Drawdowns

[theasxgorilla]

What development platforms are forum members using?

Amibroker, love it.

 

[Nick Radge]

Howard,
There are numerous inputs beyond price and volume that can be used. Think quant traders as an example. My example of Campbell was they use different tools. Their returns are risk adjusted and based on $11 billion and they are commodity/fx based. I agree in context about what you say about commodity trends, but I disagree with stock trends.

If you know of, or find, trend following systems that work for the majority of common stocks, please let us in on the techniques

I'm not going to reveal my system but perhaps you can dissect tech/a's TechTrader for the general population. However, why don't you select 10 common stocks - preferably at random - and I'll run my system over them and report back. This is not a pishing contest - I just disagree with your statements.

Nick

 

[tech/a]

dissect tech/a's TechTrader for the general population.

Please go ahead.
I personally would like to know why it works when it evidently shouldnt .
Particularly as I have traded it and 2 similar.

 

[nizar]

Would it be an accurate statement to say that the proportion of those trading mechanical systems (as a proportion of all the market participants) is far greater in the futures market than it is in the stockmarket?

Maybe this is why mechanical systems such as the Donchian channel breakouts have lost their touch in the futures market (because of more people using the same or similar system) but still a goer in the stockmarket?

 

[Chorlton]

Hi Nick --

I agree that simplistic systems, particularly when applied to long term holding periods, are unlikely to work well. They are too easily implemented. If it turns out that one is trading profitably, there are a lot of people who will notice, begin trading that system, and quickly remove the inefficiency that system recognizes.

Out of interest for someone new to System Development, is this always the case with ALL Mechanical Systems or only those which are built upon specific rules such as those designed to look for inefficiencies in the Mrkt??

Why will they always be ultimately untradable if many people look for similar buy/sell signals and trade them accordingly??

Afterall, wouldn't a system based on "solid" trading rules work over the long-term regardless of how many people where trading similar rules. For example. look at TechTrader. The system itself is very straightfoward and I'm sure that there are many systems out there which are looking for similar set-ups. Yet, it has consistantyly performed well over the last 4years.

Surely if many systems are identifying the same stocks on the same days for trading purposes, then (as long as we follow our systems without question & tampering) then this will surely only help to increase the chances of those stocks moving in our favour?

I'm probably missing some very obvious point here but would welcome comments....

Thanks,

Chorlton

 

[tech/a]

There is an abundance of reasons why a system wont work.

Very little on why a system or systems WILL work.
I find it strange that more is written/discussed on the former.

Academics tend to turn systems testing into Voodoo/Rocket science.
It aint that hard.
Same is obvious in discretionary trading,from what I have seen.

I guess I'm a very lucky builder!

 

[nizar]

Surely if many systems are identifying the same stocks on the same days for trading purposes, then (as long as we follow our systems without question & tampering) then this will surely only help to increase the chances of those stocks moving in our favour?

Well if several systems were to buy the opening price of the same stock on the same day, and as such the buy side would be stacked when the market opens (as it would open much higher than if few people were going to buy the stock) then extra consideration must be given to slippage.

But we can incorporate this into our trading simulations with tradesim (market order instead of default order).

Though it shouldnt matter a great deal if you are dealing with liquid stocks (liquidity filter) and if you are a small player (which i am, but hey you gotta start somewhere )

 

[Chorlton]

Well if several systems were to buy the opening price of the same stock on the same day, and as such the buy side would be stacked when the market opens (as it would open much higher than if few people were going to buy the stock) then extra consideration must be given to slippage.

But we can incorporate this into our trading simulations with tradesim (market order instead of default order).

Though it shouldnt matter a great deal if you are dealing with liquid stocks (liquidity filter) and if you are a small player (which i am, but hey you gotta start somewhere )

I agree Nizar, slippage would potentially be an issue but as you say this could (to some degree) be accounted for when backtesting.....

On a seperate note, I'd like to thank all for their replies on this thread and others regarding System Development. I have learned a lot simply by reading these posts....

