Controlling Complexity in Trading Systems.

The development of trading systems can achieve high degrees of complexity. The process of simulating a particular strategy, termed backtesting can be triggered in several stages of optimization, and monitored at various levels of automation.

Developers always seek out new strategies or logical combinations. A system composed of a set of modules, which are interconnected, could trigger many situations.

The algorithmic complexity that can arise from this process is very high and if not taken care of organizing the evolution of codes, it is possible that errors can be observed as recorded in Knight Capital.

Modules that, being in great quantity and interconnected, go through changes at different times, and, may generate imbalances in the flow of information.

It's not it, but we say it is, to see, how it would be, if it was!

I heard this definition of simulation when I was an undergraduate student. Because it was funny and intuitive, I found it very interesting to use this term to define simulation.

The great utility and insight behind this definition is the ability to simulate what I wish.

That is, the author of the phrase is not worried about obeying any rules or fixed protocols. Rather, he is concerned to investigate and be flexible.

BackTesting, the good guy or the bad guy?

On trading system developing, Backtesting is the process of testing a strategy on a financial historical data. However a backtest don´t represent necessarily the real performance of the system. It could easily made a overfitting, changing the parameters of the system, optimizing the system, but not realy outperforming because the parameters are overtrained to precisely this situation, with a high performance imprecision. The process of the system design needs a separate historical data for the data validation process.

The process of Optimization, the quality of the data, the indicators accuracy, for example, are details that must be looked inside. A system response, observed on the backtest graphic can present a very high sensitivity, changing drastically the performance with few parameters changes, indicating an imperfect optimization.