I am trying to do the Panel OLS Regression (which is from linearmodels.panel), but when doing the ‘fit’ function it displays the error.

Below is my code.

mod = PanelOLS(dependent = test_df['ln_entry_total'], exog = test_df[['Exist_Cap_bili', 'WX']], weights = test_df[['gdp', 'pop', 'wage', 'ld_indu', 'ld_infr', 'wind', 'tech', 'PM', 'secon_indu', 'third_indu', 'temp', 'regu']], entity_effects = True, time_effects = True)
print(mod)

res = mod.fit(cov_type = "clustered", cluster_entity=True, cluster_time =True)
print(res)

and the result displays:

PanelOLS 
Num exog: 2, Constant: False
Entity Effects: True, Time Effects: True, Num Other Effects: 0
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
~AppDataLocalTempipykernel_259601658320709.py in <module>
      7 print(mod)
      8 
----> 9 res = mod.fit(cov_type = "clustered", cluster_entity=True, cluster_time =True)
     10 print(res)

~anaconda3libsite-packageslinearmodelspanelmodel.py in fit(self, use_lsdv, use_lsmr, low_memory, cov_type, debiased, auto_df, count_effects, **cov_config)
   1838                 x_effects = np.zeros(x.shape)
   1839             else:
-> 1840                 y, x, ybar, y_effects, x_effects = self._weighted_fast_path(
   1841                     low_memory=low_memory
   1842                 )

~anaconda3libsite-packageslinearmodelspanelmodel.py in _weighted_fast_path(self, low_memory)
   1631         w = self.weights.values2d
   1632         root_w = cast(Float64Array, np.sqrt(w))
-> 1633         wybar = root_w * (w.T @ y_arr / w.sum())
   1634 
   1635         if not self._has_effect:

ValueError: operands could not be broadcast together with shapes (1105,12) (12,1)

I found out the shape should be (1105,12)(1,12) rather than (1105,12)(12,1) to solve this problem, and 12 is the number of weights that I put in PanelOLS as control variables, which produced the error.

I’d like to know how to solve this problem by modify PanelOLS, or it is even absolutely fine to consider other ways to apply control variables as weights.

I really appreciate any help.