Building Reliable Trading Systems by Keith Fitschen

Contents  ˅

Preface 1) What Is a "Tradeable-Strategy?" 2) Developing a Strategy So it Trades Like it Back-Tests 3) Find the Path of Least Resistance in the Market You Want to Trade 4) Trading System Elements: Entries 5) Trading System Elements: Exits 6) Trading System Elements: Filters 7) Why You Should Include Money Management Feedback in Your System Development 8) Bar-Scoring: A New Trading Approach 9) Avoid Being Swayed by the "Well-Chosen-Example" 10) Trading Lore 11) Introduction to Money Management 12) Traditional Money Management Techniques for Small Accounts: Commodities 13) Traditional Money Management Techniques for Small Accounts: Stock Strategy 14) Traditional Money Management Techniques for Large Accounts: Commodities 15) Traditional Money Management Techniques for Large Accounts: Stocks 16) Trading the Stock and Commodity Strategies Together Appendix A Understanding the Formulas Appendix B Understanding Futures Appendix C Understanding Continuous Contracts Appendix D More Curve-Fitting Examples

There are two requirements for a tradeable strategy. It must produce sufficient returns for the risk involved. And the strategy's real-time performance should be similar to its back-tested results.

While defining the objectives is simple, the process of creating a reliable trading system is not easy. There are at least three problems to deal with:

Greed. We all want a low risk high reward system. This attitude leads to unrealistic goals and poor decision making.

Curve-fitting. Testing a new system without enough trades or different market conditions will result in curve-fitting. The system will fit the data, but won't do as well in the future.

False beliefs. It's important to critically consider any information about the markets and trading. No matter who or what the source is, don't accept something is true, unless you test it for yourself.

The answer to this chapter's question is…it depends. A tradeable strategy must have a reasonable objective and an acceptable level of risk.

Create a reasonable objective by looking at the results professional traders consistently achieve. The report, Barclay's Top 20 CTA Performance for 7/1/2005 to 6/30/2010 offers a typical performance profile for the best CTAs.

The report showed the following:

• The 5 year compounded annual returns ranged from 21.02 percent to 44.54 percent.
• Draw-downs varied from 6.88 percent to 60.72 percent.
• 6 of the 20 had a draw-down larger than their average 1 year return.
• 16 had at least one losing year out of the 5 years.
• Most of the average yearly returns were due to 1 exceptional year.
• The top CTA had an excellent average return with a small draw-down and yet still had one year with only a 1 percent return.

The reality is, there won't be high returns every year and the winning years will have to make up for the losing ones.

To create a system you're comfortable with, you need to decide the following:

• What's an acceptable return on your money.
• What's an acceptable average and worst-case draw-down.
• What's the longest period you're willing to go without an increase in your equity.

It's important to be realistic when setting goals for a trading plan. 4 of the top 5 CTAs had a losing year of 22 to 50 percent. High returns make equally large draw-downs likely. Ask yourself how you'd feel if your account was down over 20 percent on the year. And maybe down for two consecutive years.

Curve-fitting is the biggest problem in designing a system which does well in real time. Curve-fitting occurs when the testing data set is too small, or the data set is large enough, but the system has too many rules.

Creating a system can be like trying to get out of a maze. The objective is clear, but how to achieve it, is not. Say you tried the rule, "always turn left at every maze opening" and found it worked. Then you tried this rule with several other mazes and it worked again. Eureka!. You're now confident you can get out of any maze. And you will…until you come across a maze which requires you to always turn to the right.

Simple rules for mazes or trading systems won't work if they are derived incorrectly. Fitschen created a Swiss Franc system using 10 years of prices from 1980 through 1989. It used a basic strategy of buying and selling whenever price crossed a moving average.

He started by picking the moving average which showed the most profit. He then chose the most profitable trade filter and stop-loss. The three rules produced a total profit of $84,887 on 64 trades. Even the win rate of 48.4 percent was good for a long-term trend system.

But because the system was optimized to the data, the next twenty years showed a reduced profit of $30,638 on 231 trades. The average profit per trade dropped from $1326 to only $131. And the win rate also fell from 48.4 percent to 33.3 percent. Because market conditions had changed, so did the results.

To avoid curve-fitting, an estimate should be made on how likely test results will match future results. This requires some simple statistical calculations of the average profit per trade. The estimate result will be a range.

