Wednesday, 19 of June of 2013

Quantitative Trading Systems

Howard Bandy's discussion of technical analysis

• Welcome
• About
• Articles
  • Backtesting and Cognitive Dissonance
  • Having Confidence or Having Faith
  • Trading System Development
  • Logic, Data, and Synchronization
  • Trading System Health and Position Sizing
  • Book Reviews
  • Distributions
  • Measuring System Quality
  • Why Traders Stop Trading
  • Managing Subjectivity with Objective Functions
  • Testing Stops
• Trading Systems
  • 200 Day Moving Average
  • Mean Reversion based on RSI
• Forecasts
• Code
  • AmiBroker — introduction
  • Standard deviation in a single pass
  • Equity curve feedback
• Resources
• My Books
  • Special for blog readers
  • Quantitative Trading Systems
  • Modeling Trading System Performance
  • Introduction to AmiBroker
• Glossary
• Posts

Leave a comment

Comments RSS TrackBack 16 comments

• 
  Log in to Reply

  Chum

  in February 3rd, 2012 @ 21:18

  I have all three books. As my thinking and understanding of the markets has evolved, these three have come to occupy a permanent spot on my desk. I wish your sections on statistics were available during my formal education years. Better late than never, I get smarter every time I read a section. Kudos on a high quality product! The attention to detail across all fronts — the typeface, layout, progression of topics, index and the right amount of overlap are really appreciated.

  I regularly use the Introduction to AmiBroker book supplemented by the online help. I’ve had a bit of love-hate relationship with AmiBroker. I didn’t appreciate the power it presents and I think now am quite forgiving of its idiosyncrasies.

  -Sundeep

  ---

• 
  Log in to Reply

  AndrewCR

  in June 3rd, 2012 @ 15:31

  I am just about to get started with Amibroker, that is the plan anyway, and was going to buy Quantative Trading Systems to help me through. I have no programming experience and was initially going to buy Introduction to Amibroker but read that the latest version is slightly different and I should wait for the revised version.

  I want to develop and test my own systems.

  Does this sound right?

  Any advice would be appreciated.

  Thanks

  ---

• 
  Log in to Reply

  Howard

  in June 4th, 2012 @ 13:17

  Hi Andrew –

  Yes, Introduction to AmiBroker is being revised to reflect the current version of the AmiBroker software. The new edition will be published in about August, 2012. The current edition of the book can be used with the current edition of the software. The primary change in the software is in the default interface for Automatic Analysis. There is an option in the current AmiBroker software to use the “old” interface, which is the one in the current edition of the book. But you will want to use the “new” AA interface eventually.

  Quantitative Trading Systems is primarily about the design, testing, and validation of trading systems. While it was not intended to be a book about programming AmiBroker, all of the example indicators and systems are in AmiBroker’s afl language. They are fully explained and ready to run. You can download the code from the book’s website so you do not have to type them in. Many people who have previous programming experience skip the Introduction book and find that QTS, along with the AmiBroker online tutorials, gets them going in programming. You will definitely find QTS helpful for system design. Whether it is sufficient to get you started programming, I cannot answer.

  Best wishes,
  Howard

  ---

• 
  Log in to Reply

  kazaan

  in August 20th, 2012 @ 23:48

  Mr. Bandy. I’m delighted you offer Intro to Amibroker-2nd edition, free to download. I applaud the generosity of your spirit and your tireless effort here and in forums, both on and off line, to make Amibroker and Systematic Trading a little less intimidating. Sandeep describes my experience with Amibroker the best – Like John MacEnroe or yore or Ussain Bolt now, love it, hate it, but you cant just Ignore Amibroker!! I’m an not surprised at all at the serendipity in finding your mail offering this, just when i start my 4th innings at cracking Ami. Tx for everything.

  ---

• 
  Log in to Reply

  kazaan

  in August 20th, 2012 @ 23:50

  Just to Add – I have purchased ITA edition 1, and QTS. I have put purchase of the 3rd book on hold for now, while I get to speed on the basics. I look forward to your impending book thats between ITA and QTS.

