If ATR doubles - Get Out ????

Discussions about Money Management and Risk Control.
Post Reply
Jake Carriker
Site Admin
Site Admin
Posts: 1493
Joined: Fri Sep 12, 2003 10:32 am
Location: Austin, Texas

Post by Jake Carriker »

Hi guys, I'm a long time lurker, first time poster.

I have heard this notion before, and when I saw it here I took it upon myself to test it a bit. Preliminary results show promise. I tested several versions and found that using a shorter period ATR (say 8-13 bars) and comparing it to a longer period ATR (18-21) bars worked well for me using the, "Get out when ATR doubles." rule.

Using a one day (unsmoothed) ATR needed a larger multiple (3-5 times) of the beginning ATR to be an effective exit trigger.

I tested this rule on a couple of different medium to long term trend following systems, and I found marked improvement in one and modest improvement in another.

I am not declaring this a good idea or a bad one yet, but it may deserve some attention. I will continue to play with it. Happy trading.

Jake Carriker
jimsta
Full Member
Full Member
Posts: 13
Joined: Sun Jun 15, 2003 2:45 am

Post by jimsta »

A doubling of N sounds like a volatility breakout. In 5 years trading that has been my only in signal. If the trades going in the right direction then great, if not get out.
Jake Carriker
Site Admin
Site Admin
Posts: 1493
Joined: Fri Sep 12, 2003 10:32 am
Location: Austin, Texas

Post by Jake Carriker »

As an update to my last post, I have tested several versions of this type volatility exit with several types of medium to long term trend following systems using Trading Recipes.

My conclusions are not surprising. Although an exit like this improves the backtested results of certain individual systems, I find a more robust real world way to improve results is to combine two or more simpler systems in order to smooth the equity curve.

This allows each system to retain its "pure" trend following nature by avoiding cutting trades before the system's "natural" exit point. It also allows systems that trade different timeframes or utilize different trend identification techniques to have an additive effect when traded in tendem. By that I mean that the performance metrics of the systems together have better performance characteristics than each individual system does. I notice that as I add the volatility exit and / or other "cutesy" trend filters and such, the performance of multiple systems traded together tends not to be better than each individual system.

All of this reinforces the notion that a good, robust trading methodology can be built around combining several simple ideas that work, rather than trying to build the perfect system. Of course, I don't intend for this to be the final word on this at all. I am merely sharing the results I got, and my interpretation of them. Other traders will get different results and interpret them in different but equally valid ways. Happy testing, happy trading, and happy New Year!

Jake
Chris67
Roundtable Knight
Roundtable Knight
Posts: 1052
Joined: Tue Dec 16, 2003 2:12 pm
Location: London

Post by Chris67 »

Jake,

I think your'e dead right in what you said.
I would like to run more than one system to smooth out equity. At the moment ive got , approximately , a 20 day breakout system and a 100 day breakout system .. what I would dearly love to do is have a much shorter system running off of rsi's on a 60 min or 250 min chart.. do you know how I can test this ?any useful platforms ?
Jake Carriker
Site Admin
Site Admin
Posts: 1493
Joined: Fri Sep 12, 2003 10:32 am
Location: Austin, Texas

Post by Jake Carriker »

Chris, I primarily use Trading Recipes, and it does not support testing on timeframes smaller than daily. Therefore, I cannot really make any suggestions regarding applying systematic strategies to intraday timeframes.

I am sure there are others in the forum that test intraday on Tradestation, Amibroker, Wealthlab, and other, possibly homegrown, custom testing platforms. Since I don't use them, I am not sure all of those platforms support testing on intraday data, but I think they do.

Jake
Jake Carriker
Site Admin
Site Admin
Posts: 1493
Joined: Fri Sep 12, 2003 10:32 am
Location: Austin, Texas

Post by Jake Carriker »

If one did want to use ATR as a measure of volatility for the purposes of liquidating a trade when volatility passes a certain threshold, one way to "normalize" ATR would be to simply divide it by price. That way ATR is always expressed as a percentage of price. No matter what the stock price, if volatility has not truly expanded the ATR/price value will not rise strictly as a function of the absolute value of ATR.

Not that that makes this trading rule a good idea, but it solves the problem of ATR not being a good measure of true volatility if price changes dramatically.

Jake
MarkH
Roundtable Fellow
Roundtable Fellow
Posts: 53
Joined: Fri Dec 05, 2003 10:24 am
Location: Corpus Christi, Texas

Post by MarkH »

Jake -

When using backadjusted continuous futures contracts (in contrast to stocks), I think you would agree that you need to be careful using ATR divided by price since the historical prices are not accurate but the ATR measure is. The price inaccuracy is greater in some contracts than others. It may be legitimate to use this as a very rough volitility filter in limited cases, but keep in mind that even in the backadjusted contracts which have a lesser degree of price inaccuracy there is nevertheless some degree of inaccuracy in all cases.

