Personally, I wouldn't load it. I'd grab DBI and DBD::CSV and interrogate it as if it were a database. And then, if that were too slow for some reason, I'd actually put it in a database (using the same DBD::CSV for import) - perhaps a temporary table, and query the database.

That said, thus far, I've not had to move off DBD::CSV - it is fast enough for my purposes.

my @entries;
my @keys;
while ($_ = <DATA>) {
    if ($. == 0) { # first line
        @keys = split /;/;
        next;
    }
    my %data;
    $data{@keys} = split/;/;
    push @entries, \%data;
}
[download]

Update: updated some code errors.

Split the header line into an array before entering the loop, and then use that to build an AoHs:

use strict;
use warnings;

my $file = "/tmp/index.csv";
open (FL, "< $file");
my @fieldnames = split ';', <FL>; ##
my @data;
while ($line = <FL>){
    my %temp; @temp{ @fieldnames } = split ';', $line;
    push @data, \%temp;
}

## print the 'Forex' field of the 124th line (array runs from 0).
print $data[ 123 ]{ Forex };
[download]

From the docs:

Tie::Handle::CSV makes basic access to CSV files easier. When you read from the file handle, a hash reference or an array reference is returned depending on whether headers exist or do not.
Regardless of the type of the returned data, when it is converted to a string, it automatically converts back to CSV format.