Dooku Forex Trading System
March 19, 2006 by Trader Rich
I started to work on multiple trading systems over the last week or so and realized that I was falling into the same trap of trying to do too many things at once. I decided this weekend to work on my first trading system more. The first thing I did was give it a cheesy name so that I can distinguish it from all the others. I have named it Dooku. You can figure out on your own where I got this one from.
This system was previously being tested on hourly charts with a risk reward of 1:1. It tested well with a 50 pip stop loss and a 50 pip profit target. The first thing I wanted to do was to get the risk reward to 3:1. I successfully accomplished this and Dooku tests very well on the hourly charts with a stop loss of 75 pips and a profit target of 225 pips.
I will forward test this system this week and going forward to see how it performs. From my backtesting, it has never had a losing month. I will forward test this with real money and 1 lot each trade. I will use a 75 pip stop and a 225 pip profit target. I will trade only the GBP and EUR. From my calculations, this system will generate an average of 20 signals a month for each currency pair.
Popularity: 3%
Loading ...
[b]Developing My First Trading System: A Cautionary Tale of Hubris, Humility and Hope [/b]
by: David Silverman
A veteran trader checks his trading rules against backtesting and optimization in order to develop a system.
This year I celebrate my 25th anniversary as a trader.
I am no superstar, but I make a living, which distinguishes me from the vast majority of individuals who, during this period, have tried to make it as professional traders. There is a certain amount of pride I take in having outlasted many of my competitors, but it hasn’t been easy. Like all veteran traders I’ve had my share of ups, downs and sideways (often more frustrating than the downs), and even after all these years I struggle each day to understand the vicissitudes of the markets and overcome my fears and inhibitions.
If I have developed any insight after a quarter-century, it is that trading is an evolutionary process. The trader begins life as a weak, defenseless organism. With a puny brain incapable of processing information efficiently, he acts self-destructively, fading the market, adding to losing positions, promising God to never make the same mistake again, if He will just intervene this one-time. Predators lurk and entice the novice into their traps. If, somehow, the trader survives, he may earn a place on the food chain, but he cannot rely on past successes to keep him there. With every tick or pip of the market, other traders are on the prowl. So much for intelligent design. Such is the process of natural selecton in the marketplace.
Playing by the Rules:
In that context, I have given a great deal of thought to the question, what do successful traders have in common? While there are a number of answers to this question, above all, successful traders know how to follow rules. By rules, I mean a set of fundamental beliefs so deeply held that it becomes inconceivable to think of deviating from them. It almost doesn’t matter what those rules are, only that the trader believes they work and implements them in all instances. As for me, a single rule governs every trade I make. I will never, under any circumstances, put myself in a position where an individual trade or a series of trades will wipe me out. The way I accomplish this is to limit my losses to a maximum of two percent of the equity I am trading. So, for example, if I am trading a $1,000,000 account, I can lose up to $20,000 on a trade and still be in compliance with my rule. More importantly, I would have to make 50 consecutive losing trades of two percent in order to lose all of my money. This is almost inconceivable.
In practice, my losses are almost always less than two percent. While I sometimes experience a number of consecutive losing trades–I don’t think it’s ever been more than a dozen or so–my winning trades tend to be much larger than my losers. I suppose the two-percent loss threshold is somewhat arbitrary. I could extend it to five percent or more without putting myself in too much peril. But in order to justify doing this, I’d have to believe that I could make at least two-and-a-half times as much. Because I’m not sure this is a reasonable expectation and because the additional risk scares me, my two-percent rule remains the underlying principle upon which all my strategies are based.
The two-percent rule also creates a structure for my trading decisions. For example, assuming a $1 million account value, if I want to buy Google (GOOG) and my stop loss is $5 lower than my entry price, I know that I cannot buy more than 4,000 shares (this is actually an oversimplification because I rarely execute my entire position at one price, but for the purposes of the example, using round numbers illustrates the principle easily). In other words, if I want to establish a bigger position, I need to choose a stop price closer to my entry price or wait until GOOG trades lower. Moreover, if I am long five different stocks from the same industry, this must be considered a single position with a single two-percent maximum loss. (As opposed to five separate positions with a maximum loss of ten percent). Treating the aggregate position any other way is inconsistent with the basic principle. This goes for pairs trading as well. If, for example, I am spread long GOOG versus short Yahoo! Inc. (YHOO), this is a signle position with a single two percent maximum loss. It’s not very elegant but, as I say, I make a living.
