Mechanical Trading Systems by Richard L. Weissman

Contents  ˅

Preface 1) Dispelling Myths and Defining Terms: Mathematical Technical Analysis and Mechanical Trading Systems 2) Mathematical Technical Analysis: A Building Block for Mechanical Trading System Development 3) Trend-Following Systems: A Matter of Fortitude 4) Mean Reversion Systems: A Matter of Patience 5) Short-Term Systems: A Matter of Quick-Mindedness 6) Knowing Oneself: How to Challenge Your Knowledge 7) System Development and Analysis: Benefits and Pitfalls 8) Price Risk Management: Schools of Price Risk Management and Other Considerations 9) Improving the Rate of Return: Improving Returns by Expanding the Comfort Zone 10) Discretion and Systems Trading: Discretion within a Mechanical Framework 11) Psychology of Mechanical Trading: Trading Systems and Transformational Psychology Notes References) and Further Reading

Richard Weissman says the two aims of this book are trading system development and trader development. It's a combined process which leads to not only better traders, but better lives. The trading system design process requires an open mind, patience, objectivity, and discipline. The result is skills which can be applied to life beyond the markets.

Each chapter presents a different step in the system design process and its relationship to trader psychology. The idea is to develop a workable trading system which fits the trader's personality.

Several trading systems are given to illustrate the main development concepts. Despite their simplicity, trading systems like these are usually more reliable than more complex systems. These sample trading systems also provide a framework for a trader's mental skills development.

Basic Definitions

Technical analysis: Forecasting future price trends based on price history. The forecasting method may be subjective, objective, or a combination.

Simple moving average: A simple moving average is the average of a fixed number of data points, usually prices. The average is recalculated as a new data point is added and the oldest data point drops off. Moving averages are often used for indicating trends or providing entry and exit signals.

Stop-and-reverse: A trading system which is always in the market. When a security is bought and then later sold, a new short position is also taken. And when the short position is closed out, a new buy is also made.

Mechanical trading systems: Methods which produce trade entry and exits signals without requiring a trader to make decisions.

Time frames: A time period which can define trades or the length of a trader's typical trade. In this book, short-term is less than 10 days, intermediate-term is 10 days to 1 month, and long-term is 1 to 6 months.

It's reasonable to think people act in their own best interests, especially when it comes to money. Basing their models on this assumption, economists have viewed markets as efficient. The efficient market concept explains why it's hard to make money in the markets. When buyers and sellers have the same information, neither side has an advantage and the markets are always in balance.

But years of economic theory was disrupted by studies in behavioral finance. Instead of people always acting logically when making decisions about money, they often acted irrationally. It was found that emotions and biases can cause people to often act contrary to logic and their own self interest.

Behavioral finance was a revelation to economists, but not to market technicians. They knew it's the flaws in human thinking, which allows technical analysis to work.

Whether trading with a price trend or against it, either approach can work. Because markets are not efficient, prices lack the balance which would exist if markets were efficient. Instead, price distribution favors the two basic types of trades, mean reversion and trend following.

Market prices never move just straight up or down, but zig zag up and down. This back and forth movement is called reversion to the mean (price average). Mean reversion makes trades based on these natural minor retracements ("corrections") which occur during a major move.

An example of a mean reversion indicator is Welles Wilder's relative strength index (RSI). The indicator was designed to show when prices became overextended to the upside or downside. A trade might be entered or closed out at these extremes in anticipation of prices reversing back to the mean.

Markets typically move sideways about 70 percent of the time. This leaves just about 30 percent of all price action is in up and down trends. However, while trends may exist in only a minority of the time, they can result in substantial and profitable price moves.

Moving averages are often used as an indicator of trend direction. A simple trend-following system might buy when price crosses above a long-term moving average and sell when price dips below the average.

Even though a trading system may be mechanical, a trader is not. Using a trading system requires following its signals regardless of conflicting emotions and thoughts. Ideally, using a trading system should be so automatic, so habitual, it actually becomes boring.

Trade entries and exits are two areas which require a trader to ignore natural instincts. Looking for the perfect entry technique is common, especially among beginning traders. Perfect entries and exits are only found in hindsight. This misguided search for perfection is not only unnecessary, it's counterproductive. A trader misses many good trades by constantly passing up valid entry points.

Instead of looking for the best entries, traders should focus more on trade exits. Exits are what determines the amount of trade profit or loss. Unfortunately, the force of emotions cause traders to hold losing trades too long and exit winning trades too soon.

Whether showing an open-trade loss or profit, traders convince themselves prices will soon reverse. So a small profit is taken and a loss is allowed to grow. Ironically, both happen in an attempt to avoid taking a loss.

While a trader needs to avoid responding to certain natural instincts, technical analysis is meant to profit from the same natural feelings in others. A simple example of this is how support and resistance price levels form.

