Sorry for late reply,
was little too busy.
So, I have two questions -
1. Is there a good rule of thumb in terms of '% profitable trades' to look for? Because a higher '% profitable trades' will perhaps come at a cost of a lower 'Avg Win / Avg Loss ratio'. What I'm thinking is perhaps there should be one objective function - say 'CAR / Max Drawdown', and while maximizing that, care should be taken to achieve a certain minimum level of other metrics such as '% profitable trades' etc. Does that sound right? While choosing this one, when I did, I just went for the system that gave the highest RoA, (i.e. Net Profit / Max DD)
Here I will answer with
bookish reply, It depends on your comfort level (
personality and suitabiliy). When I was new to System design and testing, I also wish someone had just given me the numbers
(rule of thumbs) which then I could aim to get and I hated such objective answers. But guess what, after thousands of testing/design/live and countless hours, I have realised there is nothing called rule of thumb ! (if it were so, it would not be the most attractive + most difficult profession of all).
I say personality because your
risk tolerance will be different than mine and different to MF managers to Insurance Companies horizon, I hope you are getting me.
there is no Rule of thumb but one commandment,
Risk is directly proportional to Gain. You must be willing to take more risk (higher DD) for more return (higher CAR).
(higher Risk means higher Risk of Ruin and going out of business).
Some people like higher win% and mediocre payoff ratio (
e.g yours truly) , others will be in b/n extremes.
Again I say
personality/trait is more important because you should have the
confidence to stick with the system during difficult times and it will not come unless you have designed based on parameters you believe in (
personally i will never trade any system with less than 50% win ratio, no matter what the expectancy or payoff or Return on Account).
I am giving you my
magic marks, but I don't want you to go by my numbers , experiment on your own, find your comfort spot (
double entendre, not intentional he he).
These are minimum criteria (
with little flexibility if other parameters are superbly well, i.e flying out of the ceiling)
- CAR/MDD > 4 (generally 5 is my target)
- %win trade > 50
- Exposure < 70% (or else i take that as overtrading)
- Avg Profig/Loss % > 2 (little flexible, depending upon strategy nature)
- Ulcer Perf. Index > 10
- Sharpe Ratio > 1.5
*For other readers too, these are
not benchmarks and moreover could even be over optimist or pessimist values. always use your own judgement, knowledge to derive any standards. Beyond basics nothing is perfectly deterministic
(in trading and in world) and there is always a trade off.
This choice makes the difference.
2. How do you know when a system is 'broken'? Let's say I have tested a system with an out of sample Pft Factor > 3, and >60% Profitable Trades. I could always be your 'millionth man' whose max consecutive losers while live trading exceed the max consecutive losers in the out-of-sample test record. At What point do I decide to stop trading the system no matter how great the historical performance?
any deviation from the backtested result should put you on the edge whether it is
devasting deviation, mild deviation or
happy deviation. If the results don't align after 50 Trades, I reduce the position size and if still not by 100 trades I take them down no matter what. If strategy was performing better than expected then I would try to find out the reason underlying this and incorporate in the strategy.
If strategy starts producing huge losses before 50 trades, I have the
Strategy Stop Loss limit, 50% of portfolio allocated to the strategy. This number is arbitrary and is based on my comfort, it is not devised on any formula or statistical analysis. Initially it used to be
around 30% but with more test my confidence on strategy has increased and i believe in not having very tight stops.
Again, perhaps my selection criteria was over simplistic, but I took that system with those parameters that maximized the RoA. Other systems, which closed overnight positions, and were not susceptible to unfavourable gaps were performing worse in the 2.5 year period tested. I made an assumption that if the system was able to tolerate a sufficient no. of unfavourable gaps in the out of sample testing, then it will be able to do so in the live trading period.
Problem with gaps is it is hard to quantify them, the SD of gaps (if you calculate and plot them over a period) is high and certainly skewed distribution thus
randomness (
read chaos) is high. It hurts specially when trading lower TF. I was once working on overnight system (buy around market close to sell next day morning),
a gap strategy, and
lost some hairs, he hehe
yeah. i thought the exit efficiency was very poor as well. I tried coding in chandelier exit in System 2, which made its performance worse. This (system 1) system has volatility based exits, but I'll try out chandelier exits for this one.
my 2 cents, don't look at stock level
volatility, rather
market level or if you can
sector level, pays good dividend, !
The constraint is not the portfolio size, it is the inability to code and backtest. Tried as I have, I have so far not been able to become proficient in coding all my ideas. I've very often read about the importance of position sizing, and that's one thing I'll definitely want to get into the system.
Our forte is our
system, platform, code, these are our weapons. As they say in fitness training, more you sweat more you lose, here I say, more you sweat less you lose.
(pun intended)
there is no hurry, take your time with mastering the code (
unfortunately will not be able to help here since have no knowledge of Easy Language, if you switch to Amibroker or Strategy Trader, feel free to ask anything)
Discussing all the deficiencies of this system, perhaps it was a bad decision to go forth and trade it live, but it reflected the best wisdom of that time, and sadly it is still the historically best-performing system that I have been able to code and test so far.
.
backtest, doubt all the results, backtest again, live test and when your confidence is high put it to real acid test and even after that don't be dissapointed if it doesn't perform, remember
Journey is the reward.
I think may be my comments aren't making you feel too happy, so for all your serotonin and feel good, a small gift.
http://www.youtube.com/watch?v=MwKYjZ_8EcE&feature=share