Turtle Trading Rules

Chapter 20 The Lies of the History Test (1)
Liars and scoundrels lurk in dark corners, waiting for unsuspecting prey.Don't be their meal.

The Stonehenge Plus system turned $5 into $5000 million in just 100 years. Stonehenge Plus was invented by Stupendus Magnificus, a NASA scientist who discovered a way to use the same programs used to launch Mars rovers in forex trading superior.With an accuracy of more than 90%, this system has not lost money in a single month in 10 years.It's so rare that we're only going to sell 100 sets.Get your set for only $1999, buy it now, don't miss out - an ad by a system peddler Trading enthusiasts included in the mailing list have seen such advertisements too.But buyers beware: there are bogus amateurs using irresponsible marketing and unrealistic postmortems to tout their latest inventions.There are plenty of peddlers who know full well that their systems will never deliver the returns they boast about.Many people also intentionally tweak testing methods to glorify their systems in an exaggerated way.However, not all vendors are so brazen.Some people do think their systems work, but they either don't realize that their underlying methodology is flawed, or they don't understand the limitations of historical testing or the drawbacks of using historical testing results to predict the future.Of course, there are some people who are very good at avoiding the pitfalls of historical testing.Unfortunately, such people are few and far between, and it can be difficult for an inexperienced trader to tell the difference between a system that has been developed with good testing methods.

Even seasoned traders often wonder why their systems perform far worse than historical simulations in real trading.They know this phenomenon exists and they will try to fix the problem, but they don't understand the root of the problem.In fact, the difference between historical test results and actual trading results is mainly caused by four factors:
Trader Effect: If a method has made a lot of money recently, other traders are likely to notice it and start imitating it with similar methods, which can easily cause the method to not work as well as it did in the first place .

Random effect: The phenomenon that the results of historical testing exaggerate the inherent advantages of the system may also be pure randomness.

Optimization paradox: The process of choosing certain parameters (such as choosing a 25-day moving average instead of a 30-day moving average), which may reduce the predictive value of a post-hoc test.

Overfitting or curve fitting: A system may be so complex that it loses its predictive value.Because it fits so well with historical data, a slight change in market behavior can cause a significant deterioration in performance.

trader effect

There is a concept in physics called the observer effect (observer effect), which means that the act of measuring a phenomenon sometimes affects this phenomenon, and the observer's observation behavior disturbs their experiment.Something similar happens in the world of trading: the act of trading itself has the potential to alter the underlying market state upon which the trade is based.I call it the trader effect.Anything that repeats itself has the potential to be noticed by market participants.Likewise, a strategy that has performed particularly well in the recent past is likely to be noticed by many traders.However, if too many traders start trying to utilize the same strategy, the strategy will no longer work as well as it once did.

Let's consider a breakout strategy.If you know a market is relatively small but there will be a lot of big traders buying on breakouts, how do you make money from them?Is there a winning strategy similar to the money printing machine?

You will buy ahead of other traders, and take advantage of the situation to push the price up to a certain level, which will trigger buying orders from these big traders.You'll then sell your position to them for a guaranteed profit.In effect, you are manipulating the price and taking advantage of other buyers.

Suppose you are a gold trader.What would you do if you knew that ACME would buy 410.5 August gold contracts aggressively at $1000 each?
If your buying volume is large enough to push the price up to this limit price point, you can sell at this point.If the current price is far from this limit, you may not have enough funds on hand to push the market to this level.But if the current price was already close to that point, say $408 a share, then a few buy orders could push the price high enough to trigger more buy orders from ACME.

Since your strategy is to buy first and sell quickly, you may change the meaning of the breakout point itself.Without the Trader Effect, a breakout could indicate that a resistance level has been broken, increasing the likelihood of a market move in your favour.But if you add in the impact of new buy orders, the meaning of the breakout will be changed, because the purpose of these buy orders is only to push the price up to the level of the breakout point.

for example.Assuming no one wants to buy above $408 per contract, but someone is willing to sell 409 contracts above $1000 per contract, these sell orders will act like a ceiling, preventing the price from breaking through $409 per contract.Without your buying order, the market would not have advanced to the high of $410.50 per share, so the breakout would not have occurred.Therefore, if you simulate a system based on the breakout method, there will be no breakouts and no trades will occur.

Now imagine that you entered the market under the same circumstances and took all of those 409 contracts at an average price of $1000 each.Now, there are no sellers at that price, so you have to buy another 411 contracts from the sellers asking $100 each to push the price up a little more.This transaction triggers a buy order from ACME, at which point you can sell those 411 contracts to ACME at $1000 each.As good as ACME feels about itself, the real success is you.The last thing to do is to dispose of the remaining 100 contracts.With no buyers left at the recent high prices, you're forced to sell low, back to $407 per copy.Not counting commission costs, you lost $100 on those 4 contracts ($4 per ounce, 100 ounces per contract, 100 contracts total), but you made $1000 on those 20 contracts ( $2 per ounce, 100 ounces per contract, 1000 contracts in total), making a total of $16.That's a pretty good result for an operation that takes a few seconds.

What about the traders at ACME looking to take advantage of the breakout method?They have a huge losing position, and they are in it for reasons that have nothing to do with their historical testing.This is a consequence of the trader effect.