Cheers...

Chorlton

 

[Temjin]

I use metastock/tradesim.
Amibroker i did consider it, but decided the formula language would be too hard to learn.
We dont all have degrees in computer science

Man, Amibroker looks so much easier than MetaTrader, which is the development platform I am using. It's lack of any simulation features but learning the ability to code such features allow me to be more flexible with different softwares / capability of transfering systems in the future.

Though it shouldnt matter a great deal if you are dealing with liquid stocks (liquidity filter) and if you are a small player (which i am, but hey you gotta start somewhere )

This is also one of the reason why being "small" has an edge over large players because of the liquidity factors.

I tend view the entire market as a pond of liquid gold that spills some of its content out regularly due to inefficiency. A large player will try to capture all the spilling by the bucket load, while as smaller players, we can take the spilling by the droplets without upsetting the whole "scavenging" so much. And droplets might seem small, but can still remain a very lucrative business when the pond of liquid gold we are talking about are in the orders of trillion of dollars every day.

This would explain why certain simple systems may still work because it is not employed by the bucket boys who hire professional traders/IT geeks to develop quantum mechanic based alorgorim trading systems. (i.e. banks/large financial insituations)

My beliefs anyway. (and out of topic too, sorry)

By the way, love the discussion here, keep it going.

 

[theasxgorilla]

Please go ahead.
I personally would like to know why it works when it evidently shouldnt .
Particularly as I have traded it and 2 similar.

I'll pre-empt a response from Howard and have a go:

* Start date bias
* Leveraged exposure to a strong and protracted bull market
* Non-cap-weighted exposure to the market leading to index out-performance
* Price filter encourages taking positions in what are likely to be lower liquidity issues with inherently greater growth potential
* Discretionary entry trigger over-ride may have added some performance edge

ASX.G

 

[stevo]

Greetings --
If I am trading a group of common stocks, the group must have been selected by some process. Selection of tickers to include must be subject to the same procedures as selection of parameters for a model. The system is a trading model that will be applied to all issues in the group, plus the price and volume series for the issues in that group.

Focusing on the system that is made up of a model and a group of common stocks. During the process of developing the system, I will see that some days there are more buy signals than I have funds to cover. The system must recognize that and handle it as part of the system. There must be some mechanical way of assigning a score to each individual signal, then taking positions in those stocks with the highest scores.

AmiBroker does this in a two step process, all within one system. The model is run over all the tickers in the group. Say the group consists of 20 stocks, each a candidate for a signal to buy, sell, or be flat. I decide ahead of time, and build into my system, how many positions I wish to hold at any given time; say it is five. Assume I am trading using end-of-day data and taking positions at the next day's open. At the end of each day, I run the system. If there are sell signals for stocks I own, those will be sold at the next open. If there are buy signals for stocks I do not own, say there are seven buy signals, each individual signal will be assigned a number called its PositionScore. If there are positions available after the sales, say three, then orders will be placed to buy three stocks -- the three chosen from among the seven are the three with the top PositionScore values. The system knows how to do this and all I see (unless I want to see the deeper details) are the Sell and Buy orders for those tickers that the system chooses.
www.quantitativetradingsystems.com

Howard
My approach is to test a system on a large number of stocks using PositionScore set to random. The system has to give positive results for all portfolio runs in a Monte Carlo simulation - any of the outcomes in the backtest must give results that I would be willing to receive. Then I might use PositionScore to rank the trades to be taken.

But if the system requires PositionScore to be considered tradeable then I will not trade it. I suppose that I could test how robust the PositionScore value is - something that is worth pursuing.

I have traded a long term system for 5 years now where if I have 5 trades but can only take 2 then I use my discretion to select the 2 I want to take. At times I have flipped a coin. Over the last 5 years I have made 153 trades using this approach - so I guess it is fair to say I have walked the system forward. I have even tested different, randomly generated stock universes to see what sort of variation I may get with different universes, because the universe I trade in the future will not be the same as the universe I tested in the past.

