Thread: System from scratch!

Ok - so as you know, I've been playing around with a system with a CCI Entry - posted in a few of my other threads.

Well The ADX hook system from scratch has been thrown out - I feel it does not apply to my want list at the start, however the more I test the DTRT entry coupled with different exit rules / money management, the more I like it.

Im currently going through testing with 0.1 lot sizing (so not at the stage of position sizing) However I am testing my exit strategy coupled to that entry.

This is a lot more complex as more has to be done by hand (so instead of using a TP / SL combo - which is a very ugly way to trade.. Reason behind this is - yes you can setup a nice risk / reward ratio - however there is no provision for letting your trades run).

So over the last few hours I have only managed to test 3 months of a year - so in the next couple of days I'll post up the results using a true exit strategy.

If I'm still impressed with the systems - and it's ticking all the personal boxes, i'll then test with the proper money management. And then we'll see what it can truely do.

Stay tuned

Looking forward for yet another system from scratch

Ok, before I go into the results of my latest run - I want to talk a bit more about stops/exits.

First of all.. Repeat after me: MY EXIT AND MY STOPS DECIDE IF I WIN OR LOOSE and HOW MUCH I WIN OR LOOSE BY!

Entry does not make your trades winners, only the exit can do that!

So when we are looking at exits we have to think of 5 possibilities.

1. Large Loss
2. Small Loss
3. Breakeven
4. Small Win
5. Large Win

The first thing we want to do when designing a system is eliminate the first point (i.e- large losses). We want NO possibility of large loss. Now this is definitely one of the trickiest things to set. When you set inital stops too close volatility can take you out of the market even though the price ends up moving where you predicted..

So sometimes using an ATR (average true range) stop is good, as this blows out with volatility. Depending on timeframe and commodity you are trading you might have to use a different multiple of ATR.

So initially when we enter a trade we want a hard stop in place which protects us from a large loss, however gives us room for the trade to develop.

So now we only have the following possibilities: Small loss / Breakeven / Small profit / Large profit.

To cover the breakeven, once the trade is in your favor - the moving of stops to breakeven is a good practice.

Next step is to cover the small profit - how about taking half the trade off at 1.5 times your risk, then adjusting your stop to just below this point. Even if your stop is hit after taking 50% of the trade off, you still end up with a small profit - larger than an average loss.

The last possibility is the KEY. Large profits can be tricky, however I like using some kind of trailing stop. Or use an ATR trailing as this takes into account volatility. So basically after you've taken your small profit, you can let the trade run to grab the large profit.

So as you can see, often just using 1 stop is not enough - often you have to throw together 3-4 stops to setup an exit strategy that suits your system.

If you do not want to spend the time on exits, one way to cover this is to run a trailing stop from the start - it will not be as profitable but its definitely a possibility. Testing will tell if it works for your system.

Ok so the system I have been playing with: Testing your systems... Here!

I went through the task of validating the entry signal: Entry Signals... SECRETS REVEALED!!!

and now finally I have added a parabolic stop. Of course I will be trying different stops to try an improve.

Im pleased as by doing this I have been able to decrease the average loss, and increase the average profit substantially. Draw down is still to a minimal and so is consecutive loosing trades. Additionally the number of trades per year is very nice. (you can find the comparison usdjpy year in the entry signals revealed thread)



This gives us an expectancy of: 8.06 (or 0.22R positive expectancy)

Oh and the other thing to notice - is by letting the winners run on we've been able to secure multiple +10R winners. These make all the difference!

Now to illustrate what I was discussing about stops - the previous run was just using a Parabolic SAR stop, activated on entry (0.02 / 0.02 / 0.2 settings) - it looked like this:



and as the SAR walked towards the price bar by bar, the stop is adjusted onto the SAR dot. It works just like a trailing stop, however accelerates to lock in profits when price goes sideways.

Now i've adjusted the entry so I have a hard stop half the distance of the sar dot and the entry - it looks like this:



The idea behind this is to minimise the losses even more. Also on high volatility the SAR dot starts a lot further away.

The result of this one action is as follows:




Notice the max draw down is less, but also notice the difference in the return.

new expectancy = 9.88 (or 0.30R positive expectancy!)

Thats a nice increase and i'm really happy with this.

Once again - we have not applied any money management. That will show a different picture again!

Just to clarify - the last system test was using BOTH a hard initial stop on entry and also the parabolic sar as a trailing. The goal is the maximise winners and minimise losses.