First, calculate the standard deviation of a sample number of trades. Then add the standard deviation to the average profit per trade. Next, subtract the standard deviation from the average profit per trade.

If the sample average profit per trade was $100 and the standard deviation was $2, then 68.3 percent of future trades should be between $98 and $102. Two standard deviations give a probability of 95.4 percent that future trade profits would be in a range of $96 to $104.

Ideally, test samples should be hundreds of trades, but that's not realistic. To derive the same estimate for small samples, divide the standard deviation by the square root of the number of trades. This number is the standard error. Take the standard error and add it to, and subtract it from, the sample average trade profit. As before, the result is an estimated per trade profit range.

Unfortunately, real world results don't have such a small standard deviation. The range will be much larger than $2 for a $100 average. The actual average might be $50 with a standard error of $75. The range is now -$25 to $125. If the sample individual trade profits vary widely, both the standard error and the profit range can increase significantly.

To reduce the problem of small sample size, Fitschen devised the Build, Rebuild, and Compare (BRAC) method.

Using BRAC to create a moving average system for 37 commodities yielded a 32.7 percent win rate with an average profit of $397. The last year (2010) had a win rate of 33.0 percent and average profit of $1089. After removing 2010 from the sample, the same 3 parameter values used before, still showed the most profit. The win rate changed from 32.7 to 32.6 and the average profit $397 to $362. BRAC doesn't eliminate curve-fitting, but it does, reduce it.

Curve-fitting is the main reason for system failure, but not the only reason. There's also:

• Assuming there's a trade for every point within a bar's range. Gaps in a shorter time frame than the one tested, won't show up in the results. To get around this, do some testing on shorter time periods to see if the initial results are confirmed.
• Errors caused by dual orders. A system may have a buy and sell order open at the same time. Which one is executed depends on the market action. But if both orders could have been filled on the same day, you don't know which one would have been first. To check for this effect on initial test results, retest on some shorter time periods.
• Not allowing for transaction costs. Much like unseen gaps, the order price may not be the actual execution price. Limit orders can fix this, but they will also miss entering some trades. The result is, losing trades will always get filled, but not all the winners.

System sellers typically say their method works the same for all markets, in all time frames. It's true the method may "work", but the results won't be the same for different markets.

Test a system on a basket of stocks, commodities, or currencies, and the results will vary widely. From the years 2000 through 2011, buying the weakest stocks at the start of each month showed an average profit per month of 1.56 percent. In comparison, buy-and-hold had an average monthly profit of 0.71 percent. If the strongest stocks were bought, the average monthly profit was a mere 0.16 percent. Buying the weakest stocks was the clear winner.

Using the same 3 trade approaches and time period, commodities were also tested. Their test results showed an average monthly profit per trade of $66 for buy-and-hold, $158 buying the strongest commodities, and $23 buying the weakest ones. In general, stocks profit most with counter-trend trading, and commodities work best with trend systems.

The average per trade results for currency pairs was -$87 with buy-and-hold, -$101 when buying the strongest currency pairs and $178 when buying the weakest. Similar to stocks, currency pairs work best with a counter-trend approach. However, U.S. dollar pairs are the exception, since they do trend.

This set of results was done with daily prices. It's suggested you do a similar test for the time frame you trade. Running tests with intra-day prices showed stocks trended, as did commodities and currency pairs. For stocks and currencies this was a reversal of their daily price behavior.

The message is clear. To create the right type of system, determine the characteristics of the markets and time frames traded.

Once the trading characteristics of a market are understood, testing can begin for entry techniques.

Price bars create the context for a trade. One bar tells very little, but if there are enough to form a distinctive pattern, that may be a sufficient reason for making a trade. The more price bars used for an entry, the closer the entry is to following a major trend. And getting in line with a major trend, means more profit potential.

However, the number of bars is not the only consideration when choosing a good entry signal. Although patterns with a few price bars offer smaller per-trade profits than patterns with more bars, they also offer more trade opportunities.

To allow for the difference in trade frequency and profit size, only compare entry methods which have the same number of price bars. Usually, when comparing technical indicators, this involves changing their default values.

To make a comparison, plot the average profit of an entry signal over a period of weeks for an asset group. A basic strategy like prices crossing above and below a simple moving average could represent a baseline entry method. Several moving averages of different lengths would be included and their average profit per trade recorded over time.