Using random, or subjective judgement, to make the final stock selection is not really similar to varying the values of a moving average in the buy and sell criteria - as long as you know what the probabilities and potential outcomes are.

Is it possible to take whatever criteria you want to use for PositionScore ranking and incorporate it into the buy signal? For example if ROC(C, 1) is what is used to rank stocks in PositionScore then determine a cutoff point below which the trade is not worth taking and use in the buy signal. This would reduce the number of buy signals potentially to the point that there are no extra buy signals that need to be scored.

I think it would be true to say that I am as close to an investor as a trader can get - I have trades that have been running for 3 years, although I have also cut trades within a couple of weeks of taking them. Many of the big winners go in excess of 1 year and my average hold time is 9 months. I also don't even look at daily charts, let alone intra-day charts. I only use the weekly time frame and I don't use leverage.

These types of time frames and holding periods are probably quite foreign to a futures trader. But it works for me and the basic principles and techniques used to develop a trading system are the same.

I look forward to reading your book - I am sure that it will get me thinking judging by your posts to this forum.

thanks
stevo

 

[tech/a]

I'll pre-empt a response from Howard and have a go:

* Start date bias
Unknown at the start date

* Leveraged exposure to a strong and protracted bull market
Thats the way it was designed it performed as expected.

* Non-cap-weighted exposure to the market leading to index out-performance
Hmm dont know that anyone would weight their positions to capitalisation---do people do that?

* Price filter encourages taking positions in what are likely to be lower liquidity issues with inherently greater growth potential
The universe approximates the ASX 300,greater growth is excellent seems then to be within the system.

* Discretionary entry trigger over-ride may have added some performance edge
May do but it still performs within the maximum parameters returned in Montecarlo testing.
ASX.G

Even so I dont think this explains WHY this system succeeds.
It only points out some conditions of the system and past conditions of the market which have contributed to its performance.Many other systems have failed even during this bull market.

The market conditions of the last 5 yrs are exactly the conditions the system was designed for.
As far as I'm concerned job done!

 

[theasxgorilla]

Even so I dont think this explains WHY this system succeeds.
It only points out some conditions of the system and past conditions of the market which have contributed to its performance.Many other systems have failed even during this bull market.

The market conditions of the last 5 yrs are exactly the conditions the system was designed for.
As far as I'm concerned job done!

I forget one last element, the tweak in the entry trigger that tries to capture only breakouts that are forming higher lows and highs improves the entry efficiency or edge ratio or whatever you want to call it. In other words, fewer breakouts ought to be false breakouts.

There is also an edge apparent in something as simple as the position sizing ie. not oversizing into risky or slow moving positions or undersizing into those stocks with the potential to make the greatest difference to the system's returns.

* Start date bias
Unknown at the start date
Understood. A linear regressed line through the equity curve might help address this.

* Non-cap-weighted exposure to the market leading to index out-performance
Hmm dont know that anyone would weight their positions to capitalisation---do people do that?
Heard of an index fund? Mutual funds look a lot more like the index than a trend following stock system with 10 - 20 equal sized positions that also avoids the ASX50 altogether, by design. You bet people do it, and it's a hand brake on performance. Therefore, not doing it is an edge and IMO a reason why a system like TechTrader works.

* Discretionary entry trigger over-ride may have added some performance edge
May do but it still performs within the maximum parameters returned in Montecarlo testing.
Does this mean you helped walk the system through a superior equity curve path as it was traded forward? Perhaps you could Monte Carlo the traded period of TechTrader and compare the public result with the distribution of CAGR and MaxDD observed during testing.

I'm still not really sure what you're holding back tech/a...do you have a why that is unclear to everyone else? The system is on the table. I don't see anything magestic about it beyond the things I've already pointed it.

 

[howardbandy]

Greetings --

Stevo said in one of his postings: "But if the system requires PositionScore to be considered tradeable then I will not trade it. I suppose that I could test how robust the PositionScore value is - something that is worth pursuing."