  ---

• 
  Log in to Reply

  edilson.floriano

  in September 11th, 2012 @ 18:09

  Howard,
  I bought the Kindle version of the Larry Connors’ book “Short Term Strategies That Work”.
  It would be great if you could launch your next book on Kindle version…

  Thanks.
  Edilson

  ---

• 
  Log in to Reply

  Keith McCombs

  in October 12th, 2012 @ 23:49

  Howard –
  I noticed in MTSP, on page 189, that Figure 10.2 had a logarithmic Y axis. Could you please tell me what software you used to display a logarithmic axis?

  Thank you.
  – Keith

  ---

• 
  Log in to Reply

  HowardBandy

  in October 13th, 2012 @ 07:21

  Hi Keith –

  That chart was created in Excel. Right-click the vertical axis, Format Axis, check Logarithmic scale.

  Howard

  ---

• 
  Log in to Reply

  RedBaron

  in November 20th, 2012 @ 11:06

  I have had an opportunity to trade systems (e.g. Connors) that require me to enter numerous limit orders to buy before the open, in hopes that a very small percentage of them would be hit. The problem with using many limit orders was you can rapidly exceed your buying power if you try to implement such a system with any broker. This is because brokers will only let you place a number of limit orders up to your buying power. I trade only within IRA’s, which I am finding out more and more people are doing. In an IRA, you cannot use margin, so you reach the buying power limit easily. So I was indeed experiencing an inability to trade these systems. So I called Interactive Brokers and asked how can I do this. I was told there was no way around this. Also, I wrote an email to IB asking the same question and got the same negative answer. Well there is a way, which you may or may not know of, but which I will tell you here. Instead of placing limit orders into the market, you use a feature of IB called Alerts. On IB, an Alert is a feature that allows you to input a “Condition”, that when met, initiates an “Action”. I think it may commonly be used to notify you when something either price, volume, or news happens, you get an email sent to you. Well, IB gave enough degrees of freedom in their alerts to allow the following: For a condition, you set the condition to be “when a specific stock meets or falls below a certain price”, then issue the following action: “issue a limit order to buy the stock at that limit price that was met in the condition”. It works!!!! Before the market opens, I took all of my Connors signals of limit orders to buy and made an alert for each one of them. Since there is not an order issued at this point, I was allowed to put in many more alerts (i.e. hiding as limit orders) than my buying power would have allowed. And it works perfectly. The only problem with this, is it is time coming to do this. It may take you one to 1.5 hours to input all of these alerts. But is is worth it. It makes trading these systems “doable”

  ---

• 
  Log in to Reply

  Jim_Stinson

  in November 27th, 2012 @ 08:36

  Hi Howard. I actually found your paper “Developing Robust Trading Systems, with Implications for Position Sizing and System Health” on the NAAIM website. That paper then led me to your site and now to your two books.

  In the paper above, you provide an example trading system using AmiBroker code. I don’t know AmiBroker, but instead use TradeStation. Do you have the TS code I can use to work my way through the remainder of your paper?

  Thanks, Jim

  ---

• 
  Log in to Reply

  HowardBandy

  in November 27th, 2012 @ 09:09

  Hi Jim –

  I had been using TradeStation as my trading system development platform for years. About 15 years ago, I began the transition from TradeStation to AmiBroker. The break became complete when I wrote the first of my trading system development books. I wanted to provide fully disclosed programs so readers could replicate my results. Among the choices for the platform to use were TradeStation and AmiBroker. I chose AmiBroker for several reasons. Two of the most compelling — AmiBroker was more powerful than TradeStation and could implement trading system development techniques that TradeStation could not, and still cannot today; AmiBroker is much less expensive, making trading system development accessible to many more people.

  I now use AmiBroker and Excel, almost exclusively, for development and analysis. I do not have TradeStation code for those programs.

