When you say "bandwidth", I think you mean a passband, with both a lower and upper frequency limit?
Or.. maybe do you mean the maximum frequency where the response has started to fall to -3dB ?
in general, the "bandwidth" of any device under test is found by the point at which the response has decreased by 3dB below the levels that were at the frequencies right in the middle of it's passband. This is also known as the "half-power point"
The insertion loss..
The interconnect might have a insertion loss which applies right across the passband and beyond. The -3dB roll-off at the passband edge is in addition to the insertion loss. We always mean -3dB relative to the level in the passband when deciding where the limit of the passband is. For a signal going through, the insertion loss is then also accounted as further loss, but the insertion loss does not have a role in determining the 3dB bandwidth.
Keep in mind that even though -3dB is a convenient way of defining band edge, you might have to allow sufficient "extra" so that the response available to the real signals does not drop as far as -3dB. You might want them all safely in a region where the response is (say) still within 0.5dB of the passband maximum.