books about online stock trading, forex, futures, stock investing, market, trading systems
The Mixed Role of Stops for Controlling Risk

Keeping losses small, which can mean using a closely placed Stop order, has always been considered sensible risk control, but it may present more problems than it solves. A series of losses from short-term trades can be greater than a single, large loss caused by a longer-term position. In addition, when you get stopped out but the trend remains unchanged, how do you decide when to reenter the maSrket?

A stop-loss order is a mixed blessing. It is not the order itself, because the act of monitoring the market and using Limit orders to carefully liquidate a trade has the same result. Large stop-losses offer some protection against cat­ astrophic events and price shocks; however, attempts to keep risk very small causes larger losses and a worse risk profile. Small risk limits are often acti­ vated by market noise; there is little relationship between getting stopped out and the direction or change of direction of price movement.

Historically, there is some level of risk protection that would have benefited a portfolio. Close inspection of lengthy test results show that the reason for improved performance is a single, adverse price jump that would have been reduced by using a stop-loss. But it is not clear that such a situation would occur again, and if it occurred, that the stop- loss order could have been executed profitably.

In some cases, a poor trading strategy is improved by using stops. But then, it would also be improved by not trading. There are occasional improvements in the return/risk ratio of performance using systematic risk control. When the program has no position, after the risk limit has been reached, the equity is not subject to the ups and downs of price movement. If a large loss is averted during this time, overall perfor­ mance improves. The alternative which is to trade a smaller portfolio, is often the better choice for risk reduction.

Need for Risk Control

Ideally, a Stop order that has been entered into the market is expected to automatically get you out of a position when prices move against you. It forces you to decide, in advance, how much you can afford to lose, so that your trading risk is under control. It avoids last-minute decisions and the temptation to hold a losing position with the hope that prices will recover. Stops are normally "resting orders" held by the floor broker to be exe­ cuted when the price is reached. For our purpose, we will include any trader who intends to limit losses to a fixed amount, whether or not the order is placed in advance as a Stop order.

Risk Protection or False Hope?

The use of stops, or the intent to limit losses to a predetermined fixed amount, may give a false sense of security (shown in Table 6-1). For example, a Stop order that is reached during an illiquid or quiet market will result in large slippage; during a fast market, or a price jump, it will often get the worst fill. Large traders cannot enter stops because they move the market and cause substantial slippage.

Setting Stops Based on What Ton Can Afford to Lose

Stops are set for two main reasons:

•  They can trigger a change in market direction, or

•  They can limit losses to a preset amount.

Table 6-1. Using a Stop-Loss

Expectations Reality

Limits losses to preset levels

Slippage and price shocks increase the size of the loss substantially

Losses will be much smaller than profits

Individual losses may be smaller, but a series of losses can be very large

There can still be more frequent profitable Market noise causes more losses than profits
trades than losing ones unless the trader has a very good strategy

Stops placed to limit the size of losses are more common and more difficult to evaluate. The use of a small risk control can be intended to produce a particular performance profile, as in Case 1.

Case 1. A different foreign exchange trader buying 10 million $/marks expects to iiold the trade for a US$20,000, or 0.20 percent profit, based on a 1-day volatility strategy. To show proper risk con­trol, the trader would like to keep losses under US$10,000, giving a 2:1 ratio of average profits to average losses. The trader would also like to have at least half of the trades profitable over a long-term per­ formance record. The total picture would show a steady profitabili­ty and reasonably small losses—a very pleasing performance.

But the market does not work this way. It is not possible to control both the return/risk ratio and the number of profitable trades.

Interference from Market Noise

"Noise" is the term given to unpredictable price movements. Normally noise is associated with small price changes, but in reality it includes large price shocks. Noise is caused by traders entering and exiting the market with different objectives and different time frames. An institu­tion adds positions because of new investment funds; Russia sells gold to pay for a wheat purchase; or an auto company liquidates part of its stock portfolio to cover foreign exchange losses. Although each event is unique, the total picture of this noise creates a very predictable random distribution with some very special properties. Figure 6-1 shows the gen­ eral rule that erratic prices (noise) which are twice as large occur half as often.

The S&P price changes appear to fall exactly on the random distribution line in Figure 6-1 (a), while the bonds and Deutsche marks have about twice as many moves of 1 percent and 4 percent than a random occurrence. It is important to note, however, that the number of occur­ rences of bonds and Deutsche marks still decline by 50 percent as the size of the move increases in steps of 0.5 percent.

