Maurice’s Journal

[LeeDanello]

AKA Moccha.
Hello everyone. I’ve just finished Section 2 Module 18 of the course and have coded up the Amibroker boiler template. You couldn’t find a better template to base your system on. I think I’ve finally come to grips with the looping code although one can never be too sure. There is where the new debugger will help. All I need to do is to incorporate a customised ranking system in there. Been reading Andreas Clenows book Stocks on the Move – he certainly doesn’t pull his punches and then I’ll give Short Term Trading Strategies That Work By Larry Connors, Cesar Alvarez a crack and if I can fit them in, I’ll try Building Winning Algorithmic Trading Systems by Kevin Day and possibly Adam Grimes book. I need ideas! Even if I skim read them and use something to start of with it might get the ball rolling. It doesn’t matter where they come from just as long as I can validate something!

[Nick Radge]

Kevin Davey is big into optimisation in single markets so tread carefully.

Try Ernie Chan: http://epchan.blogspot.com.au/search/label/Strategies

I also think he has a few books: http://www.amazon.com/Ernest-P.-Chan/e/B001JSEBVG

Connors and Alveraz have been done quite hard, especially their RSI(2) strategy. There have many versions around – I think I have 5 or so

Craig has done quite a bit of study on Adam Grimes stuff – have a chat with him.

Clenow stuff is good. I think I have about 4 versions of his rotational system. Lengthening his parameters will perform better (which is what he probably does)

[LeeDanello]

Ok I might skip straight to Adam Grimes book. I thought Kevin Daley was a bit different given he used Tradestation and the fact that he did pretty well in the Trading World Cup. So he’s in the Larry Williams mould. He pushes his own course pretty hard. Looks like my reading will be multi tasking. Will probably print out Tony Crabel’s book that is in the depository and have a look at the bar patterns in that and try and use that in conjunction with some other stuff as a stepping stone in building am mean reversion system. Seems like everyone has got one of those!

[Nick Radge]

I came across another site that uses single market systems – I’ll try and find it for you. That guy made the argument that there is nothing wrong with data mining patterns. As I say in my notes, it’s my personal opinion rather than a law that must be followed. The one thing I would suggest, which is also what Howard Bandy suggests, is using a system stop loss of some type, like an equity curve trail stop or some such.

[TrentRothall]

Nick Radge wrote:

I came across another site that uses single market systems – I’ll try and find it for you. That guy made the argument that there is nothing wrong with data mining patterns. As I say in my notes, it’s my personal opinion rather than a law that must be followed. The one thing I would suggest, which is also what Howard Bandy suggests, is using a system stop loss of some type, like an equity curve trail stop or some such.

Is the stop loss in case the system gets totally out of sync with that market? So you would stop trading it for example

[Nick Radge]

The issue is that the system is not robust – the variance of a large sample of data is not available. Therefore we really can’t know how stable the system really is. The more sample, the more idea we have of stability. The less stable, the less idea. So, you may have a system that ‘appear’ good on a small sample, but in actual fact that brief period is just a good run within a very poor system. You need a way to stop the rot.

Howard Bandy uses metrics, i.e. if tests show a maxDD of 10% and you hit 12%, the you stop. The equity curve stop does the same thing. Theory then states that should the system stabilize you can start trading it again.

[LeeDanello]

Finished the theory side of things. Now the hard part – putting into practice. At the moment it seems like I have information overload. I think I need to take a step back and find out how to assimilate the theory.

[LeeDanello]

I’ve been working on the mean reversion system for the last few weeks. Probably similar in type to others on here. I thought I had some reasonable results except the drawdowns were a little high for my liking closer to 20% even though the MAR was about 1 or slightly higher – payoff ratio was just under 1 which was also an issue as was the profit factor which was slightly less than 2 and win rate was about 60 to 65%. The problem was 2008 and even though the index filter would help I found using that reduced the CAGR so I’ve been trying to work a way around not using an index filter. Seems like the stretch is the way to go (thanks TrenRoth), but found with that it was1 step forward and 2 steps back. In other words some of the entry conditions became superflous and bloated. So I’ve had to strip the code back and find out what was working and what was not. It’s an interesting excercise, one tweak of a parameter here and there or an introduction of a condition can make the code look good but really underneath it’s probably flaky. Mind you I’m doing this on the XAO and I suspect things may be different on the S&P. The other issue I’ve found is trying to push the exposure rate. Initially it was 30% with highish drawdowns and then with the addition of a few conditions drawdown goes down but then so does exposure. Getting it back to 25% with a maximum 15% drawdown is the goal with a MAR over 1 would be good.

[ScottMcNab]

I have really struggled to get a system with metrics that I would want to trade on XAO. The frequency of trades with SP1500 seems to really help CAR. It may be worth trying some of your older versions on the SP (as they are) just out of interest ?

[Nick Radge]

Maurice,
The other thing to do is tighten/loosen the index filter which will help the trade flow. Looser will allow more trades and tightening will allow less.

[LeeDanello]

I’ve run a Monte Carlo on my mean reversion system which is based on the RSI(2) and 3 consecutive lower closes on the historical All Ords constituents and came up with the following results

These are the results of a walk forward over the same period. Some divergence in 2012

Equity curve is moving up in the out of sample so that is a positive.

Here’s the code if anyone is interested in playing around with it. No ground breaking secrets in here. Haven’t tested it on any other markets.
I used 10 days for the short MA and 150 days for the long MA. These were the only variables that I tried to optimise. These seem pretty stable.

[StephaneFima]

Thank you Maurice. Impressive!

I did not checked the code but just tried to use it “as is” (I just deleted the first few lines dealing with ASX) and I turned off the Price Filter.
From 01/01/05 till today, the CAR / MDD I get are the following:

– S&P 500 => +2.60% / -23%
– Russell 1000 => +3.30% / -31%
– Russell 2000 => +6.11% / -37%
– Russell 3000 => +2.91% / -37%
– Nasdaq 100 => +6.80% / -15%

So, without any further MC analysis, it seems that your system is still profitable on those markets. However, CAR is quite different than on the ASX…
I also tried to test the S&P 500 universe with an Index Filter on the $SPX at MA(100), but strangely it lowers significantly the CAR (+0.5%). MDD is -21%

[LeeDanello]

That’s interesting. I might have to get some US data and tweak some parameters to improve those results.

[LeeDanello]

Got some US data and ran the system with the same parameters except the price filter which had a minimum price of $5 and a maximun of $100 and a 10 day average minimum turnover of $1,000,000. Numbers that came out weren’t too bad

[LeeDanello]

Just got myself a new computer and in the process of getting all my stuff from the old computer to the new one. Problem is the new one is 64 bit and the old one is 32 bit so restoring the image probably isn’t the way to go. So slowly slowly getting Amibroker and Premium data loaded. It’s a painfully slow process. It’s been a few weeks since I was doing any stress testing and now I have to go back over the notes and find out what I need to do after the Monte Carlo. Will have to make some notes of the procedure of stress testing. Need to set up a cookbook method.

Maybe

1. Single run backtest
2. Monte Carlo
3. Walk Forward In Sample out of Sample
4. Rolling Analysis
etc etc

[Author]

Posts