Moxy's System Development Journal

I've decided to document my first system development. I have done programming in my other life for the last 20 years and have decided to take the plunge and try to develop a trading system using MQL.

I have been reading a book by Dr Emilio Tomasini and Urban Jaekle called 'Trading Systems: A New Approach to System Optimisation and Portfolio Construction'. It's a very good read and shows the step on how to go from idea, into development through to testing and deciding if the system is worth trading. It got my development juices flowing so hopefully I will pick up some MQL skills along the way whilst developing my first mechanical trading system.

I plan to use the example system given in the book and will hopefully be able to replicate the books findings and may be, just may be, improve on the system.

To start with I will describe the system and identify the basic logic that I will hopefully be able to code and test. From there I will carry out the various test described in the book and add a few of my own.

If anyone wants to join in or follow along, please feel free. The more help / pointers I have in the right direction the better

Important point to remember: I am not an experienced trading system developer!! :nerd:


System Description:
This is a simple moving average (MA) crossover system with a slight twist. Instead of the MA crossover being a signal to enter a trade this system will use the high/low of the closed bar when the MA crossed as the trigger price to enter the trade. An order in the opposite direction will close the existing position resulting in a reversal of the trade.

Notes:
Only Market orders are to be used. No BuyStop or SellStop orders.

The logic above shows that a signal in the opposite direction will close 1 position and open another in the opposite direction.

Extern variables for SL, TP and Trailing stop are to be made available and applied to the order at time of order entry.

Cheers,

Yesterday I spent some time turning the pseudo logic into MT4 code. With this being the first system I have coded it's fair to say I had some 'issues', but with perseverance I managed to complete the coding to allow me to run the first tests of the system to see how it performs compared to the results published in the book.

I said in the previous post that the book got my development juices flowing again, which gave me the drive to give the system coding a go. After running several back tests, I can now say those juices have been soaked up and i'm almost completely dry. My backtest results are like chalk compared to the book's cheese. The system performs nowhere near what is documented in the book.

What a pain in the arse.

The results are so different, it seems pointless continuing the testing exercise that I had planned. To give you an idea of the differences. Here are some of the key figures from the book compared to those found in my backtesting.


Cheers,

They really don't look that too far off to me. I'd stick at it if I were you, you'll get something good eventually.

What were you expecting for results as compared to the book?

So why are the results so different? Lets look at the 2 major differences between the 2 tests:

The book runs the backtests on GBPUSD on a 30 minute timeframe but the authors use the futures markets and not the spot forex market which I am using with MT4. Each tick in the futures market is worth $6.50, whereas in the spot forex market a one tick move on a one lot is worth $10. The actual tick value in my mind should not determine whether a system would be profitable or not, but it will affect the actual dollar profit produced from the system.
The next point to note is that the book runs the backtest with zero slippage and no commission costs.

When you consider this information, this could account for the huge difference in performace I am seeing. The futures markets often have at least a one point spread on GBP/USD and when liquidity is thin you can easily see a 2 or 3 ticks spread. On the spot forex side of things the spread is very similar but very rarely gets below 2 ticks. When I executed the backtest the MT4 Symbol properties were showing around a 2.5 tick spread. To see if we can bring the results of the two back tests into line (or at least a little bit closer) let's do some simple math to determine what the results may be if we exclude the 2.5 tick spread from the MT4 backtest results. Also note the MT4 back tests will often give slippage due to the way it calculates the individual ticks for our backtest. I'm going to give the MT4 BackTest results some leeway and account for a total of three ticks (combined spread and slippage).

The total number of trades in the NT 4 BackTest was 1892. Let's multiplied this number by 3 giving 5676, and let's multiply that by $10 giving a result of $56,760. The MT4 BackTest results showed a total net profit of -$99,119. If we add on the $56,760. This will bring our total net profit to -$42359. Hmm, still pretty crap.

One thing to keep in mind when comparing the two sets of results is that this initial test is really only a starting point and is to be used to determine if the system would have the makings of a viable profitable trading system. The BackTest results described in the book do show a nice steady rising equity curve. It does have drawdowns, but those drawdowns quickly recover, and the system achieves new equity highs on a consistent basis. The MT4 equity curve on the other hand, has a steady decline.

So, even though I am convinced, after reading the book, this trading system is profitable in the futures markets, I am left wondering whether it's worth continuing to investigate this system. Like I said above, this first initial test is really to determine if the system shows promise and can be made into a viable profitable trading system. Looking at the results of the MT4 BackTest with a declining equity curve, it doesn't look very promising at all. I don't know exactly why the system doesn't come anywhere near the performance of that described in the book. May be it is due to the quality of data we have access to in MT4 or may be the two different markets, even though based on the same underlying instrument, behave in a very different manner. Or maybe it's the quality of the tools used to BackTest the system. The authors of the book used Trade-Station for the majority of their testing and by all reports Trade-Station is held in quite high regard.

Any thoughts are very welcome.

I think my next step with system development will be to look for a different system to test. There is no point flogging a dead horse!! Although I don't want to lay this system completely to rest.

If anyone has any system idea they think is worth looking into, drop me a message and lets see what we can come up with. I am not just going to code EA's for anyone, you will have to put in some effort yourself by describing why you think a system is worth coding. Also, remember, I am not an experienced MT4 coder, so keep the systems as simple as possible (they are the best anyway).

For anyone interested, I have attached the MQ4 file I produced for the test.

Attached File(s)

MA-Twister_v01.mq4   7 KB | 565 downloads

Cheers,

Hi TraderHotch,

Thanks for stopping by. The results in the book show a steady rising equity curve with short periods of draw down. The MT4 results, even though showing similar % winning trades etc... has a very consistent declining equity curve. It looks almost the opposite of that shown in the book.

The system described in the book does go on to apply various filters, take profit and stoploss levels that result in a very consistent profitable system. I had planned to do the same with the MT4 version and probably will as curiosity will get the better of me and I will want to see how the system will perform after applying the filters.

Cheers,

The opposite would involve wild swings no?

If you had the exact same data as the book there might be reason to me distressed, but so far so good. Keep at it and you'll get somewhere eventually.