Automated Strategy Builders

[xandman]

I could use a primer on this. Currently evaluating StrategyQuant.

There are automated strategy builders that can build, test, and optimize code for as little of 2 months of Starbucks coffee.

Doesn't this imply the automated trading does not work?

On a more positive and productive note, Where does the system designer, me/you, add value to this work flow?

 

[bashatrader]

I could use a primer on this. Currently evaluating StrategyQuant.

Here is your primer. Interestingly so, written by someone who has developed a program like that.

Doesn't this imply the automated trading does not work?

No, it would be an unsound conclusion.

 

[xandman]

Thanks, Bash. Seems like I have a long way to go. I just keep generating more questions.

Attached is some Pseudo code from the Break-Out family of strategies. You are simply a right-click away from generating Metatrader or Ninjascript code. It has some way to go. Sadly, it trades way better than me.

83% annualized return on a 36% Max DD. Let's not pooh-pooh the exotic indicators, yet. The program threw it into the pot and made it work.


== Entry conditions
====================================================================
LongEntryCondition = (LinearRegression(5) > Lowest(45))
ShortEntryCondition = (LinearRegression(5) < Highest(45))


====================================================================
== Entry orders
====================================================================
-- Long entry
if LongEntryCondition is true {
if No position is open then Buy at HeikenAshiOpen(16) + (-0.3 * ABS(SMA(57) - Ichimoku(9, 26, 52, Senkou Span A))) Stop;
Stop/Limit order expires after 22 bars.

Stop Loss = (1.67 * ATR(46)) pips;

// Stop trailing (on close)
Move Stop to (KeltnerChannel_Middle(68, 2.2) + (0) * BBWidthRatio(29, 2.0))) on bar close;
}

-- Short entry
if ShortEntryCondition is true {
if No position is open then Sell at HeikenAshiOpen(16) + (0.3 * ABS(SMA(57) - Ichimoku(9, 26, 52, Senkou Span B))) Stop;
Stop/Limit order expires after 22 bars.

Stop Loss = (1.67 * ATR(46)) pips;

// Stop trailing (on close)
Move Stop to (KeltnerChannel_Middle(68, 2.2) + (0) * BBWidthRatio(29, 2.0))) on bar close;
}

 

[xandman]

Restarting. I am too caught in a chicken and egg game. For the ranking, filtering and sorting criteria for the system that I want before genetic optimization and Monte Carlo testing. I guess, if you told your 5 year old to pick a system on one criteria it would be out-of-sample Sharpe ratio or expectancy that's stable.

I will out in the field for the week. Will catch up on some reading.

 

[xandman]

Sharpe is more appropriate for portfolios. Variants of DD/Return are best for trading single instruments.

Interesting observation after loading 15m SP500 data:

A fortunate limitation, the SP500 contract size was normalized to trade just like a forex contract (in pips). So the optimizations were done purely on the behavior of the price series.

The strategy builder was able to build strategies that had significantly ( 1/20) the drawdown than forex. I did check for Long/short biases in the strategy and these top performing strategies were equally adept at trading the short side.

Although the strategies are probably curve fitted, I think it makes a case that it is significantly easier to be consistently profitable trading stocks than forex when trading with a 15m break out method.

That's all I will be saying about my explorations. I leave journaling to the people with more domain knowledge in this area.

 

[ronblack]

The strategy builder was able to build strategies that had significantly ( 1/20) the drawdown than forex.

This is because of wrong position sizing. in forex. You cannot size positions in forex the same way you do with stocks although in general forex is riskier due to the widening of the spread during fast markets.

That pseudo code is the king of curve-fitting.

 

[eurusdzn]

"Although the strategies are probably curve fitted, I think it makes a case that it is significantly easier to be consistently profitable trading stocks than forex when trading with a 15m break out method."

That alone may be worth the two months of Starbucks money.
On a side note, if the required fit for performance is the trend , it seems with these auto systems that you could design a "shutoff" when the trend changes.
So then, why does it matter if the data is fit?

Now, i know that is an over simplification of strategy design . I am new at that.
But, as Daal said in the Global Macro thread: it is the large fund managers,
long only, pressing there bets and driving this market.
Any algo can trade ths market.