[Numpy-discussion] Financial TS models
Sat Sep 18 16:45:35 CDT 2010
On Sat, Sep 18, 2010 at 10:33 AM, <email@example.com> wrote:
> On Sat, Sep 18, 2010 at 10:13 AM, <firstname.lastname@example.org> wrote:
>> On Sat, Sep 18, 2010 at 8:09 AM, Virgil Stokes <email@example.com> wrote:
>>> I am considering the development of an all Python package (with numpy and
>>> matplotlib) for the modeling and analysis of financial time series.
>>> This is a rather ambitious and could take sometime before I can have something
>>> that is useful. Thus, any suggestions, pointers, etc. to related work would be
> I should have just asked you: What do you have in mind?
> instead of writing my notes from the readings I just did into the email.
> I think any open source contributions in this area will be very useful
> with the increased popularity of python in Finance.
>> Depends on what you want to do, but I would join or build on top of an
>> existing package.
>> I just got distracted with an extended internet search after finding
>> (They use Redis as an in-memory and persistent storage. After reading
>> up a bit, I think this might be useful if you have a web front end
>> http://github.com/lsbardel/jflow in mind, but maybe not as good as
>> hdf5 for desktop work. Just guessing since I used neither, and I
>> always worry about installation problems on Windows.)
>> They just started public development but all packages are in BSD from
>> what I have seen.
>> Otherwise, I would build on top of pandas, scikits.timeseries or larry
>> or tabular if you want to handle your own time variable.
>> For specific financial time series, e.g. stocks, exchange rates,
>> options, I have seen only bits and pieces, or only partially
>> implemented code (with a BSD compatible license), outside of quantlib
>> and it's python bindings.
>> Maybe someone knows more about what's available.
>> For the econometrics/statistical analysis I haven't seen much outside
>> of pandas and statsmodels in this area (besides isolated examples and
>> recipes). I started to write on this in the statsmodels sandbox
>> (including simulators).
>> "modeling and analysis of financial time series" is a big project,
>> and to get any results within a reasonable amount of time (unless you
>> are part of a large team) is to specialize on some pieces.
>> This is just my impression, since I thought of doing the same thing,
>> but didn't see any way to get very far.
>> (I just spend some weekends just to get the data from the Federal
>> Reserve and wrap the API for the economics data base (Fred) of the
>> Federal Reserve Saint Louis, the boring storage backend is zipped
>>> Thank you,
>>> NumPy-Discussion mailing list
> NumPy-Discussion mailing list
I'm also interested as well what you might have in mind. For
econometric models, the place to start (and perhaps contribute) would
definitely be statsmodels. For other models it depends. Over the next
year or so I plan to be working as often as I can on implementing time
series models either inside statsmodels (with a pandas interface so
you can carry around metadata) or inside pandas itself (which is not
necessarily intended for financial applications, but that's what I've
used it for). I'm interested in both standard econometric models (see
e.g. Lütkepohl's 2006 time series book) and Bayesian time series
More information about the NumPy-Discussion