The index on the time dimension must be either numeric or date-like error

Question

I am trying to do a panel regression with

from linearmodels.panel import PooledOLS
import statsmodels.api as sm

but I encountered a problem with my index. For the regression I need a Multiindex, so I have a Dummy variable and the time (see below). The two indicies are a_c ( a dummy variable for all 10 countries of my analysis) and Timestamp (the date).

                i_ct    preemu  postemu  γ1             γ2
α_c Timestamp   

10  2020-06-24  -0.04   0.00    0.02    -1.110223e-16   2.000000e-02
    2020-06-25  0.05    0.00    -0.04   -1.000000e-02   1.000000e-02
    2020-06-26  0.02    0.00    0.05    0.000000e+00    1.000000e-02
    2020-06-29  0.00    0.00    0.02    -2.000000e-02   -1.110223e-16
    2020-06-30  0.08    0.00    0.00    2.000000e-02    -6.000000e-02   

When I run the regression

exog_vars = ['preemu','postemu','γ1','γ2']
exog = sm.add_constant(beta_panel[exog_vars])
mod = PooledOLS(beta_panel.i_ct, exog)
pooled_res = mod.fit()
print(pooled_res)

I get this error:

ValueError: The index on the time dimension must be either numeric or date-like

But when the timestamp is not in the index I get this error:

Series can only be used with a 2-level MultiIndex

Anyone knows why it could throw the first error?

Answer 1

Change the string date time type to numerical (I tried to change it to timestamp or other time datatype, but the error was still existing), for example: 2021/01/31 change to 20210131. Then it works.

Answer 2

The time dimension index in the example provided looks like a string. Converting it to a date format or a numerical format should work.

Also note that the entity index comes first in order and the time dimension is second in the multi-index, like the example provided has correctly done.

A small oversight that can cause errors.