how do u determine if something is statistically significant?
- Thread starter Gordon Gekko
- Start date
A sample size of 30 (minimum) is the statistically significant magic number. Seems low, but it's a fact. However, you can't go wrong by increasing the sample size to whatever you like.
Elvis
30 observations should result in a normal distribution........
statistical significance is a different matter.
consult any intro to stat book for relevant tests .
Elvis
30 observations should result in a normal distribution........
statistical significance is a different matter.
consult any intro to stat book for relevant tests .
it is not curve-fit when you don't "predict" the past but the future 
. This requires that you know the true form of the law, this is the case in physics. If you don't know it then you must be much more cautious see "Normal distributions are not the norm." http://www.elitetrader.com/vb/showthread.php?s=&threadid=25943&perpage=6&pagenumber=2
P.S.1: remember student law requires NORMAL LAW assumption also. Student law is used when the underlying law is normal but the variance is unknown it just gives a more flat distribution than normal law due to ignorance of variance but it doesn't let you forget about normal law assumption. Also the confidence interval is not the (fixed) tolerance interval: it's about giving an interval that contains the mean not an interval around the true mean.
P.S.2: that's the reason it is much more confortable to have a fundamental model like the one I have because with the knowledge of the true law of market I can use statistical infence more confortably
.
. This requires that you know the true form of the law, this is the case in physics. If you don't know it then you must be much more cautious see "Normal distributions are not the norm." http://www.elitetrader.com/vb/showthread.php?s=&threadid=25943&perpage=6&pagenumber=2P.S.1: remember student law requires NORMAL LAW assumption also. Student law is used when the underlying law is normal but the variance is unknown it just gives a more flat distribution than normal law due to ignorance of variance but it doesn't let you forget about normal law assumption. Also the confidence interval is not the (fixed) tolerance interval: it's about giving an interval that contains the mean not an interval around the true mean.
P.S.2: that's the reason it is much more confortable to have a fundamental model like the one I have because with the knowledge of the true law of market I can use statistical infence more confortably
.About Student law this is exactly what Shewart said (his book on statistical process control is originally in english but I translate badly from french version ) :
"Theorically it is possible to find a solution to make prediction for any kind of [statistical] universe with the statistical theory of distribution. But ... when the distribution is known but abnormal, the mean, the standard deviation and the size of the sample cannot make prediction, in particular prediction of Student type interval and tolerance interval with the same degree of validity than when the distribution is normal."
When the functional form of the Universe is unknown - which should be usual case in market - he explains that of course the picture worsens.
) :"Theorically it is possible to find a solution to make prediction for any kind of [statistical] universe with the statistical theory of distribution. But ... when the distribution is known but abnormal, the mean, the standard deviation and the size of the sample cannot make prediction, in particular prediction of Student type interval and tolerance interval with the same degree of validity than when the distribution is normal."
When the functional form of the Universe is unknown - which should be usual case in market - he explains that of course the picture worsens.
Quote from harrytrader:
it is not curve-fit when you don't "predict" the past but the future
P.S.1: remember student law requires NORMAL LAW assumption also. Student law is used when the underlying law is normal but the variance is unknown it just gives a more flat distribution than normal law due to ignorance of variance but it doesn't let you forget about normal law assumption. Also the confidence interval is not the (fixed) tolerance interval: it's about giving an interval that contains the mean not an interval around the true mean.
P.S.2: that's the reason it is much more confortable to have a fundamental model like the one I have because with the knowledge of the true law of market I can use statistical infence more confortably
If you overestimate the capability of your system you will undergo the Taguchi Loss function which is ^2 see for comment answer to thread
Regret: A Powerful Emotion You Must Face :
http://www.elitetrader.com/vb/showthread.php?s=&postid=394133#post394133
Regret: A Powerful Emotion You Must Face :
http://www.elitetrader.com/vb/showthread.php?s=&postid=394133#post394133
Nice to see some stat talk... "Taguchi Loss Function", it's been years since I've heard that mentioned... sometimes I miss stats (I used to be a quality engineer/statistician for Ford Motor Co. etc years ago)..
fwiw, I look at the market in terms of the inverse of our goal in a production environment, where we look at keeping processes in control, improving mfg process capability (Cpk) and reducing variation..
in the markets, I'm looking for a consolidation that then shows a process of "going out of control", eg buyers or sellers clearly taking over in the time/sales and 1-minute candlestick chart, moving the price outside of a normal curve for a breakout/breakdown..
having the bkgrd of working in production data processes for years, it's helped, since the charts and data patterns that we used to help improve mfg process control, can be used w/modifications for loss of control breakout points.
I used to publish on QFD and other quality deployment tools, here's an article mention from one of my ASQ articles back in 1992 (see the first article link under "Articles"):
http://qfdcapture.com/references.htm
There's plenty of statistical methods application to stock chart trading and the data points, I'm thankful for all the great corporate training + experience opportunities I was given to understand data patterns early in my career.
ken
fwiw, I look at the market in terms of the inverse of our goal in a production environment, where we look at keeping processes in control, improving mfg process capability (Cpk) and reducing variation..
in the markets, I'm looking for a consolidation that then shows a process of "going out of control", eg buyers or sellers clearly taking over in the time/sales and 1-minute candlestick chart, moving the price outside of a normal curve for a breakout/breakdown..
having the bkgrd of working in production data processes for years, it's helped, since the charts and data patterns that we used to help improve mfg process control, can be used w/modifications for loss of control breakout points.
I used to publish on QFD and other quality deployment tools, here's an article mention from one of my ASQ articles back in 1992 (see the first article link under "Articles"):
http://qfdcapture.com/references.htm
There's plenty of statistical methods application to stock chart trading and the data points, I'm thankful for all the great corporate training + experience opportunities I was given to understand data patterns early in my career.
ken
-------------------
Professional money managers measure by the month;
and years - plural.
Would rather err on too much stat sign than too little;
would however note differences in uptrend & bear trend which we are not in now.
would however note differences in uptrend & bear trend which we are not in now.