If your using EFS, enter the highest return you've had...

[Doji3333]

Update...

For all of you that are interested..

I have just completed the conversion of my new EMINI trading system from BT mode to RT (auto execution) mode - but I still have to test it on Monday.

The results still continue to impress me. Here is the BT results for the last 2 days (which have been rather congested with a narrow range).

I'm including a $12.50 slippage for all trades (this adjust the entry price by -0.25) - even though I'm using limit orders for all entry and PT orders. This BT is using 4 contracts ($50 per point * 4).

See for yourself..

B

[derekg]

Too aggressive?

Brad,

Your system does look impressive, but would you agree that $10K may be insufficient starting capital? Looking at the history, there were a number of $2k drawdowns and one over $4K or 40%. A few of those in a row could wipe you out.

Seems like a good system but with a high risk of ruin. Dropping to one contract would seem more realistic, but I suppose my pain threshold isn't as high as others. Even at 1 contract a 20-25% ROI ain't too shabby.

Have you been able to backtest 120 days (or longer?) If so, I'd be interested in seeing how it did. My biggest gripe with ESignal is their extreme lack of intraday data.

Derek

[Doji3333]

Starting Capital..

Derek,

You are correct. $10K may be too little for the test I ran with 4 contracts - but enough for only a single contract.

That is why I'm going to test it in RT next week. If the results of the RT system are similar to those of the BT system, then I might be on to something..

Remember I keep telling everyone TEST TEST TEST - well, I'm at that stage now.

B

[augustus]

Brad,

many thanks for the answers.

Stephan

[Doji3333]

Status Report

Fist day of testing...

At least 3 bugs in the code (Doh!!). Fixed them all and we'll see how the system does for the rest of the day.

One thing I did notice though is that the LIMIT orders are working properly and the system is capable of working within these contraints.

I have added a user variable called EntryOffset (defaulted to 0.0). But in the case the user does not want to wait for a fixed limit order to get filled, the user can enter an Offset for the limits.. This would potentially help the system get filled quicker.

Anyway, back to more testing..

More results later.

B

[soylent]

Brad (and all) --

Great thread!

I liked your discussion of profit targets vs. 'letting profits run'.

You guys have built some amazing EFS's. I've never been able to create such a smooth equity curve. Usually I find the market goes into a 'funk' that last a couple weeks where my system won't work followed by a couple weeks of bliss, and so on... I've tried to build filters but it seems, at some point, I over-fit and WHAM it stops working completely.

Anyhow, I've built many things that work great on 30days data but never on 6months worth. Sadly...

But, I have found a) Take only the perfect setups, b) Profit targets rule.

-c

[Doji3333]

Status report...

Well, today was one of those days.. the system lost $1800 trading four contracts... (just like my other tests). But, it showed a very important factor to this system...

when it is right, it hits about 70+% accuracy. When it is wrong, it goes wrong HARD. Now, after a detailed analysis of the BT results, I believe i have found the solution (or a potential solution)... A 3 LOSS LIMIT. After three losses (on any given day), the system will shut down.

Doing some manual retesting, there are the days that showed the most losses..

5/7/04 - lost $476.39 without loss limit... MADE $296 with the three loss limit

5/13/04 - lost $2188.00 without loss limit... lost $996 with the three loss limit

5/13/04 - lost $1811.19 without loss limit... lost $696 with the three loss limit

Now, the downside to this addition is that it will probably reduce the profits on some days. From a manual scan, I can see that there are 2~3 other days in the past 20 days where this rule would have shut it down early.

One option I might try is "three losses before 12PM EST". This option to the original rule may help the system let profits run and shut-down days where the system is not doing well (n the morning). All three of the LOSS days (mentioned above) have this characteristic - so there might be some merrit to it.

Anyway, the RT code (after many minor revisions) appears to be working well now. It is executing the LIMIT orders perfectly and the trading system logic appears to be nearly perfect. Will do more testing tomorrow.

I'm glad some of you are ejoying this thread - this is what I go thru every day. Others here will tell you that building a complete system is not an easy task. There is alot of work and thinking that goes into it.

The way I look at a day like today.. They're going to happen and all we can do is put into place mechanisms to try to contain losses.

More tomorrow.

B

[Doji3333]

Sorry...

That second 5/13/04 should have been 5/24/04.

B

[spike500]

Re: Status report...

Originally posted by Doji3333
Well, today was one of those days.. the system lost $1800 trading four contracts... (just like my other tests). But, it showed a very important factor to this system...

when it is right, it hits about 70+% accuracy. When it is wrong, it goes wrong HARD. Now, after a detailed analysis of the BT results, I believe i have found the solution (or a potential solution)... A 3 LOSS LIMIT. After three losses (on any given day), the system will shut down.

Doing some manual retesting, there are the days that showed the most losses..

5/7/04 - lost $476.39 without loss limit... MADE $296 with the three loss limit

5/13/04 - lost $2188.00 without loss limit... lost $996 with the three loss limit

5/13/04 - lost $1811.19 without loss limit... lost $696 with the three loss limit

Now, the downside to this addition is that it will probably reduce the profits on some days. From a manual scan, I can see that there are 2~3 other days in the past 20 days where this rule would have shut it down early.

One option I might try is "three losses before 12PM EST". This option to the original rule may help the system let profits run and shut-down days where the system is not doing well (n the morning). All three of the LOSS days (mentioned above) have this characteristic - so there might be some merrit to it.

Anyway, the RT code (after many minor revisions) appears to be working well now. It is executing the LIMIT orders perfectly and the trading system logic appears to be nearly perfect. Will do more testing tomorrow.

