Time Based Exits - the ideal exit strategy?

Discussion in 'Automated Trading' started by 4XQs, Apr 16, 2009.

  1. 4XQs


    Even though I've traded for what feels like eons, I've spent some free time the last couple of months putting together automated strategies in MultiCharts, and it's great fun. Whilst experimenting with various types of exits I've been thinking...

    There are many reasons for trading with a stop loss; keeping losses small can be one of them, avoiding a catastrophical loss can be another. The latter I'd say would be related to the amount of leverage applied to your position (or insanity like being long C last fall, even unleveraged). Ultimately, it depends on the system of course - what kind of movements you're trying to capture.

    But if you can stand the swings, I find time based exits without a stoploss to be the ultimate exit strategy - with a few assumptions, of course. So, let's say you're trading a portfolio of 10 uncorrelated systems with no (or very controlled) leverage, why wouldn't you move the catastrophic loss far, far away where it (hopefully) will never get triggered - and rely on a time based exit?

    I hope someone can provide some interesting talking points on the subject, keep in mind I'm a system developing newbie - although I've owned the Ralph Vince books for 10 years.
  2. Yes - stop assuming that an arbitrary choice must be ideal
  3. MGJ


    If you like it,
    if you've tested it,
    if you've compared it to all of the Standard Exit Strategies favored by the Orthodox Conventional Wisdom, and discovered that you like it best of all,
    if you have great confidence in your testing approach,

    then why not indeed? You don't have to please anybody except yourself; and if YOU like it, who cares whether other people like it too? You're risking your money; do it your way.
  4. 4XQs


    lol guys, just what I was waiting for :)

    To elaborate a bit further: One of the ways I measure the quality of an entry setup is to chart where the trades are at after n bars. Sometimes this "edge" (if you will) disappears after a certain time and it (for sure) becomes random. Thus, you'd want to exit before that happens.

    In retrospect, I think I'm just advocating that time based exits can have their place in addition to, sometimes even replacing a normal stoploss.
  5. Your will blow your account sooner or later if you don't use a stop loss.

  6. heech


    To me, this smells like curve-fitting.
  7. I am more unconvinced that just picking one and saying it is best has any value. As another poster above basicaly said, IF YOU TESTED THEM ALL AND IT IS TRUE.

    Otherwise, it has no meaning.
  8. ej420


    Interesting comment. My intuition tells me that a time-based exit should actually lead to _less_ curve fitting. Think of it this way: when you are using more complicated exits such as volatility-based stops, and profit targets, you are basically looking for certain spikes/dips in your data (since those are what triggers the entry/exists), effectively reducing the size of your dataset, and making it more likely that you would overfit. So, basically I'm saying that from the perspective of curve fitting, the simpler the exits, the better. Any thoughts on this?
  9. If you want to study entries in isolation, you can plot a curve of a metric based in 1,2,3,...n day exits. For example, a curve of CAR with days on the x axis and the CAR value for each exit on the y axis. You can then compare this with curves from other entry methods or even random entries. If you see a clear trend in your curve but none for random entries it's an indication that you may have found something useful and it also gives you an idea of how the "power" of your entry method changes over time. If you're comparing multiple entry methods just be careful the metric you choose will work for that purpose. Net profit would be a bad choice for example.

  10. 4XQs


    Thanks for that, yes that's how I'm thinking as well. I usually look at Return on Account which takes into account the drawdown as well. But there's so many factors to consider that I'm finding it very hard to be confident that a system works - even with walk forward testing.

    What I'm doing to boost confidence in the systems is to split the data into three sections, and making sure that the systems perform at least OK in all three. However, since I'm working with intraday data (I use 1-min data for "accuracy" but work on higher time frames like 5m-12m-20m-30m) I find market internals changing too much from the different periods. Thus you end up with systems able to work well in say volatile markets and then drain and die in ranging/trending markets.

    Someone posted that you need the stoploss or your account will blow up, but I have to remind you that if you're trading for instance FX and don't use leverage you're not going to blow up even with at 1500 pip move against you.

    I've also found that realistic slippage is super important, and can kill the most promising of systems - forcing you to higher timeframes.

    You can probably tell from my posts that I don't have a math/stat background (unfortunately). I don't trade my systems yet for what that's worth - I'm a discretionary trader that's started to play with MultiCharts.
    #10     Apr 17, 2009