Programs that find systems

(login for full post details)

I recently discovered that there are some programs that find systems and also write code. I thought the idea was excellent and I downloaded one called Adaptrade. This program runs fine and crashes only occasionally but I wasted two free weeks of testing trying every possible combination of metrics, settings, etc. with no results. The system code would reveal random systems, most with complicated exits based on triangular MAs or functions of ATR that obviously attempted to fit price series. Most, if not 100% of the trading systems involved very simple entries and that alone made me skeptical about this product. This is example code for ES daily futures for a system that had best performance:

Code

{ Strategy inputs }
Inputs: NL1 (97),
        NL2 (43),
        NS1 (5),
        TargFrL (2.0242),
        MMStopSz (1490.00),
        NATRTargS (54),
        TargFrS (1.2240),
        PSParam (1.00),
        RoundPS (true),
        RoundTo (1),
        SizeLimit (100);

{ Variables for average true range for entry and exit orders }
Var:    ATRTargS (0);

{ Variables for entry and exit prices }
Var:    TargPrL  (0),
        TargPrS  (0);

{ Variables for entry and exit conditions }
Var:    EntCondL  (false),
        EntCondS  (false);

{ Variables for position sizing }
Var:    NShares  (0);

{ Average true range }
ATRTargS = AvgTrueRange(NATRTargS);

{ Entry and exit conditions }
EntCondL = Volume crosses above Average(Average(Volume, NL1), NL2);
EntCondS = DayOfWeek(date) = NS1;

{ Position sizing calculations }
NShares = PSParam;

If RoundPS and RoundTo > 0 then
   NShares = IntPortion(NShares/RoundTo) * RoundTo;

NShares = MinList(NShares, SizeLimit);

{ Entry orders }
If MarketPosition = 0 and EntCondL then begin
   Buy("EnMark-L") NShares shares next bar at market;
end;

If MarketPosition = 0 and EntCondS then begin
   Sell short("EnMark-S") NShares shares next bar at market;
end;

{ Exit orders, long trades }
If MarketPosition > 0 then begin

   TargPrL = EntryPrice + TargFrL * TrueRange;
   Sell("ExTarg-L") next bar at TargPrL limit;
end;

{ Exit orders, short trades }
If MarketPosition < 0 then begin

   TargPrS = EntryPrice - TargFrS * ATRTargS;
   Buy to cover("ExTarg-S") next bar at TargPrS limit;
end;

SetStopShare;
SetStopLoss(MMStopSz);

Does this make sense to you? Have you tried any other similar products? Is there any value in using them?

(login for full post details)

I just downloaded Multicharts and Price Action Lab. The latter finds patterns and writes EL code you can use in Multicharts. I have tried it before but I did not have a lot of time to play with it then. I think patterns of this kind have more chances than indicators and data mining bias is decreased a lot since all patterns depend on OHLC only. I think genetic programming is a fancy term for curve-fitting. At least this is the purpose we used it for in college. I have heard that most hedge funds have given up on neural networks and genetic programming and now are using other methods for finding edges. Someone I know who worked for an investment bank before told me that genetic programming was a disaster for them. Maybe some fuzzy logic expert systems are better for this task.

(login for full post details)

Interesting to hear that example of hedge funds dropping neural networks and genetic algorithms. Makes me wonder what those in charge of those MFGlobal and JPMorgan overleveraged failed bets based their trades/hedges on. Now all they need to do is to simulate a trader with panic monkey mode and overleveraging and averaging in until margin called haha.

I guess one could make another ol'/new system by using ATR and combining conditions with one or a few of hundreds of varying oscillators, indicators, or MA crosses. Could you backtest the systems your strategy generator came up with dryg?

(login for full post details)

After spending many days and nights testing Adaptrade I was not able to find a single system that after testing profitable in the out of sample, continued to be profitable in paper trading mode. What a waste of time. That software is just a curve-fitting optimizer.

(login for full post details)

Cloudy

Interesting to hear that example of hedge funds dropping neural networks and genetic algorithms. Makes me wonder what those in charge of those MFGlobal and JPMorgan overleveraged failed bets based their trades/hedges on. Now all they need to do is to simulate a trader with panic monkey mode and overleveraging and averaging in until margin called haha.

I guess one could make another ol'/new system by using ATR and combining conditions with one or a few of hundreds of varying oscillators, indicators, or MA crosses. Could you backtest the systems your strategy generator came up with dryg?

I gave up on genetic programming. The results look like permutations of some kind. I am working now towards developing something along the lines of Price Action Lab. Just OHLC patterns with a a couple more indicators.

Programs that find systems

Discussion in Trading Reviews and Vendors

Programs that find systems

NinjaTrader Indicator Challenge!