I'm glad some of you are ejoying this thread - this is what I go thru every day. Others here will tell you that building a complete system is not an easy task. There is alot of work and thinking that goes into it.

The way I look at a day like today.. They're going to happen and all we can do is put into place mechanisms to try to contain losses.

More tomorrow.

B

Hi,

if it goes wrong HARD than it means to me that you trdae against the trend. Have you included a trendfinding part in your system?
Trendfollowing will never let you take 3 consecutie losses. That's also a way to limit losses.

I'm daytrading since 1990 on a Tradestation. Just to give you an idea of my level of "experience".

Regards

[shogun]

many wise words spoken here, it seems to be the curse of the efs that what works in back test will not work in real time, my experiences are very similar to soylent , they work great over a month but when you start going back 6 mths the kill you dead, the ones that seem the most profitable over the long run are countertrend but i could not stomach the monster drawdowns,

so all eyes are on brad,

[spike500]

Originally posted by shogun
many wise words spoken here, it seems to be the curse of the efs that what works in back test will not work in real time, my experiences are very similar to soylent , they work great over a month but when you start going back 6 mths the kill you dead, the ones that seem the most profitable over the long run are countertrend but i could not stomach the monster drawdowns,

so all eyes are on brad,

To me there are several rules to follow:

1. establish the trend
2. never trade against the trend
3. define your stoploss before you trade
4. never optimize your system by backtesting
5. always follow the rules of your system
6. your system should be selfadaptive to the trend


regards,

spike

[Doji3333]

Dang....

In the spotlight - eh??

Well, I'm still testing the RT code. It is doing what I want, but there are still some minor issues to clear up.

Also, I've created a version based on Spike's comments that I will test today... The test with this variant shows less trades and about the same returns and the first version. One thing I tought of trying it to MERGE the two versions (thus allowing for short-term trades to be taken while a long term trade is on).

As a side note, it took two losses this morning, then took 4 profitable trades since then.

One comment regarding system and how they work... It is very unlikely that you will find a single system that works well all the time. There will be drawdowns or periods of time where it fails (I call these "Market Phases").

The way we resolved this for the hedge fund is to have three different systems that all work together. Thus, if one or two fail for a period of time, we'll have the other(s) to diversify our trading attempts.

More later,

B

[soylent]

Hi Spike --

This thread just gets better!

Ok, so, lets talk a bit about possible scenarios.

TREND systems:
A) What is a trend?
I define a trend as anything above/below the 34 period EMA. W/ a 10 (simple) as the 'momentum' line. When above 10, we have momentum.

So, when you take that trend, it works GREAT (10 turns up) MOST of the time! You can trade nearly any timeframe w/ that setup. Until.... The dreaded oscillations. Then, we start chopping back and forth over the 10.

So, how do you block out those scenarios?

I've tried making WIDE stops so we can chop big over the 34 and I don't look for a setup until 10 cross 34 and we break the xover price. But, when that holds true, those are the worst whip-outs because we will immediately oscillate above the 34 for a big stop loss.

And, just when you have given up all hope (5 or 6 10pt drawdowns), it all comes together but you sat out because you know those buggers are just trying to trick you.

So, you try to add ADX or some other 'trend-icator' into the system. But, those are so lagging that by the time you DO take the trend, it's over. So, now you wait for a retrace to hop on the next leg only NOW it reverses & you get blown for another 10pts.

So, you look for a low-lag indicator (e.g. CCI). CCI is very nice for entry/exit triggers. But, as you study CCI...

That's when you discover:

COUNTER TREND!

Yes, that seems to be where the best bets occur. Look @ AGET -- EWaves.... Basically, they look for extreme conditions with or w/o the trend and take those.

So, essentially, I've found:
a) Trends are nice when they work. And, they are awesome on DAILY and WEEKLY timeframes. But, on highly leveraged intraday stuff...
b) I just watch those big 20+pt all day rallys that painfully move up 1 pt every 15mins and wait until they falter and then I hop on and pickup 15pts in 30mins as all the money gets dumped back out in a counter-trend 'blow out those stops' move...

Anyone else have thoughts on this?

Maybe my definition of TREND is wrong? Certainly you could make a case that TREND is anything above/below the 3EMA (that actually works well, too, I've found, if you're nimble)

-c

[gspiker]

Hi Brad,

Well, today was one of those days.. the system lost $1800 trading four contracts... (just like my other tests). But, it showed a very important factor to this system...

That was bound to happen. Murphy says the first time you take your system live you will get hosed, and hosed even worse if people know about it.

It's even more likely if they backtesting has been limited, but in reality - it will not matter. You can backtest for 20 years and see great results, but when you go live that will be the single worst day for the system.

Its good to see you forward testing the system though, some issues with backtesting just will not be seen until you simulate it live. You can make a system that will backtest near 100% accurate and is all but untradable in RT.

Good luck!

Garth

[gspiker]

Shogun,

many wise words spoken here, it seems to be the curse of the efs that what works in back test will not work in real time, my experiences are very similar to soylent , they work great over a month but when you start going back 6 mths the kill you dead, the ones that seem the most profitable over the long run are countertrend but i could not stomach the monster drawdowns,

I don't think this is EFS specific...it has to do with developing systems and backtesting vs RT trading results.

The big problem (which is now slowly being fixed) is the lack of intraday data in eSignal. But even then, backtesting and real trading are different beasts and sometimes results will differ no matter how much you backtest.

btw, there was an interesting similar thread to this on the ethinos group if anyone wants, I can post a link to it.

Garth