Turtle Trading Rules

Chapter 21 The Lies of the History Test (2)
Luck, or the element of randomness, plays a big role in the performance of traders and funds, even though the best traders don't want to admit it to their investors.Historical performance is loud and hard evidence in the eyes of investors, but in fact it is not that hard.For example, if you invest in a fund, you generally hope that the future performance of the fund will still be as good as it has been in the past.The problem is that historical performance is also a matter of luck.Some funds are managed really well, but luck is average; some funds are managed poorly, but luck is very good.If you only focus on historical records, you can't tell the difference between strength and luck.The random effects are too large and pervasive for you to draw firm conclusions.

Consider the best result in the 100 tests mentioned above.If your trading style is conservative, say your risk level is only 25% of that of the Turtles, one of the tests will give you an average annual return of 10% over 25.7 years and a maximum drawdown of only 17.7%.We all know that a trader who enters the market randomly is unlikely to perform at this level in the future, because the random strategy has no advantage.Unfortunately for a person who believes only in the historical record, there are always some lucky ones among the many traders who look good but are actually extremely mediocre.

Lucky

We can also recognize random effects from natural phenomena.Human qualities such as intelligence, height, physical ability, and singing ability are all products of random effects.If you have good genetics for a trait (that is, both of your parents have it), you are more likely than most people to have it, even though you may not have the trait to the extent of the parents.If both your parents are tall, you are likely to be tall, too, but the more above-average parents you have, the more likely you are to be shorter than them.

In genetics and statistics, this pattern is known as mean reversion or the regression effect.Your tall parents also had the genes for being tall, and have a very lucky combination of genes for height.But a person who is lucky enough to have a tall genetic combination can pass on the genes but not the luck, so their child is more likely to be close to the average height because the child is less likely to be as tall as the parents" lucky" genetic combination.

When you use performance measures to distinguish good funds from bad funds, you can easily run into random effects problems.Because there are more mediocre traders who are lucky than good traders who are unlucky.Suppose there are 1000 traders, 80% of them are close to the average, and there are only five or six real masters.Well, only five or six people have the potential to be good traders with bad luck, while 800 mediocre people have the chance to have good luck.If 800% of these 2 people are lucky enough to have a good record for 10 years (as you can see from the test mentioned above, the actual rate may even be higher than 2%), which means that there are certainly 21 people with a good record. individuals, but only 1/4 of them are really good traders.

really good trader

Time favors the really good traders over the mediocre lucky ones.Even if 800 of those 16 had a good 10 years, their performance would likely be mediocre over the next 15 years.On the contrary, if you only consider the record of the past 5 years, there will be a sharp increase in people who seem to be good but are actually just lucky.This is because the impact of random effects is more pronounced in the short run.

In our test, if we shorten the test time, for example, only look at the situation from January 2003 to June 1, how will the difference level change?According to the test results, the average performance of the random entry system during this period is mediocre, with a return rate of 2006% and a MAR ratio of 6.This performance is far worse than those of the real system: the return of the triple moving average system is 35%, and the MAR ratio is 1.06.The return of the Bollinger breakout system is 48.5%, and the MAR ratio is 1.50.The dual moving average system also has a return of 52.2% and a MAR ratio of 1.54.

那么，有多少幸运儿从那100次随机测试中产生呢？有多少人仅凭好运就击败了我们的最佳系统呢？在100次测试中，有17次的MAR比率高于1.54；在这17次中，有7次的回报率超过了52.2%。最好的一个随机交易者获得了71.4%的回报率、34.5%的最大衰落和2.07的MAR比率。如果你还想靠3年历史记录寻找优秀的交易者，请想想这些数据吧。

When you look at short-term history, you should understand that there is a large element of luck in the performance you are seeing.If you want to know whether a trader is just one of the lucky mediocrities or one of the few true masters, you should dig deeper than the record and take a good look at the people behind it.

Good investors invest in people, not history.When they observe traders, they know which traits predict future superior performance and which traits reflect mediocre ability.This is the best way to overcome random effects.There is good news for those who are doing historical testing: if there is a chance that the test results are due to random effects rather than systematic dominance, you can easily spot this.We’ll get to that later, but for now, let’s look at two other reasons why historical test results don’t match actual trading results.

optimization contradiction
There is also an effect that creates discrepancies between historical test results and actual trading results, which I call optimization paradox.This contradiction creates a lot of confusion, especially for those new to computer simulation.Some trading systems need to use specific values ​​for calculation, and the process of selecting these values ​​is optimization.These values ​​are called parameters.For example, the number of days for calculating the long-term moving average is a parameter, and the number of days for calculating the short-term moving average is also a parameter.Optimization is the process of selecting the best or optimal values ​​for these parameters.There are many traders who believe that optimization is a bad thing because it leads to curve fitting phenomena and poor performance.I say this is nonsense!