Developing a Trading Model:
With these rules in mind, I decided last year to conduct an experiment, in which I would take a single trading idea and develop it into a formal trading model. This, of course, is done all the time by sophisticated trading firms using custom-built technology and by retail traders working with comparatively simple platforms such as TradeStation, MetaStock, Fidelity’s Wealth Lab Pro and other off-the-shelf software from brokers or independent vendors. Everyone seems to be searching for the perfect trading system, the elusive Holy Grail of systems.
Not me. I feel fortunate when I can find my keys in the morning, so I have no delusions of grandeur. My more modest goal was simply to discover whether my cherished rules held up when subjected to the model developer’s standard techniques of backtesting and optimization. In a way, the prospect intimidated me. For one thing, I was concerned with the potential cost. I know of traders and firms who have spent hundreds of thousands or even millions of dollars developing systems. Because such amounts are far more than two percent of my trading equity, I vowed to enter this project with a stop loss, no different than if I were trading GOOG. Additionally, a part of me feared the outcome for an entirely different reason. What if Burton Malkiel was right when he said, in his classic text A Random Walk Down Wall Street, “a blindfolded monkey throwing darts at a newspaper’s financial pages could select a portfolio that would do just as well as one carefully selected by the experts.” Simply put, what if I found that the principles upon which I have based my career were without susbstance? It’s not like the “Random Walk police” were going to come to my house and lead me away in handcuffs in the middle of the night, but how would I deal with the blow to my psyche? What would this mean for my trading strategies? Could I come to work on the day after and pretend nothing had changed?
[b] Calling the Geek Squad [/b]
This would not be my first attempt at system development. About 15 years ago, I tried to learn how to write code to develop trading models, first using relatively primitive software from a company called Logical Information Machines and then, when that didn’t turn out well, using TradeStation. The promotional literature from both companies implied that the software was so simple to use that anyone, except perhaps a village idiot, could be programming in a few short hours. After a hundred hours of unsuccessful attempts, imagine my humiliation. Well, not this time. This time I knew what I needed: I needed a geek. I turned, reflexively, to the University of Chicago, my alma mater. They allowed me to place an ad on their website, and within days, I had thirty-five resumes. I eventually interviewed fourteen impossibly young, impossibly bright graduate students. One of the candidates had a bachelor’s degree in mathematics, a master’s degree in computer science, and solved a Rubik’s Cube in front of me in less than a minute, blindfolded. While this demonstration was as sad as it was impressive–imagine how many Saturday nights this poor boy spent along playing with his Rubik’s Cube–I was awed by his impressive display of right-brain thinking. I had found my geek.
I began spending money immediately. While I believed my new employee–let’s call him Rube–had the technical skills I was looking for, I was concerned that teaching him the basics of the market, as well as my particular approach, was going to consume a great deal of my time. I was right about that. It cost me approximately 40 hours of one-on-one tutelage, but I viewed this as a necessary tradeoff. Time spent with Rube at the outset would surely produce a dividend when the programming process commenced. There were additional upfront expenses. Two PCs and four monitors cost about $3,500. Two broadband connections addded $100 per month. Various pieces of software, books and training manuals added a few hundred dollars more to the tab. His salary cost me $1,200 per week. He also drank a lot of Red Bull and a can of that stuff costs more than a gallon of unleaded gas.
[b]Finding the Right Software [/b]
Once we got past the initial learning stage, I instructed Rube to investigate obtaining system-writing software and quote data, to determine the best way for us to conduct our research. While there are many software packages available containing preprogrammed functions and actual black-box systems, I did not believe they met my needs. First, it is very difficult and time consuming to ascertain that the definitions and assumptions the vendor made in creating the software are acceptable. Second, these packages tend to be expensive. Some require that you open a brokerage account. Others are sold for a monthly fee.
So instead, I decided to use a program called Wealth Lab Pro (ATP), the front-end trading system I use to execute my transactions. Fidelity provides the software for “free” to its active traders, although, of course, nothing in the financial services world is really free, and were I not paying commissions on ATP, Fidelity would certainly charge a monthly fee for the model-building software. Rube informed me that in WLP he could write programs in C++ and easily export data to and import day from Microsoft Excel, where he could manipulate the database and do additional programming. I had no clue what he was talking about, but it seemed to make him happy, and when my geek is happy, so am I. One of the best features of WLP is the the quote feed from Fidelity comes directly from the exchanges and is fairly reliable. This is extremely important because when it comes to modeling, if the quote data is defective, so too will be the results of the research. In other words, garbage in, garbarge out.
