Search | User Listing Vidsite | Calendars | Quotes
Home Page ->  OmniTrader HowTo's -> BackTesting Methods -> View ThreadLogon (or Register, or Join TradeTight)

You are logged in as a limited-access Guest.To join TradeTight, first read the info in the Organization & Content room, then click the link above. 

Psalm 37:4 (NAS) ... Delight yourself in the Lord; and He will give you the desires of your heart.


Sticky Years to Use for EOD Backtesting
Jump to page : 1
Now viewing page 1 [50 msgs/pg]
Jump to room :
JimDean
Posted 8/9/2011 3:43 PM (#2756)
Subject: Years to Use for EOD Backtesting



Owner/Admin

Posts: 3925
2000100050010010010010025
Location:
USA: GA, Lawrenceville
"How far back should testing cover in history" ...

IGNORING the strawman-issue of "survivorship" ... I've attached a chart for everyone to think about. It shows the DJI over hte past twenty years, in LOG scale. There are a bunch of eyeball-best-fit lines, and a red square drawn on it.

The point I am trying to illustrate is the "character of the market" over that time period. Here are my simple observations from that chart:

1. A long-extended moderate bullish run (starting prior to 1990), shown by the dashed violet line, ended somewhere around winter of 1995.

2. A shorter and steeper bullish run (shown by the dashed orange line) picked up in early 1995 and ended approximately in the winter of 1999.

3. There is a "trading range" (from a very long timeframe point of view, shown by the red box) that started in the beginning of 1996 during the steep run, and continues through today.

4. Within the period of 1996-2010, we see MANY DIFFERENT market conditions. I've drawn in short solid line segments where blue is flattish, red is bearish, and green is bullish ... and the lighter shades are the more extreme examples of those three cases.

From my point of view, again ... as an "observation" ... that period of time represents a PERFECT representative backtest study, for the purpose of establishing good parameter-boundaries (using SW), and also covering many different market-conditions in roughly equal proportions of time.

NOW, from my point of view as a GUESS (not an observation) ... my "reading" of the general state of things is that the market will likely be CONTINUING in the red-box kind of trading-range-mix of activity for VERY MANY YEARS to come. Y'all may agree or may not ... that's to be expected.

But there is one other factor. The symbols that we work with don't all cover the same number of bars ... of course they all are present today, but as you go back in time, the total count gradually drops off. So, if you want the statistics to "equally weight" the earlier phases of the market, represented by the orange and blue and red lines that are pre-winter-2002, you need to take advantage of the fact that the orange and blue lines are "long enough" that if we cover a bit more history of orange, the smaller number of symbols available during that time will get representatively enough extra trades during that market condition, so that the relative contribution, statistically, of the orange and blue matches that of the later years in which the line-segments are shorter, but the number of symbols covering them is larger.

Whew - that was a long sentence. What I'm getting at is that:

1. if you restrict the testing to the violet-boxed area from winter-2002 on, then we MISS OUT on including the meaningfully-different market conditions that occurred before that time.

2. in order to properly and equally "weight" the earliest market conditions, we need to account for the reduced number of symbols (and therefore trades) during that time, by incorporating more of that market-type bars in the range. Fortunately, the long steady runs are early in our window, so this is possible.

My Conclusion:
For long tests, use 1996-present
For medium tests, use 2003-present
For short tests, use 2007-present


(Twenty year view.png)



Attachments
Attachments Twenty year view.png (33KB - 7 downloads)
Top of the page Bottom of the page
JimDean
Posted 3/22/2013 3:55 PM (#4413 - in reply to #2756)
Subject: Years to Use for EOD Backtesting



Owner/Admin

Posts: 3925
2000100050010010010010025
Location:
USA: GA, Lawrenceville
Here is an update to the original post, regarding useful backtesting ranges

a. to get a good sampling of all market types, use 2000 thru 20012
b. med period that focuses on mainly-bullish would be 2003 thru 2006
c. med period that focuses on mainly-bearish would be mid-2007 thru mid-2009
d. med period that focuses on very volatile bullish would be 2010 thru 2012
e. med period that focuses on very volatile bearish would be 2000 thru 2002
f. med period that focuses on flat-performance would be late 2003 thru late 2006

the chart illustrates this:


(Market Zones.png)



Attachments
Attachments Market Zones.png (52KB - 5 downloads)
Top of the page Bottom of the page
Jump to page : 1
Now viewing page 1 [50 msgs/pg]
( E-mail a Link | Printer Version | Search Room )

Owner of site: Jim Dean -- Forum content is confidential, and may not be distributed without written permission.