In general, I backtested for the longest period the platform would allow, which is unfortunately only back to about 2002 (I really wish it went back pre-.com boom, to 1997 or so). I would then often do separate tests for 2007-2013, 2012+, etc, just to see how returns were affected by market timing- you can kind of eyeball this, but over long timeframes, sometimes unacceptable underperformance during a volatile period can be a bit hidden over a long period.