Then different types of moving averages, such as exponential or geometric, with the same periods, could be compared. In fact, any technical indicator using the same parameter periods as the moving averages, could be compared.

Like testing for market characteristics, this technique shows how different entry methods perform over time. It's a simple way to find out which entry methods work best, and over what time frame.

There are four common ways to exit a trade:

A reversal signal
This is used in systems which are always in the market. A typical example is a simple moving average system. When price crosses above a moving average, a long position is created and a short position is closed out. When price crosses below the average, the long position is sold and a new short trade is entered.

A stop loss
This exit has two purposes. Its initial role is to limit the size of a losing trade. Then once a trade is in profit, the stop can be moved to limit profit loss due to an adverse price move.

A channel system creates buy and sell signals when price crosses the highest high, or the lowest low, for a set time period. Usually, a channel system sell stop is the lowest low. But sometimes, the range between the entry at the highest high and the sell stop at the lowest low, creates too much risk. In that case, a stop loss can be used to reduce the initial risk.

The downside of an initial stop loss, is that it reduces the average profit per trade. This happens for two reasons. The stop loss eliminates a few trades which would have been profitable if only the regular system exit was used. The stop also takes a few losses, which would have been smaller, if the trades were closed out with the system's signal.

A time-based exit
System testing may show at a certain point in a trade, there is usually very little additional profit. This makes holding the trade beyond that point, not worth the additional risk.

Or the case may be that if a trade isn't in profit within a given amount of time, it usually turns into a loss. In either case, a time-based exit can be appropriate.

For intra-day traders, the time of day may also be a time-based exit. Certain hours in the trading day may offer the best point to exit a trade. Or maybe getting out at the close, regardless of a trade's profit or loss, avoids the overnight holding risk.

A profit stop
Some systems come with automatic profit targets. When prices are moving in a sideways pattern, the highs and lows roughly form two parallel lines. This pattern has a price objective of the difference between these two price level lines.

If price exceeds the upper price line, the price objective is added to the entry point to find the trade's target price. If price falls below the lower price line, the trade's target price is the entry price minus the price objective.

Both stop losses and profit stops may be either a dollar amount or a percentage of the entry price.

Exit signals are tested in a similar way to entry testing. Start with a chosen entry signal and use a range of exit types to see how each performs. The combination which produces the most profit, is not necessarily the best choice. Fitschen uses the gain-to-pain ratio. This ratio is the average annual dollar return, divided by the average maximum draw-down.

The combination with the highest gain-to-pain ratio, may also not be the best choice. It may have an average profit per trade which is too low, in comparison to transaction costs. The gain-to-pain ratio has to be weighed against what's an achievable, real-world average profit.

Trade filters are used to avoid an entry or possibly an exit in a particular situation. Typical filters include:

• Day of the week. Some days of the week show better profit potential in the stock market, than others.
• Longer-term trend. A system may benefit from using a long-term trend indicator which supports the direction of a short-term trade signal. For example, the trend of a long-term moving average used to confirm a short-term moving average entry signal.
• Volatility. Trades in markets without significant price movement lack profit potential, and should be filtered out. But also avoid very volatile markets, since their risk is excessive.
• Seasonal. Some commodities and stocks are subject to seasonal effects, which can change how a trade performs.

This chapter completes two systems started in the entry chapter. One of the systems is for stocks and the other is for commodities. They're both based on a system created by Richard Donchian in the 1960s. It's a breakout system covering the last n bars. A buy signal is given if price exceeds the highest high or close of the previous n bars. When price goes below the lowest low of the last n bars, that's a sell and sell short signal.

Using the methods from chapters 3 through 6, a portfolio of a hundred stocks was profitable for 11 out of the 12 years tested. 2008 was the only down year for the stock system. The commodity system was tested with 56 futures (commodities and financial assets) over 32 years and had only one losing year.

The gain-to=pain ratio is aptly named. On paper, it's easy to ignore the impact of large draw-downs. But when a draw-down happens in real time, day after day and month after month, the financial and emotional pain can be significant.

The gain-to-pain ratio is a simple metric of the average profit per year divided by the average maximum draw-down per year. Since this ratio can be calculated at every step of the system creation process, it avoids a system-killing surprise at the end.

