Since the initial test of the 65sma-3cc model was encouraging, we can now do more testing. The first item of business is to insert an initial money management stop into this model. Our detailed analysis of the MAE showed that we could safely set bur stop at $1,000, or even as high as $1,750, and capture substantially all profitable trades.
However, we should insert another condition into the formulation of the model before testing for the effect of initial stops. If our stop is too “tight” during testing, we will be stopped out right after the first signal. Then, there may be a succession of trades, all in the same direction (all long or short signals), that will also result in losing trades, before one of them kicks into the major trend. Thus, the analysis would be distorted. What we want is to pick off exactly the same trades as we did without any initial stop. To achieve this goal, we must insert rules that do not allow successive trades of the same type, to ensure that we will not have two back-to-back long or short trades if we get stopped out after the first signal. In effect, with this rule, if we get stopped out, we must wait for the opposing signal before getting in. Of course, you do not need this condition for actual trading.
Inserting an initial condition should have two effects. (1) It should reduce the maximum intraday drawdown, since some potentially large losing trades will be cut off. (2) It should also reduce the number of profitable trades and the total paper profit, since the same stop will also cut off some potentially profitable trades. Some calculations will show if we can verify these expectations.
The results of these calculations are shown in Table 4.3, which can be compared to the results in Table 4.2. The markets and test periods are identical in both tables. Adding a $1,000 stop reduces total paper profits by 21.5 percent, from $1,386,747 to $1,088,804. Similarly, the number of winning trades fell to 689 from 810, or by 17.6 percent. As expected, the average maximum drawdown and its standard deviation also decreased, showing the desired smoothing effect due to the initial stop. The reduction was about 18.5 percent in the drawdown, and 40 percent in the standard deviation. Thus, adding a hard dollar initial money management stop had the desired effect of reducing drawdown and smoothing out the variation in system performance. There was also a resultant reduction in total returns.
We chose the $1,000 initial money management stop from the MFE plot. Calculations for a $500 stop result in an even greater reduction in profits, drawdown, and volatility.
We can continue this line of thought by looking at the U.S. bond and deutsche mark markets. Our analysis of 777 profitable trades showed that once the drawdown exceeded -$1,750, few trades ended with a profit. Hence, the initial stop is varied from $250 to $1,750 in the following tests to look at the effect on the total number of profitable trades. As the initial money management stop increases, the number of profitable trades increases and then levels off (see Figure 4.12, page 90). This shows that the initial stop acts as a filter, and as the stop widens, it allows more trades to pass through. Eventually, the filter is too big, and does not cut off any trades. This allows the number of profitable trades to level off.
We have so far placed our stop using a dollar figure without ac-counting for market volatility. However, whereas in the coffee market, a $1,000 stop may seem too tight, in the corn market it may seen too wide. Thus, in some markets, a given stop will work like a stop near the left edge of Figure 4.12, and, conversely, in other markets, the same dollar stop will work like a stop on the right side of the figure.
We can get around this problem by using a volatility-based initial money management stop. For our calculations, we can set an initial money management stop as a multiple of the 15-day SMA of the daily true range for measuring volatility. We use the same continuous contracts as in Table 4.2 to test the U.S. bond market with volatility-based stops ranging from 0.25 to 3.0 times the 15-day SMA of the daily true range.
Figure 4.13 shows that a stop set at less than 1.25 times the average volatility is too tight. Once the stop increases past 2.00, the paper profit increases and the drawdown increases. The drawdown is minimized at a 1.50 stop. This means there is a balance between being too tight or too loose. The same behavior can be seen very nicely in the live hogs market (see Figure 4.14, page 93).
As might be expected, when we increase the money-management stop, the largest losing trade will probably increase. This happens be-cause our stop is farther and farther away from the entry price. The sugar market shows this nicely (see Figure 4.15) when tested over the same period as Table 4.1. Other calculations (not shown) show that the largest winning trade is affected only a little by the initial stop, since these trades usually are profitable from the very beginning. You may set a volatility-based stop or a hard-dollar stop with equivalent results. You may have to set a different dollar stop for each market, although you could use the same volatility stop across all markets. Note that with a volatility stop, the actual dollar amount changes over time, aad hence you must ensure that this stop is within your overall hard-dollar limits for risk control.
You should note some limits on how the initial money-management stop can be tested. In most cases, the amount of the stop must be larger than the daily trading range. The software cannot determine if your stop could have been hit intraday if the stop is smaller than the daily trading range. Unless you have intraday data, you cannot test the effect of, say, a $250 stop using daily data.
In summary, adding an initial money-management stop is useful from a risk-control point of view because it reduces the largest losing trade and the maximum drawdown. But, it also cuts off some winning trades, and hence total profits are lower over the long term. You may add a dollar stop or a volatility-based stop, but both must follow sound guidelines.