Despite Abigail-II's notes, it is possible to do this in linear time. Hashes are the answer; here, you need a nested one. The following will consume some memory in exchange for runing in O(n).

#!/usr/bin/perl -w
use strict;
use Data::Dumper;

my @AoA = (
    [ qw(AAA BUY 98 0) ],
    [ qw(BBB SEL 27 1) ],
    [ qw(FFF BUY 43 4) ],
    [ qw(AAA SEL 98 0) ],
    [ qw(CCC SEL 98 0) ],
);

my %directive_for;
push @{ $directive_for{$_->[0]}{$_->[2]}{$_->[3]} }, $_->[1] for @AoA;

my @grouped;
for my $i (keys %directive_for) {
    for my $j (keys %{$directive_for{$i}}) {
        for my $k (keys %{$directive_for{$i}{$j}}) {
            push @grouped, [ $i, $directive_for{$i}{$j}{$k}, $j, $k ];
        }
    }
}

print Dumper(\@grouped);
[download]

The result is an array of unique arrays where element 1 is itself an array with all the values that element had in the various non-unique copies.

Makeshifts last the longest.