My initial view:

Pros:

• Rapid development (at least initially)
• Leverage a lot of free prebuilt libraries (graphs, numerics, etc)
• You can easily integrate a phyton intepreter and have everything scriptable in an easy way. Cool feature.
• A lot of people have at least heard about phyton

Cons:

• If need to specify that it is a phyton based product that might scare some people away (policy reasons, etc)
• You might run into scalability problems later on in the development. I'm no expert of phyton, but that is a tendency for these types of languages

Manta,

 Can you you kindly post your code for the trading platform? It would be pretty exciting to see it.

Many Thanks in advance

Can I receive the code as well ? I had plans to learn Python for a few months already, but did not find time yet to really dig into it. Would be nice to have some finance-related code to look at.

Thanks

My goal is to have a time series toolbox in python that is useful and easy to use.  I'd like it to be flexible to handle various data frequencies (starting with daily), with flexible sized data sets going along with any set of dates.

As you will be able to tell, I am new at this, planning to learn OO for a later version.  I would have thought this would have been already done by now but I couldn't find it in the cheese shop.  I'm using Enthought 2.4 on a winXP lapdog, and have not been able to get gnuplot or VTK to work either.  (I believe I have the paths set correctly.)  For references I am using the python docs and this book.

Attached File: py code.zip

pytso.py - time series functions

pyopt.py - a first attempt at python, some dodgy options pricing/greeks functions

test.py - use to test pytso.py, dumping output to a csv file for comparison to excel.

In order to use pytso or test, the first step is to upload data.  See calcrets in pytso for this.  You may need to modify it to fit your data.  My raw data (NYMEX: CL) came to me like this from Unfair Advantage:

Attached File: CL.zip

"Another approach is that you might want to keep your data and methods together in a class or a closure so that there is no need to pass anything around."

I'll google "closure" to find out what those are.

signalseeker and tristanreid, many thanks.  There is much to learn.

Very cool.  Thanks for the info.

Can we keep this thread going with more simple examples, illustrating the language and how it's changed? Besides being a python dilettante, I've been working with much older versions of python, I'm largely unaware of what's gone on.  I only recently learned the 'yield' keyword.

If you think about it, adding the yield keyword was very unfriendly to those with lots of fixed income code.  I don't think they were fully consulted about it.  Anyway, here's an example of the yield keyword; it lets you create a function that may be called multiple times, and maintains a state.  Like this:

>>> def pows2Until(n):
...     i=1
...     while i<n:
...             yield i
...             i*=2
...
>>> for i in pows2Until(100):
...     print i
...
1
2
4
8
16
32
64

I know there are other python sites out there, maybe this is unnecessary.  But I'm hoping we can build up to something more complicated than illustrating the basics of the language, something domain specific?

-t.

tristanreid,

I am all for idea of keeping this thread going with examples. But like you suggested, I think it will be better if we build something useful. How about building the timeseries tool box that tradenator is writing?

I'm all for it.  One question though: at what level do we start?  We could build everything from scratch or we could use open-src libraries like numpy (which has linear alg. routines), and scipy (includes fft and some stats stuff) and build time series specific interfaces on top of that.  I don't have experience with either of those libraries, I'd love to learn, but I'd also love to just write stuff.  Whatever's clever, I'm happy either way.

One thing that could be fun is to pick a particular time-series problem, and learn how to use numpy/scipy to solve the problem, then to expand the solution to make it as generic as possible, then to argue about a better way to do it.  Then repeat. 

-t.

<sniff> <sniff> I was once unemployed, living with my new mother-in-law in a 2 bedroom apartment in a foreign country, with my dog in quarantine.  I was so desparate I accepted a job doing y2k project management in the end.

I selected Enthought on the advice of the book I have (link earlier in the thread) and because they have things packaged up together nicely for a noob like me.  I'll put biggles on the list of things to try (thanks signalseeker).

I think scipy's distribution functions and numpy/scipy arrays are worth using.  I'd be happy to collaborate on this, but I am not the right person to be leading it. Following tristanreid's approach, we could start with the movav and movsd functions I posted below.  Assuming I have a series of dates,data what would a better approach to these be?  Also, I was thinking of having a third item in the tuple carrying descriptive info like market, what the data is, and so on.  This could be useful if/when data becomes a 2-D array, such as a set of returns for my universe of commodity markets that I can pass through the movav function all together.

Should this be tagged as a new thread or does it matter?