For most traders, the development of a strategy is just a means, a necessary path to profit. The process of developing a trading system is more difficult than the final operation: in general, actual trading is much more interesting than developing a trading strategy. Translation: Xu Shengjie
However, this can lead traders to take shortcuts, which can lead to all sorts of mistakes when developing their trading strategies. Although in the short term, the trading method is easier, the mistakes caused by shortcuts will eventually affect live trading.
Fortunately, most traders make similar mistakes when developing a strategy, and once they are identified, they can be corrected and improved. Still, it's not easy to rule out these issues; When shortcuts are not available, it becomes more difficult to move from strategy development to proper execution.
So, what are some of the most common mistakes and shortcuts? There are roughly three types. Below, we'll explain how each of these three types of mistakes can have a negative impact, and how to find and correct them. Avoiding these common mistakes will help build a better trading system.
Pitfall 1: The strategy is too complex to understand
In the process of trading, you will inevitably encounter some complex trading strategies.
For an autonomous trader, there may be a situation of "analytical paralysis": it is obvious that his charts are full of technical indicators, technical lines, and areas of support and resistance.
For an algorithmic trader, a complex method will consist of thousands of lines of code and require the optimization of dozens or even hundreds of variables.
Both methods have one thing in common: they are both extremely complex and involve a lot of things. A lot of inexperienced traders will think that this is a way to develop a system; They believe that the more indicators, the better the algorithm fits past data and the better the trading strategy will be. But that's not the case.
There is a fallacy when weighing a complex strategy that leads us to the wrong conclusion that this strategy is superior to the average strategy:
Just because a strategy can achieve better results on historical data does not mean that it will be equally successful in real trading.
In fact, constantly adjusting the strategy by adding indicators, or adding new rules to the algorithm makes the trading strategy more complicated; This practice usually only gives traders a false sense of confidence. Improving and adding more rules to a strategy doesn't mean it's going to get better.
It may be hard for many people to believe that a simple trading strategy is usually best.
For self-directed traders, a relatively concise chart with only one or two technical indicators, combined with a deep understanding of price movements and market dynamics, is often much better than a chart full of technical lines and indicators.
For algorithmic traders, a simple entry order is usually better than a rule that requires 5 to 10 conditions to be met before a trade can be executed.
Pitfall 2: A trading strategy that doesn't take into account market frictions
If you compare the trading systems on the market, you will find that most of them will list a small disclaimer: commissions and slippage spreads are not included. Similarly, many people who develop their own systems ignore commissions and slippage costs; Even when they take these costs into account, they often underestimate the actual amount.
You'll hear that the system doesn't include commissions and slippage for a variety of reasons. The most common excuse is that "different brokers usually charge different commissions"; Another common reason is "my system only uses limit orders". But the real reason is:
This can make the system look better. If you take into account the actual transaction costs, it is even more difficult to find a system that can make a profit.
Take, for example, a system that trades E-Mini S&P 500 futures (CME:ESM14) that uses a technique to capture very small market movements through micro-trades and frequent intraday buying and selling hedging for speculation.
Excluding commissions and slippage, the system makes 20 trades per day, earning an average profit of $15 per trade. Traders see a profit of $300 per day and will think that these trades are not bad. But if you add a commission of $5 per round and a slippage of 1 basis point (which is probably already an optimistic estimate), a profit of $300 per day turns into a loss of $50 per day.
One potential effect of a trading strategy that doesn't take into account commissions and slippage costs is that the system may cause traders to make too many trades.
Here's an example. Let's say System A is the small speculative system mentioned above, and you can make a profit of $300 per day without taking into account commissions and slippage costs. Comparatively speaking, System B only makes one trade per day, regardless of transaction costs, and can make an average profit of $50 per trade. Anyone comparing these two systems would choose System A.
However, when commissions and slippage are added, the result is the opposite; System B is the only strategy that is worthwhile.
The number of trades is less, and the proportion of commissions and slippage to total profits is much smaller.
Therefore, it is crucial to take into account a certain amount of commission and slippage costs at an early stage of system development. For E-Mini S&P 500 futures, the assumption of a commission of $5 per trading round and slippage of one or two basis points is relatively reasonable.
Pitfall 3: System testing uses all historical data
The third mistake that many traders can make when developing the system is to use all available historical data when testing the system. Most inexperienced traders will optimize and analyze all the historical data up to today. They do this because they want to make sure that the strategy has reflected adjustments to the latest data.
Of course, if the first test fails, the trader will add some rules or filters to the system (so that it is more likely to make the mistake of the previously mentioned strategy being too complex) and run all the data again.
Eventually, the trader will find a viable trading strategy system and apply this system to real trading. But when something goes wrong with a trading strategy, it's almost always a surefire response.
A better, but also more difficult, method is to validate a trading system by testing out-of-sample data.
For example, a trader may use data from the last 10 years to develop a trading strategy, but keep the data from the most recent year. When the system is developed, test with unknown (i.e., out-of-sample data); If the system works well, then the system can be used for real-time trading.
In addition, a moving pane test can be taken. This approach, which uses multiple out-of-sample time periods, is more likely to be successful because the backtest net worth curve already fully includes out-of-sample optimization results.
One of the drawbacks of the moving pane test or out-of-sample test is:
Once an out-of-sample test has been run once, any further testing is no longer based on true "out-of-sample data".
Therefore, if the test has been run many times, it is easy to inadvertently turn an out-of-sample test into an in-sample test. However, the out-of-sample testing method is still better than the one that optimizes all the data.
There are no shortcuts to system development
Designing a viable trading strategy is quite difficult. In fact, a lot of traders never really do this; Many times, they take shortcuts or make oversimplified mistakes in the development of the system. Of course, it's much easier to add rule after rule, filter after filter in the system, than it is to find a suitable and concise rule.
Similarly, it's fairly easy to find a viable strategy without taking into account the frictional costs of commissions and slippage.
Finally, optimizing all available historical data is a relatively simple approach compared to moving pane testing or out-of-sample testing, and the results will appear to be better, but actually make no sense.
The crux of the matter is that if we use a method of optimizing all the data to build a trading strategy system, it is equivalent to designing a trading strategy that can only be profitable if it is applied within the tested time sample. Of course, this requires having a time machine, which is obviously more complicated than building a profitable system.
The takeaways we take away from the common mistakes we make in the development of the above systems are:
If one approach makes the system design easier, or if the success rate of backtesting is significantly higher, then this is actually a warning sign that something may be wrong. A proper and proper system development process is always difficult.