When I run the same model, is it possible to have lower theoretical mean in second order approximation than in first order approximation?

I thought it should be higher due to constant (positive) risk adjustment.

For example, if I run second order approximation, I see “correction” term(let’s say A) below “constant”(B) in policy function.

What I understand is that this correction term is related to the risk(i.e. variance of shocks).

But “A” has negative value, not positive value…

Anyway, I can see that “B” - “A(negative value)” = steady state value in first order approximation.

But I still don’t understand why the theoretical mean is lower in the second order approximation…