Another important consideration is that data, especially historical-intraday and real-time data, can cost thousands of dollars. Using WLP, which includes a huge amount of data at no extra charge, allowed me to keep my costs relatively low. Finally, Fidelity supports its products extremely well. The ATP/WLP help desk is open before, during and after trading hours. I have found the people who answer the phone to be knowledgeable and patient. Knowing there is someone available to curse at when you get frustrated is an important value-added service. None of this is to suggest that WLP is the only or best way to conduct this sort of research. But after I fully investigated the readily available software, it seemed like WLP was the program that best met my specific needs. Nor am I suggesting that one nees to hire a computer programmer in order to successfully test trading ideas. Unless, of course, you are a village idiot. In that case, it is highly recommended.
[b]The”Rubber Band” System [/b]
By this point, we were ready to begin programming. It would be impossible in this short space to describe everything that went into building my model because we performed thousands of tests using ten years of data for the 500 stocks in the S&P 500 and the 100 stocks in the NASAQ 100. I will say that the concept is based on a type of trade I have been making for 25 years that somewhat simplistically can be characterized as my “rubber band” strategy. The idea behind it is that as the market moves, it stretches–like a rubber band–and when it can stretch no further, it will snap back. This approach requires one to fade the immediate trend. So, for example, one query goes as follows:
1. Stock XYZ is up at least ten percent between the
close of day one and close of day two.
2. Sell short 100 shares of XYZ on close of day
two.
3. Buy back 100 shares of XYZ on close of day
three.
4. Or, on day three, buy back 100 shares of XYZ
if XYZ moves up more than five percent from
close of day two.
There are three possible outcomes to this trade on day three: cover the short position with a profit on the close, cover on the close with a loss of less than five percent, cover before the close with a stop loss of five percent. And the verdict? The rubber band system is profitable. It generated more than 24,000 trades over ten years, an average of about 40 trades per stock, four trades/year/stock and overall, about ten trades every day from the universe of 600 stocks. The system made about $1 million, an average of slightly more than $50 per trade. Fifty-five percent of the trades were from the long side and 45 percent were from the short side. Long trades were far more profitable, generating about $800,000. As I dissected the data, I found that certain stocks seemed to respond well to the system and others lost money every time the signals were present. Illiquid stocks (stocks with less than 100,000 shares per day traded) performed especially poorly. I could go on, but I just want to give you a sample of the information one finds in an analysis of this kind.
[b]Don’t Forget About Commissions or Spreads [/b]
So you ask, why not just make the daily ten trades in perpetuity and collect the winnings? Well, we haven’t looked at the entire picture. Commissions on the completed transaction are $16, which brings the average per trade down to $34. Then, we must factor in slippage. Let’s say that amounts to $25. We’ve now reduced the average to about $10. The margin of error is now so slim that what looked initially like an interesting system now appears to be a breakeven proposition.
But wait! We’re just getting started. What if we cherry pick, acting only on signals for stocks that make money with this system? Or what if we simply buy 200 shares of every stock that responds well and only 100 shares of those that don’t? What if we only execute trades on the long side, which produced most of the profits? What if we filter from the system stocks that trade less than 100,000 shares per day because they are a drainon profitability? What if we alter the original query variables? Buy XYZ after a 20-percent move? Get out on day ten? Sell stop of 15 percent?
I have no idea whether there is any individual change or series of changes we can impose that will salvage the rubber band system. The only way to know for sure is to test. This optimization–the attempt to find the variable that will produce the optimal results–can, theoretically, go on for infinity. At some point, however, one has to decide if the system is worth trading or if it is a dud?
Even after months of study, I remain unsure. While some of the work Rube and I have done looks promising, I do not believe we have tested our results rigorously enough to justify committing my trading capital to them. But we are getting there. The next step is to take our findings to a professor in the Graduate School of Business at the U of C, an uber-geek, who will use all of the mathematical tricks in his financial engineer’s bag to break my model. If he cannot, I may be on to something after all. This exercise will cost big money, but I have discovered that model-building is like raising a child. Shoes, braces, college, weddings–you just keep writing checks and hope to God that one day they will support you in your old age. In the meantime, I continue to trade the old-fashioned way and continue to make a living.
Also, I have become surprisingly good at Rubik’s Cube.
SFO Magazine (March 2006)
what software are you using for backtesting?
I have my own database and write frontend backtesting scripts using python or perl. I found it to be much more flexible and faster than software platforms. It takes some work initially to write formulas but I consider this a good learning experience to really get intimate with the calculation of a certain formula.