xplorer
257 posts
msg #34012
- Ignore xplorer |
11/19/2004 8:08:32 PM
Backtesting is a necessity !
... but if it isn't done right it can skew reality.
So what is the most reliable and accurate way to back test our filters ?
Hopefully someone has the answer so we all can use a known "standard".
Should a simple line at the end of our filter work ?
OFFSET nn days (nn being an number to max)
like this :
(but the results are always biased to the current sentiment)
Or should every command end with the
LAG nn DAYS
(this won't give you the % gains ... you have to calculate it)
(but I think it probably is the only REAL way to do it !)
... there has to be a better way !
.. or a combo of both :
(yikes ... get me out of here !)
..or maybe the PERFOMANCE listed on the SF results ?
.. or another way ?
Any opinions ... or info would be appreciated !
THANKS !
...by the way ...
this filter gives fantastic results,
if you backtest the "CORRECT" way.
|
jpistell
123 posts
msg #34015
- Ignore jpistell |
11/20/2004 12:26:04 AM
Backtesting needs to withstand the 4 seasons of trading.
SPRING, SUMMER, FALL & WINTER.
Summer = bulls rally
Winter = Bears rally
Spring & Summer join the 2 phases together.
|
jclaffee
81 posts
msg #34016
- Ignore jclaffee |
11/20/2004 3:27:52 AM
xplorer writes
"Backtesting is a necessity!
. . .but if it isn't done right it can skew reality.
So what is the most reliable and accurate way to backtest our filters?"
IMHO, this question goes like an icepick to ones survival as a trader and the answer lies beyond the range of any of the alternative answers proposed.
Consider this:
On Monday, 11-1, my filter gives me an entry signal on an equity which closed the session at $10 a share. . .on Tuesday, 11-2, it closes up 1.00 to $11.00. . .on Wednesday, 11-3, it gains another 1.00 to $12.00. Two days, 20% gain. . .right? One great filter. . .right? Maybe. . .maybe not.
Going a little further, let's say that on the morning of Tuesday, 11-2, my issue either
1) gapped up 0.75 to open at 10.75 and traded above 10.60 all day so -- since my trading discipline includes "don't chase price higher" -- I never had an entry. (Perhaps I'd be insightful enough to conclude that if the "pros" supported this advance in the final hour an entry at the higher level would not be "chasing" -- perhaps not.)
or
2) opened at 9.80, giving me an entry, then traded lower to 9.00, triggering my initial stop, before registering the advance to 11.00.
So, I had the right issue at the right time and, in alternative 1), failed to get on board and, in alternative 2), took a loss. In either event, a "backtest" of my filter would say that I had a 20% gain in two sessions.
Again, IMHO, meaningful backtesting has to be done by following the chart of an issue day by day according to ones own trading style. Anything else is backtesting a unreal world.
Jim
|
xplorer
257 posts
msg #34020
- Ignore xplorer |
11/20/2004 9:30:40 AM
Here is an accurate ...is burdeonsome, convoluted, and restricted
to short term trades... unless you are willing to accept the gaps with
the "weeks" testing...if that is your trading personality.
This is way to get an accurate backtest.
Assuming I have a filter in which the results will
return stocks that I want to hold less than 6 days....
a quick return.
for 18 days ago:
...or for 118 days ago:
... or for the 8 weeks:
(though this screen wasn't designed for this type of "hold")
Your thoughts ?
Is there a better way ?
|
xplorer
257 posts
msg #34021
- Ignore xplorer |
11/20/2004 9:35:03 AM
Just to add clarity to the results ...
the "gain" columns are a ratio ...
less than 1.0,and you have a loss...i.e .94 is a 6% loss.
a 1.15 would be a 15% gain...
|
xplorer
257 posts
msg #34022
- Ignore xplorer |
11/20/2004 11:46:32 AM
Jim,
I agree with you ... backtesting is not a cut and dry science ... but within the limitation of SF, we need to have a way to evaluate our screens. I like the idea of this site do the number crunching... Of course there are anomalies, picks that will deviate from the norm ... noise is expected.
In the end, trading skills will seperate the true traders from the dreamers.
|
cegis
235 posts
msg #34165
- Ignore cegis |
11/28/2004 8:19:24 PM
My suggestion was posted in the thread "Technique: Calc your own returns - for any period" in the General Discussion forum. xplorer seems to be using some of that technique.
Basically, what I do now is use "days ago" (e.g., 10 days ago) to offset the conditions in the filter I'm backtesting, then using "future" days ago (e.g., 9 days ago, 8 days ago,...) to get (using "add column") open/high/low/close of the following days. I then download the results into a spreadsheet and analyze the results. There, I can "program" assumptions to most closely match the trading I'm contemplating with the filter, take commissions into account, and anything else I can think of.
HTH,
C
|
xplorer
257 posts
msg #34178
- Ignore xplorer |
11/29/2004 5:02:15 PM
Hey CEGIS ...
I took a look at you post you referenced .... and it is similar. I like the way you incorporated the HIGH and LOW within the last xx days ... nice ! Technical writing seems to be a gift of your ... nicely written.
|
cegis
235 posts
msg #34181
- Ignore cegis |
11/29/2004 7:51:58 PM
Gee, I blush! :-) Glad you liked it, and found it useful...
C
|