Categories
Forex Educational Library

Designing a Trading System (V) – Testing Exits and Stops

The importance of exits

It’s not possible to create a system based only on almost perfect entries and using random exits and stops. As we could observe in the entry testing example, the percent profitability of an entry signal is very close to 50%; a coin flip would reach that score. The real money is made directing our efforts into proper risk control, money management, and adequate exits.

Test methods for exits and stops

Testing exits independently are much more difficult than testing entries. Sometimes the entry and the exit is mingled together in a way that’s difficult to separate, for example, when trading support-resistance levels. If that’s the case, the best way is to test the entire system at once.

When it’s possible to evaluate exits by themselves Kevin J. Davey proposes two approaches:

  • Test with one or several similar entries
  • Random entry

The ideas by LeBeau and Lucas move near the same path: Their method of testing exits is to create a very generic entry signal to feed the exit method. If each exit method is tested using identical entries, it should be possible to make valid conclusions about their relative value. Not all exits work equally well with different entries, but if the entry is generic enough, we may get some feeling for the relative effectiveness of a set of exits.

Their method of testing exits is a general approach that encompasses both ways enumerated by Kevin J, Davey.  Nothing is more generic than a random entry, and a non-random entry is improved if we make it close to the type of entry we intended for our system.

Test with similar entries

The fundamental idea behind testing exits, is, of course, to see if it has an edge over a random exit or a benchmark exit. A reliable exit should make a wrong or random entry into a profitable system, and for sure, it should not make an entry system worse.

To create a generic entry signal similar to the one we have in mind for a new system, we need to classify that system. Usually, new systems fall into two categories: Trend following and return to the mean, or counter-trend.

For trend-following systems, Kevin uses an N-bar breakout entry, while for counter-trend systems he uses a relative strength index entry. An exit strategy that works well in similar entry types should work equally well with the intended entry.

Random Entry

The Idea of a random entry is to see if the exit shows an edge by itself. The way to do that is to use a random entry as the optimising parameter with more than 100 runs while keeping the exit fixed and observe what percent of the optimisations are profitable. A very successful exit might have a high percentage of successful series. That is, in my opinion, the best criteria to know if an idea for an exit is valid and has an edge.

Fig. 1 shows the EasyLanguage code for a random entry. This code is a modified version of the one that Bill Brower has explained in his book EasyLanguage Learning-by-Example Workbook. This piece of code can easily be transposed to MQL or another language.

  • Fakecount is used to iterate the optimiser to get several random entry results
  • Initbars is the minimum number of bars the system wait till a new entry signal can be triggered.
  • Randvar is the maximum of bars a random generator can deliver, and it’s added to initbars.
  • BuySell is used to test long and short positions separately 1 for longs -1 for shorts

Fig. 2 shows the 3D graph of a trail-stop exit. We observe that around 0.15%, there is the sweet spot in all random iterations (50 in this example).

In this case, at 0.15% trail stop, the number of profitable iterations were more than 30 out of 50, that is, 60% were positive.

Evaluation criteria

Besides profitability, it’s good to assess the quality of an exit signal based on Maximum Favourable Excursion (MFE) and Maximum Adverse Excursion (MAE) standards.

Those two concepts, developed by John Sweeney, define the maximum adverse excursion (MAE) that a sound signal reaches before it proceeds to run in our favour.

Fig. 3 Show a trading system without stops. We observe that there is a drawdown level beyond which there is almost no winner and mostly are losers. That level defines the MAE optimal stop.

The maximum favourable excursion is the profit level that maximises the result on average of a trading system before it starts to fade and lose profits. MFE is the complementary concept to MAE, but it is a bit harder to visualise.

In Fig. 4 we observe that for the trade in the circle the actual run-up was 0.54% while the real close was 0.3%, the difference is a substantive profit left on the table. The goal of the exit is, on average, to make that distance as short as possible. That means the trading points should be, on average, closer to the upper line, as Fig. 4 shows.

Those two concepts applied to exits mean that the exit must avoid levels beyond MAE, and it should not give back too many profits and get close to the maximum possible excursion.

Total testing

Once we have assessed the goodness of an entry system and have a collection of useful exits to compliment it, we need to test them together and see how entries and exits interact.

The main rule at this point is getting profitability in the majority of combinations of entry and exit signals. For example, if we have ten parameter values on entries and ten on the exits, we would like positive results on more than fifty. Also, I love to see smoothness on the 3D surface. Too many peaks and valleys are a sign of randomness, and it’s not good for the future behaviour of the system, as in Fig. 5

 

Then, we choose a point on the hill that’s surrounded by a convex and smooth surface with similar profitability. That way we are trying to prevent a change in the character of the market which would spoil the system too soon.

Fig. 6  and 7 shows the equity curve and drawdown of a mean-reverting system, and its summary report after having passed the different stages up to the final testing and selection.

With that final assessment, we have achieved the limited testing stage. Therefore, this strategy has passed the test and is a candidate to further forward analyses and optimisations.

 

 

Further readings from this series:

Forex Designing a Trading System (I) – Introduction to Systematic Trading

Designing a Trading System (II) – The Toolbox Part 1

Designing a Trading System (III) – The Toolbox Part 2

Designing a Trading System (IV) – Testing Entries


 

References:

Building Winning Algorithmic Trading Systems, Kevin J. Davey

Campaign Trading, John Sweeney

Computer Analysis of the Futures Markets, Charles LeBeau, George Lucas

Images were taken with permission from Multicharts 11, trading platform.

 ©Forex.Academy