Support and resistance levels are established when prices reverse at approximately the same level as they had previously. When there's a price level where prices previously bottomed, this encourages traders to buy at this support level. And when prices move back to a previous peak, it induces traders to sell, which makes this a resistance level.

Prices may also find support or resistance at certain percentage of a previous price move. Common percentages are 38, 50, and 62. If a stock bottomed at $10 and then doubled to $20, a 50 percent retracement of the move would be back to $15. It's likely some buying would come in at this level and it might be enough to make $15 a support level.

Like price, technical indicators can trigger a market reaction. A 200-day simple moving average is a popular measure for determining the major trend of a market. If prices are above the moving average, the trend is considered up and if below, the trend is down. So due to its widespread use, when price first closes above or below the 200-day moving average, this can produce some buying or selling in response.

Mathematical technical indicators can be divided into two categories. Indicators which follow trends are one category and those which indicate price reversion to the mean, the other. Both types can be the basis for creating a mechanical trading system.

Trend-Following Indicators
Trend-following indicators work for two reasons. When prices are not going sideways, they move in up and down trends. Since some of these trends result in large price moves, they provide the trend trader with significant profit.

The second reason trend indicators work is because they are designed to take large profits and small loses. This is contrary to the natural human instinct which results in taking small profits and large losses. Trend following provides the discipline to automatically let profits accumulate when holding winning trades. Conversely, losing trades are closed out quickly to avoid growing losses.

Moving averages are the most popular type of trend-following indicators. They can be calculated in a variety of ways, but the most common is the simple moving average. Despite its popularity, the construction of a simple moving average presents a problem.

All the prices or data points in a simple moving average are given equal importance. So in a 200-day simple moving average, today's price is given no more significance than the price 200 days ago. This equal weighting of all data points makes long term moving averages slow to respond to recent data.

To counter the lag of a simple moving average, exponential and linearly weighted moving averages give proportionally more weight to recent data. Giving greater weight to prices which happened on larger volume is another way to adjust a moving average.

Giving greater weight to more recent data is one way to reduce a moving average's inherent lag. A shorter period moving average will also reduce the lag. Since a short period moving average consists of only recent data, it naturally responds quickly to recent changes.

When used for following trends, a moving average must be responsive to recent prices, but not overly so. If a moving average responds to easily, it results in frequent small profits and losses without the occasional large win. Adding a condition or filter to validate moving average signals can help avoid excessive price sensitivity and the resulting bad trades.

Adding Moving Average Trend-Trade Filters
One type of trade filter consists of a time factor. System entries and exits are typically triggered by a single price bar's action. A buy might be signaled if price closes above a moving average and a sell signal on a close below. Instead of only a single close, two or more consecutive closes could be required for a valid signal. Another type of time filter would be to adjust the moving average's period in response to changes in price action.

Using a moving average envelope is another method to help avoid bad trade signals. A fixed percentage of a moving average is both added to and subtracted from the moving average. This forms an envelope of two bands, above, below, and parallel to the moving average. A trade could be entered upon penetrating one of the bands and then exited after penetrating the moving average.

A third means of filtering out bad trade signals is to use two or three moving averages instead of one. The moving averages would have different periods. With two moving averages, trades could be signaled when the shorter-term moving average penetrated the longer-term moving average.

One or two moving average systems are typically always in the market, every exit is also a reversed position. However, three moving average systems can allow for closing out a position without reversing into a new trade.

A system could consist of a short-term, an intermediate-term, and a long-term moving average. A trade might be signaled when the short-term moving average closes beyond the intermediate-term moving average which is also beyond the long-term moving average. The trade is closed out when the short-term moving average moves back through the intermediate term moving average. A new trade is made when all three are in the same alignment.

The easiest way to avoid moving averages problems is simply to eliminate them. The channel breakout system is a method similar in concept to the moving average envelope. Trades are signaled when price exceeds the previous highest high or lowest low for a given period. Like the moving average envelope, the price channel helps filter out minor random price action.

Some technical indicators can both confirm other trend system signals or used as a stand alone system. Gerald Appel's moving average convergence divergence (MACD) along with Welles Wilder's directional movement indicator (DMI) and parabolic are examples of dual purpose indicators.

Mean Reversion Indicators
While trends happen only about 30 percent of the time, mean reversion can work in all market periods. Mean reversion relies on times when emotionally fueled trading drives markets to minor or major extremes. Trend trading seeks to avoid the problems caused by human emotions, but mean reversion depends on them.

Most mean reversion indicators are oscillators. Their purpose is to identify counter-trend entry points in order to trade against ("fade") the underlying trend. Weissman categorizes mean reversion indicators as percentage, differential, or statistical oscillators.

Two popular percentage oscillators are stochastics and the relative strength index (RSI). Stochastics operates on the concept that price extremes are indicated by a sequence of prices closing towards one end of a price bar's range. Stochastics takes the latest close and expresses it as a percentage of the range for the previous 9 or 14 prices. Different combinations of raw figures and moving averages are used to create the indicator. Price extremes are identified by their proximity to the upper and lower end of a scale from 0 to 100.

