Ken, I think you're taking too much advantage of your hindsight. Consider the same Timely chart: 207.61.23.99
On Tue, May 30, the 6th bar at one time was well below the moving averages and would've likely given a false MAX signal. Looking at the hard right edge at that time, the stochastics were toppy and OBV had level off. Would you have shorted?
The same thing could've easily happened on your May 31st example. The price might've dipped only to return, although the OBV and stochastics were dropping at that point.
These and other examples look very different when staring at that chart in real time. Others on the same chart:
The 7th bar on Monday, June 5, may have presented a long signal that didn't confirm. Same with the 6th bar on Thursday and 1st bar on Friday. This week, if any, would've been the week to mechanically trade the system rather than anticipating the signals. (My scanner was going nuts this week finding signals that never confirmed.)
The problem with backtesting, as I see it, has to do with order execution and fill price. In my particular test, the machine assumes a fill at the closing price for the period which is not always possible. What if the bid ask spread were a point and the closing price reflecting a seller at the bid? What if the price was running hard when the signal came? A fill at the ask would've been tough. All I want to see is relative performance among stocks, I don't realisitically expect my results to be the same as the backtests assuming I had actually played the signals I'm testing.
BTW, I'd still really like to try MAX with fibonacci retracements for stop points but have a lot to learn first.
Sam |