PositionScore is just a part of the trading system. It can be included or excluded, as you wish. If it is excluded, AmiBroker will use a default process that functions as PositionScore just as though you wrote it that way and intended it that way -- see my note later in this post.

When I am using PositionScore, I think of the system as operating in two phases.

Phase one. Process all the tickers in the list of possible issues. If you are trading all the stocks in the S&P 100 index, that list is 100 long. For each ticker, use whatever rules or indicators are need to generate a Buy signal. If 14 of the issues generated Buy signals, pass the shorter list of 14 on to phase two.

Phase two (only needed if there are more Buy signals than there are "slots" available). Rank the issues that have current Buy signals according to PositionScore. Buy the ones listed at the top of the list, starting at the top and continuing down until you have filled all the slots.

Setting PositionScore to a random number is fine. What that says is that all of the alternatives are of equal goodness as measured by your objective function, and the choice should be made at random.

But, if there is some additional information that can help select those alternatives that are more likely to be profitable, or more profitable, or have lower drawdown, or add diversity, or whatever else is important, PositionScore makes it possible to identify the best of the alternatives.

If there are more Buy signals than slots and PositionScore has not been specified, AmiBroker sorts the list into order alphabetically, then starts at the top, just as though you specified PositionScore to be Alphabetical rank. If there are often more signals than slots, you will see positions being taken in tickers starting at the beginning of the alphabet much more often than tickers later in the alphabet. Be sure you mean that.

Note: Setting PositionScore to Random sets up a MonteCarlo process. You may want that or you may not. If the initial random number seed is random -- such as the lower digits of the clock -- and it probably should be random rather than specified in the code -- then the sequence is not repeatable and the trades cannot be replicated. In order to evaluate a trading system like that, many runs should be made and the resulting distribution(s) evaluated using statistical methods. One run is not enough.

Thanks,
Howard
www.quantitativetradingsystems.com

 

[tech/a]

I don't see anything magestic about it beyond the things I've already pointed it.

Nor do I.
The biggest kick I get out of it is that its simplicity and public performance pissses the hell out of a lot of people who would love to have seen it fail miserably.

I'll say it again its not that hard.
Humans have the capacity of turning the most simple into the most complex.

 

[stevo]

Nor do I.
The biggest kick I get out of it is that its simplicity and public performance pissses the hell out of a lot of people who would love to have seen it fail miserably.

I'll say it again its not that hard.
Humans have the capacity of turning the most simple into the most complex.

It may not be that hard for some but there are 2 things that make it hard for us normal irrational humans. Firstly our emotions and, secondly, we don't really know what is going on - what the psychologists call "cognitive difficulties" .

The lines on a computer generated equity curve don't convey the emotional turmoil when a trader hits a 20% plus drawdown - "what if my system doesn't really work", "what will happen tomorrow, is this 1929 / 1987 all over again" or "I am going to lose all my money" etc. We see this sort of navel gazing on these forums.

tech trader is a great exercise. Ideally people get to understand it, do some system testing themselves, run through the signals it has given in the past and get a handle on what they are getting into. I am sure that lots of people have gained a huge amount of knowledge from understanding tech trader better.

I would rather not be tech/a to all those traders that see it as a pot of gold but have no idea how it works. The first big drawdown will see them blaming everyone else and probably abandoning the system because it doesn't work - probably just before it takes off again!

I have seen it happen in a small way with a friend, although he just over-rides the system because he has a "feel" for the markets and doesn't need it anymore. Supreme confidence, most of us think that we are god when it comes to the markets

I know my ability to trade by gut feel and animal cunning - no chance!

I personally believe that simplicity is vastly over-rated, but then I ignore most of the market information and look for a solution based on incomplete knowledge.:

Enough of my ranting.

stevo

 

[tech/a]

It may not be that hard for some but there are 2 things that make it hard for us normal irrational humans. Firstly our emotions and, secondly, we don't really know what is going on - what the psychologists call "cognitive difficulties" .

