Lossy formats such as AAC, MP3, AC3 and DTS do not strictly have a bit depth as such which is why they offer high compresssion rates.
http://en.wikipedia.org/wiki/Audio_bit_depth
The general rule of thumb when decoding them is to use a high quality codec that uses a high internal number crunching algorithm (32bit float/int for example) to get the most accurate results, although accurate does not necessarily mean better sounding as it is all maths and totally subjective but this is the reason I prefer to use the LibMad MP3 codec when playing back MP3 at 24bit output via a 24bit DAC.
Lossless formats like FLAC only use file compression (a zipped wave file crudely speaking) and so maintain the original bit depth/waveform and are hence lossless (no audio is tweaked/removed).