Mark
Jake Carriker
Site Admin
Site Admin
Posts: 1493
Joined: Fri Sep 12, 2003 10:32 am
Location: Austin, Texas

Post by Jake Carriker »

Hi Mark,

Since I was putting forward an idea that might help solve the problem proposed by King_Tiger, but I don't use the idea personally, I guess I did not think through all the ramifications of testing the idea on backadjusted contracts.

It seems to me that the ATR values from backadjusted prices would be based on those prices and be in proportion to such prices, just as today's ATR is based on today's price and is in proportion to it. Please explain how an error would be introduced by using backdjusted contracts. I guess I don't understand what you are getting at.

Perhaps there would be errors if the price resolution was not fine enough to represent trading activity in an issue, but that is generally a problem inherent to stock data rather than futures. The Veritrader documentation shows an example of this in Worden Brothers stock data. When a stock splits multiple times and the smallest trading increment is a penny, at a certain point the trading activity will be distorted.

However in the case we are considering, if a crude oil contract has a backadjusted price of $2.37 as opposed to it's actual price at that time being $24.68, the ATR in either case should be the same X% of price. I would not dream of calculating ATR based on the actual historical price and then divide that number by a backadjusted value. That would be quite counterintuitive.

I must be missing something somewhere, so I apologize in advance for my ignorance. Thanks for your more rigorous treatment of the subject.

Jake
MarkH
Roundtable Fellow
Roundtable Fellow
Posts: 53
Joined: Fri Dec 05, 2003 10:24 am
Location: Corpus Christi, Texas

Post by MarkH »

Jake -

Crude is a good example because the backadjusted prices have the additional complication of also going negative as well as the problem of the backadjusted price level being very different from the actual historical price.

In any event, the 1990 ATR of crude that is calculated on the continuous backadjusted futures contract that includes all of the data through today (2004) will be exactly the same as the ATR that is calculated on the 1990 actual contract. This is so because the range of prices on the daily price bars are not affected by the backadjusting method. The absolute price value does change. Therefore, the ratio of ATR to price level will change as well.



Mark
Jake Carriker
Site Admin
Site Admin
Posts: 1493
Joined: Fri Sep 12, 2003 10:32 am
Location: Austin, Texas

Post by Jake Carriker »

Point well taken Mark.

I fired up Unfair Advantage and checked several examples using crude. I used a 21 day ATR as calculated by UA. The ATR does seem to stay proportional to price for actual contract data and backadjusted contracts that are raised to prevent negative prices, thus giving an actual price close to the historical price.

Example:
Backadjusted raised price series for 01/02/1990 : Close = 26.88, ATR = .46, ATR/Price = 1.71%

Actual contract data (March 1990 contract) for the same date : Close = $22.41, ATR = .42, ATR/Price = 1.81%

I would propose that this small difference is primarily due to the fact that CSI only calculates ATR to 2 decimal places, but wait...

Backadjusted contract NOT RAISED to account for negative prices. (Same date):
Close = -3.44, ATR = .46, ATR/Price = Don't even bother.

So, you are correct. Good work. No wonder I don't use that method :) .
MarkH
Roundtable Fellow
Roundtable Fellow
Posts: 53
Joined: Fri Dec 05, 2003 10:24 am
Location: Corpus Christi, Texas

Post by MarkH »

Jake -

Good example. Note that CSI simply adds an appropriate round number to the data series to lift the entire series above zero. For crude, my program is adding $30.00. It could just as easily add 40, 50, 60, etc., and I have seen the amount added change from time to time as the continuous series is rebuilt with new data over time. Obviously, this affects the denominator in the ratio too.

Also, I think the reason you got the diffence in ATR (.46 on 1/2/1990 for the continuous contract and .42 for the March 1990 contract) is that the appropriate comparison would be to the February 1990 contract since it would be the front month in the continuous series on 1/2/1990.

Mark
1tudor
Contributor
Contributor
Posts: 4
Joined: Tue Jul 22, 2003 10:56 pm
Location: New York City

If ATR doubles - Get Out????

Post by 1tudor »

I have recently started to think of this topic as well....

My thinking goes down this path:
If your initial position size is 10,000 shares of XYZ and the ATR is 1.....fine. XYZ goes parabolic and the ATR is now 2.

ATR can actually benefit you when it gets "juiced"; however, you go from trading 1 unit to now having 2 units on.

If I was trading for my own acount, you can accept the increase in volatility, however, when trading outside money, I believe that selling off half the position would be the most prudent path. I have yet to backtest this idea, but I'd love to hear others ideas.

Thanks
Post Reply