RSI divides the total price gains made by higher closes by the total price losses made by lower closes for the previous 14 periods. This value is converted to an index plotted on a scale of 0 to 100 with price extremes indicated by values near either end of the scale.

Differential oscillators are created by subtracting one data set from another. Typical examples are the moving average convergence divergence (MACD) and the directional movement indicator (DMI). Both create an indicator from subtracting one moving average from another.

MACD uses an exponential average of the differences between two other exponential moving averages of closing prices. DMI is the simple difference between the positive directional movement average (+DI) and the negative directional movement (-DI) for the previous 14 periods.

The momentum and rate of change (ROC) oscillators compare the difference between two prices instead of two moving averages. Momentum subtracts a fixed period close from the current close. The ROC divides the current close by a fixed period prior close.

Unlike percentage oscillators, differential oscillators have no fixed scale. This makes determining price extremes more subjective and difficult.

Statistical oscillators compare current price activity on a statistical basis with how prices varied in the recent past. Bollinger Bands and the commodity channel index (CCI) are two examples of statistical oscillators.

The Bollinger Bands' reference for standard deviation is the previous 20 periods. Two standard deviations values are added to and subtracted from a simple 20-period moving average. Price extremes often occur when prices reach or exceed one of the bands.

The CCI subtracts a 20 period simple moving average from the current price and divides the result by 0.015 times the mean deviation. Extreme readings can range from +200 to -200. Although originally designed for trend following, it's more commonly used for counter-trend trading.

According to the author, traders have either a trend-following, mean reversion, or day-trading personality. Who you are determines how you should trade.

A mechanical trend-following system simplifies the trading process. Determining entries, exits, and managing risk become easy. But those are the only easy things. Sitting through account drawdowns and strings of losing trades requires mental toughness and utmost confidence in the system.

Planning for Indicator-Triggered Trades
If an indicator signals a trade, you need to determine when the trade order is executed. The signals may be valid during the day or only at the close. If a trade is signaled during the day, there's the risk the signal won't remain valid at the close.

If signals are only valid at the close, the trade may be executed either at the close or the next day's open. At the close orders run a similar risk to a intraday trade.

Prices near the close might be swinging above and below the signal point. So just before the close, the price might signal a trade and a trade is entered. But if prices swing back at the close, there is no signal and the trade is invalid. The only way to make sure a trade is valid at the close is to wait and take the trade on the next day's open.

Planning a Portfolio
It's important for system reliability to backtest multiple assets, but they need to be uncorrelated. To avoid correlated assets, choose futures from different commodity groups, stocks from different sectors, and forex pairs which are unrelated.

The markets chosen should have good volume and sufficient volatility. The position size for the different markets should be roughly the same, so one or two markets don't skew the results.

Another consideration when testing a portfolio is its maximum drawdown. To be accurate, it's necessary to measure the portfolio drawdown of open positions on a daily basis. Since not all testing software can do this, this lack of testing evidence should be recognized when planning live trading risk.

Planning the Data Used
When backtesting futures and stocks, an important consideration is how to handle discontinuous prices. Normally, it's the nearest futures contract month which is used for testing. But when the contract expires, usually it's at a different price than the next contract month's price. Similarly, there's a price gap between the day before and the day of a stock split.

Data providers can supply price histories which are adjusted to close the gaps. But this can result in a couple of other problems to be aware of. The first problem is, all the adjusted pre-gap prices are not the same as what they originally were. Secondly, the accumulation of multiple downward gap adjustments could make the oldest futures prices negative and stock prices close to zero.

Two Moving Average Crossover System
In general, moving average trading systems can give robust results with a few simple rules. The two moving average crossover is an example of a basic yet effective trend-following system. Like the other systems tested, each trade had $100 deducted for slippage and commissions.

For the ten year period tested, 8 of the 10 portfolio markets were profitable with an annualized return of 8.48 percent. The profit to maximum drawdown was 4.24. The results had the typical traits of a trend-following system. The win rate was less than 40 percent, the portfolio had 10 consecutive losses, and a maximum drawdown duration of 635 days.

Ichimoku Two Moving Average Crossover System
This system adds a whipsaw trade reduction rule to the basic two moving average crossover method. The system requires the 26-day moving average to turn in the direction of the crossover.

The results of this system show how adding a logical rule doesn't always translate to better results. Only 6 of the 10 markets were profitable and the profit to maximum drawdown was only 0.16. If whipsaws trades get reduced the average number of trades should be fewer and the trade duration longer, but neither happened. None of the Ichimoku test metrics were better than those for the basic two moving average crossover.

