IT Setup

I wanted to show you my IT setup - nothing spectacular I think - just for the sake of completeness as an algo blogger ;-) 

Hardware:

• IMac 27 Inch (built late 2012) /  2.7 GHz Intel Core i5 / 4 GB 1333 MHz DDR3 / AMD Radeon HD 6770M 512 MB / 1TB HDD
• 2 times 24 Inch Samsung SyncMaster 245B
• External harddrives: LaCie Porsche Design P9230 3 TB (managing the tickdata) / Apple Time Capsule 2 TB (saving the OS)
• Power supply: APC Back-UPS 700 / APC edge connector

Software:

• MAC OS X Lion 10.7.5
• Parallels Desktop 7
• Windows XP

VPS:

• Dell PowerEdge or HP Proliant
• Windows Server 2008 64 Bit
• 2 Cores / 8 GB DDR3 / 500 GB HDD

Tickdata backtesting

Metatrader 4 was not build for tickdata testing. Nevertheless there are serveral ways to get a backtesting quality of 99% or more. Here you'll find excellent descriptions of several methods. Next to that there are several brokers offering free tickdata. Everyone has it's own advantages and disadvantages. Dukascopy kindly offers free tickdata too. But therea several things you have to consider like the data type offered by Dukascopy. In the following I'll describe my experience with several ways of downloading and importig Dukascopy tickdata into Metatrader 4:

1. Download dukascopy tickdata with tickstory and start MT4. I managed downloading Eurusd tickdata from 2003-2013 with tickstory an got a backtesting quality of 99,9%. Perfekt! Easy download and no problems in MT4. But soon it became a bit rough. When I tried the same with Gbpusd (and other pairs) it did not work - no download - no backtesting. Some weeks ago there was released version tickstory 1.2 which I'll test soon. Perhaps this will solve these problems.
2. Download dukascopy tickdata with Strategyquant Tickdata Downloader. After that you have to copy the csv file to MT4 tester folder and apply this process. It's quite easy when you've done it once. Unfortunately the history starts not until 2007 and you get a backtest quality of 99% (perhaps a bit less due to the script which converts the .hst file). Up until now this is my prefered process.
3. Download dukascopy data with this process. Then apply this process. I will add my comments when I've tested this method.

Anyway you always should pay attention to these comments. If you want to do longer backtests (longer than two years = file size is larger than 2 GB) the easiest way is to use Birt's tickdata suite.

Strategyquant user guide

Today I managed reading the user guide of strategyquant. It's well structured and even if you do not use the program you could read it for educational purposes too. Things like walk forward analysis, genetic evolution in programming, robustness test and monte carlo simulation are explained. One thing I am missing that is already abvious a this stage is the function to import custom EAs into strategyquant. Although it sounds a bit complicated you can import custom indicators but you cannot import custom EAs. This is strange. So you'd have to build an indicator of your EA and import it in strategyquant or build the strategy itself in strategyquant with it's built in strategy builder. But I do not expect to be able to build complex strategies within the program. I have to emphasize that I've just tested the program and not seriously used it at this stage. Nevertheless I think the program will be of great help. Optimally you can find promising strategies with it and fine-tune them later in the mql editor with your programming skills.

Data

Choosing the right data is critical for the strategy development process. The data's accuracy should determine the meaningulness of the backtest's result. Nevertheless there are systems which can be backtested and there are systems which cannot be backtested like newstrading and even some high frequency trading strategies.
In Forex there is not one price because of it's decentralized structure. Therefore the best approach in my opinion is to develop and test the strategy with different data sources. I prefer tickdata because it's most precise. Nevertheless I am not convinced that this data will always produce the most realistic results. I expect a lot of variation in the results of backtests, forward tests and backtests with different data sources and accuracy. Therefore I will take the average result over different brokers into consideration - not paying too much attention to one single result. On the other hand testing with different data sets offers the opportunity to pay attention to possible feed dependency of the system already within the backtesting stage. This could save a lot of time in the forward testing stage. I will be using the following data sources:

1. Built in data of strategyquant
2. Dukascopy tickdata
3. Alpari data
4. TrueFX tickdata
5. Forextester data (free and paid)
6. GAIN Capital data
7. Histdata

Genetic Programming

For genetic programming purposes I will be using the software strategyquant. I've tested it and found it quite helpful. It should help me to develop strategies, run robustness tests and apply walk forward analysis. The strategy development is done by genetic evolution. Custom indicators can be imported. I will use a VPS to run the development processes according to my ideas.

Execution and backtesting platform

My plan

I am used to metatrader 4. I know it's pitfalls especially in backtesting. Regarding third party plugins I am curious how metaquotes policy will pay out. Nevertheless I think it's a good starting point. It's free and widely used. In the future I want to move to ninjatrader and preferably matlab.