Medium to Long term system development

[tech/a]

Just a few things.

Can I have a look at the Stats?
Have you delayed the entry and exit by 1 day?
What is your universe of stocks?

Thanks.

 

[doctorj]

Have you limited your universe of stocks? I've just copied the formula into amibroker and deleted a couple of zero's to reflect my data being in dollars and tested it over this year (-4% net profit,36% Max DD) and from 1/1/2004 until Friday which didn't show anything much more encouraging (3.35% annualised return, 33% winners of 18 trades executed with a max DD of 46$).

Testing back to 2000 is a little more promising, but is primarily the product of two trades. See stats for testing back to 2000 below.

 

[Milk Man]

Try this (attached). There was something wrong with it but I don't know what. Just tested at about 55% pa 1/1/00-26/8/05

Tech-
The universe is just the xao- uncut. All the universe mods I made didn't seem to work any better.

I set buy and sell delay to 1. This just buys next day on open and sells next day on close.

 

[doctorj]

Have so far been unable to replicate results anything like yours. I'll keep plugging away.

 

[tech/a]

Ok a few things.

The Reward to Risk Ratio is only 1.25 this is very low 2.5 is the expected minimum.I trade 3 systems the worst with 8. best 12x.

Try the method using 50K rather than 1 million.

Also try various dates 2000-2005,2001-2005 etc.

It is possible (And Montecarlo Analysis will bear this out) That the Standard deviations are very wide and there maybe of 5000 tested portfolios 1000 that win and 4000 that lose---you have a winner--if you traded it on another set of dates it may be a loser.

I'll also try and code it to Metastock and see how it goes there.

Good work though its got to stand all tests so dont think everyone is ganging up!

 

[doctorj]

Good point. Loakglen's system will only benefit from the rigors of thorough testing.

Tech, most of us have studied techtrader and its evolution in detail via the threads at reefcap. Could we convince you to share the other systems you trade for educational purposes?

 

[tech/a]

Doc.

I really do want to have some things for me.Ive spent 100s of hrs and dont wish to give it all away.After all the best learning is that which you experience.

But I will say that one is a weekly system which is a hybrid of the method Steve adopted on Reefcap.

The 3rd is another longer term method.
All are based around Margin Stocks thats what I trade.

I'm yet to find a medium term method which stacks up.
By that I mean a stock trading portfolio type method.
I really think if your going to trade in this manner you should be trading Futures.(Medium or short term).

I also like the idea of trading one or 2 favorite stocks heavily.
I also think we trade with blinkers---we should be trading the bourses which are out performing---not just the ASX.

 

[Milk Man]

I'm seeing even better results with 25% positions. Ive also altered the min position size, to account for initial drawdown possibilities, to $1350 and the max s/price of stock to $50. Getting bad return on 1/1/2004 to present and 1/1/2005 to present. The main issue seems to be cash managment.

I have tried portfolio stop buy and sell criteria with some success but I felt they were optimized too much for the periods I was testing.

Since the reward/risk ratio's are so low the cash management has to be spot on. What it boils down to ATM is "do I trade my free capital or is it not enough for a new position".

PS- tech, my data is in cents so portfolio size is 1 million cents not dollars.

 

[tech/a]

Loak.

I'm trying to code ami to M/S so have a few questions---most is OK.

What is the code for Priceboundcheck?

What does Null mean? (Zero?)

Think I'm OK with the rest.

Ta Muchly.

 

[doctorj]

"PriceBoundChecking" - if set to False - disables checking and adjusting buyprice/sellprice/coverprice/shortprice arrays to current symbol High-Low range.

IIf(C>O,Sum(C-O,60),Null);

Translated:

IF Close is greater than the Open
THEN Pos is equal to the sum of (Close - Open) for the last 60 days
Else Pos = null

Null is to have no value - null does not equal zero.

 

[tech/a]

Thanks doc.

Metastock when using the IF function is a little different.

I should be fine.

Thanks

 

[Milk Man]

Sorry for the delay guys, first day off at home since last attempt. The joys of dairy farming!

I'll attach a copy of what i'm up to. The % gains are down but the drawdowns are way down. This leaves the door wide open for those wanting to gear. Keep in mind that data is in cents (ezycast sucks! do not subscribe!) so dollar gains are cents and liquidity is cents etc.

Please pick it to bits. One thing i'm wary of is the 'universe'. The sooner we get something to use the sooner we can test it in real time; eliminating the crystal ball effect that using XJO or margin lists might have on backtesting.

Has anyone else done anything? Not making any headway or found something too good to share ?

 

[Milk Man]

Please find attached the results, formula and settings for this puppy. Sorry, I didnt realise you guys couldnt view all this on previous post . I'll also attach all the watchlists in the amibroker formula. If you import them into amibroker and number them 10-21 from oldest to newest it will work straight up. Again, remember data is in cents so youll probably have to knock the last two zero's off the numbers that are too big. Ezycast was bought by Bodhi so my data should be fixed soon anyways.

Let me know if your numbers arent stacking up the same as mine- but ill be buggered if I know why they have been in the past.

 

[Milk Man]

Here are the amibroker files for the watchlists. "jan00" corresponds with watchlist10 in the formula. So you can start watchlist 10 by adding a common stock to all watchlists in there (BHP or something) then importing the unzipped jan00 file. I use the "whole lot" file to test from instead of "all files" (heaps quicker) so just save it to a watchlist thats not 10-21 and select it. You dont have to do it this way but otherwise youd just have to change the watchlist numbers in the formula.

Good luck.

PS- You all owe me a beer; that watchlist crap took me ages on the S&P site.

 

[Milk Man]

Heres some more: they wouldnt all fit.

 

[Milk Man]

Heres the last of em. PHEW!
:

 

[Artamon]

G'day all,

Has anyone been forward testing this strategy since it was posted back in November???

Cheers,
Andrew.

 

[tech/a]

For those interested there is some good work happening here.
Both Amibroker and Metastock code.

http://lightning.he.net/cgi-bin/suid/~reefcap/ultimatebb.cgi?ubb=get_topic;f=15;t=000264;p=6

Different to what was discussed here but promising in the medium longterm.

 

[doctorj]

Have been following the thread, but as of yet been unable to replicate their results on my end.

Anyone else following?

 

[Milk Man]

Should have returned about 5% since it was posted. My data was corrupted so it wasnt actually as good as was tested initially. Forward testing is the most important part IMO to check the formula isnt "data mined", to use nick's expression.