Capital Market Investments are known as being risky process,
requiring an adequate Risk Management (see Financial Risk Management , Volatility vs. Risk, HFT Risk ). Trading Systems are not
out of this group, but trading systems have a particular element: It´s a
systematic approach. Systematic doesn´t means profitable, however means
tractable (see Financial Automation and Control: a new age).
When I say tractable, I’m referring to the mathematical model,
where all the variables and parameters are disposal, ensure tracking.
1 - properties of the asset traded (mean, standart
deviation, kurtosis, skewness or any kind of mathematical model output, like
any technical indicator, filters, wavelets, etc)
2 – trading system intern or output variable.
Owning various models and knowing that we have
to manage them looking maximizing profit and minimizing risk, the automation
process is monitoring all the systems output performances indicators (profit,
return, growing factor, stability factor, sharp ratio, etc) and choosing the
systems and the managed capital of each one, habilitating the systems that we
believe that will perform based on historical comparison of the indicators with
the historical profit result.
One simple choice is pick the systems that are performing
well right now, but the performance of the models can depend on specific market
conditions, like volatility.
The backtesting of this group of systems need more attention,
mainly because of the historical data. The group of systems could have
different optimization windows and it could present some degree of overfitting.
The model optimization need a validation data, as well as any other model
design process.
All of this control system is one level above of complexity
of a single model, presenting a diverse logic possibility of systems choice. The
models can be managed based on their strategy groups, isolation trend models
from pair trading ones, scalping, etc.
The more number of trading systems, more diverse and fewer
risk, like any other investment. Sometimes it´s more useful work on diverse kinds
of models than on the best one. Models that are not so complex and modern could
present high profit, high loses, being a poor model alone, but could be good
models for this kind of approach.
