|
PureBytes Links
Trading Reference Links
|
At 08:57 PM 8/23/2007, Simon Dawson wrote:
>Of course, being on a TS list, using TS, there's nothing you can do about it. I mention it in passing, if you're wondering why systems that are faster than your data might be misbehaving.
>
>Short version: Don't trust TS to accurately estimate sub-bar signals.
TradeStation has long had back-testing "within the bar" to test your strategy to any resolution you would like such as every minute, 10 minutes, etc., of a daily bar, for example.
>I'm looking at thousands of trades on many markets over 20+ years so these errors really jump out - and in some markets result in ridiculously significant differences.
TradeStation now has 64 bit floating point precision so this problem is gone.
I am always amused when I hear people talk about minute errors in back-testing. The only really important error you need to worry about is that your future data will not be at all like your past data. All other errors are trivial by comparison.
>To recap:
>
>With MaxBarsBack 50, on bar 50, Average(Close,15) calculates, but Average(Close,15)[anything] doesn't - even though there IS data there that could be used.
>
>So, if you want accurate signals, but you're referring to, say, 5 bars ago, you actually need to wait until MaxBarsBack+5 before you can trust anything that comes out of TS.
What TradeStation does with initialization is well specified. For most cases it works well automatically and does not require the user to do anything. You can certainly find cased where it isn't perfect and fix it with a line or two of code, if you care.
I think this is a lot better than lots of other platforms that force you to write all kinds of separate initialization code yourself.
>The XAverage settling is a completely separate nightmare, in this regard :)
This is a totally ridiculous statement. There are lots of simple ways to initialize the one-line XAverage code to whatever value you would like. If you do not like the way the built-in function does it, then write your own version. It will take you less time than you spent writing about it...
>oh yes, this is completely expected & reasonable. The point I raised above was simply in the case of "odd data values" in back adjusted data.
That has nothing to do with TradeStation. It is a function of how you choose to back-adjust data.
Bob Fulks
|