Modeling over Null values

[TIGTAD]

Why does maxent output expand prediction to null values (-9999) of my rasters? 

Is there a solution to this? 

Thanks,

Tigran

[TIGTAD]

Let me explain the problem in more details:

I prepare my variable data by masking in ArcMap, and then convert it to DIVA grids for Maxent to accept. Values outside of the mask are all - no data, DIVA gis shows them black, and in the metadata null values are set to -9999. So inputs look good. However, the maxent outputs (produced with version 3.4.2, 3.4.4) show the background as monotone blue, and after importing .asc' back in Arc, I am getting the areas over those values as predicted unsuitable, raw values in one model were something like 0.0000138. Also this may be affecting response curves and AUC values. Any thoughts?

Thanks,

Tigran