Three Moving Average Crossover System
The two moving average crossover system is always in the market since it reverses positions on every exit. The addition of a longer-term moving average allows positions to be closed out without initiating a new trade.

Overall, the three moving average system did worse than the two moving average crossover. With a lower profit total and a higher drawdown, the portfolio had a profit to maximum drawdown of 2.69. Adding the third moving average reduced the time in markets to under 68 percent. In general, this is a good indication since money is not always exposed to risk. But in this case, it wasn't good enough.

Ichimoku Three Moving Average Crossover System
Adding a moving average to the two moving average Ichimoku system creates two changes. The moving average addition allows the system to close out positions without always initiating new ones. Secondly, the new moving average changes the whipsaw rule.

The Ichimoku three moving average crossover system was a significant improvement over the two moving average version. The portfolio profit was 6 time greater and had a profit to maximum drawdown of 3.01. All the test metrics were also better than the plain three moving average crossover system. Compared to the plain two moving average crossover system, this system version achieved about 90 of the profits while in the markets less than 64 percent of the time.

MACD System
Like the other systems, this example is meant to be a starting point for further development, not a refined, ready to use system. The number of parameters and their relationship to each other with MACD, allows for an infinite variety of possible systems. In this system, the default parameters were the ones used. The two exponential moving averages were 13 and 26 days. The signal line was 9 days.

This system compares favorably to the two moving average crossover system. Its profit was larger but so was its drawdown. The result was a profit to maximum drawdown of 5.16. The most significant difference between the two systems was a much longer trade duration. The two moving average crossover trades lasted an average of 25 days and MACD trades averaged 143 days.

DMI System
There are several methods to reduce whipsaw trades with a trend-following system. One approach for oscillator based systems, is to require the oscillator to go further than just crossing the zero line. That is the method used for this system.

Systems like this show why choosing which rules to use is often a balancing act. Requiring more significant oscillator change can help to reduce whipsaw losses and drawdowns. But it also may reduce overall profit since it tends to enter a trade late. That may have been the case with this system. The drawdown was relatively low, but so was the profit. The profit to maximum drawdown was 1.93. The maximum drawdown duration of almost 5 years would have made the low profit even more difficult to accept.

DMI with an ADX Filter System
This system uses ADX as an additional trade filter to help reduce the number of bad trades made by the previous DMI system.

Instead of eliminating trades from the original system, the ADX filter actually added one more trade to the portfolio total. Both the profit and drawdown were reduced which resulted in a profit to maximum drawdown of 1.65. The system did manage to reduce the maximum drawdown duration to about 3½ years instead of almost 5.

Channel Breakout System
This system is based on the method created by Richard Donchian. The only difference is the trades are signaled exactly at the 20-day highs and lows instead of when they are exceeded. Weissman suggests varying the channel period and to also try using a exit signal period shorter than the entry period signal. Different periods for entry and exit allow the system to sometimes be out of the market.

Like the two moving average crossover, this is a simple system and yet effective. The profit to maximum drawdown was 3.54. Although the overall results were good, this system had 19 consecutive losses which tied it with the Ichimoku two moving average crossover as the worst among the systems tested.

Bollinger Bands System
This system uses a close beyond the upper or lower band to signal a long or short entry. Exits are triggered when price moves back to the center moving average between the bands.

The Bollinger Bands system proved to be very efficient. It produced a good profit while in the market only 51.18 percent of the time, the least of the tested systems. The average trade duration was also the fewest at 15.3 days. The profit to maximum drawdown was 3.79.

Comparing Systems
Testing systems is more than finding a profitable method. The comparison of system test metrics enables finding a system which matches a trader's personality. You may like a system's profit to maximum drawdown ratio, but not if the drawdown is too big for your account. Or maybe the consecutive losses and average trade duration are higher than you'd like. And even if all the other metrics look good, a system might not trade enough to suit you. In the end, the "best" system is what works best for you.

When comparing systems, it's important to compare the different asset class results. Testing multiple asset classes gives an indication how reliable a trading system is. But since stocks tend to be mean reverting and currencies tend to trend, eliminating an under performing class is reasonable.

When comparing systems, you might want to see the effect of taking one system's filter and applying it or a similar filter to the other systems. And filters can work equally well for both entries and exits. For instance, a stop-and-reverse system may be improved with a fixed percentage stop loss instead of the normal exit.

However, setting a fixed amount to exit a profitable trade is not recommended. Trend trading has most of the profits coming from only a few trades. And setting a fixed profit limit is a sure way to miss out on an exceptional trade. But however you decide to use a filter, the foundation advice to keep it simple always applies.

Traders use many ways to determine trends, but really it's a simple definition. If current prices are higher than previous prices, the trend is up. If current prices are lower, the trend is down.

But when are prices too high or too low? Unfortunately, there's no simple way to define when prices are overextended. This lack of simplicity combined with shorter time frames is what separates mean reversion trading from trend trading.

