Directional edge testing

Discussion in 'Strategy Development' started by dom993, Jan 24, 2013.

  1. dom993


    Hi there,

    I am working on a new system, which actually re-uses old parts (trend-identification) in a fairly novel way (for me).

    One of the things I routinely do, in the course of developing a new system, is to test its directional edge. To that effect, I use a combination of Target & Stop of same size (actually, Target 1-tick less than Stop, so that a move of exact same direction from entry will trigger either).

    I then vary entry placement, as well stop/target distance from entry, and look at the win%.

    In the attached graph, the X-axis represents the distance stop/target to entry, expressed as a percentage of a key characteristic of the entry pattern. The only caveat here, is that the exits can be before Target/Stop is hit, in case of a bona-fide exit signal.

    To my very surprise, the damn curve is flat around 56.5% ... this is for a sample set of about 3500-4000 trades.

    This is the very first time I see that ... usually, that curve would have a more pronounced bell curve .

    Can someone shed some light on what this flat curve really means ?

    Thanks in advance
  2. kut2k2


    Not sure what you mean by directional edge. To me a system generates both entry and exit signals. But you're randomly varying the entries just to test the exits alone? Why? Or have I completely missed your point?

    Perhaps a concrete example of what you're doing will help you get the answer you seek.
  3. jb514


    I think he's just testing percent profitable if the targets and stops are equidistant.

    Doesn't it make sense that the chart should flatline? Isn't it charting number of trades and % winners?
  4. jb514


    I think he's just testing percent profitable if the targets and stops are equidistant.

    I think flatline means no longterm directional edge.
  5. kut2k2


    Maybe that's it.

    When I see the word 'directional', I think of directional trading, which is pretty much the same as trend trading. We trend traders don't have profit targets, we let our winnings run. Not sure what type of trading requires both targets and stops but if it works for him, more power to him.
  6. dom993


    Directional edge - I hope this is clear, but anyway, I define it as the win% when using Target-size = Stop-size (minus 1-tick), so that really indicates the % of "good" signals.

    That understood, I currently only vary the size of those Target/Stop (not the entry placement). The idea is to find-out :

    - the average level of noise that follows the entry (stop must be large enough to resist that)
    - the average distance price travels in the intended direction after "good" signals (when the target becomes too far on average, the win% falls back from its peak)

    without any impact from trade-management (both stop & target are set at the entry and left alone after that).

    Obviously since Target-size=Stop-size, where win% peaks is the optimum ... and a win% of 60% translate roughly in a P/F = 1.5, before trade management.

    Now, the graph I presented does include some form of trade management : it actually obeys to continue/flat/reverse exit signals generated by the strategy, on top of the initial target/stop.

    I have since then run the same test w/o those continue/flat/reverse exit signals, and it is worse than I expected.
  7. My turn to say "I may have misunderstood", but I interpret the above to mean that "target is 1-tick closer to entry than stop is". No? Are you assuming something different on account of other unstated slippage/fill assumptions?

  8. what do you mean..."why is it flatlining?" :confused:

    the flat line is the law of large numbers kicking it
  9. dom993


    Correct ... this is just to ensure a move of exact same size either triggers the stop or fills the target. So win% becomes a convenient way of keeping track of the percentage of correct calls.
  10. I'm guessing you're
    1. testing with NT
    2. with a 1-tick/1-second secondary time period
    3. with BetterThanDefault fill type
    4. with some slippage assumptions

    I'm guessing you’re doing something like the following calculation... take, as an example, case of a long trade, with assumptions that target order is a limit order, stop order is a stop market order.

    Say targetPrice = entryPrice + targetPrice
    So, stopPrice = entryPrice - (targetDistance + 1-tick) = entryPrice - targetPrice - 1-tick

    Target order will fill at target price (i.e. entry + targetDistance) once price moves at least targetDistance + 1-tick above entryPrice (but only netting targetPrice per unit traded).

    Stop order will fill at stop price - slippage once price has moved at least targetDistance + 1-tick (losing net targetDistance + 1-tick + slippage per unit traded).

    So the basis of what you are saying is that in either case (target or stop) the order triggers once price has moved at least targetDistance + 1-tick.

    Is that correct up to there? If not, then ignore the rest! LOL!
    = = = = = = = = =
    ... otherwise ... read on ...

    The fact that target and stop moves are of equal size would suggest win rate should be approx 50% (in deference to HurricaneUS, I won’t provide another link here to that paper I brought to your attention in a previous post …).
    I would suspect therefore that the result you are seeing is coming from the trade management activities rather than the initial target/stop.

    Perhaps these trade management activities may have the net effect of shortening your average winner/lengthening your average loser?

    What does NT report about average winner and average loser? If the main effect is the initial target/stop, then avg winner should be abs (entryPrice-targetPrice), and avg loser should be abs(entryPrice - (stopPrice + slippage effect)), and you could calculate the expected ratio.

    If on the otherhand the trade management effects are the cause of the >50% win rate, then this ratio (avg winner / avg loser) would be lower than the above calc would suggest.
    #10     Jan 25, 2013