When we look closely at the larger moves (called the "tail" of the dis­ tribution) in Figure 6-1 (b), the S&P has more extreme changes than either of the other markets and much more than if it were a random dis­ tribution. These large moves are caused by price shocks.

The interesting phenomenon concerning the size and frequency of erratic price noise is that the number of occurrences times the size of the move is always the same. Although the bonds and Deutsche marks varied from a random distribution, they still conformed very closely to the same rule.

Specifying the Performance Profile. Based on the normal distribu­ tion of prices shown in Figure 6-1 and Table 6-2, the foreign exchange trading example, Case 1, could not work. If a trader targets a US$20,000 profit with losses one-half the size, then there should be twice as many losses. As the size of the profits increases compared with the average loss, the num j ber of small losses must also increase. If a naive trading model gives four profits averaging US$20,000 and eight losses of US$10,000, then a good strategy must capture higher profits, reduce the losses, or turn one or two losses into small profits.

Expectations

In these cases, the real prices are so similar to random movement that you could not tell the difference without a careful mathematical study. We can assume that decreasing the size of a stop-loss will cause a pre­ dictable additional number of trades to be stopped out, always con­ forming to Table 6-2, with the possible exception of large stops. . Our expectations of improved performance using a small stop should be low. As the stop-loss gets smaller, the number of trades that are stopped out gets larger. Each time a stop-loss is activated, there is additional slippage because the order forces an execution in the direction of market movement.

The larger stops may be different. If prices are exceptionally volatile, then a stop-loss might save some of that loss. If you plan to get out when a price shock hits, then a stop-loss could not make things worse— and might make it better by acting as a major risk control. Large stops only work with longer-term trading; a fast trend will always give a reversal at the same time a larger stop would have been reached. Therefore, small stops will generally be ineffective, but larger stops will sometimes improve performance.

Capitalizing on Distribution. A successful system will need to cap­ ture larger than normal profits, or cut losses. The use of a stop-loss is the place to start. If a few of the unusually large price moves shown in Figure 6-1 (b) can be controlled when they generate losses and captured when they produced profits, the overall system performance will improve.

Testing a System with a Stop-Loss

A simple trend-following system was tested on daily data using stop-losses to see whether they helped profits and controlled risk. The system included the following main features:

•  An exponential moving average, @ Exp_MA, was used for the trend.

•  A buy signal was given when the trendline turned up,
@Exp_MA> @Exp_MA[1].

•  A sell signal was given when the trendline turned down,
@Exp_MA< @Exp_MA[1].

•  The trade was exited if the loss, at the close, exceeded a preset percent­
age level (the stop-loss), or when a reverse buy or sell signal occurred.

•  All orders were executed on the close of the local trading day.

•  There were no commissions or slippage charged to any trade.

•  Risk was defined as 1 standard deviation of equity.

In actual trading, the more active systems will show additional losses due to relatively larger transaction costs. Because all trades are executed in the direction of the price movement, individual estimates of cost can be subtracted from the profits per trade. Without costs, the results of this test will still give a reasonable comparison of the use of different stop-loss limits.

Table 6-3 shows a 10-year optimization of trend speed versus percentage stop-loss for Chrysler, Siemens (for only 5 years), the Deutsche mark, and Eurodollars, ending June 1993. Only the adjusted rate of return is shown to allow comparisons (see in Chapter 4, "Choosing Between the Currency and Bond Portfolios," for an explanation of adjusted rate of return). The far right column in Table 6-3, marked "None," shows the results of the trend system without any stop-loss. The far left column, marked ".02," shows a very small .02 percent stop-loss.

The conclusion is that improvements are inconsistent. The best results for this test set seem to be Siemens, which has higher adjusted returns for trends below 100 days, and scattered improvement in the middle of the table, centered near a stop-loss of .50 percent.

Because the major currencies are said to have more trends, we expect­ ed a stop-loss in the Deutsche mark to produce better results. Prices that reach the stop-loss level in the D-mark should continue in the same direction until it activates a new, reverse trend signal. That did not hap­pen. The results from the smallest stops were uniformly worse than the use of no stops in the 10-year test. Improvement occurs in a small area in the center of the table.

Larger stops, or no stop, show a more uniform result. When using an optimization test to select the best trading rules, it is preferable to see smooth, consistent results, rather than alternating profits and losses. Both Chrysler and the D-mark have poor returns in the bottom left cor­ner of the tables.

Another View with a Shorter Test Period

Table 6-4 compares the cash returns with risk-adjusted returns for the Deutsche mark and Eurodollars over the 4 years ending November 1992, using comparable stops. The Deutsche mark cash returns appear much better than Eurodollars but have higher, inconsistent risk. When adjusted for risk, the Eurodollar returns are clearly better.

