General Topics
Markets
Technical Topics
Brokerage Firms
Company Specific
Community Lounge
Site Support

Statistical Analysis Programs

Discussion in 'Strategy Building' started by Hello_Dollars, Mar 31, 2004.

1. Hello_Dollars

I'm developing several stats arb related strategies and need some off-the shelf software that will allow me to run price dispersion, time series and other analyses on various markets. I am not a programmer and do not desire to become one. Hence, ease of use, in addition to robustness, would be my main criteria. Also, the ability to painlessly download historical prices from one or more of the popular data vendors would obviously be a requirement.

I recall using Minitab when I took stats, but that was several years ago (I believe it was dos based at the time) and my memory of it is hazy at best. Would that suit my needs or are there better alternatives? Thanks.

2. Cheese

".. that will allow me to run price dispersion, time series and other analyses.." Hello Dollars

Would also be interested in software for above/related areas.

5. MrJohnGalt

MatLab...

But QuantStudio maybe another...

6. BKuerbs

Quite interesting to see someone else use Statistica. Do you use any elements of the Time Series module?

I use Statistica, without the NN, mostly for exploratory data analysis. Simple things, like distribution of ranges etc.

Regards

Bernd Kuerbs

7. theplumber

Spectrum (Fourier) and Cross-Spectrum Analysis.

At least their manual is not trustable as for statistical process control (spc): shewart - the statistician who invented this method - has never said what they pretend he had said : sure they didn't read him but just report from "modern" gurus who have completely forgotten the original epistemology !

From their manual what is said is not true :
"X-bar Charts For Non-Normal Data. The control limits for standard X-bar charts are constructed based on the assumption that the sample means are approximately normally distributed. Thus, the underlying individual observations do not have to be normally distributed, since, as the sample size increases, the distribution of the means will become approximately normal (i.e., see discussion of the central limit theorem in the Elementary Concepts; however, note that for R, SÂ¸ and S**2 charts, it is assumed that the individual observations are normally distributed). Shewhart (1931) in his original work experimented with various non-normal distributions for individual observations, and evaluated the resulting distributions of means for samples of size four. He concluded that, indeed, the standard normal distribution-based control limits for the means are appropriate, as long as the underlying distribution of observations are approximately normal. (See also Hoyer and Ellis, 1996, for an introduction and discussion of the distributional assumptions for quality control charting.)

However, as Ryan (1989) points out, when the distribution of observations is highly skewed and the sample sizes are small, then the resulting standard control limits may produce a large number of false alarms (increased alpha error rate), as well as a larger number of false negative ("process-is-in-control") readings (increased beta-error rate). You can compute control limits (as well as process capability indices) for X-bar charts based on so-called Johnson curves(Johnson, 1949), which allow to approximate the skewness and kurtosis for a large range of non-normal distributions (see also Fitting Distributions by Moments, in Process Analysis). These non- normal X-bar charts are useful when the distribution of means across the samples is clearly skewed, or otherwise non-normal. "

What Shewart said is in fact THE OPPOSITE : NON NORMAL LAW CANNOT GIVE TRUSTABLE RESULTS or to quote him exactly (I translate from french version book translated itself from english )

"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."

This is just an illustration of what I said about "modern" gurus: