which model did you backtest it?
if not "open prices" and do not have low level timeframes in your metatarder for the whole period of time then you may get unrealistic results.
i think for the last month you see more or less real test as you have data for small timeframes, which are used for modelling bars of larger timeframes.
read articles here with explanations:
http://www.metaquotes.net/experts/articles