Contour Map of Eurodollar Stop-Loss Tests. Figure 6-2 shows a contour map of the Eurodollar tests. The white areas have the highest profits and the black parts the worst losses. As in the 10-year tests, the use of small stops, seen at the left edge of the figure, is inconsistent, alter­ nating between light and dark, with neither the best or worst results. As stops become larger toward the right, performance is more uniform and predictable. We can conclude that small stops are sensitive to specific price patterns and noise; therefore they are erratic.

Table 6-3. Stop-Losses: Results of Longer Term Tests

Values are annuallzed rate of return, adjusted to a 25% maximum drawdown.

Pattern of Results. The size of the stop-loss must be based on the speed of the trend. A small stop with a slow system will usually be hit and rarely allows the trade to reach a profit. A large stop with a fast system will never be reached before the trend signal reverses the position. It should not be surprising that only a narrow band of stops applies to one trend speed.

Intraday Stops with a Daily System

It is tempting to react quickly to an adverse price move to keep losses small. Although the trend may be determined using daily closing prices, a stop could be activated when prices move badly during the trading session. Unfortunately, the market noise will cause many more stops to be reached. At the end of the day, you will have captured losses that would have disappeared had you waited for the closing price. The accumulation of intraday losses and the increased number of trades will be far worse than basing the stop-loss on the same daily data as the system trend calculations.

Table 6-4. Stop-Losses: 4-Year Test Results Comparison of cash returns vs. risk-adjusting returns. Deutsche Mark: 1030 Days ( 11/22/88 to 11/23/92 )

Benefits of Being Out off the Market

Being out of the market may reduce equity fluctuation, even though a smaller profit is taken at the end of the trade. If the use of a stop-loss neither helps or hurts profitability, then the time spent out of the market will avoid some erratic price movement and improve the reward/risk ratio of trading performance. Holding a position that alternates from a profit to a loss without ultimately ending up profitable does not benefit your trading.

Comparison of 4-Year Results: The pattern in the "NONE" column (indicating no stops) shows that the risk-adjusted results are, in general, more consistent than unadjusted returns. Risk-adjusted returns show some large improvements due to the use of stops, but the total picture is very inconsistent. The variation is especially apparent by comparing the smallest .02% stops in the far left column with no stops in the far right column. Large stops affect the longer-term trends first (seen along the bottom of the table) and slowly work their way to the faster trends as the size of the stop gets smaller. Deutsche mark returns are uniformly better with 2% stops but inconsistent when stops become small. Eurodollar returns are best with no stops, which confirms their strong trnding character.

Figure 6-3(b) is the contour map of the return/risk ratio (annualized return divided by 1 standard deviation of equity changes) over the same 10-year period. The white area, showing the best performance, is larger than in the previous map and includes areas in the bottom left. This means that the use of a small stop with a slow trend does improve Deutsche mark performance by exiting trades that become volatile and unprofitable. The far left edge remains inconsistent, and the right side is very uniform.

A Stop-Loss May Conflict with the Strategy

The nature of a stop-loss order is often contrary to the system with which it is used; it may "fight" with a trend-following program. The purpose of a trend is to smooth out and ignore market noise; the trend- line is substituted for prices to represent a better approximation of price direction. A stop-loss that is too close will be reached by an erratic price move, offsetting the value of the trend.

Trend Systems

When a stop is reached, the trend has not yet reversed. The system is saying that the trend is still intact. If that is true the trend will continue and the loss will turn into a profit, but you would not have a position because you were stopped out. If the stop works most of the time, then the trend changes whenever the stop is triggered. That is the same as saying the trend is too slow. A faster trend would catch the change of direction sooner. In either case, the solution does not seem to lie with the use of a stop-loss.

Countertrend Systems

Trading against the trend requires frequent small profits. To get more profits, it is necessary to hold larger losses, waiting for prices to move your way whether because of noise or good forecasting. A stop-loss will cause more frequent losses and prevent profits from developing. The two will not work together.

Apparent Improvement

Tests show that stop-losses improve results, either outright profits or reward/risk ratios. Usually, that is not the case. Tests of short intervals may not be representative of the long-term picture. The use of intraday stops can easily misrepresent the fill that is received from a stop. The lack of liquidity, or a price shock during the trading session cannot be seen by most computerized testing packages.