Ok, so you might look at the equity curve and say: "That looks nice", however how do you know how robust the system is?

So what can we look at to evaluate the system.

Win/Loss

Lets say we have two systems.

System A has an win ratio of 90%
System B has an win ratio of 10%

well you can evaluate this and go: System A is the way to go... however what if the 1 loss was a 100 times your risk looser. You'd blowout big time. And what if system B's 1 winner was 1000 times the risk. You'd be laughing!

So Win/Loss is not a good way to evaluate a system. The next way is to look at expectancy.

Expectancy

(Formula = ((average profit x probability of win) - (average risk x probability of loss)) ÷ average risk amount)

This tells us how much to expect per dollar risked. Now please note the formula I use is using uses the Average loss (which is an actual loss for a trade) not the average risk (which is the money you allocate to loose, not what you actually loose), and doing this I am again a bit more conservative.

So lets look at system A & B again.


System A has an expectancy of 0.25R
System B has an expectancy of 2.5R

You'd be saying system B is the way to go. Well no. This doesn't show the full pictures. How bout if I told you System A traded 2000 times per year and system B trades 10 time. System A is going to be a hell of a lot more profitable in the same period of time.

Expectunity

This is where the next evaluation is made, and that is the amount of Expectancy it has the chance to produce per period of time. This is known as Expectunity. Say system A can produce 200 trades in a month, it would have an Expectunity of 50R per month, System B has 10 trades per month and that gives you an Expectunity of 25R.

This is not a bad way to evaluate a system, however what about possibility of largest draw down?

This is a hard thing to measure as you need a large sample size to evaluate this (10000+ trades) so I don't use this, however I do work out the system quality number. Or SQN.


System Quality Number

What this tells us the variability of a system over a certain sample size. It is basically the ratio between the expectancy and the systems standard deviation. In all my calculations I use a sample size of 1 year.

The formula for this is below:

SQN = (Expectancy / Standard deviation) ÷ Square root number of trades.

From here we will get a number that can be evaluated in the following way:

Less than 1.0: Hard system to trade
1.01 - 2.0: Average system
2.01 - 3.0: Good system
3.01 - 5.0: Excellent system
5.01 - 7.0: superb system (few exist)
7.01 - above: Holy Grail.

When i run systems through my software - I am constantly looking at all these factors. Win/Loss tell me if psychologically I can take the system , Expectancy gives me an idea of average reward per trade, Expectunity lets me know how often I will be trading, and lastly SQN gives me an idea on the overall quality of the system.

Every time I do a run through forex tester, the trade log is transferred to an excel sheet that works everything out for me- It takes me a second to see if i'm heading down the right track.

All the Thoery I discuss here is Van Tharp's work, and I recommend all his products (as many will).

As a matter of interest - this is what the system looks like from the last run through:





So.... where are we going from here? Well I still haven't finished with my exits. Next run i'm going to do will add in a third exit method. I will include: Initial stop (as per above), Modified parabolic sar (im going to slow down its acceleration to try get into longer trends), and im also going to add in a small profit exit (just above break even).

Stay tuned

Now, something of interest. I think that adding in the stop just above breakeven has my changed the system in a way that grabs a lot of wins at the point of breakeven.

44 trades were closed out like this (under 0.5R profit).

This has had a negative effect on the overall position of the system in terms of SQN.

The modified sar has helped 6 trades run on to over 5R a further 8 trades were over 3R, the biggest loser was 3R.

In comparison to the last run: 4 trades over 5R and a further 13 over 3R, biggest loser was 4R.



here is how it looked




The next run I will be doing is with only the reduced entry stop and the modified SAR. I want to see how that looks. Please note all tests are on 1hr USDJPY 2008. I will be walking the test forwards (or to another year) when I am happy with my exit strategy.

Remember:

Entry - Gets you into the market.
Exit - designates your loss / win and the size of your loss / win
Money management - Leverages the system to make exception profits.

We still haven't touched the MM side of this system.

Ok, so the whole intention of changing the parabolic sar setting was to let my winner run a bit more. It has worked and you can tell by the average gain. However its worked in reverse as it has not cut my losers short.

Here is how it looks:





So we need to walk forward test to make sure I haven't curve fit it to a certain year. I'll do that next before thinking about how to bring the SQN up. Im looking for +3.0 SQN.