Getting Started
Preparing the Software
Dependencies
The GME package requires pandas and statsmodels packages.
Installation
Download and install the GME package.
Tip
Easily copy the code below by using the copy button in the upper right corner of the code block.
pip install gme
A Basic Example: Run PPML estimation
Step 1. Import the needed packages
import gme as gme
import pandas as pd
Step 2. Create EstimationData
The GME package is built upon a specialized data object called EstimationData. EstimationData contains data in a Pandas DataDrame as well as a collection of additional information and tools that are useful for gravity modeling, such as
- A log of the history of the data, such as the location of the file it was read from and modifications made to it,
- Metadata, such as names of the columns containing importer, exporter, and year information, so that they need not be continuously supplied,
- Several tools for producing summary statistics or other types of commonly sought descriptive information.
Begin by loading example trade data. The dataset used in the following code is available for download or can be accessed directly with python and pandas, as shown below.
sample_data = pd.read_csv(
'https://www.usitc.gov/data/gravity/example_trade_and_grav_data_small.csv')
sample_data.head()
importer exporter year trade_value agree_pta common_language \
0 AUS BRA 1989 3.035469e+08 0.0 1.0
1 AUS CAN 1989 8.769946e+08 0.0 1.0
2 AUS CHE 1989 4.005245e+08 0.0 1.0
3 AUS DEU 1989 2.468977e+09 0.0 0.0
4 AUS DNK 1989 1.763072e+08 0.0 1.0
contiguity log_distance
0 0.0 9.553332
1 0.0 9.637676
2 0.0 9.687557
3 0.0 9.675007
4 0.0 9.657311
Next, create an instance of the EstimationData object using the sample data. To create an EstimationData instance (called gme_data in the example below), you need to supply a Pandas DataFrame and identifiers for certain key columns, such as the trade flows, importer/exporter, year, and sector (if applicable).
gme_data = gme.EstimationData(data_frame = sample_data,
imp_var_name = 'importer',
exp_var_name = 'exporter',
trade_var_name = 'trade_value',
year_var_name = 'year')
print(gme_data)
number of countries: 62
number of exporters: 62
number of importers: 62
number of years: 27
number of sectors: not_applicable
dimensions: (98612, 8)
Step 3. Create an EstimationModel
After creating an EstimationData object, you need to create an EstimationModel object, which will be used to produce gravity estimates. To create an EstimationModel instance (called gme_model in the example below), you need to supply an EstimationData object and a specification.
gme_model = gme.EstimationModel(estimation_data = gme_data,
lhs_var = 'trade_value',
rhs_var = ['log_distance','agree_pta','common_language','contiguity'],
fixed_effects = ['importer', 'exporter'],
keep_years = [2015])
The initialization of EstimationModel establishes a reference to the EstimationData object rather than a copy of the data.
Step 4. Estimate the model
What is a method?
A method is a function that is connected only to a particular object. Click here to see the relevant Python documentation.
Once the EstimationModel is defined, it can be estimated by applying the method .estimate()
gme_model.estimate()
The code provides some information while it is running:
select specification variables: ['log_distance', 'agree_pta', 'common_language', 'contiguity', 'trade_value', 'importer', 'exporter', 'year'], Observations excluded by user: {'rows': 0, 'columns': 0}
drop_intratrade: no, Observations excluded by user: {'rows': 0, 'columns': 0}
drop_imp: none, Observations excluded by user: {'rows': 0, 'columns': 0}
drop_exp: none, Observations excluded by user: {'rows': 0, 'columns': 0}
keep_imp: all available, Observations excluded by user: {'rows': 0, 'columns': 0}
keep_exp: all available, Observations excluded by user: {'rows': 0, 'columns': 0}
drop_years: none, Observations excluded by user: {'rows': 0, 'columns': 0}
keep_years: [2015], Observations excluded by user: {'rows': 94952, 'columns': 0}
drop_missing: yes, Observations excluded by user: {'rows': 0, 'columns': 0}
Estimation began at 09:29 AM on Jun 13, 2018
Estimation completed at 09:29 AM on Jun 13, 2018'''
The results are stored in a collection (called dictionary in Python) with each sector having its own set of results. If no sectors were supplied or used, there would be only one set of results in the dictionary, labeled 'all'.
A simple table with regression results can be produced with the following command:
gme_model.format_regression_table(format = ‘txt’,
path = “C:\\Documents\\regression_table.txt”)
Variable all
a_agree_pta agree_pta 0.338***
a_agree_pta_se (0.088)
a_common_language common_language 0.063
a_common_language_se (0.071)
a_contiguity contiguity 0.211***
a_contiguity_se (0.085)
a_exporter_fe_ARG exporter_fe_ARG -0.444
a_exporter_fe_ARG_se (0.365)
a_exporter_fe_AUS exporter_fe_AUS 0.619
a_exporter_fe_AUS_se (0.500)
... ...
a_importer_fe_USA importer_fe_USA 30.791***
a_importer_fe_USA_se (0.562)
a_log_distance log_distance -0.784***
a_log_distance_se (0.051)
b_aic AIC 1806014725995.581
b_bic BIC 1806014667986.696
b_llf Likelihood -903007362899.79
b_nobs Obs. 2040