I'm new to algo trading world. I wrote few scripts and ended up with one which seems to provide consistent returns for 5 years of backtest for AUDUSD 60, 120 and 240 hours. But it performs poorly for other currency pairs or the same currency pairs with different time frames.

Do you think that the algorithm should be able to perform similar for other currency pairs with at least same time frames to make sure that the returns shown are not due to random reasons?

I wouldn't expect the strategy to perform similarly on other pairs. If it does, great. If it doesn't, then that doesn't mean that it is bad on AUDUSD too.

Here is a good article on testing whether a strategy performs better than a randomised version of itself. http://www.financial-hacker.com/whites-reality-check/

Thanks for the great article!