  One of my books, “Introduction to AmiBroker,” is in its second edition and is a free downloadable pdf file, available from the book’s website. “Quantitative Trading Systems,” also in its second edition, explains my thoughts on system development. You get a brief introduction to that in one of the chapters of the Introduction book. QTS uses AmiBroker to illustrate the points I make. “Modeling Trading System Performance” is platform independent. It begins with a set of trades (real or hypothetical) and analyzes the risk associated. From risk, determines position size, then profit potential. It discusses the need to continually monitor system performance and be aware of system health. Position sizing is based on recent performance, is adjusted in a Bayesian manner, and must be dynamic. As the performance of a system deteriorates, position size will be reduced, eventually to zero if the system is broken.

  “Mean Reversion Trading Systems” is in preparation and will go to the printer in a few weeks. It discusses trading systems that trade frequently and hold a few days. All of the concepts and programs are fully disclosed, with AmiBroker code.

  I hope you find my materials useful.

  Regards,
  Howard

  ---

• 
  Log in to Reply

  brucew

  in May 22nd, 2013 @ 16:15

  Hi Howard,

  I have purchased all your books, and have enjoyed immensely, the information you have provided.

  Of late, I have been spending a lot of time using Amibroker and doing back-testing of Trading Systems. Amibroker, I believe is an exceptional charting package, nevertheless, I have my reservations on its usefulness as a good back-testing platform.

  In Quantitative Trading Systems, I read the chapter covering Monte Carlo, and this is different to how I have approached Monte Carlo in the past.

  I have used TradeSim for my portfolio back-testing for the past decade, and the approach TradeSim has for Monte Carlo is as follows.

  Let’s say I am working on a Stocks system using explorations on end of day data, with buy or sell signals entered into on the next trading day.
  In TradeSim, suppose you have three buy signals on a day, however you only have sufficient capital to enter one trade. In a mechanical trading system, any stock meeting the buy criteria should have equal standing as a prospective position, therefore TradeSim MonteCarlo analysis will produce “Monte Carlo” reports on all three variables, had either of the three stocks been selected. Going forward, each of these three stocks will have different terms before the positions are closed out (on potentially different types of stop loss / exit triggers) and as capital becomes available, other new buy signals will have new positions taken up, on very different time-frames. TradeSim will analyse a “best profit” and “worst profit” results, which should accurately reflect true results. There is no predictive analysis in this, everything in the analysis is supported with substantiated values. In short, TradeSim is just analysing all combinations which meet the Trading system criteria.
  So, if I am diligently operating a trading system, I should be able to run a TradeSim simulation over the trading period, and my actual
  profitability must fall within the TradeSim profit range. Where my actual profit is in relation to the TradeSim profit range will determine how good my random stock selections have been, compared to the “best” or “worst” choices.

  Where I am getting very disillusioned with Amibroker, is that in the example above, when you have three buy signals, and only sufficient capital for one trade, Amibroker will select one entry, and ignore the other two buy signals. By not analysing the possible impact of the other two stocks, results in analysis which is very inaccurate and misleading.

  With a simulation over a long-term, or a trading system which generates a lot of trades, these inaccuracies will be very significant. I question whether the Amibroker back-testing reporting has any value at all?

  Howard, do you support my views?

  Best regards
  Bruce

  ---

• 
  Log in to Reply

  HowardBandy

  in June 4th, 2013 @ 16:25

  Hi Bruce –

  As I mention in Quantitative Trading Systems, Monte Carlo simulation, MCS, is useful in many different ways — some of which it is appropriate to do in AmiBroker, others not.

  Within AmiBroker, MCS can be used to add or subtract small adjustments to the price and / or volume data to observe the sensitivity of the model to noise in the data. MCS can also be used to perturb parameter values to observe the sensitivity of the model to noise in the parameters.

