localprojections 0.0.5

This module implements the local projections models for single entity time series, and panel / longitudinal data settings, due to Jorda (2005).

localprojections

This module implements the local projections models for single entity time series, and panel / longitudinal data settings, due to Jorda (2005), and based on codes available here.

Installation

1. pip install localprojections

Implementation

Panel Local Projections Model

Documentation

localprojections.PanelLP(data, Y, response, horizon, lags, varcov, ci_width)

Parameters

data :
Pandas MultiIndex dataframe with entity as the outer index, and time as the inner index.

Y :
List of column names in data to be used in the model estimation

response :
List of column names in Y to be used as response variables when estimating the impulse response functions (IRFs)

horizon :
Integer indicating the estimation horizon of the IRFs

lags :
Integer indicating the number of lags to be included in the model estimation

varcov :
Variance-covariance estimator to be used in estimating standard errors; refer to the linearmodels package.

ci_width :
Float higher than 0 and less than 1, i.e., (0, 1), indicating the width of the confidence intervals of the IRFs; ci_width=0.95 indicates a 95% confidence interval

Output

This function returns a pandas dataframe of 6 columns:

1. Shock indicates the shock variable
2. Response indicates the response variable
3. Horizon indicates the response horizon of the IRF
4. Mean indicates the point estimate of the IRF
5. LB indicates the lower bound of the confidence interval of the IRF
6. LB indicates the upper bound of the confidence interval of the IRF

For instance, the estimates of the 6-period ahead IRF of y from a shock in x, can be found in the row with Shock=x, Response=y, and Horizon=6

Example

from statsmodels.datasets import grunfeld
import localprojections as lp

df = grunfeld.load_pandas().data # import the Grunfeld investment data set
df = df.set_index(['firm', 'year']) # set entity-year indices (as per requirements in bashtage's linearmodels)

endog = ['invest', 'value', 'capital'] # cholesky ordering: invest --> value --> capital
response = endog.copy() # estimate the responses of all variables to shocks from all variables
irf_horizon = 8 # estimate IRFs up to 8 periods ahead
opt_lags = 2 # include 2 lags in the local projections model
opt_cov = 'robust' # HAC standard errors
opt_ci = 0.95 # 95% confidence intervals

irf = lp.PanelLP(data=df, # input dataframe
                 Y=endog, # variables in the model
                 response=response, # variables whose IRFs should be estimated
                 horizon=irf_horizon, # estimation horizon of IRFs
                 lags=opt_lags, # lags in the model
                 varcov=opt_cov, # type of standard errors
                 ci_width=opt_ci # width of confidence band
                 )
irfplot = lp.IRFPlot(irf=irf, # take output from the estimated model
                     response=['invest'], # plot only response of invest ...
                     shock=endog, # ... to shocks from all variables
                     n_columns=2, # max 2 columns in the figure
                     n_rows=2, # max 2 rows in the figure
                     maintitle='Panel LP: IRFs of Investment', # self-defined title of the IRF plot
                     show_fig=True, # display figure (from plotly)
                     save_pic=False # don't save any figures on local drive
                     )

Single Entity Time Series Local Projections Model

Documentation

localprojections.TimeSeriesLP(data, Y, response, horizon, lags, newey_lags, ci_width)

Parameters

data :
Pandas dataframe

Y :
List of column names in data to be used in the model estimation

response :
List of column names in Y to be used as response variables when estimating the impulse response functions (IRFs)

horizon :
Integer indicating the estimation horizon of the IRFs

lags :
Integer indicating the number of lags to be included in the model estimation

newey_lags :
Maximum number of lags to be used when estimating the Newey-West standard errors

ci_width :
Float higher than 0 and less than 1, i.e., (0, 1), indicating the width of the confidence intervals of the IRFs; ci_width=0.95 indicates a 95% confidence interval

Output

This function also returns a pandas dataframe of 6 columns:

1. Shock indicates the shock variable
2. Response indicates the response variable
3. Horizon indicates the response horizon of the IRF
4. Mean indicates the point estimate of the IRF
5. LB indicates the lower bound of the confidence interval of the IRF
6. LB indicates the upper bound of the confidence interval of the IRF

For instance, the estimates of the 6-period ahead IRF of y from a shock in x, can be found in the row with Shock=x, Response=y, and Horizon=6

Example

from statsmodels.datasets import grunfeld
import localprojections as lp

df = grunfeld.load_pandas().data # import the Grunfeld investment data set
df = df[df['firm'] == 'General Motors'] # keep only one entity (as an example of a single entity time series setting)
df = df.set_index(['year']) # set time variable as index

endog = ['invest', 'value', 'capital'] # cholesky ordering: invest --> value --> capital
response = endog.copy() # estimate the responses of all variables to shocks from all variables
irf_horizon = 8 # estimate IRFs up to 8 periods ahead
opt_lags = 2 # include 2 lags in the local projections model
opt_cov = 'robust' # HAC standard errors
opt_ci = 0.95 # 95% confidence intervals

# Use TimeSeriesLP for the single entity case
irf = lp.TimeSeriesLP(data=df, # input dataframe
                      Y=endog, # variables in the model
                      response=response, # variables whose IRFs should be estimated
                      horizon=irf_horizon, # estimation horizon of IRFs
                      lags=opt_lags, # lags in the model
                      newey_lags=2, # maximum lags when estimating Newey-West standard errors
                      ci_width=opt_ci # width of confidence band
                      )
irfplot = lp.IRFPlot(irf=irf, # take output from the estimated model
                     response=['invest'], # plot only response of invest ...
                     shock=endog, # ... to shocks from all variables
                     n_columns=2, # max 2 columns in the figure
                     n_rows=2, # max 2 rows in the figure
                     maintitle='Single Entity Time Series LP: IRFs of Investment', # self-defined title of the IRF plot
                     show_fig=True, # display figure (from plotly)
                     save_pic=False # don't save any figures on local drive
                     )

Plotting Impulse Response Functions

Documentation

localprojections.IRFPlot(irf, response, shock, n_columns, n_rows, maintitle, show_fig, save_pic, out_path, out_name)

Parameters

irf :
Output from PanelLP(), or TimeSeriesLP()

response :
List of variables contained in irf's Response column whose IRFs is to be plotted

shock :
List of variables contained in irf's Shock column whose IRFs is to be plotted

n_columns :
Integer indicating the number of IRF figures per row in the overall figure

n_rows :
Integer indicating the number of IRF figures per column in the overall figure

maintitle :
Strings to be used as the title of the overall figure; default is ''Local Projections Model: Impulse Response Functions'

show_fig :
Boolean indicating whether to render the overall figure

save_pic :
Boolean indicating whether to save the overall figure in the local directory; if True, a html file and a png file will be saved

out_path :
Strings indicating the directory at which the overall figure should be saved in; only used if save_pic is True

out_name :
Strings indicating the name of the file in which the overall figure should be saved as; only used if save_pic is True, and default is IRFPlot

Output

This function returns a plotly graph objects figure with n_columns (columns) x n_rows (rows) subplots. Depending on arguments passed, the figure may be rendered during implementation and / or saved in the local directory.

Examples

See above.

Requirements

Python Packages

• pandas>=1.4.3
• numpy>=1.23.0
• linearmodels>=4.27
• plotly>=5.9.0
• statsmodels>=0.13.2