whether the regressions should be performed with entity effects, time effects or both (entity_effects, time_effects or all_effects arguments respectively). Methods for Spec class include .run, which runs the regression and .rename – a method to rename variable according to the dictionary passed.

The Model class is a wrapper around the compare function of linearmodels. When creating Model, one has to specify the baseline regression specification, passed as a Spec object. Optional arguments include passing a rename_dict, according to which the variables are going to be renamed, as well as setting an all_effects Boolean variable, which will add the four versions of baseline Spec object with all possible combinations of entity and time effects to the model. The Model class has .rename, .add_spec and .remove_spec methods. The latter has a mandatory index argument and second optional index argument, which, if passed would work as a end point for slice. The .run method executes all Spec objects within Model and outputs them to a table. Optional argument coeff_decimals allows to specify the number of decimals for coefficient estimates and t-values, while latex_path allows to save the output table to a disk.