Loop modeling in nucleic acids (i.e. in RNA) is done for the same reasons as loop modeling in proteins: to find out the possible 3D structure of certain regions in a biomolecule (here, RNA). For instance, a structure resulting from an X-ray experiment might miss a couple of bases/nucleotides, possibly because they couldn't be traced properly during processing (one example would be the PDB entry 1u9s of an RNAse P domain that misses the atomic coordinates for a complete tetraloop hairpin). Another use-case would be to check which loops could fit into a specified site in an RNA molecule, e.g. to find alternative loop structures that might serve (at least from a geometrical point of view) as replacement for an existing structure.

In any case, "loop modeling" in RNA is a 3D structure problem. Here, secondary structure prediction can only help you with finding the stem and loop regions in a sequence with unknown 3D structure. Since you already know the 3D structure of the molecule (with the exception of your target loop), secondary structure prediction does not yield any useful information for the loop modeling at all.

References:

Schudoma et al., 2010, Nucleic Acids Research http://nar.oxfordjournals.org/content/38/3/970

Schudoma et al., 2010, Bioinformatics http://bioinformatics.oxfordjournals.org/content/26/13/1671