The causes for a temporary price extreme are different for each instance. This variation of causes makes it hard to choose the best parameters for an effective mechanical mean reversion trading system. In fact, some level of trade discretion may make sense.

Changes Made for Testing the Mean Reversion Systems
It's possible for a mean reversion system to trigger the stop loss on the same day it hits the profit target. Because 10 years of intraday price data was not available for testing, on the days when both prices were hit, the trade was counted as a loss.

Percentage based stops can work well except for assets with back-adjusted price data, like futures. The solution chosen was a fixed dollar stop for only the first system tested. If prices vary widely over a test period, Weissman recommends the use of an indicator based stop loss, such as a Bollinger band.

The initial portfolio had ten markets. Nine were from asset classes which trend well and one which is prone to mean reversion. A new portfolio was created which changed the markets tested. The new portfolio (9 markets instead of 10) reversed the market trading characteristics. Eight tend to be mean reverting (non-dollar forex pairs and S&P 500 E-mini) and one which tends to trend (euro/U.S. dollar pair).

The first system was tested on both the original portfolio and again on the new portfolio. The remaining systems were tested on only the new portfolio.

RSI with Moving Average Filter System
This system looks for Relative Strength Index (RSI) extremes with a 200-day moving average as a trend filter. Since mean reversion trades don't always immediately revert to the mean, a separate stop is included to limit the maximum loss.

As mentioned, this test was the only mean reversion system which was tested on the original trend favoring portfolio. It's no surprise the test results were not good. The profit to maximum drawdown was 0.71 and the maximum drawdown period lasted over six years.

When the same system was tested on the new portfolio, the results improved but were only fair. The profit to maximum drawdown was 2.27 and the maximum drawdown duration was cut almost in half.

Bollinger Bands with a Moving Average Filter System
Some indicators, like Bollinger Bands, can be used to identify both trends and price extremes. This system uses the upper and lower bands to signal price extremes. A 200-day simple moving average is used to indicate the trend. This filter allows only taking trades in the same direction as the trend.

Minor price extremes have only a short duration. Because of this, mean reversion systems should be designed to keep trade duration also relatively short. The Bollinger Band system kept the trades short (8 day average) by using conservative exits. Exits were made at the mean (20-day moving average) instead of holding out for an assumed continuation of the trend.

The results were a mix of good and bad. The win rate was 68.26 percent with only 4 consecutive losses. But the profit to maximum drawdown was 1.95 with a maximum drawdown duration of 7 years.

Bollinger Bands with an ADX Filter System
Minor price extremes can happen both with the trend and against it. When a moving average is used as a filter, trades are only taken in the trend's direction. By using ADX as the filter, trades are made in both directions, regardless of the underlying trend. Instead of exiting at the mean, there's a profit target of 1.25 percent (7 of the 9 markets) or 2.5 percent (2 of the 9 markets).

The only bright spot for this system, compared to the previous version, was a reduction of the maximum drawdown duration to 3 years. The profit to maximum drawdown fell to 1.60 and the win rate dropped to 55.62 percent. These results reflect how usually it's best to take mean reversion trades in line with the trend direction.

Slow Stochastics with a CCI Filter System
This system requires both the slow stochastics and the commodity channel index (CCI)indicators to reach extreme readings. Both the slow stochastics and a percentage stop loss are used to exit a trade.

The system did make a profit, which was the only positive of the test results. The win rate was 50.00 percent, the profit to maximum drawdown was 0.46, and the portfolio had 8 consecutive losses.

Slow Stochastics with a CCI Filter and Time Exit System
Due to the short term nature of mean reversion systems, limiting the trade duration can sometimes improve profits. This system adds a 15-day automatic exit to the previous system.

The addition of the time exit did improve the total profits but the other metrics were little changed. The profit to maximum drawdown was 0.57.

When assembling a trading portfolio, it's important to remember the price action is not the same in all markets. Some markets have a bias towards trending and some have a bias towards mean reversion. In addition to matching markets to the system type, the markets chosen must have enough liquidity and volatility to produce a reasonable return.

Sufficient liquidity and volatility are especially critical with short-term systems. If the liquidity is inadequate, the bid/ask spreads get larger. If the intraday volatility is low, there's not enough price movement to make a good profit. Short-term systems are sensitive to these factors because they reduce the system's already small trade profits.

Changes Made for Testing the Short-Term Trading Systems
To provide the needed liquidity and intraday volatility, only 4 assets were tested. Most of the following systems were tested on only the Nasdaq 100 index futures. The first test was done with only the T-bond futures and the euro/U.S. dollar pair. Due to the greater liquidity of the assets chosen, the slippage and commissions expense was reduced from $100 to $75.

2-Hour Price Bar Channel Breakout System
This system uses 2-hour price bars instead of daily prices. Entries are made using a channel of 15-day highs and lows. 8-day highs and lows signal exits. There's also a time exit beyond 7.5 days.

