Overview: The Brand New approach to Trading System Design
The common approach to the trading system design
is the following routine:
Begin (*)
Choose the parameters:
- Asset.
- Time Frame.
- Concept (trend following, patterns etc.).
- Formalization via indicators.
- Feasible sets of indicator’s parameters.
- Signal generation rules.
- Order types.
- Look through parameters in sample data.
- Estimation of parameters according to a
criterion (Net Profit, Profit Factor etc.) to
choose
the best parameters’ set.
- If the results of item 9 are satisfactory
load out-of-sample data and testing the trading
system for the best parameters’ set.
- If the result of item 10 differs from the
result of item 9 within acceptable bounds then go
out
else go to (*).
End.
It is possible to include the additional
parameters:
- Position size calculation.
- Asset share calculation, or Portfolio
optimization.
In fact all steps are led by the random search
except for item 8 where parameters’ values are
defined by enumeration of all possibilities. At
the same time sorting is led by one of the
TradeStation performance criteria (Profit Factor
for example). The cycle stops if the satisfactory
result is got. Using the algorithm man-hours are
tremendous. The resulting trading system is
hardly
optimal.
Following the principle “human should thinking,
machine should working” we suggest replace the
above routine to the following:
- Choose the parameters:
- Admitted asset candidates.
- Possible time frames.
- Admitted concept set.
- possible indicator set by which admitted
concepts can be formalized.
- Feasible sets of indicator’s parameters.
- Elementary rule set from which any signal can
be generated.
- All possible orders with its parameters.
- Assign optimization criterion of any
complexity and with any constraints.
- Set Genetic Algorithm parameters.
- Set out-of-sample data interval.
- Run TS GO Genetic Optimizer.
For a few minutes we shall get a set of optimal
trading systems, according to our fitness
criterion, tested in out-of-sample data.
All we need is to choose reasonable parameters’
bounds and meaningful optimization criterion.
|