As a test of the effectiveness of the gain-to-pain ratio, it was compared with using the average profit per trade in system development. The same two systems developed in the previous three chapters, were redeveloped. The parameters were chosen on the basis of the highest average per trade profit.

The stock system developed using the gain-to-pain ratio, had an average annual profit of $44,591 with an average yearly maximum draw-down of $19,789. The redeveloped system, on a profit per trade basis, showed an average profit of $31,854 and an average maximum yearly draw-down of $33,613.

Using the profit per trade metric, the commodity system improved the average annual profit from $86,639 to $128,805. But that increase came at a cost to the draw-down. The original gain-to-pain derived system had an average yearly maximum draw-down of $31,369 but the redeveloped system had an average yearly maximum draw-down of $123,968. Almost 4 times as big.

The power of a trading system is in its objectivity and consistency. The system rules make sure all the trade signals are the same. However, not all trades are the same. The rules don't show how strong a trade signal is. A signal which meets all the rules, may actually be weaker than a signal which meets only some of the rules.

To overcome this difficulty, Fitschen developed the process of bar-scoring. This process can be used to rate entries, exits, or any factor in a trade. The bar score is the expected profit at a certain number of days after trade entry. A high positive or negative score, indicates a strong upward or downward move is likely.

The bar-scoring process begins with selecting a criteria. This could be a technical indicator or a trading characteristics such as volume or volatility. The next step is to decide how many days to track the average return of a trade.

An example of bar scoring could be rating the Relative Strength Indicator (RSI) over a period of 5 days. The RSI's range is divided into segments with each segment containing the same number of trade samples. The results would look something like this following table:

A long-term trader might want to extend the period covered from 5 days to 20 days or more. Also, several independent criteria can be rated and their daily scores combined. Divide the combined daily scores by the number of criteria to get the expected daily return. Positive return scores represent a possible buying opportunity and negative returns show a potential short sale.

When creating the criteria segments, the more the better, as long as the same sample size is kept for each segment. And to avoid curve-fitting, there's no substitute for a large sample size.

Bar-scoring can be used to rate the strength of entries and exits, or modify their application. The Build, Rebuild, and Compare (BRAC) process of chapter two can be used to verify the extent of curve-fitting and the reliability of the bar-scoring results.

There are a number of trading concepts which get repeated so often, they seem to be true, even though they're not. Trading claims can be tested with bar scoring to check their validity. Fitschen tested the following trading concepts:

Divergence
Divergence occurs when price makes a new high or low, but an indicator such as momentum, doesn't also make a new high or low. A divergence is supposed to indicate the price high or low is a top or bottom and a trend reversal. When tested for the Nasdaq 100 over 12 years, no profit was shown for 5 holding periods of 5 to 30 days. A group of 56 futures tested over 32 years also showed no profit. In fact, the per trade losses were large enough to reverse the divergence buy and sell rules.

Gaps
An opening price gap occurs when prices open above or below the previous day's range. Buying higher and lower opening gaps, especially those gaps below the previous day's low, tested profitable for stocks. Commodities mainly showed small losses when buying higher gaps and selling lower gaps.

If an opening price gap is not filled by the day's close, that's considered a bar gap. Bar gaps for stocks showed the best results when shorting bar up gaps and buying bar down gaps. If not in a congestion range, commodities had profits when buying bar up gaps and shorting bar down gaps.

Fibonacci
The Fibonacci sequence is a series of numbers created by adding each number to the previous one, starting with 0 and 1. The initial sequence is: 0, 1, 1, 2, 3, 5, 8, 13, 21… and so on. Traders use various combinations of the numbers to come up with ratios which are supposed to be significant. For instance, 55/233 = 0.236 and 233/55 = 1.236. Ratios of 0.236 and 0.382 were tested as price retracement levels for the stock and commodity groups. Buying both retracement levels showed profits, but not when shorting.

To see if Fibonacci ratios are better than other ratios, 0.05 to 0.30 were tested for stocks and 0.20 beat 0.236 for 3 of 4 holding periods and beat 0.382 for all 4.

Commodities showed profits for 2 of 4 holding periods for 0.236 retracements and no profitable holding periods for 0.382.

