Archive for the ‘Prediction’ Category


Well here you go, a small C# class for Kalman filtering (https://sourceforge.net/projects/autospreader/files/C%23/KalmanFilter.cs/download)….it has a dependency on the GeneralMatrix (http://www.codeproject.com/KB/recipes/psdotnetmatrix.aspx) library but any matrix library would suffice….Enjoy and as always if you have questions ASK!

Advertisements

On our last post, there was a brief introduction to Eureqa.  In this post, I am to show you sample output from its findings for our trusty wheat data:

First outright = price level

Second outright = price level + 13.75

third outright = price level +25.75

fourth outright = price level*0.96+59.78

It took 5 minutes to find these relations in the package and here are the sample mis-pricings:

This is in no way a complete picture; it was only my aim to show you the type of pictures that the user may paint with this tool.  In the next Eureqa post, we will show how to use the software to create these numbers and make a more complete wheat market model.

Posted: April 16, 2010 in Forecasting, Prediction
Tags:

I was surfing around and found a new software package that is called Eureqa (http://ccsl.mae.cornell.edu/eureqa_download) from Cornell Computation Synthesis Laboratories.  It is used as a robotic scientist to discover physical laws in large data sets.

For example, I was bored today and ran it on a wheat dataset as can be seen in the screenshot below and found the following relationships in the wheat term structure:

Note the above relations are in regards to the last contract in the term structure which is something I often like to market make.  First note, it is organized by increasing error.  Notice that the first relation is that it found that the last wheat contract in the term structure is typically around 15 cents of the contract preceding it during this particular contract.

Checking that relationship yeilds:

Play around with it!  Ask any question that you may have as I am rapidly becoming an expert with it.


I have been working to create some files for predictive methods in finance.  The first of these will be a double exponential smoothing prediction.  Hopefully followed by Kalman’s approach and a few others.

The basic method may be found in this (http://www.ljmu.ac.uk/Images_Everyone/Jozef_1st.pdf) paper by Dunis.

Initial results (below) where as expected without optimizing the MSE of the set:

As always, the related file is in the SourceForge repository (https://sourceforge.net/projects/autospreader/files) under DESP.xls.  Optimize it if you are feeling ambitious and make sure I made no errors; threw it together in about 5 minutes for fun after reading the paper.