I was wondering about the efficiency of this solution so I compared it with GrandFather's solution:

# x.pl
use Benchmark qw(timethese cmpthese);

cmpthese(
    -2,
    {
        GrandFather => sub {
            my $idxMax = 0;
            $ARGV[$idxMax] > $ARGV[$_] or $idxMax = $_ for 1 .. $#ARGV
+;
            $idxMax;
        },
        reisinge => sub {
            [ sort { $ARGV[$b] <=> $ARGV[$a] } 0..$#ARGV ]->[0];
        }
    }
);
[download]

These are the results of the benchmarking:

$ perl x.pl $(seq 1 1000)
              Rate    reisinge GrandFather
reisinge    3742/s          --         -3%
GrandFather 3864/s          3%          --
$ perl x.pl $(seq 1 10000)
             Rate    reisinge GrandFather
reisinge    377/s          --         -5%
GrandFather 395/s          5%          --
$ perl x.pl $(seq 1 100000)
              Rate    reisinge GrandFather
reisinge    34.1/s          --        -18%
GrandFather 41.6/s         22%          --
[download]

GrandFathers's solution seems to be more efficient, i.e. it's faster (to run) and scales better. But mine is shorter :-).