I agree its a good toy if u just want to hack basic ideas but it quickly becomes limited when you want to get over the basics.. all the packages are basic. proprietary modeling is impossible unless u want to write up ur own code which is far better done in other languages.. the packages u mention are pretty much econ101 and hardly anyone uses that bullshit in reality. much of pros use R, S, Matlab etc. to hack ideas but implementation is done in C++ usually. so sure if u want to test stuff quick it can be useful. but if u want to push boundaries actually theoretically test models (one reason i asked about GARCH) than u need something that allows u to do anything u like. what trader has been successful with R? none that i know. most prop traders in banks actually prefer EXCEL more than anything else.. on the hedge find side there is alot of demand for statiticians with S/R skills but only to analyse large data banks and develop models. I once read a very good book on data mining and knowledge datbases which made it clear the problems we face today with analysing data and implementing such applications. (highly recommend) http://www.amazon.com/Advances-Knowledge-Discovery-Data-Mining/dp/0262560976 I PERSONALLY love delphi and think it's the best language out there for doing anything (esp. requiring database). many of the people I meet develop their own models (not off the shelf garbage ie. CAPM, garch1.1. etc..) so I don't think these basic R packages are useful at all. what's more valuable is if u understand theory e.g. OLS and want to implement ur own way of optimising/testing a model.. R is certainly useful but my view is you'll soon want to experiment/explore ur data more and you might think maybe u had invested ur time in using something else. btw it might take less time coding Blackscholes in R (who the fuck uses BS these days anyways) But it would actually take far less time coding it in C++/# and directly socket trading it.. the reason i feel u fear OOP language is u probably lack the theoretical foundations of what u want to examine.. there is no point in hittin buttons to return some time series result if u don't truly understand what the software is doing with the data. by coding stuff urself u will learn far more. in any case do explore it. such quant traders know more than 1 language anyways and use a variety of tools..
Batman28, Very informative post, thanks! I agree with almost everything you've written. The only thing I think we were disagreeing on is how much is available in R, and I think I understand why; I was referring to stats *in general*, whereas perhaps you're referring to stats that are typically applied to finance; as you mention garch, etc. I'll guess that over half of the stat tests publicly available in R have never been applied to any financial data whatsoever. I'm not saying they SHOULD, either; >95% of them will probably lead down rabbit holes, which is a good explanation for why hedge funds, as you say, hire R guys to just play with data and/or develop new tests in R, as it does allow for a very succinct statement of statistical models. I agree completely that the commonly used OOP languages -- Delphi, C++/#, Java -- are far better than R for implementing something that performs real-life trades. Thanks for sharing your thoughts; you're clearly one of the best-informed algo traders on this site.
agree, great stuff Batman28. In some sense there is alot in your posts that while I agree with, part of me has had a hard time accepting it. From what I've found, it shouldn't be that hard to collect data, spit out some type of analysis with R or Matlab, then either get matlab up to trade direct or program the strategy in off the shelf retail software. I think the most important point in Batman's posts are though that if you don't understand the model at a deep level then its worthless to bother with this stuff. Then if you do understand the model it should be easy to code up in a real language and no reason to go to matlab/R. Just getting R to spit out data with GARCH is worthless. I had already signed up to take math classes in the fall but I'm not going to bother getting to into this now until I'm closer to Stochastic Calculus. Also, from what I've found, if you can accept what Batman wrote, HDF5 and Python is probly a great combo.