My system

[RhinoTrader]

Quote from MarkBrown:

you know even a bull system will work in a bear market because a bear market has to rear its head up every now and then just to shake loose the weak shorts right...

so if you have a trend filter that identifies the long term trend ok then you make your system buy on a certain criteria when its a bull market and switch to a different type of buy which would be quicker to enter and faster to get out of longs in a bear market right..

see switches is what you need split the system up so it can do different things based on what the market condition is, dont try and make on thing fit all data.

you could even split up the data so only work on a kick ass bull market model ignore what it does in a bear market and then build a kick ass buying model for a bear market and i can help you stitch it together..

mb

the whole idea is to isolate small portions and perfect that then worry about combining it all. otherwise you will just endlessly be lost trying to curve fit.

Interesting post Mark. This raises the interesting question of whether switches can be effective - obviously your view is that they can.

How did you get comfortable that your switches are robust? My problem with this topic has been that (a) the market does not switch from bull to bear so often (if looking at daily or weekly charts) and (b) the switch introduces a whole new rule-set to my trading. This means that I now I am relying on the fact that my switch rules will continue to work and then my entry / exit rules will also hold up.

My research has shown some interesting and simple switch-type methods for intraday trading which is my main focus anyway. However, for daily stuff, it is oh-so-tempting to filter out the 01-02 period for long-only stock systems. Designing such a filter is easy but achieving statistical significance is not-so.

Would be interested to hear your views.

 

[GTG]

That system looks a lot like curve-fitting. Adding a trend-filter would probably just make the problem worse.

 

[Bo_D_]

hi GTG

do you say this because of the parameters used? e.g. 3 and 4 day moving average? instead of blocks of 5 or 10.

could you be a little more specific?

and ive got amibroker now so once i get the data i can look at 3D parameter peformance charts, which will be a big help.

cheers.

 

[MarkBrown]

Quote from RhinoTrader:How did you get comfortable that your switches are robust?
Would be interested to hear your views.

a simple method is to make a system which counter trend buys and sells in trendless markets do that to perfection and focus on that.

then just add a breakout switch so that the counter trend trade can reverse on a risk parameter when it has been reached.

think buy sell buy sell buy sell and failure "ok' now we have identified the sell failed right so if it ain't going down then its probably going up so if it breaks the channel we buy it. otherwise we set and wait for the next buy.

mb

 

[Hook N. Sinker]

Quote from Bo_D_:

system is as follows-

cross above 3 day high

volatility between 1-6 (ATR10)

moving average angle scan- 200MA must have increased 5(ATR200) in the last 45 days

(lots of other stuff deleted)

I think you have more rules than necessary. You might write a moving average system and compare results. Simple trading systems can work. Try buying if closing price > 100 day average and sell if closing price < 100 day average. Use 100 share lots, don't worry about position sizing now. Use 30 years of price data from some of the older Dow Jones Industrial stocks. Experiment with different averages, say a 50 day average or 200 day average.

I think your equity graph in tcl_system_report.doc is OK. There are flat periods during years 1997, 2001 and 2002 but you did not lose money and not losing is a good thing. Your capital in trades during the flat period years is small. Maybe you made money in money market funds while waiting for the change to a bull market. That is OK.

 

[Bo_D_]

Quote from Hook N. Sinker:

I think you have more rules than necessary. You might write a moving average system and compare results. Simple trading systems can work. Try buying if closing price > 100 day average and sell if closing price < 100 day average. Use 100 share lots, don't worry about position sizing now. Use 30 years of price data from some of the older Dow Jones Industrial stocks. Experiment with different averages, say a 50 day average or 200 day average.

I think your equity graph in tcl_system_report.doc is OK. There are flat periods during years 1997, 2001 and 2002 but you did not lose money and not losing is a good thing. Your capital in trades during the flat period years is small. Maybe you made money in money market funds while waiting for the change to a bull market. That is OK.

thanks hooknsinker,

i dont see how i can have a profitable short term system by just those few parameters? the entry points would have no edge.

i think i might look at simplifying it though. thanks for the tip.

im also trading aussie stocks and dont have U.S. data. do you find it a good practice to test your system on other markets to see if its robust or not?

 

[MarkBrown]

SPYDERTRADER AND JACK HEARSHEY PAID POSTERS!

RIGHT PAID BY ET TO POST!

 

[MarkBrown]

the mop up by the mods has already begun but just remember spydertrader and jack hershey are paid posters on et they get paid to type and thats why they are here and so defensive..

this is where their money comes from not trading..

 

[GTG]

do you say this because of the parameters used? e.g. 3 and 4 day moving average? instead of blocks of 5 or 10.

could you be a little more specific?

and ive got amibroker now so once i get the data i can look at 3D parameter peformance charts, which will be a big help.

First of all, I'm not expert on trading systems, like some of the other contributors to this thread, but here is what I see:

The first red-flag that suggests curve-fitting is the equity curve, I see that it is dominated (especially early on) by a few large gains, like for example what the heck happened in 1999 where you have that one huge jump?

Then when I look at your entry criteria, I see a lot of very specific rules, using a bunch of different indicators that are all derived from basically the same time-series (which is price). This also suggests curve fitting to me as well.

Furthermore, the entry rules look kind of arbitrary. It looks like what you might have done, is picked an indicator and then jiggled the parameters to the indicator around until you found parameter values that generated good trades, and then "improved" it by adding another indicator and doing the same thing over and over again, until you ended up with your 7 different entry criteria. If you’re using in-sample data I would think that would almost guarantees that you would have ended up with curve-fitting.

What are your reasons for picking each rule? What hypothesis did you set out to test when you first began working on this system? A good type of hypothesis would be something like:

"I wonder if stocks with long-term up-trending volatility tend to break out more strongly on the short-term than other stocks?...now I wonder what would be a good indicator to use for measuring 'long-term up-trending volatility’…maybe an upward sloping long-term moving average of the ATR would be a good indicator to use for measuring ‘up-trending volatility’?"

as opposed to hypothesis like this, which would not be as good:

"I wonder if filtering by the 10-day moving average of the ATR would improve a break-out system? What minimum and maximum values for the 10-day ATR will maximize the profits for the system?...between 1 and 3 seems to work best, I’m going to use those values!…I wonder if adding a moving average filter on price would make the system better….hmm if the stock is above it’s 12-day moving average the system gets a lot worse, but the 10-day works awesome and it even gets me out of that really bad trade in march of 2001, so I’m going to use the 10-day MA!..."

 

[MarkBrown]

toby crabel uses a factor or percentage of the 10 day avg true range in his trading and i have systems that use it as well so its not a bad thing at all to use.

the one problem is that if you have an anomaly like a large day it can skew future readings until its out of its system and what i do is ignore any reading outside a 3rd quartile reading.

also you said something that a more complicated method would be needed to make money dont think that way ok be open minded. there are some really simple stuff making money.