Stock Splits
Buying a few days before and selling a few days after a stock split, showed profits for 3,400 highly liquid stocks tested over a 12 year period. Reverse splits resulted in losses for all holding periods tested.

Dividends
Using the same 3,400 stock splits group, stocks were bought 1 month before their dividend date. The stocks were sold 5 days after the dividend date. Adding the dividend to the change in stock price produced a 1.48 percent profit. Profits more than doubled when the dividend was 2 percent or more of the entry price.

Are trade exits more important than entries? Van K. Tharp's book, Trade Your Way to Financial Freedom said that they were. Tharp came to this conclusion from the results of a test. In the test, trades were entered randomly and closed out with a defined exit. This strategy was profitable for the 10 markets tested.

However, Fitschen points out the conclusion was in error. All the entries were not random, only the initial ones. Since the system was always in the market, once the trade was started, all the following entries were determined by the exit.

Fitschen did his own test with a random entry and the same volatility stop from Tharp's test, using 56 futures with a total of 700,000 trades. The result was an average per-trade profit of $0.000025 or essentially 0.

The same futures group was then tested with a random entry and the trade was reversed on the exit stop. This time there was a profit of $239 per trade. This test showed the stop was the deciding factor for the entry, not the exit.

Kelly Criteria
The Kelly criteria comes from the gambling world and is a formula for deciding bet size. When used in money management for trading, it can produce large returns. However, what is not usually mentioned when the Kelly criteria is discussed, is the high risk involved. When tested for a hypothetical system having a 60 percent win rate over a 1,000 runs, it did produce good profits. But the average biggest draw-down was 77.3 percent. Draw-downs of 90 percent or more showed up for 16 percent of the runs. Even the best run had a draw-down of 36 percent.

Monte Carlo Analysis
Monte Carlo simulations are often used for statistical analysis in various science fields. When used for trading system analysis, it has two main faults. Since the number of trades used in the simulation is small, it's hard to avoid curve-fitting. Also, the simulation doesn't show the in-trade draw-downs which can be significant. Monte Carlo simulations generally ignore the effects of good money management which further reduces its value.

Artificial Data
To avoid the problem of curve-fitting, a system should be tested with tens of thousands of trades in every market situation. But since this is not possible, a suggested solution is to test with artificial price data.

The problem with this approach is that typically, the data is a random price series. This means there's no correlation between the action of a price bar and the previous or following bar. It also eliminates the correlation of stocks or commodities within the same category. Artificial data is unreal and creates unreal price series.

Losing Trades
"Never add to a loser", is common trading advice to avoid making a bad trade worse. But is this always the right approach? Not necessarily.

In a test of Nasdaq 100 stocks over a 12 year period, a system generated $37 profit per trade with a 68.9 percent win rate. The system bought the stock when it closed 1 standard deviation below its 10-day average price. When a second position was added if price dropped to 2 standard deviations below the average, the profit per trade was $46 with a win rate of 71.9 percent. When draw-down is acceptable, it's possible some losing trades are only temporary losers and adding to them can make sense.

Winning Trades
If adding to a losing trade can sometimes make sense, certainly adding to winners is even better. Or maybe not. Adding to a profitable position, automatically increases the risk to reward ratio over the same time frame. And if shortly after adding to the position the trade direction reverses, it's possible a winner is turned into a loser.

Taking a Profit
"No one ever got hurt taking a profit", is a feel-good idea some traders follow, especially after a series of losing trades. But the reason for creating a system, is to avoid arbitrary decisions like this. The best course is to have a plan and take the exits when they're supposed to be made.

30 Trades
It's often stated a 30-trade sample is enough to validate a system. 30 trades will only show how that particular sample distribution performed, not how most trades are distributed. Trade profits and losses can vary widely. That's the reason to test large samples. This ensures there is an accurate representation of trade distribution.

The Best System Statistic
There's a wide range of opinion about what the best statistical rating of a trading system is. Fitschen thinks the best system metric is not a formula, but a picture. A smooth rising equity curve, without large draw-downs or long periods of inaction, is the perfect picture. Simply put, the ideal system.

Dogs of the Dow
The Dogs of the Dow is a strategy which buys the stocks in the Dow Jones Industrial Average according to their dividend. Each year, the strategy buys the 10 Dow stocks with the highest dividend to price ratio. These stocks are held until the next year, when the process is repeated.