Thats why we design systems.But yes if you dont design it yourself and have confidence in your testing,UNDERSTAND why a system works,UNDERSTAND that the numbers generated are your ultimate stop,you will fall into the psychological war within yourself in times of under performance.
I think that being in Business for 30 yrs and having the ability to make instant decisions,often when vast sums of Money are at stake (Plus having been on the brink of Bankrutcy twice) has stood me in good stead to UNDERSTAND what makes sound business decision making,and in a timely fashion.

The lines on a computer generated equity curve don't convey the emotional turmoil when a trader hits a 20% plus drawdown - "what if my system doesn't really work", "what will happen tomorrow, is this 1929 / 1987 all over again" or "I am going to lose all my money" etc. We see this sort of navel gazing on these forums.

Ive learnt too that there comes a time when you'll be exited and stopped,pretty close to where pain is at its most.Armed with your exits,Stops and Blueprint of systems parameters which tell you if your trading in a market not seen in testing,you should never be in panic mode.

tech trader is a great exercise. Ideally people get to understand it, do some system testing themselves, run through the signals it has given in the past and get a handle on what they are getting into. I am sure that lots of people have gained a huge amount of knowledge from understanding tech trader better.

Well its taught me a lot.

I would rather not be tech/a to all those traders that see it as a pot of gold but have no idea how it works. The first big drawdown will see them blaming everyone else and probably abandoning the system because it doesn't work - probably just before it takes off again!

T/T is at an interesting phase now with only a few open positions, a great deal of cash and still within its blueprint.More learning will come I'm sure.

I have seen it happen in a small way with a friend, although he just over-rides the system because he has a "feel" for the markets and doesn't need it anymore. Supreme confidence, most of us think that we are god when it comes to the markets

From the years Ive been on 2 forums Ive seen it year in year out with many many traders coming and going.Few return.Some are brilliant analysts both Fundamental and Technical.
BUT APPLICATION of your trading method is the hardest of all.Application of some method youve come up with that logically should return great profit (Buy low sell high as an example) isnt as simple as doing just that.
Of ALL the reasons traders fail this is the single biggest reason for failure that I have seen.

I know my ability to trade by gut feel and animal cunning - no chance!

I personally believe that simplicity is vastly over-rated, but then I ignore most of the market information and look for a solution based on incomplete knowledge.:

Enough of my ranting.
stevo

The question should be simplicity in WHAT!
Simplicity in WHY your business will succeed and others fail.
Thats the SIMPLICITY I speak of!
It--- Business of any sort is THAT simple.

 

[Chorlton]

T/T is at an interesting phase now with only a few open positions, a great deal of cash and still within its blueprint.More learning will come I'm sure.

Tech/A,

Is the above a result of the system simply not taking any current signals (and is therefore still running) or have you now turned the system off???


Regards,

Chorlton

 

[tech/a]

The System that is traded on Radges site is still business as usual.
Its right up to date and there to veiw if people wish.
It took a new trade last week.

Personally is another matter.
I did trade 3 similar systems to T/T.
All I exited on July 27.
The reasoning behind this right or wrong is posted on the techtrader thread.
I am currently only taking a few short term trades.

 

[motorway]

* Non-cap-weighted exposure to the market leading to index out-performance
Hmm dont know that anyone would weight their positions to capitalisation---do people do that?
Heard of an index fund? Mutual funds look a lot more like the index than a trend following stock system with 10 - 20 equal sized positions that also avoids the ASX50 altogether, by design. You bet people do it, and it's a hand brake on performance. Therefore, not doing it is an edge and IMO a reason why a system like TechTrader works.

Someone who is very critical of mechnical trend following is Victor Niederhoffer ...

Victor Niederhoffer Reviews 'Does Trend Following Work On Stocks?'

