I think it highly doubtful that any time soon you'll be able to point at a security or at a watchlist and provide some highly generalized criteria like Max Trades Year less than X, Car greater than Y & MDD less than Z and push a button and have AB ( or anything else ) generate a system and get decent results without anything else ...

What I discussed briefly at the end of my presentation at the first conference that I was at was the additional capabilities that IO allows because of the fact that the number of variables that are being optimized has virtually no impact on run time. This allows optimization to be used for other things that you might not immediately think of ...

One of these things is what I referred to at the time as rule creation, selection and combination ...

In a simple example let's assume we have a few potential indicators we might want to use like a Stochastic, MACD, RSI and an ROC. Combining these in some way into an effective system while seemingly straight forward might take a very long time using hunt and peck and because of the limitations of exhaustive search normal optimization processes won't help us much.

I won't code the whole thing and this is off the top of my head so it may not be syntactically correct but the purpose here is to demonstrate the concept.

For the first component ( Stochastic ) we have the following basic parts ...

Len1 = Optimize("Len1", 1, 1, 100, 1);
Len2 = Optimize("Len2", 1, 1, 100, 1);
Len3 = Optimize("Len3", 1, 1, 100, 1);

RHHV = HHV(C, Len1);
RLLV = LLV(C, Len1);
StoX = IIf(RHHV == RLLV, 50, 100 * (C - RLLV) / (RHHV - RLLV));
StoEMA = EMA(StoX, Len2);
StoTrg = EMA(StoAMA, Len3);

StoBuy = StoEMA > StoTrg;
StoSell = StoEMA < StoTrg;

Let's assume that for the other components we have the appropriate code in place that result in:

MACDBuy, MACDSell ... RSIBuy, RSISell ... ROCBuy, ROCSell.

There are of course lots of ways the results of these individual buy & sell signals can be combined but for the purpose of this post lets assume that some number of selected components will be used to generate buy and sell signals and those may not be the same components on the buy side as on the sell side.

So ...

Lets add ...

UseStoBuy = Optimize("UseStoBuy", 1, 0, 1, 1);
UseStoSell = Optimize("UseStoSell", 1, 0, 1, 1);

... and we'll add other on/off (Use) flags for the other components ...

This will then allow us to write additional optimize statements and Buy / Sell statements like what are below ( I Used GTEQ to represent greater than or equal to in what's below becuase the forum s/w is sometimes "fussy" about how it treats things like this ...

BuyTot = Optimize("BuyTot", 1, 1, 4, 1);
SellTot = Optimize("SellTot", 1, 1, 4, 1);

Buy = UseStoBuy * StoBuy + UseMACDBuy * MACDBuy + UseRSIBuy * RSIBuy + UseROCBuy * ROCBuy GTEQ BuyTot;

Sell = UseStoSell * StoSell + UseMACDSell * MACDSell + UseRSISell * RSISell + UseROCSell * ROCSell GTEQ SellTot;

When optimized this combination of statements then will find the best combination of buy side and sell side components to use and how many of them need to be in place to generate respective buy and sell signals with the best results.

IO in this case allows a type of problem to be solved that can not really be solved any other way ...

In sample results, sensitivity and out of sample results ... A tale of two optimizations ...

Lets take the same system ( RuttrFT ) and optimize for RYOTX from 1/1/92 - 12/31/99 to get the best parameter values we can in terms of CAR / MDD and then see how the system performed from 1/1/2000 forward ... In the first optimization we won't take the sensitivity of those parameters into account at all, in the second we will use parameter sensitivity as a means to influence where in the solution space optimization looks to arrive at parameter values ...