The backtest data periods for T-bonds and the euro/U.S. dollar were different, so there was no combined portfolio result. But most of the metrics were similar with a win rate of about 52 percent and trade duration of about 7 days. The most significant difference was the profit to maximum drawdown ratio. T-bonds was 1.02 and euro/U.S. dollar was 2.55.

RSI with a Moving Average Filter System
This is the same Relative Strength Index system from Chapter 4 using 2-hour price bars instead of daily prices. The system was tested on only the Nasdaq 100 index futures.

The test results were not great, but all the metrics except the win rate were improved compared to the Chapter 4 daily price system. The win rate was 36.23 percent and the profit to maximum drawdown was 1.38.

This same system was also tested on the Nasdaq 100 index futures using 60-minute, 30-minute, 15-minute, and 5-minute price bars. The original 400-hour moving average filter was reduced in length proportionately. The 60-minute price tests had a 200-hour moving average and the 5-minute moving average filter was down to 16.67 hours.

Listed from longest (60-minute bars) to shortest (5-minute bars), the profit to maximum drawdown was 1.5, -0.53, -0.33, -0.04. The 5-minute bar win rate of 67.06 percent was the only one of the four over 50 percent.

Both the 15-minute and 5-minute price bar versions were also tested on the S&P 500 index futures. The profit to maximum drawdown was 0.44 for the 15-minute price bars and 0.27 for the 5-minute bars.

Slow Stochastics with a CCI Filter and Time Exit System
This is the same Chapter 4 system with 2-hour prices bars replacing daily price. The Nasdaq 100 index futures was the only market tested.

This system's results were much like those of the previous RSI system. Compared to the daily price bar version in Chapter 4, some metrics improved, some were worse. The profit to maximum drawdown ratio increased to 1.01, but the win rate dropped to 41.82 percent.

Seven-Period Reversal System
Like the previous mean reversion systems, this system tries to identify periods of price extremes. But unlike the others, this system waits until price moves in the intended trade direction. Exits are made at a seven-period reversal or profit target or stop loss. The Nasdaq 100 index futures was the market tested using 60-minute price bars.

As with the other systems in this chapter, the results were neither exceptionally good or bad. The profit to maximum drawdown was 1.27 with a win rate of 48.8 percent.

RSI Crossover System
Instead of taking a trade at the RSI indicator extreme, it waits to confirm the market is reversing. The market tested was the Nasdaq 100 index futures using 60-minute price bars.

The test results showed the effects of a profit target 3 times the size of the stop loss. The win rate was only 26.4 percent with 14 consecutive losses. The profit to maximum drawdown was 1.10.

This system was also tested using 30-minute price bars and a 1 percent profit exit, The win rate dropped slightly to 25.00 percent and the profit to maximum drawdown also fell to 0.18.

Experienced traders know there's no perfect system and no system that's right for everyone. Trading successfully is not easy, and if you're uncomfortable with the system you use, it becomes that much harder. So it's vital to your trading success that your preferences and personality align with your system.

Long-Term Trend-Following System Factors
System Example: MACD.
Usual Trade Duration: 100 to 200 trading days.

Intermediate to Long-Term Trend-Following System Factors
System Example: Channel Breakout.
Usual Trade Duration: 30 to 100 trading days.

Intermediate-Term Trend-Following System Factors
System Examples: Bollinger Bands, DMI, and Moving Average Crossover.
Usual Trade Duration: 10 to 40 trading days.

Intermediate-Term Mean Reversion with Trend-Following Filter System Factors
System Examples: RSI Extremes and Bollinger Bands with Moving Average Filters.
Usual Trade Duration: 30 to 50 trading days.

Short-Term to Intermediate-Term Non-directionally Biased Mean Reversion System Factors
System Example: Bollinger Bands with ADX Filter and Slow Stochastics with CCI Filter.
Usual Trade Duration: 15 to 40 trading days.

Trend-Following Swing Trading System Factors
System Example: Channel Breakout with 15-Day Entry, 8-Day Exit, and 7.5 Day time Exit.
Usual Trade Duration: 1 to 15 trading days.

Mean Reversion Swing Trading with a Trend-Following Filter System Factors
System Example: RSI Extremes with a Moving Average Filter.
Usual Trade Duration: 1 to 15 trading days.

Non-directionally Biased Mean Reversion Swing Trading System Factors
System Example: RSI Crossover.
Usual Trade Duration: 1 to 15 trading days.

Mean Reversion Day Trading with a Trend-Following Filter System Factors
System Example: RSI Extremes with a Moving Average Filter.
Usual Trade Duration: minutes to hours.

Non-directionally Biased Mean Reversion Day Trading System Factors
System Example: RSI Crossover.
Usual Trade Duration: minutes to hours.

