Creating a bootstrap sample by group in python

Question

Consider the below dataFrame

         y   X1  X2  X3

ID year

1  2010  1   2   3   4

1  2011  3   4   5   6

2  2010  1   2   3   4

2  2011  3   4   5   6

2  2012  7   8   9  10

...

I'd prefer to make a few bootstrap test from the first df, figure a fixed-effects panel regression on the new bootstrap tests and than store the relating beta coefficients. The methodology I found for "normal" linear regression is the accompanying

betas = pd.DataFrame()

for i in range(10):

    # Creating a bootstrap sample with replacement

    bootstrap = df.sample(n=df.shape[0], replace=True)

    # Fit the regression and save beta coefficients

    DV_bs = bootstrap.y

    IV_bs = sm2.add_constant(bootstrap[['X1', 'X2', 'X3']])

    fe_mod_bs = PanelOLS(DV_bs, IV_bs, entity_effects=True ).fit(cov_type='clustered', cluster_entity=True)

    b = pd.DataFrame(fe_mod_bs.params)

    print(b.head())

    betas = pd.concat([betas, b], axis = 1, join = 'outer')

I have few questions:

1. Does the overall approach make sense for panel regression at all?
2. How do I adjust the bootstrapping so that the multilevel / panel structure is taken into account and complete IDs instead of single rows are "picked" during the bootstrapping?

Answer 1

Look at the below code:

companies = pd.DataFrame(df.reset_index().Company.unique())

betas_summary = pd.DataFrame()

for i in tqdm(range(1, 10001)):

    # Creating a bootstrap sample with replacement

    bootstrap = companies.sample(n=companies.shape[0], replace=True)

    bootstrap.rename(columns={bootstrap.columns[0]: "Company"}, inplace=True)

    Period = list(range(1, 25))

    list_of_bs_comp = bootstrap.Company.to_list()

    multiindex = [list_of_bs_comp, np.array(Period)]

    bs_df = pd.MultiIndex.from_product(multiindex, names=['Company', 'Period'])

    bs_result = df.loc[bs_df, :]

    

    betas = pd.DataFrame()

    

    # Fit the regression and save beta coefficients

    DV_bs = bs_result.y

    IV_bs = sm2.add_constant(bs_result[['X1', 'X2', 'X3']])

    fe_mod_bs = PanelOLS(DV_bs, IV_bs, entity_effects=True ).fit(cov_type='clustered', cluster_entity=True)

    b = pd.DataFrame(fe_mod_bs.params)

    b.rename(columns={'parameter':"b"}, inplace=True)

    betas = pd.concat([betas, b], axis = 1, join = 'outer')

Here, My entity variable will be my Company and Period will be my Time variable.

Looking for a good python tutorial course? Join the python certification course and get certified.

For more details, do check out the below video tutorial...