  External to AmiBroker, after a system has been developed, tested, and validated, the set of trades that the developer believes to be the “best estimate” of future performance can be used to estimate the risk and profit potential of the system, and to establish safe position size. This is a very powerful technique that leads to determination of system health, estimation of equity drawdown, and determination of the safe position size based on recent performance. I discuss this in detail in Modeling Trading System Performance, including providing the MCS code and step by step examples.

  The application you describe is to use MCS to choose trades to execute when there are more trades signaled than funds or tracks available. I strongly feel that this is a poor practice. I imagine myself running my system before the market opens, seeing many signals, say 20, and having limited funds, say enough for 10 positions. I will not throw dice or flip a coin to decide which to take, and I do not know a trader who will; rather, I will evaluate the 20 possible trades and rank them using some criteria. AmiBroker uses the system variable PositionScore for precisely the purpose of ranking alternatives. A better technique is to write AmiBroker code that represents the ranking rules, evaluate that for each potential trade, assign the resulting value to PositionScore, and let AmiBroker allocate funds to trades in the order I prefer them.

  Best regards,
  Howard

  ---

• 
  Log in to Reply

  brucew

  in June 12th, 2013 @ 05:43

  Hi Howard,

  Thank-you for your comments.

  I do need to do more work on the value of applying the “PositionScore” process. After you have designed a Trading System on set of criteria, which back testing results shows gives you positive expectancy, you then need to design this PositionScore, which appears to be a seconadary level to your Trading System, which will only apply when you have more buy signals, than available funds.

  Having used TradeSim for all my backtesting in the last decade, currently I am trying to replicate my Trading System backtesting results within Amibroker. Tomasz tells me that Amibroker can do everything that TradeSim can do, and I am trying to align my backtesting results.

  I still have a lot to learn with backtesting in Amibroker, and even basic things like how to apply Pyramiding has been a challenge.

  Currently, when I have to apply a ranking process for “buy signal candidates” , I do apply subjective rules to this process. It is not a toss of the coin, but rather other considerations such as what sector the stock is in (try not to have the portfolio over-weighted in certain sectors, as well as ensuring the stock is in a sector that is performing) and I will also look at the chart to look for consistent price action. Whether I do a good or a bad job of this ranking process, my profitability must be within the TradeSim best / worst profit range. I need to work towards designing a PointScore process to see if it will deliver a better result?

  Thanks again Howard – I appreciate your advice.

  Best regards
  Bruce Wood

  ---

• 
  Log in to Reply

  nooby123

  in June 16th, 2013 @ 17:01

  Hello Mr Bandy – I was curious why did you present the “3 Day High Low” system as two different systems (Long/Flat and Short/Flat) in your book “Mean Revision Trading System”? I downloaded the code, and combined them back in to a single system, which seemed to work very well. I also made some of the variables “optimizable”, and it did even better. Thank you for writing such an excellent book, and making the source code available!

  ---

• 
  Log in to Reply

  HowardBandy

  in June 17th, 2013 @ 08:51

  Greetings —

  The reasons for creating two separate systems — one long/flat, the other short/flat — are explained on page 18 of the Mean Reversion book. In summary, the advantages of separating the code include:
  1. The program for long/flat will be shorter, clearer, have fewer parameters, be easier to program, and run faster than the program that is long/flat/short.
  2. The rules and parameters needed to decide whether to be long or flat are fewer and simpler than the rules and parameters needed to decide which of three categories.
  3. The accuracy of identification of long signals will be higher when there are only two categories.
  4. Separating the long/flat portion of a system from the short/flat portion removes the temptation to treat signals as being symmetric.

  Additionally, as the number of rules and parameters in a system increase, the problem of “the curse of dimensionality” (discussed in Chapter 10, and in an article on this blog) becomes greater.

  I agree that there are parameter values that produce better results than those listed in the book. I recommend using the walk forward technique to be certain out-of-sample performance is satisfactory.

  Regards,
  Howard

  ---

Quantitative Trading Systems uses WordPress as content manager.