One event can appear to make the use of a stop-loss worthwhile. Reducing the loss from a major price shock to a reasonable level may be a good exchange for small give-ups. A trader must realistically assess whether a resting stop or visual stop would have offered the protection needed. It is difficult to design a system that continually gives up prof­its in expectation of possibly reducing risk in a single case.

Once a trade has been stopped out, but the underlying trend or counter- trend position does not reverse or exit naturally, the trader faces another problem. Do you reenter the market and chance another loss in the same trade? Then the compounded risk would be much larger for each trade. If the decision is to reenter, what criteria should be used? Because the trend- following signal remains the same, there is no new signal to buy or sell; therefore, the trader must have additional rules for reentering the market. That makes the system much more complicated.

Managing Risk with and without Stops

If small stop-losses are not predictable, then the best choice for controlling risk is to deleverage, reducing the risk to acceptable levels. Deleveraging is always the safe alternative. It can be achieved with the following simple procedure:

•  Find the long-term risk level of the trading program, using a combi­
nation of maximum drawdown and standard deviation of equity
changes (see Chapter 4).

•  Adjust the system risk to your acceptable level. You probably want
less than a 1 percent chance of losing more than 10 percent of the
invested capital during any year; therefore, set 3 standard deviations
of the annualized risk equal to 10 percent of your investment.

•  Determine the investment size or the amount of capital to be traded,
based on the adjusted risk in step 2.

•  To give protection from major price shocks, use a larger stop-loss
that will not be easily reached due to noise.

For example, you have a trading program that returns 40 percent per year with a 95 percent chance that losses will be under 15 percent during the 10-year test period (1 standard deviation of the equity changes is equal to 7.5 percent). However, you want to keep the risk under 10 percent, which is two-thirds of the current level. If the required capitalization is US$1 mil­ lion, then increase the investment by one-third to US$1.33 million.

Because control of risk is more important than higher profits, we accept a 20 percent expected return to keep the risk at 7.5 percent. By using a large stop-loss, the returns on the test optimization are more consistent, and we have greater confidence that the stop-loss will not interfere with the expected returns.

Institutions allocate their professional traders a specific amount of trad­ ing capital on which their performance will be judged. To maximize profits and minimize risk, they tend to use high leverage and small stop-losses. The result is often modest profits and larger risk.

As discussed earlier, the amount of noise in foreign exchange markets causes small stops to be triggered frequently, preventing traders from reaching their profit objectives. But small stops are seen to be necessary because of the high leverage and potential risk. Most foreign exchange traders would be under pressure to explain losses in excess of 10 percent of capital.

Therefore, the best performance profile is not achieved by using small stops. It is best when the program is deleveraged and only larger stops, or no stops are used. But deleveraging means using a smaller amount of capital. As was pointed out in Chapter 4, if you achieve better profits trading a smaller amount (which can be seen from the size of the positions taken in the market), the head of the Foreign Exchange Trading Division might say, "If you can produce 20 percent return using only half the money, then use all of it so that you can get a 40 percent profit!" Unfortunately, the reserve capital is needed for a period of unusually high risk. It does not occur often, but you must be prepared for it anyway.

What do you do? If you don't trade more of the capital, it might be taken away. If you do trade more and the market is hit with a price shock, you can lose an unacceptably large part of the capital. The answer is don't trade more. You must be able to explain that deleveraging to achieve risk control is safer for the investment portfolio and for the company. Artificial risk controls are counterproductive over the long term and often over the short term. It is a corporate problem that must be resolved. The trader must only produce returns with the best return /risk ratio.

Siimmtuy

It is difficult to trade without a clear idea of risk, and a stop-loss, or fixed-value limit, is the most agreeable to traders. A stop-loss based on the number of points you are willing to lose is not a good choice. But performance using a stop-loss is inconsistent and, during a fast market or a price shock, when risk protection is most important, a resting stop- loss could result in the worst fill. Stops based on logical price levels, such as support and resistance, or volatility, are much more likely to improve performance. Tests confirm that larger stops give better per­ formance than small stops.

If a relatively small stop-loss consistently improves a trend-following system, the trader should see if a faster trend might work even better. If a larger stop-loss improves performance, it may have been the result of a single event—it would be difficult to build a system around one situ­ation.

Reducing risk by increasing the investment size, or lowering lever­age, is the safest method of all. With smaller exposure, risk is always proportionally less. If you increase the leverage of your system because of the perceived safety of a stop-loss, one price shock is enough to cause serious damage. A stop-loss does not guarantee risk protection. If you need to deleverage to avoid occasional large risk, then a stop-loss may no longer benefit performance.

 Back to contents page