First Attempt at Modeling Crime in Chicago, Part 3

I've been trying to keep my attempts to model crimes in Chicago relatively simple. I was hoping to be able to generate some decent predictions by examining a few months of crime data as a signal using a simple autoregressive moving average model (ARMA). And I've finally made some progress!

I found that a ARMA(2, 3) process acts as a reasonable descriptor for the first difference of the crime data between July 1st, 2011 and March 31st, 2012. The graphs above are simulations for April 2012 that were generated with my regressed ARMA(2, 3) parameters. My choices of using an ARMA model and fitting the first difference of the data is explained in a previous post. The Mathematica code that performs the fitting and diagnostic checks is quite extensive so I may release the work for this post as a package in a later post and document all of the functions then. But if you want to play around with my code and do not mind working with Mathematica, you can find everything used to create this post here (Warning! It is still a work in progress).