[python-uk] London Python Roles

Andy Robinson andy at reportlab.com
Thu Dec 16 00:02:01 CET 2010


> Report Lab does a fair bit of work in the financial sector in a rather different field.
Sorry, the light took a while to reach the batcave tonight...

I was pushing Python in finance back in 1997/8, and there have been
many, many people using it (usually under the radar at first) in the
City for a long time.  In the old days it excelled at gluing systems
together, scripting other peoples' C code, and prototyping algorithms.
 There were many times when people needed a solution faster than "IT"
could organise it, and being freely available and able to work with
Excel, it helped a lot of quants.

Now, of course, Python is mainstream, and other languages have got a
lot better at the 'glue' and web stuff and caught up to some degree.


On 15 December 2010 14:02, Jonathan Hartley <tartley at tartley.com> wrote:
> On the other hand, if you're running across the
> internet, then any slowdown due to using Python verses another language
> would be vastly swamped by network and other IO delays. Am I very much
> mistaken?

There are no general answers to that question.  There are indeed many
cases where people needlessly worry about the language when network
and IO dominate.  There are also lots of cases where you want to do
some kind of "atomic job" on one machine, and find it's an order of
magnitude too slow in a high level language.  There are some "Monte
Carlo" approaches to pricing securities which have no analytic
solution; and in the retail sector it's fashionable now to show
'clouds of outcomes' about where your pension might end up, needing
10000 random walks to plot a chart and spit out a 2-3 page PDF
including it.

Two of the beauties of Python in this area are that (a) you can afford
to rewrite your algorithm several times and be sure it's the best
approach, and (b) if you really need to, it's easy to shift the 'inner
loops' into C.

Putting it in perspective, I even know some people in the City who
find hand-coded C too slow for their simulations, and who are
basically writing microcode for chips in order to put a supercomputer
under their desk!

The bigger question is whether all this horsepower ultimately leads to
better investment performance.  I will stay out of that one ;-)


-- 
Andy Robinson
ReportLab


More information about the python-uk mailing list