My Perspective On Algorithmic Trading

I replied as follows:

I am following the algorithmic trading talk and the Saxobank guy is dead wrong. The future is algorithmic trading, but right now people don’t seem to understand the issues.

I am a mechanical engineer with a specialty in Artificial intelligence, Automation, and control systems. When I completed my degree in 1992 we had the same argument when FEM (Finite Element Modeling) came out. People said you would not need engineers because the software would design for you. The reality is that engineers are needed more than ever. What has changed is the role of the engineer in that they have to be more creative. Engineers need to come up with ideas that push the envelope.

The same will happen with traders. I do predict the death of the traditional trader as per the Saxobank person was talking. Though it does not mean traders are dead. It means trader better adapt to the new environment and focus on honing their descriptive strategy skills. The traders that do that will survive the others will fail like the the engineer who draws on the drafting board with pencil and ruler.

Right now the problem, and I have fixed it with my system, though it helps I am a programmer is the ability to code and develop new strategies within minutes, hours, or days. Right now algorithmic trading systems get stuck with single strategies, and the new d’jour is going through the news. That’s not what I do…

The saxobank guy replied that I will still be caught and my system will have serious problems with pullback.

WRONG, WRONG, WRONG…

I have feeling that there is a comprehension gap regarding algorithmic trading. You can create a model, but you should never predict. I have come to the conclusion that traders can trade any market, and thus an algorithmic system must be able to trade any market, with any security. 

I think my system has achieved that, and I want to show the following trading results for my algorithmic trading system.

The spot Trained Trade is June 1, 2007, and I consider any calculations before that point as hind sight bias. In other words if I was to calculate my earnings in the hind sight I would be re-enforcing what I already know. What I want to know is how well my trading system will do in forward.

For the most part I have been doing my experimentation and tuning for the month of June since that is when I traded live and have a way to correlate my theoretical with the real market condition. And as you can see the trader did fine.

Though what surprised me, and I was hoping that my algorithmic system would do this is that it could trade through the rough times. At the point where I label Peak DOW the data and trading is pretty plain vanilla. It is after the Peak DOW where things get nuts. And it is here where I wondered if my algorithmic trading system would work.

If you look at the label Current DOW that was the DOW level on August 15, 2007, and you see that my algorithmic trading system is making a profit. Yes there was a big whoop pullback, but that was a bug. I kept the bug in the results because bug is a bug is a bug and I have to live with it. It’s fixed, but that would not change my profit levels. Though the system fights back and to being profitable, while the DOW has sunk like a rock.

What I am ecstatic about is that my trader can trade a situation that it did not know about. The drop in the DOW and the volatility was not in the training set, and hence by being profitable in that situation I can feel comfortable that I have developed a robust algorithmic trading system. I have developed a trader, and not a prediction program. The reason why the bug did not cause a further pullback is because I have built-in “stop trading” mechanisms just in case my code has a bug.

This is what algorithmic trading is about, at least this what I think algorithmic trading should be about. When am I taking my algorithmic trading system live (again)? October 1, 2007. June was fun and I am happy I had a trial run. But now I want to start trading some serious levels!