I've never really played with Parse::RecDescent until now, but here's what I came up with. Your date and time regexes had problems, you're inconsistent about returning an array reference or a hash reference, and my first 'line' rule (below) has an example of how to label the 'type', if that's the sort of result you want. Also you don't have start and end dates on every rule, so I leave that up to you to fix if necessary. I sort of stole merlyn's idea of how to organize the top level rules and ran with that :)

use Parse::RecDescent;
use strict;
use warnings;
use Data::Dumper;

# Make sure the parser dies when it encounters an error
$::RD_ERRORS = 1;
# Enable warnings. This will warn on unused rules &c.
$::RD_WARN   = 1;
# Give out hints to help fix problems.
$::RD_HINT   = 1;

# Create and compile the source file
my $parser = Parse::RecDescent->new( q(
    comma      : ","
    date       : /\b\d{1,2}\/\d{1,2}\/\d{1,2}\b/
    start_date : date
    end_date   : date
    time       : /\b\d\d:\d\d:\d\d\b/
    rate       : /\b\d+\.\d{4}\b/
    rates      : rate comma { $item{rate} }
    start_rate : rate
    end_rate   : rate
    change     : rate
    whitespace : /\s*/

    lines : line /\z/ { $item{line} }
    line  : "G017RATEBRKRL" comma rate comma start_date
      comma end_date comma time { $item{type} = $item[0]; \%item }
    line  : "G017CP111 D" comma start_rate comma end_rate comma
      change comma date comma time { \%item }
    line  : "G017RPAGO/N" comma rate comma whitespace comma
      whitespace comma date comma time { \%item }
    line  : "G017ONFD" comma rates(6) date comma time { \%item }
    line  : "G017PDFF" comma rates(4) date comma time { \%item }
) );

while ( my $quote_data = <DATA> ) {
    next if $quote_data !~ /\S/;
    my $result = $parser->lines( $quote_data );
 
    if ( defined $result ) {
        print Dumper $result;
    } else {
        print "Failed!\n";
    }
}
[download]

All that being said, I'm not sure I'd actually use Parse::RecDescent for this problem. I might just quickly get the first field, and use that as a key to a hash of subroutines which use regexes to parse the data and return the results. I'd consider how much you care about efficiency in this routine anyway.