Subject: Correlation Analysis
From: "salvucci@xxxxxxxx" <salvucci@xxxxxxxx>
Date: Tue, 3 Jun 1997 23:55:58 +0000
Comments: Authenticated sender is <salvucci@xxxxxxxxxxxx>
Reply-To: Michael Salvucci <salvucci@xxxxxxxx>
Resent-Date: Tue, 3 Jun 1997 23:20:00 -0600
I'm still a newbie on Metastock and have been reading up on Correlation Analysis. I really appreciate everyone's help in the past. I would like to get everyone's opinion that has used correlation. It seems to me that if you compare the correlation of an independent variable (an indicator) against a dependent variable (a stock's closing price) and receive a positive correlation, you should have a leading indicator. Therefore, if you have a leading indicator, you may have a system or part of a system.
In this simple system test:
you look to see if there's a positive (greater than 0.9) correlation between MACD() and the closing price, that MACD() is less than 0, and that MACD() is increasing. If all of these things agree, you enter long. Do the reverse to exit long.
The Metastock manual does not get in-depth in its discussion about the PERIODS and SHIFT. Let's say, I wanted to maximize my profits within 20 days. Should I set the SHIFT to 20? Does anyone know the effect the results have when you increase or decrease the period and shift? I've performed the above changes, but I'm not sure how to interpret the results when I start changing numbers. I don't want to just 'optimize' something I don't fully understand or 'optimize' something for the sake of optimizing.
Increasing the SHIFT seems to put alot of uncertainty into the correlation - it starts decreasing. This may be the result of the farther you look into the future, the more unlikely your prediction will be. Is this a correct interpretation? Should SHIFT always be set between, say, 1 and 3 ?
Increasing the PERIODS seems to smooth out the line (if plotted as an indicator), but what does it actually smooth? This is where I'm losing it. Does it mean that if PERIODS is set high (eg. 50 to smooth it over a long term) , the stock is following a long term upward trend, and the indicator is following an upward trend, then the correlation should be close to 1 because it is smoothing a result that coincides with the price increase?
Correlation, to me, seems to validate or invalidate an indicator and/or system test. Because of this, I 'think' it's a great tool to use in TA.
Something to think about: Only use indicators in a system test when they report a positive correlation to price. What do you think?
Your help is greatly appreciated!
Lawrenceburg, Tennessee USA
ICQ User Number 1035286