Originally Posted by CreditViolet

Yes, if stop size is based on account equity risk then either its impossible to set stops using the SAR method and if one does use SAR for stops then one will have to ignore account equity risk many times. Can't have your cake and eat it too.

Absolutely correct scientifically,however on cult & emotion scale this may be wrong.

Originally Posted by Saurabh

Good question , reminds me of a group trading on 60 min frame, and always in the market. Hmm... you mean that SAR tells when one is right or wrong... For me each trade is a new one... my stop price may not necessarily indicate a good entry for reverse position. Again, often I do not use the same size while reversing.

So you are actually Proposing a New Set of Rules for 60 min Flow ?

I will add my observations from a statistical point of view.

Analyzing where to place stop-losses depends on the characteristics of the trading strategy. If a strategy is discretionary, then stop-losses should be discretionary as well, but portfolio level draw-downs and stop-loss (max equity risked) can be fixed and well defined. If entry and exit criteria are discretionary, using mechanical stop-loss will not make the results predictable (it can profitable, but not predictable).

My approach for a non-discretionary strategy is to first back-test the strategy without (hard) stop-losses. This provides me with the natural characteristics of the strategy - purely based on the underlying indicators (e.g. RSI based system etc).

Next (sample screens), I take a look at the MAE (Maximum Adverse Excursion) tables. In an ideal world, it should exhibit classical bell-curve distribution - with one or two "fat-tails" or black swans for extreme events. If the MAE distribution is not bell-curve, there will be very little value addition trying to optimize stops. If there is a high distribution towards the extreme, it is virtually useless to identify good stop-loss levels - on could rather trade in opposite direction of the signals with a better chance of success.

Then, I study the MFE (columns) and MAE distribution matrix. This shows the distribution of potential profits and possible worst-case losses. (These are not where the trades were exited, but the maximum draw-down and profit while the trade was open). Looking at this, the objective is to take profits at MFE levels without getting stopped out at MAE levels. A greater concentration of trades in the center to upper-right area is desired (greater MFE smaller MAE - i.e. greater reward/risk ratio).

Based on this data, I then approximate my stop-levels. Notice that in the example, nearly 55%(36.36+18.56) of the trades have an MAE of less than 2%. This means, if I place a stop at 2%, the W/L ratio is 55%, and my winning trades have to make 1.65% to break even. Likewise, nearly 45% of the trades have a MFE >= 2% and MAE <=2%. This makes the system at break-even in the worst case. But, the system may experience draw-downs from time to time, and a sequence of few losing trades in a row may erode my confidence. Last, but now the least - this is back-tested and not walk-forward tested. So, 2% stop-loss may not be optimal when trading in the future.

A signficant mistake that traders of mechanical systems make is too much reliance on back-testing and lack of walk-forward testing.

Let us say, I have back-tested a strategy from say 2000 to 2008, over 9 years and 500 trades.
Based on my analysis, if I come to a conclusion that 2% stop is ideal, I would be fooling myself. The right approach would be to divide the trading data into at least 4 subsets, re-analyze and perform walk-forward testing.

So, I must take 2000-2001 records, identify what would be an optimal stop-loss. Then, I must test the 2002-2003 range using this stop-loss value and see whether the system performs (or blows up). Now repeat this using 2000-2003 (you may find a different value for optimal stop-loss), and test it on 2004-2005 and see whether the system survives, succeeds or blows-up.

Irrespective of the methodology you come up with to identify optimal stop-levels, unless it is reasonably successful (and consistent) in walk-forward testing (and optimization) it is unpredictable and perhaps risky.

Using specific methods like SAR, ATR, Fixed % of entry etc are irrelevant - their utility varies from markets to markets, time-horizon etc. Eventually whatever methods you use, their effectiveness will be reflected in the diagnosis (bell-curve distribution) and the result.

If a trading sytem has fundamental flaws, not amount of optimization of stop-losses will help you. Perhaps a better approach would be to investigate whether trading the signals in the opposite direction would be profitable ?

Last but not the least, all trading systems have a shelf-life. If a trading system has approached its expiration date (or has expired), however hard you try to identify optimal stops, it will be a worthless effort.

Attached Images

	sl_mae_mfe_matrix.jpg (65.1 KB, 53 views)
	sl_stats_mfe_mae.jpg (96.5 KB, 65 views)

Good points

Let me add to this point - A signficant mistake that traders of mechanical systems make is too much reliance on back-testing and lack of walk-forward testing.

Yes, the mistake they make is by not studying the market data itself before beginning the system design process.

__________________
Adapt,Overcome,Improvise

Any Site Issues/Problems? PM Me

An e.g. of MAE/MFE plotted against the Profit/Loss of trades

__________________
Adapt,Overcome,Improvise

Any Site Issues/Problems? PM Me

Adheer,

An excellent article,thks,infact an insight for members like us who are unaware of how a System is made ,tested & accepted .
But here again you are mostly on 'Models Of Mkt' and you have slightly inclined towards those QAnalyst who believes Mkt is Structured.

How about i make a Random Entry then what would be my Stop ?

As you are well aware than many of us here ,that Models Of Mkt are generally measured and perceived in two ways
1) Model #1—Dollar change
2) Model #2—Percent change

Here the QAnalyst starts with these broad assumptions:=
a) Change in price follows a probability distribution.
b)Each day’s change (Xi) is independent and identically distributed.

Can there be a better 3rd Way ? than these so in ALL aspect my Stop determination becomes more Definitive irrespective of which camp's QAnalyst i may belong to,is my Quest.

Asish

hi
what i am trying to say is
first and most imp is to identify a trend
how the trend can be identified; every one has its own method to find the trend
trend line
mov averages
Elliot
profiles (as now in season) etc
many indicators can be used to identify the trend no matter small/ Minor/ inter/ major!!
once u identify the trend up / down and u place according to that and if Ur identification is rgt u will only hit stop profit
there stop loss is still assumed just to safeguard abnormals
but what i think before it will hit the abnormal values it will hit Ur stop profit

now the stop, it varies lot in intra/ minor/ inter/ major

in 5 min trading set up obviously we will not carry position unless we are comfortable in 60 min and the scenario outside our mkts

in identifying the trend i ma still novice in learning process!!
Kf
i still lay stress on knowing what the trend is neg / pos and the time frame

__________________
The world can only be grasped by action, not by contemplation...The hand is the cutting edge of the mind

Originally Posted by CreditViolet

Good points Let me add to this point - A signficant mistake that traders of mechanical systems make is too much reliance on back-testing and lack of walk-forward testing. Yes, the mistake they make is by not studying the market data itself before beginning the system design process.

Jesse,

If i remember correctly the reason for adopting Neural or Genetic Algorithm was introduced by you 2 yrs back and we had few discussion on these aspects.

Asish

Originally Posted by joy

WoW how did you knew what i was feeling !!!!!!!!!!!