Optimization is a good thing if done well, because it is better to know the effect of changing parameters than to ignore it.When we examine the effects of varying parameters, we often find indications that the behavior of the system is the result of random effects or curve fitting, rather than a reflection of system dominance.The so-called optimization process is nothing more than observing the impact of adjusting parameter values ​​on transaction results, and reasonably deciding what parameter values ​​to use in actual transactions.

Some traders think optimization is harmful or dangerous simply because they don't understand the optimization paradox and have seen the consequences of improper optimization - which is what statistics call The source of the overfitting phenomenon.

The so-called optimization contradiction means that the parameter optimization process has two contradictory effects: on the one hand, it can increase the probability that the system will perform well in the future, and on the other hand, it will reduce the probability that the system’s future performance will meet the simulation test results.Thus, while parameter optimization improves the expected performance of the system, it also reduces the predictive value of historical simulation metrics.I believe it is because of this lack of understanding that many traders shy away from optimization out of fear of over-optimization and curve fitting.But in my opinion, proper optimization is always wise.

Using proper optimized parameter values ​​can increase the likelihood of the system achieving desirable results in real trading.An example helps us understand this.Consider the Bollinger breakthrough system, which has two parameters: one is the long-term average price, and the other is the standard deviation. The standard deviation of the long-term average price plus or minus a certain multiple within a certain period of time is the volatility channel of the system.Figure 11-1 reflects the MAR ratio of this system under different standard deviation parameter values. The horizontal axis represents the channel width, which is the standard deviation multiple, ranging from 1 to 4 times.

As can be seen in the figure, 2.4 standard deviations correspond to the best simulation results.Any entry criteria less than or greater than 2.4 standard deviations lowers the MAR ratio.

Now let's see if optimization is actually beneficial.Suppose we did not consider the optimization of the channel width, but chose a parameter value of 3 times the standard deviation based on subjective feeling - because we remember that the statistics textbook said that for the normal distribution, more than 99% of the values ​​will be Fall within the range of plus or minus 3 standard deviations of the mean.If the future wasn't too different from the past, we'd be missing out on a lot of profits, and we'd have a much bigger drawdown than 2.4 standard deviations.How big is the gap?Just look at a few data: in 10 and a half years, assuming the same level of decline, the profit under 2.4 times the standard deviation is 3 times as much as 8 times the standard deviation, and the ratio of the average annual return between the two is 54.5 % to 28.2%.

Not optimizing means being completely at the mercy of luck.Discovering the impact of adjusting this parameter, we better understand the role of the entry criteria parameter and the sensitivity of trading results to this parameter.We now know that if the channel is too narrow, there will be too many trades, which will weaken the performance of the system; if the channel is too wide, you will miss a lot of trends while waiting to enter the market, which is also not good for the system.If you forego optimization out of fear of over-optimization and curve fitting, you are not gaining the insight that could greatly improve your trading results and provide you with the tools to design better systems in the future. Some new ideas.Several other parameters will be introduced below, and you will see that their changes also correspond to changes in the peak or hill shape of the system performance.

It reflects the influence of the calculation days of the moving average closing price on the MAR ratio. The calculation days of the moving average price can determine the center line of the Bollinger Band volatility channel, ranging from 150 days to 500 days.

As shown, 350 days corresponds to the best test results.Any parameter value greater or less than 350 days will reduce the MAR ratio.

It reflects the MAR ratio under different exit standard parameters.The exit criterion is a parameter that specifies the exit point of the system.When we introduced the Bollinger breakout system in the previous article, we said that when the closing price crosses the moving average (that is, the center line of the channel), the system exits the market.My purpose in this test was to see what would happen if the system exited the market after or before this crossing point.In the figure, a positive exit criterion parameter value represents the number of standard deviations above the moving average for long trades and the number of standard deviations below the moving average for short trades.Conversely, negative parameter values ​​indicate below the moving average for long trades and above the moving average for short trades.

让我们看看退出标准参数值从–1.5逐渐变动到1.0的影响。如图11–3所示，参数值达到–0.8时，测试结果最好。任何高于或低于–0.8的值都会降低MAR比率。

(End of this chapter)