We normally select the test period so that it covers about 50..100 trades per degree of freedom of the strategy. This is an empirical rule - shorter test periods tend to produce less significant results.

There are exceptions, for instance Z4 and Z5 were only tested with data after 2011.