Having worked through the processes to either develop and test your own trading system or perform detailed due diligence on a commercially available trading system using the steps discussed over the past weeks, this week we will look at what constitutes a winning trading system. A winning trading system will give you a profitable ‘edge’ over the market and result in a profitable outcome over time. To ensure your system has an edge, there are a few more questions that need to be answered and some more data that will need to be analysed. These questions include:-
• How many completed historical trades are required to demonstrate that an edge has a high probability of working in the future?
• How will an edge perform with limited capital when only a subset of trades are traded out of the full available list?
• What metrics can be used to determine which set of trading system research results is better than the next?
When you run your research program against the historical data, a list of trades will be generated with data recorded for each trade. Look for typical data like:
• entry and exit date
• entry and exit price
• percentage profit
• hold period for the trade
• MAE & MFE (Maximum Adverse and Maximum Favourable Excursion)
To start with, always test your edge with the same unit position sizes. For testing and analytical purposes don’t use any money management until you have established that you have a statistical edge. Use a position size that will allow at least one share of the highest priced instrument in your ‘in sample’ historical data. Typically, a position size of $10,000 should do the trick for researching stocks. Index research for CFDs typically requires a larger minimum unit. For example, the DJIA has had values above and below 10,000 as has the Nikkei and ASX 200 Energy Index; therefore a position size of $10,000 would not be large enough to trap the trades when the index was above 10,000 because it is not large enough to trade one CFD. Hence, valid trades would be omitted from the trade’s database.
The research results are then measured statistically across the entire database of historically generated trades according to the rules of the system that you have coded. More data for analysis of the results is calculated from this database of trades:
• total number of trades
• number of winners and losers and hence the percentage winning rate
• the average percentage size of the profit trades and loss trades—will be the same as absolute because a unit position size has been used for all trades
• the average percentage move per trade across all trades—immediately tells you whether you have an edge
• the average hold period for winners, losers and per trade across all trades
The diagram below shows a solid black line that depicts a breakeven edge. Note that the final edge that you trade should include brokerage, commissions and slippage.
The breakeven edge is computed from the percentage of winners and the profit ratio, which is the relative size of the average winner to the average loser. A profit ratio of 2.0 means that the average winner is twice the size of the average loser.
The leftmost dot on the breakeven line is positioned at a winning rate of 20%. A profit ratio of four will be required to break even, net of brokerage and slippage. The rightmost dot on the breakeven line is positioned at 50% where a profit ratio of one will be required to break even, net of brokerage and slippage.
Any system with an edge that falls below the breakeven line has a negative mathematical expectation and hence virtually no probability of being profitable over a large sample of trades. Through luck, over a small sample of trades you may make money, but the more trades you complete with a negative mathematical expectation system, the higher the probability that you will lose money. Absolutely no type of money management rules can save such a system over a large sample of trades.
Any system positioned above the breakeven line can make money. The further away from that line, the higher the probability that not only will it make money but also that it will be profitable from a smaller sample of trades. Also, the further away from the breakeven line, the less volatile the system’s equity curve should be and the steeper the equity curve should be. This means lower drawdown’s while making money faster.
A good system does not need to be that far away from the breakeven line to make good money as the power of compounding is on your side when you have a positive mathematical expectation working for you in the market. I use the word ‘can’ rather than ‘will’ because poor money management rules will undo an edge with a high positive mathematical expectation as will poor trading psychology.
In reality, generating a system that is much further away than the ‘Excellent System’ will be difficult to achieve. If you do, there is a high probability that the system is either overly optimised, and hence curve fitted to a narrow price action dataset, or has too many criteria and filters which means the same thing. It is overly optimized by constraining its degrees of freedom (covered in next week’s blog on optimisation). This may also occur with a system that trades infrequently, i.e. the historical testing generated a small sample of trades. Again, this usually results from too many criteria and over optimising.
Other metrics that should be established to compare one system to the next include:
• pessimistic rate of return
• average hold period per trade
• average profit per trade
• average size of profit and loss trades
• maximum consecutive profit and loss trades
• Sharpe Ratio
• Recovery Factor, Ulcer Index, etc.
Your research tool should include these metrics and explanations as standard features, or you can program them yourself, or use Excel. The definitions of these can also be found in books by Perry Kaufman, Van Tharp, Joe Krutsinger, Thomas Stridsman and others.
The steps of choosing a concept, programming it, establishing a testing data set, measuring the research results, modifying the concept or choosing another are an ongoing iterative loop that will continue until you have met the original objectives that you stated for your system.
An important point needs to be made regarding the ‘best system’. Simply stated, there is no ‘best’ system. Most people are led to the markets in search of larger returns than are on offer elsewhere, which creates a paradigm of evaluating systems from a single metric—those with the ‘best’ returns. Are the returns measured as raw percentage return on capital? Or return per daily/weekly/quarterly bar? Or return per trade?
A system with the ‘best’ raw percentage returns will probably not be able to be traded by the great majority of people because it would likely require many hours of effort every day: trading a leveraged instrument, having a volatile equity curve, having an average hold period of less than a day and requiring a trading mindset only achieved by a select few who engage the markets. You see, the more often you watch price action, the more negative feedback you will get from the market, which psychologically makes it far more difficult to trade successfully. Randomness takes care of this. The less often you watch price action, the more positive feedback you will get from the market.
A system with the ‘best’ return per trade will probably have an average hold period of many months or even years. This shows that using a paradigm of measuring a system by its returns will determine the timeframe in which you trade and how much time you will require to trade. This is the tail wagging the dog. You need to start at the top and specify how you want to engage the market and then find/build a system that fits your lifestyle requirements such as time availability, ability to handle equity curve volatility, absolute returns on capital requirement, etc.
As you can now appreciate, the process of designing and developing a professional system that works and delivers a positive edge to the trader is a complex process. It is not one to be taken lightly as the number of variables that need to be considered and examined is large. Coupled with that are the psychological issues and paradigms that also need to be observed and confronted by those wanting to develop a system that has no personal bias associated with it.
In next week’s blog I will look at the traps of over optimisation and curve fitting that can create a false sense of security and unobtainable profit objectives for the unwary developer.