MOODS scans both strands by default, so a - hit just means the PWM matched the reverse complement at that position. It's a real occurrence, not an artefact.

Whether it matters comes down to your input orientation, which I suspect is the actual source of the confusion. If you reoriented each promoter to run 5'->3' in the gene's direction before scanning, then for something directional like TATA the + hits are the ones that could serve that gene and a - hit is antisense. If you scanned raw genomic plus strand without reorienting, then a minus-strand gene's genuine TATA legitimately turns up as a - hit and that one is the relevant one. So "plus strand gene but the only hit is minus" is completely normal in the second case and a warning sign in the first.

Quickest way to work out which situation you're in: take one promoter you're sure about and search it for the TATA consensus TATAWAWR on both strands, then look at whether the plus or the minus hits land ~25-35 bp upstream of your TSS. That's consensus matching rather than PWM scoring so it doesn't replace MOODS, it just tells you which way your input is facing. I work on SeqBench so discount accordingly, https://seqbench.com/tools/motif-finder does IUPAC codes with both-strand scanning; any pattern searcher does the same job.

One other thing worth flagging -- MA0001 is AGL3 in JASPAR, a plant MADS-box factor, not TBP. TBP is MA0108. If TATA boxes are what you're after then you're scanning with the wrong matrix and the strand question is moot until you swap it.