In this campaign, I let the system take all entry signals, and there are a lot more of them than exit signals - so the win-rate for entry signals is just that, the % of entry signals with a positive exit, where the win-rate for exit signals is the % of exit signals where the total sum of P&L for entries exited at that exit signal is positive.
If my understanding is correct, that your "entry" measurement shows the "%-of-entries-that-hit-the-equidistant-target-rather- than-the-equidistant-stop", then yes, I think there is statistically sound basis for being confident of an edge with the entries. 31k+ trades, average trade > 0 ($67, after slippage, commissions, and conservative fill assumptions), win % bigger than 50% (55.5%) ... that's all good, in my book ... I still don't understand the "exits" side of the measurement, though. These exits are exits other than the "equidistant-target-and-equidistant-stop"?
The 31k trades results are just using all system entries & exit signals, no equidistant target/stop. The almost identical average Win & average Loss is not by design on that run. The system really only generate Long/Short/Flat signals ... depending on the current position, any signal can be an entry signal (either new position or increasing position size), a pure exit signal (Flat) or an exit + opposite side entry signal (in which case, all of the prior position is exited and the new position is entered at minimum size, ie. 1 contract).
Ah, OK, that was a point I'd misunderstood. If the 31k trades use all entry and exit mechanisms (so not just entry signals hitting equidistant targets or stops, but also entry signals being reversed by other entry signals, or being flattened by flat signals, etc), then I don't think you can read as much as I'd initially understood into the validity of the edge from the win % alone. Sorry for the confusion; I'd misunderstood ... I think it's always good to attempt to validate the edge as much as you can; but IMO this is best done outside of any trading strategy (so that you can isolate and follow price behaviour post signal without interference from targets, stops, reversals, other signals, other trades, etc). See following link for a general discussion of how a trading strategy is a difficult system within which to identify an edge clearly and unambiguously http://www.priceactionlab.com/Blog/2011/09/curve-fitting-and-optimization/ I try to distill the signal and then study that in isolation away from any strategy. If that works out OK, then go back and build a strategy around the edge you've uncovered. Initially, I thought this was what you'd achieved with the "equidistant-target-and-stop" experiment. Perhaps you can re-work how you do that?
Abattia, This is quite interesting. Could you please elaborate a bit on this? I have been trying to do analyse the edge of my entry signals separately from the exit signals too. The only way I found so far was published in the book "Way of the turtles" from Curtiss: It consists of plotting the average price trajectory following the entries and to also plot the average MFE (Maximum Favorable Excursion) and MAE (Maximum adverse Excursion) versus time after entry... But I did not manage to use this information yet to build a trading strategy... In addition, I think the actual edge can only be quantified once the exit has been defined. For example, a very short term moving average or channel breakout would probably increase my directional edge but for a very limited time. If I try to capture a very large move, I need to use a long term trend or indicator. A 5 minutes moving average has probably no idea where the price is going to be in one day or one week but can provide some indications for the next minute. Also the entry signal seems to be a trade-off between quality and quantity : The entry signals with the best "directional edge" do not seem to be as profitable as the signals of lower quality. Indeed, higher edge means more filtering and less opportunities to trade... Sometimes it pays more to trade more often lower quality entry signals thant to trade few "high quality" entries. In addition, more filtering applied to the entry means less trades and more noise + higher risk of curve fitting. So far, The only way I found to discover a profitable trading strategy is data mining and backtesting : there must be a better way! The approach requires a lot of processing power, is time consuming and at the end, many strategies are doing nothing but trading noise and fail in live trading...