CHEMCEPT LIMITED
Mathematical Modelling, Chemical Engineering Software and Engineering Consultancy
Crays Pond, Reading, England
Mathematical Modelling, Chemical Engineering Software and Engineering Consultancy
Software Products
Finance
Over ten years ago, we were approached by a leading Day-Trader and requested to
develop a system to predict the maximum and minimum prices of stocks within one
day of trading. Initially, the request was to build on the Finsight software
developed jointly by the Financial Times Group and Intera Information
Technologies Limited. The system had been developed from a classical approach
to handling Chaotic Systems (State Space Embedding). We negotiated rights to
the system, but it proved impossible to access. Accordingly, we developed a
similar system based on open literature publications. However, the resulting
system did not perform better than the simple yardstick of the opening price
+/- half the average daily price movement. We built a marginally better system
based on a statistical tool that we had developed for the process industries.
The objective of the tool was to assess the impact of random variations in raw
material specification on product quality. The resulting system showed promise
in that the predictions were better than random. However, we judged that they
would be of only marginal value to day traders. We went back to first
principles to study the actual statistics that describe stock-market price
movements. We found that they are not described by any of the standard
statistical distributions (Gaussian, Log-Normal, Negative Exponential, Weibull
etc). Accordingly, it is not surprising that standard statistical tools applied to stock market movement give little better than random predictions. Instead of these standard statistical distributions, we find that they the movements are correlated with a statistical distribution that has not previously been described in the literature. To date, we have found that every one of hundreds
of price series that we have studied are described by the same 2-parameter
distribution function. We have processed millions of stock market statistics
over the years to draw our conclusions. We find that, contrary to the usual
perception, prices do NOT move as a random walk.
For nearly ten years we offered a software tool "TomorrowsPrices" that successfully gave useful predictions of stock market price movements. However, stock market price movement is not one of our core interests. Accordingly, the software has been withdrawn. It is our intention to publish the correlations and relationships in due course. In the meantime, if any other party is interested in the technology, we could discuss our findings.
|
|