The Flaw in Trading System Reasoning...?

[capitalMan]

the belief that system will stop working is one of the main reason why some system will continue to work.

when a system stops to work, ppl stop using it, and the competition becomes less and system start to work again. and of course, as soon as it start working again, everybody jump on the bandwagon, and as the competition increases, it stops working again. the cycle continues, and the system drift in and out profitability.

In the end, the money lies in the marginal revenue that the system gains as it drifts in and out of its profitability cycle. as in econ 101, even in perfect competion(efficient market), a firm still earns a "normal profit".

it's a good idea to stick to your system for a very long time...

 

[judgejimmy]

There are robust systems that work forever.
If you are able to spot the most undervalued stocks with hard work and reasearch, your will always be a winner.

The problem is, that the best working robust systems are the most difficult to trade psychologically. Furthermore you need to apply money management rules.

I designed robust systems that have returns over 1000% annually, but have drawdowns over 90%.
Nobody trades stuff like this, because nobody can stand the heat, even you could get 100% with a 9% drawdown, if you apply a fractional size strategy (10% Invested, 90% Cash).

The problem is everybody looks for a system that gives the perfect capital curve with no drawdowns. Thats the wishfull dream that breaks your trading!

 

[AMT4SWA]

Do two things to keep your systems robust (the ones that are working for you now), at the end of every month re-backtest your system with the previous months data added in. Every month that your system trades then go back and add this new period to your overall backtesting data, but remove one month (the first month) of your previous backtesting data set. This way your total time period of backtested data remains the same but you are continuously adding the current trading environment to your data base. The next thing to do is have your system adjust for current volatility (for instance, use a "volatility based" trailing stop for any portion of your position that is remaining after your initial profit target is hit). Find ways to adjust your overall system to the current period of volatility (mean values for 30/60/90 days etc) for stops and targets....or adjust the position size of each trade to the current volatility values.

These are just some of the things my trade group does that seem to work well....hope this helps.

 

[instantrader]

We all have to be green at some point.

However for the short term trader the first lesson is to decide to go it alone because there is no other way.

If you dont have a history of doing things alone,then you will not make a short term trader.especially not a daytrader..


How do I know?

I traded the SP from the beginning,and then once i realised that all charts look and behave the same way,I worked out FOR MYSELF how id use this.

Since the time I also started to share my skill with others about very simple ways to read price and little else, I watched those who succeeded and those that are still wandering around in chat rooms (and in some cases running their own chat rooms as resident guru..the blind leading the blind)

The first lesson to becoming a trader, pick one market one time frame, turn off all distractions and LOOK at the bars(or candles ) for one month..

The alternative is to be drifting around in here for 6 losing years(the average time it takes to finally quit or master it the long way)

This month I watched two 3yrs losers gain control of the mini chart and not lose on any one day,after doing the 1 month test plus a few days of pointers to kick them off,They also made longer term scalps on crude and can dollar both falling..do they come to my student room now..no way!!

This post will find a load of smart alec repsonse, but if you sit around in here during market hours look in the mirror before you "smartass me".

 

[Holmes]

Quote from AMT4SWA:

Do two things to keep your systems robust (the ones that are working for you now), at the end of every month re-backtest your system with the previous months data added in. Every month that your system trades then go back and add this new period to your overall backtesting data, but remove one month (the first month) of your previous backtesting data set. This way your total time period of backtested data remains the same but you are continuously adding the current trading environment to your data base. The next thing to do is have your system adjust for current volatility (for instance, use a "volatility based" trailing stop for any portion of your position that is remaining after your initial profit target is hit). Find ways to adjust your overall system to the current period of volatility (mean values for 30/60/90 days etc) for stops and targets....or adjust the position size of each trade to the current volatility values.

These are just some of the things my trade group does that seem to work well....hope this helps.

Having worked in IT with artificial intelligence I can tell you that this is a risky aproach and will one day bite you in the back. There is a very fine line here to be walked in regards to curve fitting.

You will need to split the data up: one piece for development, one piece for testing and another piece for checking real time functioning. BUT... when you arrive at the Out Of Sample (OOS) phase you are no longer allowed to go back! (othersie you are still curve fitting!)

Once developped a system: you tune until xx time back and then check it over the OOS. If fine, then you can trade it. If not, then you'll be better off not to use it at all.

Hope this helps,
Sherlock

 

[Remiraz]

Quote from judgejimmy:

heat, even you could get 100% with a 9% drawdown, if you apply a fractional size strategy (10% Invested, 90% Cash).

9% drawdown, 100% returns is DAMN BLOODY GOOD!
Its double your cash every year at almost no risk!!

Quote from Holmes:



Once developped a system: you tune until xx time back and then check it over the OOS. If fine, then you can trade it. If not, then you'll be better off not to use it at all.

How far back to go when backtesting is also impt, me thinks.

Too far back and we risk adding market conditions that does not exist today.
Too little back and we're curve fitting.

 

[AMT4SWA]

The "automated" system adjustments made (with newer data added in) have always maintained or increased the systems yield curves. It might be totally wrong from an AI IT basis but it is working very well in the real world??? Who knows why....as long as it keeps the systems robust. Interesting points you bring up though....I have read about and seen an "automated trading" systems CD video that talked about what you are talking about also.

 

[Holmes]

Quote from AMT4SWA:

The "automated" system adjustments made (with newer data added in) have always maintained or increased the systems yield curves. It might be totally wrong from an AI IT basis but it is working very well in the real world??? Who knows why....as long as it keeps the systems robust. Interesting points you bring up though....I have read about and seen an "automated trading" systems CD video that talked about what you are talking about also.

Aircaft Mechanic Technician for SWA,

I can understand your reaction. After I did no longer trust AI in trading I build a 100% mechanical system for the ES based on 3 years and 3000 trades.

It was "walked forward" one month at a time, doing what you did. Add the latest month, drop the first month. Used 1 year of data to "tune" the parameters. It exactly lasted three months and then went into a severe drawdown and I was fortunate enough that I pulled the plug in time.

It has not recovered.

It basically stopped working a month after the Iraq war started and I am suspecting that the market changed due to that a number of people must have been "flat" at the beginning of the war and then developped new systems. (time on their hands) One may look at the "spoofers" and automatic order placements that seem to have gained momentum from that time onwards.

I have gone now one step further: I do no longer rely on indicators that need "tuning". I have only one indicator that I use, but the setting could just as well be 3, 5, 10, 50 times as long and it would still work.

Respectfully yours,
Sherlock

 

[jbt]

Quote from Socrates:

The true flaw in backtesting is thinking that there are a myriad of systems. In fact there are only two: reversal and continuation.

no better words said.
I guess that's why you are socrates.

 

[Holmes]

Quote from jbt:


no better words said.
I guess that's why you are socrates.

No, there is only one. They are two faces of the same coin.

I leave it to yourselves to work out why that is so.

Sherlock