drawdown and number of losses in a row are derived from 1&2. But you are right, there are a lot of parameters, and the question is certainly not easy to answer, given interlinked parameters. But don't you use some kind of benchmark / performance measure?
One of my worst enemies are my own false assumptions
Trend following : > 3
Mean-Reversion : < 2
( Trend following ) Trade % Profitable : 50% +
Its also very related to the drawdown. Hows your comfort level and could you handle 6 months of drawdown when running a trend following strategy ?
Using the calmar ratio is probably better than any of the default ratio's supplied with ninja for example.
Minimum 9 years backtest with 3 years of out of sample testing (that confirmed should exhibit similiar traits). Personally i set the bar as minimum 4 for trendfollowing however i have 1 strategy that trades maybe 3 times per year and is at a PF of 10-12 with 50% odd's of success.
At present i only have trend-following strategies live and my live report states my calmar ratio runs at 2.89 sortino 1.92 . Drawdown period of 118 days in last 450 or so days. Positive periods of 49.13% 50.87% negative periods. A max loss of -7.63% (in 1 day) A max gain of 8.51% (in 1 day). Max drawdown of 16.29% % Return over 14 months = 32% or thereabouts.
Id expect my calmar ratio to go lower when i add mean reversion techniques etc. Id also expect my backtest and out of sample test to be off by about 20%.
It also depends on the timeframe used if you watched al brooks webinar last week you could see he was stating a 1:1 risk ratio on 50-60% on a 5 minute chart not a style i would really like as its such a low timeframe that there is a lot of noise there but it obviously works for him!
The following 2 users say Thank You to leinster for this post:
The strategy I'm currently running live was in a forward testing stage for 6 months. 6 months was not something anyone recommended or a goal I set or based on any metric. That just happened to be the point where I was satisfied that my strategy was performing bug free and doing the same in live trading as it was in back testing. I have others that have been in forward testing for longer and will be indefinitely. Not sure if they will ever go live even though they are doing ok. Just not doing as well as my live strategy. Oh yes...even though i have strategy running live, I still consider it in a forward testing mode which will continue for full life of the strategy.
First thing, I'm concerned about is the average profit per trade after commission and slippage. The bigger the better but for example, my current live system averages about $63 per trade. I'm comfortable with that. Should probably track that on an on-going basis.
In development and testing, I want the equity curve to be as smooth as I can get it and be as close to a 45 degree angle as I can get including costs, commission and projected slippage. Unless you are building a high frequency trading system, those equity curve characteristics indicate all the other performance metrics will be sufficient, imo. Look at the other performance metrics to see if anything sticks to see if they tell you something about your system that will make it less than desirable for you watch in a live environment. Beyond the equity curve, look at draw down. Plot the draw down for the entire testing period and see what it looks like. Can you financially and psychologically withstand those periods?
I look at all the other performance metrics but I don't dwell on them unless there is something that sticks out.
This may seem oversimplified but that's the way I like it and it seems to be working well for me so far.
The following 6 users say Thank You to MWinfrey for this post:
The most important thing is to make sure you're testing all possible scenarios, especially the worst possible ones (often the ones people dismiss and don't test, because they don't think it can happen to them). I noticed the OP qualified their statement as "depending on timeframe" which implies they havent' done this, or that they have and are only reporting the "good" results and ignoring the "bad ones" which of course does not work in real life because you don't have the luxury of ignoring any results.
In my opinion, drawdown should be less a matter of comfort and more a matter of how much it threatens your ability to continue trading. Make sure it doesn't get to that point and in some sense it doesn't matter how much drawdown you incur. Strive to quantify it as much as possible through backtesting and in "what-if" scenarios (the more outrageous, the better).
This needs no restating for experienced traders, but for the benefit of those who may not have thought about this, the goal of paper trading is anything better than a breakeven strategy after commissions and transaction fees, while the goal of real trading is anything several times better than the interest rates you get at a bank. A successful paper strategy is basically something that proves it is not breakeven. A successful real strategy is something that makes the risk worthwhile.
The following 3 users say Thank You to mwtzzz for this post:
I said depending on timeframe as your stop relative to the timeframe will change depending on your target and initial stop and entry.
I actually use time frames in and around what i settle on for live and the reason for doing this say you have 45 minute bars as entries i would test 30 minutes and 1 hour timeframes and the reason is this can be a candidate to prove whether curve fitting has been done.
If you dont factor timeframe alongside your stop you will probably be stopped out by generally noise in the market or are placing ur stops in an easy to judge / wrong place.
Faking bar's at lower timeframes is very easy and well done by the bots!