Fitschen tested an alternative counter-trend method using only price instead of dividend. All Dow stocks with a month-ending price below the closing price 5 days earlier, were bought on the first day of the new month. This process was repeated on a monthly basis from 1980 through 2011. This group was compared with buying all 30 stocks at the start of the month and selling them at the end of the month. This modified version of the Dog strategy outperformed the buy-and-hold by an average of 5.5 percent for a annual return of 23.3 percent.

Trends
It's often said markets are in congestion ranges much more frequently than they are in trends. The Average Directional Movement Index (ADX), is an indicator often used to show when a market is trending. In a test for trend frequency, the ADX was measured for 14, 20, 40, and 80 days using 56 futures over 32 years. The results showed the shorter the time frame, the more futures trended. For the 14-day ADX, the markets trended an average of 62 percent of the time, but only 8 percent of the time for the 80-day ADX.

Systems are tested to see if they are reliable. Trading information or advice should also be tested for the same reason.

An important aspect of money management is trade position size. This is the amount of money at risk when placing a trade.The two basic ways to determine position size are a fixed dollar amount or a percentage.

Fixed Risk and Fixed Fractional
The fixed risk method uses a set dollar amount for each trade, no matter the size of the account. The fixed fractional method sets the dollar amount per trade as a fixed percentage of the account equity. As the account grows, so does the dollar amount for each trade. Most professional traders use some variation of these two approaches.

Optimal F
Most money management methods look to balance risk and profitability. Optimal f looks to grow account equity as fast as possible. Through testing, the optimal f is determined for a series of system trades. The optimum position size is the largest loss for the sample trades divided by the optimal f value.

Optimal f is a very aggressive strategy. Testing may show the calculated position size is too large to avoid an account-killing draw-down. It's a high reward method, but also high risk.

Fixed Ratio
This money management method was created by Ryan Jones and described in his book, The Trading Game. This approach increases position size as account equity grows, but not at a proportional rate. The relative position size is greatest early on and decreases as equity increases. The returns and draw-downs are not as good as the fixed risk or fractional methods.

Money Management and Account Size
Size matters when it comes to account equity. If you risk 1 percent of your account on each trade, that's $100 on a $10,000 account. This situation leads to either placing stops which are too close, or assuming greater account risk by increasing stop size. Sound money management is always important, but especially so for small accounts.

A small account trader can estimate system risk using the unique start-trade draw-down metric. For each trade, find the lowest point on the open and closed equity curve below the initial level. With 1000 trades, there would be 1000 equity curves. This process finds the maximum initial draw-down, but it also can do more.

Two probability charts can be derived from the draw-down curves. Using the range which the curves cover, a chart can be made for the probability of the initial draw-down size. Another chart can show the probability for the size of the first-year profit. These two charts give the relative risk to reward for the tested system.

Small account traders are most concerned about making it through the first year. Large equity traders are more concerned about long-term account survival. To this end, large account traders use the largest percentage draw-down of the open and closed-trade equity curve. When this percentage is compared to the annual return, it gives a good idea of the risk to reward for the system.

Traders with small accounts should be most focused on limiting risk when applying money management. For the purposes of this chapter, which deals only with trading commodities, a small account is $20,000 to $100,000.

Diversification
A first step in risk reduction is to trade a variety of commodities which generally move independently of each other. One way to diversify is use a first-N-in-a-group strategy. This involves taking the first N trades in a commodity group. The first N trades could be one of the grains and the next set could be one of the metals. This provides trade diversity without having to maintain positions in the different groups at the same time.

Trade Risk
The first way to limit trade risk is to avoid certain trades. If testing reveals using a certain stop level hurts profits or draw-down, simply don't take those trades. Testing may also show when open-trade risk exceeds a particular point, it's best to exit the trade. Otherwise, waiting to get stopped out loses some of the unrealized profit. This same technique can be applied when there is more than one open position. If the combined open-trade risk reaches a threshold level, take profits on one or more of the positions.

When choosing which commodities to trade, go with the one or two best performers in each group. In this case, as always for small accounts, best performing means lowest risk. As account size grows, add more from each group. After an initial portfolio is assembled, run the tests from chapter 11 to see performance expectations.

Whether trading stocks or commodities, the number one priority for traders with small accounts is to maintain good risk control. The money management strategies in this chapter apply for trading stocks with an account of $20,000 to $100,000.

