Looks like a pretty standard relative strength with datamining approach, not at all very unlike how a lot of screens - including the ones we have here - are developed. It's basically following the trend of recent strength (momentum), hoping it will continue (which it does until it does not), with extra spice to get rid of the picks that make the strategy look dumb. Again, it is not uncommon to develop strategies this way, but there is lots of additional methods one can use to minimize the probability of a pure datamined result, like using the out-of-sample testing, more frequent data (to get more data samples), longer backtests and more.
You also need to be very vary about graphs that shows great performance back in time if they do not state what period is "in-sample" (datamined) and what period is out-of-sample (typically post-performance).
Looks like a pretty standard
Looks like a pretty standard relative strength with datamining approach, not at all very unlike how a lot of screens - including the ones we have here - are developed. It's basically following the trend of recent strength (momentum), hoping it will continue (which it does until it does not), with extra spice to get rid of the picks that make the strategy look dumb. Again, it is not uncommon to develop strategies this way, but there is lots of additional methods one can use to minimize the probability of a pure datamined result, like using the out-of-sample testing, more frequent data (to get more data samples), longer backtests and more.
You also need to be very vary about graphs that shows great performance back in time if they do not state what period is "in-sample" (datamined) and what period is out-of-sample (typically post-performance).
Keelix