Hi everyone,

I perform panel analysis where cross-section dominates the time dimension (N around 15,000 and T around 17). My panel is unbalanced and the estimation method is RE/FE including time dummies. When I run panel unit root tests (LLC, IPS, Fisher-ADF, Fisher-PP) all but one variable is stationary. This independent variable is yield-to-maturity and follows a trend which I deem to be deterministic. The variable is proven to be non-stationary (IPS, Fisher-ADF, Fisher-PP) when entered in level form and with intercept. Performing unit root tests with intercept and trend the variable becomes partly stationary (IPS, Fisher-ADF, Fisher-PP). Only the LLC-test shows non-stationarity. However, according to my research, this test should be more suitable for balanced panel. I therefore conclude, that yield-to-maturity is (deterministic) trend-stationary.

In my field of literature yield-to-maturity is usually used in levels and not first differenced (which would make the variable stationary according to the unit root test mentioned above). However, most papers do not mention that any unit root tests have been performed.

Is it sufficient to use yield-to-maturity in levels and account for the trend with a full set of year dummy variables as controls in order to solve the spurious estimation problem?
Or are the results of the unit root test with this kind of data (large N, small T) not reliable and should be ignored?

Thanks in advance!!