Some of the best tests can be synthetic data tests. If your system can survive the rigors of stringent synthetic tests that push the limits of every possible scenario, your system might survive and even thrive in the real world. Of course, if the system passes these synthetic tests then it should be able to past historical data 'backtests' with flying colors.
Could work but only if statistical properties of original series are preserved in synthetic data otherwise the result could be misleading.
very good point about length of testing a system . i have tested a swing system for a year and dat trading for 6 months and works great. any shortening would hurt
and guys, backtesting can't do curve fitting, but only optimiser does as backtest, parameter values are finalised and set, and can't change like optimiser do you all agree?