Universal and Backtesting At The Tick Level

[swansjr 10]

As I stated in another post I would like to make a series of posts over time highlighting my experience with the Universal Clone automated system. This is the first post and it deals with backtesting.

 

Below is a 5-minute ES chart with the first date at the beginning of 2009 and ending on 4/24/2009.

 

http://www.traderslaboratory.com/forums/attachment.php?attachmentid=10325&stc=1&d=1240781966

 

It’s a great looking curve and the settings are as follows:

 

StartTime 845

EndTime 1500

CloseTime 1505

Cntracts 2

Target 1 Cntracts 1

Initial Stop Loss $150

B1 20

S1 11

B2 7

S2 13

Max Trades 1000

Day Loss $600

 

(all other inputs are not used)

 

Now, slippage and commissions are not included. But that’s not what really worries me. What does worry me is how do you back test this? Well, in TradeStation (TS) is you can set how a bar (in this case a 5-minute bar) is generated at the tick level. For example, I can back test this at the 50-tick level (and that’s what this chart is). Presumably TS then calculates the strategy appropriately down to the 50-tick level as each bar is generated. You can select a 1-tick level for even more precise results, right!

 

Well, not necessarily. According to Clayburg TS does not have this tick level backtesting down to a science. Thus, it’s not accurate. He has demonstrated that tick level backtesting is only accurate for about two weeks of history. After that, the results are not reliable.

 

Thus, the only real way to demonstrate that a given strategy is effective is to let it run on live market data. That is, you don’t trade real money but leave that chart active during normal market hours and the system will generated signals and process those signals in real-time as the data streams in from your internet connection. At the end of the day note the progress of the strategy. Repeat for a week or two and then evaluate the equity curve and system report. Does it hold up? Yes, then maybe consider trading real money. If not, back to changing the parameters.

 

So, testing can be a time consuming process.

[swansjr 10]

Here is another chart with the same ES 5-minute settings but the backtesting was performed down to the 10-tick level. Notice the huge difference in the equity curve from my pervious post. The feature I’m using in TS is the Look-Inside-Bar-Back-Testing.

 

 

And one more with this look-back feature disabled.

 

 

Again, backtesting is not a 100% science. Just because you have a great looking equity curve does not mean you have a tradable system. These curves can be altered by a simple parameter being changed within TS.