This paper, by Cole Wilcox and Eric Crittenden of Blackstar Funds, makes a worthy and thoughtful effort to answer the question of whether it's possible to devise a trend following method that works in stocks. Their method is to buy stocks at all time highs then they sell them after a 20% decline or so using the true range as a cut-off to sensitize. In order to do a valid study of such a phenomenon they have had to be careful to adjust properly for survival bias. They do this taking account of all NYSE, AMEX, and NASDAQ stocks listed and de-listed from 1983-2004.

Their results show that $1000 would have grown to some $30,000 during the period. Results for the years 2003 and 2004 of 55% return and 27% return are particularly impressive.

The defects of the study are that they do not show the statistical significance of their results. Small stocks performed much better than big stocks during the latter part of the period. It is not clear that buying the average stock in the category that they found the new highs from, and selling an average stock or a matched stock next to it in price would not have yielded very similar results. More to the point, I hypothesize that buying stocks that were 10% off their high, or at a new low, with similar inclusion criteria, would have come up with as good results. Finally, the inclusion of companies with lower than $250,000 daily trading volumes in the study makes the results unobtainable for those who would wish to implement it with reasonable new money.

While these are serious objections, they are of the armchair variety. We have tested similar strategies on big groups of stocks like the S&P500 and found it produces random results. That is just the point the researchers say, the small stocks are the ones that give the superior returns.

We must compliment the authors on a true attempt to find out the nitty-gritty of the market . Further work in the field by them and us will undoubtedly clarify the issues.

Eric Crittendon Replies:

I believe the following statement is in error:

"Finally, the inclusion of companies with lower than $250,000 daily trading volumes in the study makes the results unobtainable for those who would wish to implement it with reasonable new money."

The last paragraph of the paper goes into detail. Stocks with daily dollar volume that low would not have passed our filter.

The origin of this project was a data mining test on NASDAQ only stocks to find effective stops using a random entry method (that was back in 2001). We found several methods that were effective through various market cycles; all were derivatives of volatility, however. The stops we present in the paper are the most bland and easiest to understand. It is my nature to know what our "total wash out" risk (a.k.a. total portfolio risk) is and be comfortable with it. Without stops, and their aggregate remaining risk across all positions, I can't quantify this value. Under the assumptions in the paper we would have nearly realized this loss, being stopped out of almost 90% of our positions during the crash of 1987. I expect crashes to happen and wish to live through them. The entire system depends upon this reference point, total portfolio risk. It is calculated daily for existing and new (buy tomorrow) positions and becomes an independent variable input into a utility function that, in turn, specifies how much should be risked on each position. New positions are sized accordingly and existing positions are resized (both up and down) if they are out of alignment by a significant margin (determined by yet another utility function). The circular reference repeats every day. In this way we are able to honor every trade and control total portfolio risk. So, you can see my inability to provide portfolio performance results without the use of stops.

When restricted to the S&P 500 we found an inverse relationship between the tendency to have substantial and prolonged % moves and market capitalization. Intuitively this made sense to us as index members have already experienced the market cap growth necessary to get into the index. Also, index members tend to offer transparency that is communicated in real time by an army of analysts and research reports. Furthermore, their business models tend to be overly diversified relative to small/medium companies. Additionally, there is typically millions of dollars bid and millions asked just cents away from the prevailing price at any given time. It seems only an accounting scandal, speculative mania, or major market shock can provide the fuel for outlier moves. That being said, we don't discriminate against them; we are just happy that there are so many more small and mid-cap companies to buy.

Alston Mabry Adds:

The chart comparing their Trend System to the S&P over the period January '91 to January '95 reminds me of a chart comparing the VAY to the S&P over the same period. Most importantly, the bulk of the Trend System outperformance occurred in the period January '03 to January '05, during which the Trend System essentially doubled. The VAY essentially doubled in that same period, so one could make the case that Trend System results are hard to distinguish from the results of a randomly-selected, equal-weighted portfolio, or at least that the mean and standard deviation of the returns of such a portfolio should be the benchmark for the Trend System, rather than a cap-weighted index.