A trade should never be made before weighing the potential profit versus the risk. Similarly, before using a mechanical trading system, its benefits and limitations should be compared. While system advantages are usually easy to see, the deficiencies are less often recognized.

Even if a trading system and data have no defects, there remains one inescapable limitation. Systems try to predict the future based on the past. History does repeat itself, but never in the exact same way. This is the reason a system should be based on a simple idea so it is likely to work in the future. And a completed system should be spot checked to see how well the idea followed expectations, other than simply how profitable it was.

Trading System Optimization
Computers can speed up the process of producing many things, including errors and misleading results. The process of backtesting and optimization have been greatly simplified due to the use of computers. But neither one is without problems and it's important to understand the dangers and possible traps.

Optimization can easily show which parameters and their values best fit the tested data. Various market conditions can be cheeked to see how consistent and reliable a modified system's results are. If testing a range of parameter values shows similar results, it's an indication the parameter is a valid system element.

The optimization process also allows changing parameters or filters to produce a system which fits a trader's personality. Optimization is simply a process which can determine the best fit for the data and the trader.

The ease with which optimization can make changes to a trading system is both its benefit and downside. If a trading system is optimized to fit past data too closely, it's unlikely to do well in the future. There's a natural desire to add rules or adjust values to increase profits. But the pursuit of perfection should be replaced with an attitude that a good enough fit is good enough.

Weissman ran optimization tests of the basic two moving average crossover system over a ten year test period. Each test used a different combination of short-term and long-term moving averages. The results showed how often the best moving average combination didn't stay the best. What worked the best in testing was often not even in the top ten results in a year of new data.

The lack of consistency for one parameter or value points out the need to have broadly different parameters with an acceptable range of values. In addition, optimization should include eliminating suboptimal values since they usually remain consistently less profitable.

Besides showing the inconsistency of parameter values, the testing showed the importance of choosing new data. The process of testing a system on new data, or out of sample data, is called walk-forward testing.

The out-of-sample data period should equal about 10 to 20 percent of the in-sample data. Because of its importance and more limited size, the new data should be carefully examined for any errors or potential problems, like a change in market conditions.

Once a trading system has been adequately tested, it time to analyze the results. There's more to system analysis than finding what produces the most profitable results.

Trading system test data analysis mainly involves looking for consistency. Do the testing results show similar returns versus risk? Is the profit to maximum drawdown ratio consistent across asset classes? Are the yearly results for similar market conditions roughly the same? Do the out-of-sample results match the backtested results?

Attention should also be directed at any individual asset or yearly result which skewed the overall results. This applies to both in-sample and out-of-sample data.

Are the out-of-sample results good or bad? If bad, it may be the system, changing market conditions or due to the results of one or more assets. If the results are better than average, it's most likely market volatility has increased. In either case, individual assets should not be eliminated or given extra weight because of their out-of-sample performance.

An optional part of trading system development is a design philosophy statement. Similar to a business mission statement, the philosophy statement defines the system's goals and how they are measured.

Trading is probably more art than science. But risk management is all probability, math and preferences. You have no control of what markets do, but you manage what risks you take.

When considering risk, the two main types are individual trade risk and portfolio risk. Trade risk deals with position size and stop-losses. Portfolio risk deals with value at risk (VaR) and stress testing.

Individual Trade Risk Management
Position sizing can take one of two basic forms, Martingale and anti-Martingale. With Martingale, position size is doubled after every loss. Doing this ensures two things. The first certainty is eventually you'll win back all the losses. And secondly, at some point the string of losses will become so large, you'll go broke.

A similar process with equally negative prospects is the idea of averaging down. This technique adds more to a positions as prices go lower, typically doubling the position as price drops in half. This may not result in bankruptcy, but there can be long periods of holding the total position at a loss.

Anti-Martingale methods do the opposite of Martingale. They add to a position after a win instead of after a loss. A simple way to do this is with fixed fractional money management. This approach risks a fixed percentage of the account equity on each trade. As the account equity rises and falls, so does the per-trade dollar amount risked.

The second way trade risk is managed is with stop-losses. The ideal goal is to risk as little as possible without placing the stop-loss so close, random price fluctuation can trigger it. A good place to start is using 1 or 2 percent of the account equity.

Portfolio Risk management
WHen more than one asset is traded, there exists an overall portfolio risk. Value at risk (VaR) tries to determine how likely a portfolio could lose a certain amount within a fixed time period. This is helpful risk information, but VaR is not without its faults.

In order to derive its equations, VaR has to simplify market reality. In doing so, it makes several faulty assumptions. It views each day as an independence event with no effect on the following day. It doesn't account for periods of illiquidity, such as during market crashes. VaR also assumes past market correlations will hold in the future.

The problem with these assumptions is that they break down for the same events you want to plan for, These are the times when prices don't follow the statistical norms. And even when price activity is normal, VaR can be incomplete.

