Dear Popov,
this happen always in my test.

By generating strategies with the following data:

TEST 1
4 weeks (but it could be another data period)
M15
OOS 50%
Method: SQN or Net Balance

the portfolio results are impressive for the OOS 2 weeks. The balance is growing and it seems to have find a profitable portfolio generation approach.

TEST 2
BUT, if you generate only in SAMPLE for the 2 weeks (no until today but let' say until 2 weeks ago), the portfolio result is highly negative, compared to test 1.

Of course I've used the sames acceptance criteria: Test 1 In Sample = Test 2 Complete Backtest.

Same weeks of course.