Designing Trading Systems

1
Designing Trading Systems

• A systematic approach to trading by
• Todd Kellenbenz
• toddk63_at_yahoo.com

2
Designing Trading SystemsTopics Covered

• What is Mechanical Trading ?
• Benefits of Mechanical Trading
• Buy or Build the System ?
• Tools Needed
• First Step The Design Criteria
• Backtesting and Optimization
• Pitfalls Traps
• Resources Books

3
What is Mechanical Trading ?

• Trading based on a set of rules that completely
  defines the life of the trade.
• Entry, exit, stop loss, position sizing are all
  clearly pre-defined.
• A complete mechanical system leaves no room for
  discretionary decisions.
• Rules are developed by rigorous research and
  testing of historical data.
• Each individual trade has a random outcome.
• Extensive backtesting can demonstrate a positive
  outcome over the long term.

4
Benefits of Mechanical Trading

• Takes the emotion out of it.
• Dont have to question the signals.
• System can be evaluated on probabilities.
• Expected performance (profits, drawdowns,
  win/loss ratio) is roughly known.

5
Benefits of Mechanical Trading

• Some of the best trades are the ones that your
  emotions tell you not to take.
• Allows going against the herd.
• Analogous to a casino. Not as the player
  (gambler), but as the owner, who knows that over
  time the odds are in his favor.
• Brings some certainty in an uncertain environment.

6
Buy or Build ? - Buy

• Only option if you lack time or computer
  programming skills.
• Beware of black box red light / green light
  systems.
• If you do not intimately know what makes it work,
  you may not trust it completely and may hesitate
  to act on signals. Emotions then take over.
• I personally do not recommend this option.

7
Buy or Build ? - Build

• Requires strong computer programming skills and
  LOTS of time.
• Having designed, built and tested it yourself,
  you will have complete faith in it and will not
  hesitate to take its signals.
• No trading system works forever. Having built it
  yourself, it may be easily fixed with only a
  minor tweak.

8
Tools Needed

• Charting / backtesting software
• Amibroker (this is what I use)
• Tradestation
• Esignal
• Others?
• Good historical data. Avoid free data such as
  Yahoo. You get what you pay for.
• Futures data must be continuously backadjusted.
  I use Norgate Premium Data for this.

9
Tools Needed - Amibroker

• Very powerful and comprehensive.
• Inexpensive
• Portfolio level testing
• Easy to customize
• Optimization engines include simple brute force
  as well as genetic algorithms.
• Good metrics report (Fig.1)
• Vibrant online community
• Open source. Develop your own add-ins (dlls) with
  C, Excel VBA or similar. Many available online.
• Steep learning curve

10
First Step The Design Basis

• Define what you want from the system.
• How much time dedicated to trading it. Do you
  already have a full time job?
• Time frame to trade. Intraday, daily, weekly
  charts ?
• Which markets to trade. Stocks, futures, others ?

11
First Step The Design Basis

• Desired return (CAR)
• Expected maximum drawdown (MDD)
• Win / Loss ratio. A system with a low of
  winning trades can be psychologically difficult
  to trade.
• Optimization metric. Use something more than
  just CAR (Compounded Annual Return)
• CAR / MDD (Calmar Ratio) is a good one. K-ratio,
  Ulcer Index, others ?

12
First Step The Design Basis

• From David Buffalo Blog
• 60 Win / Loss ratio
• 1.60 profit factor (gross profits / gross losses)
• 1.00 payoff ratio (avg trade profit / avg trade
  loss)
• Use of Calmar Ratio (CAR / MDD). Once this is
  defined, it is simply a matter of position sizing
  (leverage from futures !) up to your max pain
  threshold (MDD)

13
Backtesting and Optimization

• Steps Choose concept, program it, measure
  results, modify or choose different concept,
  repeat.
• Generate ideas from reading books, trading
  magazines, online blogs, etc.
• About 30 trades needed to be statistically
  significant. Much more is better.
• Make sure profits are not from just one or two
  trades.

14
Backtesting and Optimization

• Watch Portfolio Equity line (Fig. 2). Nice smooth
  slope.
• Watch drawdown line (Fig. 2)
• Reverse the sign of a rule for confirmation or to
  possibly make an unprofitable rule a profitable
  one.
• Be suspicious of rules that work but that just
  dont make logical sense.
• Solid ideas rise to the top. Valid across many
  different markets. i.e. Rising MA (trending)

15
Backtesting and Optimization

• In-sample / Out-of-sample testing
• Optimize with in-sample data. Verify system with
  out-of-sample data.
• More important for single market systems.
• Portfolio systems have inherent randomness making
  them more forgiving to curve fitting.
• Walk-forward testing periodically re-optimizes
  with new in-sample data. Risk of curve fitting?

16
Backtesting and Optimization

• Initially test without slippage and commissions.
  No compounding.
• Later, tighten up and add slippage and
  commissions.
• Compound results only as a final check of overall
  system performance.

17
Pitfalls Traps

• Optimization is a double edged sword. Need some,
  but not too much.
• The more rules that constrain the system, the
  more likely it is curve fit to the data.
• Pick optimized parameters around clusters, not a
  single optimum point (Fig. 3).
• Add random noise to data set to test for system
  robustness (Howard Bandy Quantitative Trading
  Systems).
• A robust system should handle small amounts of
  noise without greatly impacting profitability

18
Pitfalls Traps

• Too frequent of walk-forward testing could lead
  to over optimization.
• Future Leaks. Coding error where tomorrows price
  is known today. Example ref(Close,1) instead
  of ref(Close,-1)
• Survivor bias of stock indexes (http//home.comcas
  t.net/toddk63/)
• Successful trading systems can become
  unprofitable unless new and innovative methods
  are constantly applied.

19
Resources Books

• Books
• Alexander Elder Trading for a Living.
  Triple Screen 1) go with a trend 2) look for
  pull backs 3) Enter on confirmation i.e. breakout
  of previous high
• Mark Douglas Trading in the Zone. View
  every individual trade as a random event. System
  testing gives confidence that over a large
  population of trades, outcome should be positive.
• Howard Bandy Quantitative Trading Systems,
  Modeling Trading System Performance .The nuts
  and bolts guide to developing a trading system.
  I cannot recommend Howard enough!

20
Resources Books

• Web Resources
• The Gary Stone Journal Designing Trading
  Systems
• The Buffalo Trader David Buffalos Blog
• My homepage http//home.comcast.net/toddk63/

21
Figure 2 Portfolio Equity Drawdown
22
Figure 3 - 3D Optimization