VaR tells you about probable maximum losses, but remains silent about cumulative losses. A week of losses all under the expected daily maximum loss could easily total several times the daily maximum.

Another method for determining portfolio risk is stress testing. It helps compensate for VaR deficiencies. Instead of calculating the risk of likely events, stress testing can see what happens with unlikely, but still possible, events.

Using either a real historical event or a hypothetical situation, the stress test results show where any portfolio weaknesses are. Any necessary adjustments could then be made to reduce the effects of the tested event.

Portfolio risk, like trading in general, is about survival. Assuming above average risks can result in above average returns, but only in the short term. In the long term, the result is almost certain bankruptcy.

So if you want to last as a trader, you need to control risk. And a mechanical trading system is just the tool to do that. It provides the discipline and means to control the risk for each trade. With valid backtested results, periods of losses and drawdowns can be handled with confidence.

The basis of trading success is simple… win more than you lose. Get a good return on the money risked and limit that risk. Diversification is a way to get higher returns for the same amount of risk.

There are multiple types of diversification. Portfolio diversification is the most common type. A diversified portfolio trades different asset classes instead of only one. A portfolio trading only one asset class of securities is vulnerable to all the securities losing at the same time. Trading a diverse portfolio of securities greatly reduces, if not eliminates, this possibility.

In addition to diversifying the assets traded, the trading process can also be diversified. Similar to trading more than one asset class, trading with more than one set of system parameters reduces trading risk. Making trades with different parameters reduces the negative impact any set of parameters can have on the portfolio.

A second way the trading process can be diversified is with a different process. In this case, a different and uncorrelated trading system. Combining a trend-following system with a mean reversion system can produce a higher profit to maximum drawdown than the ratio of either system alone.

There are a couple of warnings for traders who trade multiple systems at the same time. Since two different systems could be on opposite sides of the same trade, it's best to have each system in a separate account.

Trading more than one system can clash with a trader's personality. It's recommended to first achieve success with one system before adding another. This will establish the good habits needed to support trading a new system which may not be a personality match. Trading multiple systems is both a challenge and an opportunity fro personal growth.

One of the reasons for using a mechanical trading system is to eliminate decisions which might be based on emotion and bias. So discretionary mechanical system trading seems to be a contradiction in terms and negates the system's purpose. But it can be done and still remain true to a mechanical trading system's process.

Why would a trader want to apply discretion? Richard Weissman says there are two reasons, paradigm shifts and price shocks. A paradigm shift is a long-term market change which reduces the viability of a trading system.

Mechanical trading systems don't eliminate discretion, they limit it's application. Discretion is allowed and needed when planning a system, but not when trading it. So discretion can be applied on planning how to react to a paradigm shift.

There should be an objective standard to signal a shift has occurred. Weissman suggests a shift may have occurred if trading results are worse than the backtested metrics. This could be higher than expected consecutive losses or a drawdown larger than the testing maximum drawdown. The action at that point would be to either stop trading the system or significantly reduce the position size.

A price shock is a sudden and lasting change in volatility. Some price shocks are temporary and can be ignored. The price shocks to account for should have a long-term effect. Allowing for price shocks can be treated like a paradigm shift. Decide on a metric, such as a 50 percent yearly volatility increase. If it occurs, either stop trading or reduce the equity traded.

On the surface, trading is a way to make money like, any business. If a business sells its goods or services for more than their cost, the business is successful. If they fail to do that, they can't survive.

But trading can be more than just conducting financial transactions. It can literally be life changing. The skills necessary to be a successful trader require a combination of traits such as discipline and flexibility. These traits seem to oppose each other, but in reality they reflect the dual nature of the markets.

Imagine a market in a long-term up trend. As it reaches an intermediate peak, who is the smarter trader, the one who buys or the one who sells? Short term it's the seller, but long term it's the buyer. They're both right when measured by their intended time frame.

The mental lessons learned in trading can also be applied generally to life. Mechanical systems trading is not about eliminating emotions, it's about managing them. Acting out of anger instead of logic in a discussion benefits no one. But if the same anger sparked action to prevent harm to another, that could be a well managed emotion.

Ultimately, learning to trade successfully means learning about ourselves. Mechanical trading systems can show where we lack discipline or when emotions like fear or excitement are inappropriate. When we feel uncomfortable with a mechanical system trade, it's our natural response in conflict with market reality. But with repetition of profitable trades, what once seemed unnatural can become natural.

Chapter by chapter references.

Books and articles related to trading, risk, and Buddhism.

PJ Nance

Previous	Next
Naked Forex by Alex Nekritin and Walter Peters Trading off false beliefs is a sure way to lose. Each of the book's three sections covers a trading belief and shows why it's false. Alex Nekritin and Walter…	Day Trading the Currency Market by Kathy Lien Based on her experience teaching others how to trade, Kathy Lien has two objectives for the book. She wants to cover the fundamentals of the foreign exchange (FX)…