When trading commodities, it's easy to diversify by trading among the different groups which have uncorrelated prices. But with stocks, even different stock sectors can have various degrees of similar price behavior. This correlation eliminates some of the risk-reduction tactics presented for commodities trading.

A constant with any trading is the open and closed-trade draw-down. Test a system for different position sizes. When the average and maximum draw-down is found for a given position size, it shows how much can safely be risked on each trade. And as with all small accounts, when it comes to choosing between alternatives, choose the lower risk alternative.

Small trading accounts have to emphasize risk reduction over finding the best gain-to-pain ratio (average annual return/average annual maximum draw-down). But with large accounts (over $100,000) there isn't this restriction. Large account traders are able to put less of their total equity at risk per trade and can seek to maximize their gain-to-pain ratio.

Position Sizing
One of the most often used ways to determine per trade risk, is a simple fixed percentage. Testing will indicate what percentage of an account can be risked on each trade. This amount is typically about 2 percent. With a $200,000 account, $4,000 would be available for each trade. If the stop was $500, then the position would be 8 contracts. The actual percentage chosen comes from testing to find the best gain-to-pain ratio.

Risk Reduction
There are three ways to limit overall risk:

• Trading different commodities to balance risk. A large account can trade all the commodities simultaneously to achieve portfolio diversification. The only commodities to leave out are those which weren't profitable in testing.
• Using discretion on which trades to take. Some trades have stops too far from the entry and these trades should be avoided. Risk is further reduced by limiting the total open trades at any one time.
• Controlling your risk exposure with open trades. Limits can be set for individual and total open trade risk as a percentage of account equity.

Long-term account survival depends on the balance between risk and profits. With a large account (over $100,000) for trading stocks, there are a few simple things which can ensure this balance.

How much risked on each trade is the first place to start for proper money management. A sound and easy way to determine trade position size is to set a fixed rate of the account equity. With a $500,000 account and limiting risk to 1 percent of the account, would allow $5,000 per trade. If the stop on a trade was $500, the position would be 10 units of the stock.

Other ways to reduce account risk include limiting the total open trades, avoiding trades with larger than average size stops, and hedging long or short trades with the opposite side. Using ETFs can be an effective way to hedge.

Throughout this book a stock system and a commodity system were developed by applying the ideas from each chapter. Like other ways of diversification, trading multiple strategies can also lower overall risk.

The two basic ways to trade multiple strategies are segregated equity and a combined equity account. A segregated account splits the equity into one piece for each strategy. Combined equity simply trades all strategies in one account.

For either account type, a different equity amount can be applied for each strategy. As with any other decision, how much equity is assigned should be made after testing different levels.

Standard deviation, measures of volatility, and correlation are fully described.

This is a basic guide to futures. A futures contract is described, how it's priced, and its relative risk. Various statistical information is given for the commodities within each group.

One of the problems with testing a strategy is dealing with the price data. In a simple world there would never be a need to adjust price data. Unfortunately that's not our world.

Every time a stock splits or issues a dividend, the share price changes to reflect the difference in value. And when a futures contract expires, there's a price gap between the old and new contracts.

Stock data streams can be raw unadjusted prices, split-adjusted prices, dividend-adjusted prices, or split/dividend-adjusted prices. None fully reflect the reality of how a stock traded, but the split/dividend-adjusted data stream is the best to use.

Continuous futures pricing operates similar to a stock split. The prices of the old contracts are adjusted up or down to match the current price level of the new contracts.

Curve fitting is always a concern when testing a strategy. Fitschen experienced losses due to curve fitting in three different ways:

• Creating a correlated stock pairs strategy for pairs which turned out to not always be correlated.
• Using limit orders which got more or fewer orders filled than testing had indicated. This turned the strategy from good to only "so-so".
• Bar-scoring using profit-per-trade showed good results only because some open trades were mixed in with the new trades. When corrected, the results were only "marginal".

PJ Nance

Previous	Next
10 by Michael Turner Michael Turner started his professional life as a civil engineer, but his first love was computer programming. After working a number of engineering jobs, he created a software company…	New Concepts in Technical Trading Systems by J. Welles Wilder Before the personal computer became a household fixture, there were programmable calculators. These calculators could carry out a set of simple…