|
PureBytes Links
Trading Reference Links
|
Al,
Great observation! Data reliability is a big headache for all traders. Most
proprietary trading desks keep a small army just to maintain the data
integrity of their database. Unfortunately, we don't have that luxury but
that's still no excuse not to stay vigilant about data reliability. Part of
your backoffice work should include a way to flag spurious data for manual
verification. This is especially important if you are trading mechanically
on tick data. All this creates yet more work for the trader to concern
himself with, but no one said this was an easy job ;-).
That said, your EOD data provider really should default to reporting the
close from the stock's primary exchange. You want data from the most liquid
market afterall. The obvious solutions to the risk of dirty data are data
collection solutions like the Downloader's that can test for possible errors
combined with trading systems that are less sensitive to data errors. Time
limitations will thus dictate your trading universe be limited to a
manageable number of securities that can be regularly verified manually.
Nevertheless bad data will still get through, and if you are daytrading on
tick data where accuracy is always suspect you need to have an action plan
that takes this into account so you can quickly take corrective measures to
exit/enter a missed trade. The upside is that you'll be forced to use a
pretty simple and brutish system which usually ends up being more robust and
desirable anyway.
System testing basically boils down to understanding tool limitations and
balancing them with your trading edge and objectives, account size, and how
you value your time. I realize that even trading over the net, I'll never
have a speed edge over a floor trader or market maker and so never enter
into trades with expected time frames of less than a day. Being halfway
around the world, I'm not inclined to stay glued to a quote screen either
unless I'm really bored. The US market opens for me at 10:30pm JST so I need
"fire & forget" systems. For example, my most profitable mechanical daytrade
system for the SPY has been to simply buy Monday at 10am EST with a "sell on
the close order" or when my money management stop-loss is hit. Data is easy
to verify and test and no indicators to mess with. The fact it runs counter
to the general market perception of Mondays being down days is ironic and
perhaps why it works so well for me.
I think most trading systems fail because people don't realize they are only
trading their beliefs; Beliefs in trading systems usually proven only with a
few well chosen examples. Beliefs that in manipulating price data with all
kinds of hokey indicators they are somehow getting an edge over what is
truly unpredictable. Beliefs in looking for rationality in the market when
there is none, e.g. E-Wave theory. Beliefs that the markets are random under
the Bell-curve (Doo doo happens, and it happens more often than the
statistics predict! The LTCM fiasco a prime example.) Beliefs that data
feeds are reliable. Beliefs that a bar chart provides all the available
market information for decision-making (What's a daily price bar afterall,
other than a filter of the days trading activity? It doesn't show you
activity at a particular price or what time during the day it traded at that
price or even the external factors, like who's in the market or a surprise
rate cut by the Fed.) All these beliefs come together in our trading ideas
and even when faced with overwhelming evidence to the contrary, are hard to
shake. Combined with the fact even most profitable systems lose more times
than they win and you have the perfect recipe for system failure. And yes, I
realize the irony that all this is just MY belief!
cheers,
Rick
----- Original Message -----
From: Al Taglavore <altag@xxxxxxxxxx>
To: MetaStock List Group <metastock@xxxxxxxxxxxxx>
Sent: Tuesday, April 20, 1999 5:57 AM
Subject: Back testing values
| I have read many posts about backtesting methods, results, etc. Also, the
| values and curses of optimizing. The question I now pose is how valuable
| these methods are, and perhaps a reason why some systems do not work in
the
| real world.
|
| Now, the question I ask is how accurate can the various backtests be if
you
| can have 2 13/16 difference in the closing price because of a regional
| exchange last price (I have no way of knowing the volume of this last
| trade.) It would seem to me that these type of occurances would skew any
| results. I will add that today, SLB opened at 64 1/2 which will show a
| large gap up when one reads the charts, but actually, based on NYSE
results
| the open was 1/2 above Friday's close.
|
| How do other traders/investors handle this situation or does anyone factor
| these occurances into their formulas?
|