I see that the formula for AWATT, assuming only the fundemental part should be:
AWATT=Vrms x Irms x COS(V-I) x PMAX x (1/2^4)
In your case, where everything is at the fundamental, at full scale, and in phase, I think the first 3 terms all go to 1, it's not Vrms but Vrms/Vfs. I think at full scale, the power calculation is expected to come out as PMAX (33,516,139)
How do Pmax fit into a 24-bit signed register value?
I think that's where the x (1/2^4) comes in. That right shifts the calculated value 4 bits as it puts it into AWATT so you loose a little bit of resolution.
But I'm not sure any of that explains why you're getting a negative value. A negative values suggests your two inputs are out of phase from each other, in which case the cos becomes -1 rather than 1. Any chance there's a reverse-wiring on one of the inputs compared to other (i.e. on I compared to V).