Is there a rule of thumb regarding low within-FE variation?

[Chan Ge]

Dear all, could you please educate me regarding when we say the within-FE-group variation of explanatory variables is too low to have reliable estimates?
I intend to specify an OLS model with industry fixed effect. Below tabulates the pooled and within-FE variations of a set of explanatory variables.
Do you think low within-FE variation is an issue for my analyses?
Thank you!

	Pooled Sample Summary Statistics			Within-FE (ioSecTRG) Summary Information (among 13 groups)
	N	Mean	Std. Dev.	Singletons	# of FE groups excluding singletons	Zero Within-FE Variation	Within-FE Std. Dev.	Within-FE Std. Dev. ÷ Pooled Std. Dev.
High	502	0.50	0.50	2	11	1	0.20	39%
Complem	502	0.99	0.18	2	11	0	0.11	59%
GeoDist	502	5.75	2.17	2	11	0	0.65	30%
CBank	502	0.02	0.15	2	11	5	0.04	27%
CAudit	502	0.24	0.42	2	11	3	0.11	27%
AcqSize	502	8.55	1.83	2	11	0	0.72	39%
AcqRunup	502	0.00	0.11	2	11	0	0.01	13%
AcqQ	502	1.75	1.01	2	11	0	0.56	56%
AcqReturn	502	0.04	0.28	2	11	0	0.07	25%
TrgSize	502	6.45	1.68	2	11	0	0.60	36%
TrgRunup	502	0.04	0.19	2	11	0	0.02	12%
TrgQ	502	1.77	1.23	2	11	0	0.63	52%
TrgReturn	502	0.00	0.42	2	11	0	0.08	20%
Cash	502	0.37	0.48	2	11	2	0.25	51%
Stock	502	0.25	0.43	2	11	5	0.19	44%
Compete	502	0.04	0.20	2	11	5	0.05	25%
Tender	502	0.17	0.37	2	11	7	0.18	48%
Crisis	502	0.09	0.28	2	11	7	0.05	19%
PostCrisis	502	0.68	0.47	2	11	2	0.09	20%

[Carlo Lazzaro]

Chan:
as for all the R_square (and pseudo cousins) that populate the regression realm, there's no such a hard and fast rule: most depends on what values are "reasonable" in different research fields.
As far as the fixed effect estimator is concerned, once it has wiped out al the time-invariant variables, what remains should have "enough" (again, whatener this may mean in different research fields) variation to lame it work at its best.

[Chan Ge]

Chan:
as for all the R_square (and pseudo cousins) that populate the regression realm, there's no such a hard and fast rule: most depends on what values are "reasonable" in different research fields.
As far as the fixed effect estimator is concerned, once it has wiped out al the time-invariant variables, what remains should have "enough" (again, whatener this may mean in different research fields) variation to lame it work at its best.

Got it. Thank you for insights, Carlo. Highly appreciated.