A different time for winners and losers, I hadn't thought of that, I'm going to have to contemplate it for awhile
Ok, I've had a think about it. Having a different time limit for winners and losers has a flaw, in that it assumes we can, at any point, classify a trade as a winner or a loser. Whilst this may be roughly possible, it's going to get inaccurate if a trade is hanging about break even, also, I'm sure we have all seen the trade which is a loser all day then suddenly rallies to a winner in the later stages of the day. My approach is going to be the following, the first post attachment shows two PDFs, one for winners and one for losers, the axis along the bottom is in seconds, so the x-axis roughly maps to a single trading session. At any point along this axis we can calculate the probability of the trade being a winner or loser by the EOD by integrating the area under the curve between the current point and the RHS end of the x-axis. So the optimal time to get out will be when the P(winner by EOD) < P(loser by EOD). That's the plan anyway...
So you have a win rate of 80% and a .59 win/loss ratio? So let us say you make 100 trades, 80 of them winners, 20 of them losers. Each winning trade nets you $300 and each losing trade costs you $500 (.60 ratio). That's a profit of $14,000 and a profit factor of 2.4. Did I do the math correctly? Increase your trading capital and you can make serious money. Email me your strategy and I'll optimize it for you
Ok, from the distribution of trades and the win rate, it appears as if you have a non-normal distribution with few outliers. Most of your PnL comes from a stable mean. That's good, but, my concern would be how any new exit condition will affect your original exit condition. My guess is you have a very specific profit exit condition and implementing a basic time stop might shift your mean significantly. Can you backtest a time based stop? Ideally, you'll want to allow the trade to work as long as it has a chance of profit, even it's < 50/50. But, not at the expense of taking a loss. Maybe rather than a strict time based exit, you can tighten exit parameters after time(t) by cutting the trade if goes negative, or already is negative. Something to consider is the volatility of open (unrealized) PnL at each point in time. For example, if this is a mean reversion strategy, you'll find that after the initial entry, the volatility of open PnL will be high, giving you the best chance for profit. After enough time in an mean reversion trade you'll find the price action begin to settle and "flat-line" in terms if vola. These types of trades are strongly reliant on vola for edge, adding a delta(volatility/time) based exit condition might help.
The y-axis represents the frequency at which winning or losing trades closed @ x minutes after being opened. There is some normalizing and smoothing going on as well, it's not the absolute levels which are important, but the relative levels.
I don't really have the ability to back-test this stuff, but I may be able to cook something up a terms of forward testing. Any new criteria is bound to effect the mean slightly, but using the method suggested in the last post I would imagine the stop would be out somewhere beyond 1500 so hopefully the impact would be minimal on the winning distribution, but you are right, I will have to test. Good idea, I'll have to think on it though, I'm a little bit loath to introduce extra parameters to be honest. Given the general nature of the strategy, which you have already guessed at, vola, time etc are all tightly woven and important factors of the trades. I am also going to consider vola, but from a slightly different angle