The Automated Trading Machine

The way to the

Automated Trading Machine ^TM

In 1985 I studied the financial news paper and thought myself “… if I just could make money from this numbers, by shifting them back and forward …”.

By then, all I had was an idea and an Atari PC. The programming language back then was called GFA Basic. Computing power was there, but still limited – compared with the now.

A GFA Basic programmer book in 1986

The idea was a tool or method, which would produce monthly payments from a capital, like a normal monthly salary.

It should take a little while until the work would start towards the idea. In 2000 I started to implement formulas and functions into Microsoft Excel. Soon I realized if this should work the way I sought, it will need more than just a script and function language.

At the end of 2001 the idea was ready to take off. With a bunch of money, a laptop, a ticket to warm, cheap country, and lots of ideas, I went on, to start realizing the automated trading machine.

The ideal place to live … or work … or both …

In 2002 we were manly busy with programming the main parts for the AutoAnalyser. Soon I realized that historical data and current data is needed to go on. Up to then we were testing on saved static historical data, and basically ignoring the current market. Since historical and current data was not just free and easy to get, I started the module easyStockDater, the stock quote downloader.

We had some modules together. We started to test via the back testing method. Some of the models would work great for some time, just to loose the gain shortly after..

Also, the back testing method was not satisfying me. I realized that, twisted and turned indicators, would produce great results – in the back test window – but would fail mercy less in out of sample tests.

To solve the “good looking back tests” problem, we developed the so called Real Walk Forward Test (RWFT).

One could set back the program in time and let I run forward, day bay day. The back tests then had to sustain all testable parameters in the RWFT.

From highs to downs we created own indicators, some great, some useless.

In 2003 we started to marketing just the downloader module, with a surprising success.

Time went on through 2004 to 2005, with developing here and there.

In the middle of 2005 we had a break through with a systematic approach how to filter out reliable trading signals with AutoAnalyser. We finally could produce Real Walk Forward Tests which would produce average yearly returns of around 13%, with a risk of 8 (One standard deviation) over a testing period of over 50 years (gains and losses not reinvested !).

S&P500 +50 year Real Walk Forward Test (RWFT)

Note: AutoAnalyser equity curve is cumulated profit only, without reinvesting the profits (which is the realistic approach).

A important rule is also to take profits out of the system. Also the below result is looking tempting, it is not realistic. The profit is reinvested into the trading capital. Over time the capital rises and more money per trade is gained or loosed. Two main drawbacks will occur over time. First, you need to eat once a day and can not starve for +50 years. Second, the bigger the invested capital, the more difficult it gets to place the trades. And of course, no body wants to live a life for work – we want to live a life for leisure.

Note: AutoAnalyser equity curve with reinvested gain and loss.

I insisted to the team, to have very long testing periods, in order to make sure, that things are as solid as possible.

However, returns in the above mentioned range were not enough. The goal to reach is set to 20% annually return, no negative year, and a risk around 8 (One standard deviation).

Since only few instruments (Stocks, indexes etc.) have +50 years of historical data, we were limited in the choice. We decided quite early that the S&P500 would be representative. The SP500 also has the longest ETF tracker, the SPY. Note that you can not directly trade the raw index numbers. A trader needs to trade an index through an tracker, futures or options.

After the first working model were found, we needed to real life test trade. Since all signals were produced exclusively on the SP500 Index data, we executed them on the SPY tracker. Even though the SPY tracks the Index almost neatly, there are differences.

The devil is in the detail, some say – we agree too. And of course the odds are against us. The small differences were great enough to reduce a average 13% yearly return on a SP500 index data, to a around 8% yearly return on the SPY tracker data.

Note the difference in ‘Signals’ and ‘Execution Signals’. No matter what you trade, you need first a signal to execute a trade. So there is always a delay.

More modules followed, some originated from customer requests.

Then in 2006 we integrated all modules into one, and started a new marketing campaign. It went on to 2007 where again we started to raise money by selling the completed platform to the market.

As per now, we still keep working. We are not in the goal, but it is in reach. With this writing we like to inform the trader about the complex problematic of creating a real working trading machine – which really will work - and keep working in the future. Since many tools are around, only few (probably) work. The one which work (by nature) are kept secret.

To keep a working system secret, is understood different. There is of course the source code, the formulas etc. But also how you trade and apply the system to the market has to be done the right way. To avoid “insider effects”, you need to place your trades in matter, that over time it does not make rumors in your broker firm etc.

Keep in mind that all actions have re-action. No matter the size. Even if you have a working system, its still a task to run it.

If you have questions to this article, we are happy to answer.