Here is another example.There was one system a few years ago that became very popular due to its excellent performance over the years, and many brokers started offering it to their clients.At one point I heard that hundreds of millions of dollars have started following the system.But not long after its peak of influence, its adherents suffered a protracted decline of such length and severity that its 20-year history had never been tested.This system has a weakness that can be easily exploited.According to its rule, if the closing price of the day exceeds a certain level, then buy or sell at the opening of the next morning.Since other traders know what price levels trigger these buy or sell orders, they can simply buy before the day's close and sell right after the next day's open.The selling price is usually much higher than the buying price, because all the buying orders generated overnight are entered into the market at this time, which is determined by the laws of the system.

To make matters worse, the market mix chosen by the system inventors also included some less liquid markets, such as lumber and propane.For such markets, a relatively small volume of transactions can cause considerable volatility to the market, yet so many system followers flock to these markets.I believe that one of the reasons for the system's sudden and unprecedented decline is that this anticipatory buying once destroyed its edge.Other traders are not so stupid.They take advantage of any repetitive patterns they notice.Because of this, it is better to develop your own system than to follow someone else's.If you can develop your own system, your advantage is less likely to be destroyed by other traders, because they will not know when you buy or sell.

When we were playing for Rich, we used to get in the market around the same time.Other traders know that if they start getting big orders from us, those orders are likely to continue for a while.So, from time to time, floor traders and brokers will be front-runners, leading to early market moves.Since we are using limit orders, they do it risky (which is one of the reasons we use limit orders), because in case the market moves ahead of time, our order may not be filled, so we will withdraw one.Sometimes, when I want to buy or sell, but know that the market is easy to move ahead of time under the influence of traders' expectations, I will deliberately place a false order in the opposite direction.In this way, if the market really moves after hearing the news, I will cancel the initial hypothetical order and issue a real order with a limit price that is close to the market price or even more favorable than the market price.For example, if I want to buy 100 contracts, I might first place a fake sell order.Suppose this fake sell order asks to sell 415 contracts at a price of $100 each, and the current market situation is that the bid price is $410 and the ask price is $412, then the appearance of my fake sell order may change the market price to a bid price of $405 U.S. dollars, selling at $408.At this time, I can cancel the fake sell order and place a real buy order with a limit price of $410. This buy order is likely to be filled at $408 or $410 each, which is lower than the market sell price before I placed the fake order.

But this method should not be used too often, as long as it can make other traders guess what we are doing.In some ways, it's a bit like the "bluffing" technique in poker, where you deliberately bet big to confuse your opponent when you have a bad hand.You can't bluff endlessly, or you will be found out sooner or later, and you will lose in the end.But it's useful to bluff once in a while, as it tends to confuse your opponents and instead risk calling when you have a winning hand, giving you more chips for nothing.Also, bluffing itself can be a surprise and increase your odds.

Just as the occasional bluff can throw opponents at the poker table into a tizzy, the Turtles will find a way to slightly confuse those trying to figure out Richard Dennis.Some of us use a small stop loss standard, some use a large stop loss standard, some buy when the breakthrough occurs, some buy after the breakthrough, and some buy before the breakthrough.Taken together, we let off quite a few smokescreens that probably helped Dennis execute his trade a little bit.

Note that the trader effect can occur in any situation and is not necessarily the result of some traders deliberately preempting it.As long as too many traders simultaneously try to take advantage of a market phenomenon, the advantage of that phenomenon is destroyed, at least not for a while, because the orders of many traders will weaken its advantage.This problem is especially prevalent in arbitrage trading where the upside is relatively small.

random effects
Most traders have no idea how much pure randomness can affect their trading results.At this point, the understanding of ordinary investors is not even as good as that of ordinary traders.Even very sophisticated investors, including those who make decisions at pension funds and hedge funds, often don't know how large this effect can be.In fact, the impact of random events alone can make a world of difference in trading results.When random events are included, the level of variance across a series of historical simulation tests is surprisingly high.This section will address the issue of pure random effects in relation to long-term trend-following strategies.

When I mentioned the concept of odds, I said that I had simulated a random entry strategy that used only a computer-simulated coin toss to determine whether to open long or short.At that time I designed a complete system with a coin toss based entry strategy and a timed exit strategy - exiting a number of days after entry, anywhere from 20 to 120 days.I then tested the system 100 times using the same data that we used in Chapter 10 to compare different trend-following strategies.Of the 100 tests, the best one earned an average annual return of 16.9%, turning $10.5 million into $100 million over a 550-year test period, but the worst one lost an average of 20% per year .This illustrates that purely random events can make a huge difference.

What if we add a little edge factor?What if we add a trend filter like the one in Donchian's trend system and make the system resemble a trend-following system?With such a change, our market entry decision is still random, but the premise is that the market entry action can only be consistent with the direction of the general trend.It's an interesting question, because no matter what trend-following funds you look at, there's a lot of variation between good and bad.If a fund outperforms others, its managers will of course say it is the result of superior trading strategy and execution.In fact, superior performance may also be due to random effects rather than superior strategies.You can understand this better if you consider how much such a random effect can still have in a situation where the system has an advantage.

If we add a trend filter with a positive advantage to this fully randomized system, the average performance over 100 tests improves significantly.According to my tests, the average return increased to 32.46%, and the average decline decreased to 43.74%.But even with the filter in place, there is still considerable variation between the results of each test.Among the 100 random tests, the best one achieved an average annual return of 53.3% and a MAR ratio of 1.58, and the largest decline was only 33.6%; but the worst one had a return of only 17.5% and a maximum decline of 62.7 % big.

(End of this chapter)