The net transform of each node is derived by premultiplying the parent's net transform with the node's local transform.
That is, Mn = Mn-1*Mn-2*...M0, where M0 is the root transform.
This gives the intuitive composition of transforms, and provides for standard forward kinematics. It is not likely that an animation file format would use the opposite convention, which is not very useful, so I would be suprised if you find you have to make a conversion for this.
Note that transform matrices in Panda are represented row-major, which does reverse the meaning of premultiply and postmultiply from a column-major representation.