5 ms·

Forecasting with Julia

68 points by matthjensen 9y ago

matthjensen 9y ago
Most policy decisions these days are heavily influenced by proprietary forecasting models. Just look at the fuss that was made because the House passed a health bill without a Congressional Budget Office score. The CBO score will certainly play a large part in the Senate's rewrite. The problem is that the CBO and other organizations like it are quite secretive about their modeling. Most of the time they only produce point estimates, and they don't publish many of the assumptions behind their modeling. When there is a bill that contains both taxes and spending, the Joint Committee on Taxation models the tax part, the CBO models the spending part, and they just smash the results together because even those two organizations aren't willing to share and integrate their models. The NY Fed is serving as an important leader in this field. Policy analysis should be transparent and scientific, and that's what the NY Fed is moving the field towards.
- matthjensen 9y ago
  The Open Source Policy Center, where I work, is focussed on this issue. www.ospc.org Most of OSPC's work is focussed on fiscal policy rather than monetary policy.
  - Gtifn 9y ago
    What languages do you use?
    - matthjensen 9y ago
      Primarily Python. See, for example, https://github.com/open-source-economics/tax-calculator https://github.com/open-source-economics/tax-calculator. We also make many of the models available through webapps like https://www.ospc.org/taxbrain https://www.ospc.org/taxbrain and https://www.ospc.org/ccc https://www.ospc.org/ccc. The source code for those is available at https://www.github.com/opensourcepolicycenter/webapp-public https://www.github.com/opensourcepolicycenter/webapp-public
Phithagoras 9y ago
code at https://github.com/QuantEcon/RBA_RBNZ_Workshops https://github.com/QuantEcon/RBA_RBNZ_Workshops
- matthjensen 9y ago
  That's for workshop materials. Here is the model itself https://github.com/FRBNY-DSGE/DSGE.jl https://github.com/FRBNY-DSGE/DSGE.jl
morley 9y ago
I was wondering why they specifically chose Julia for this (since I know little about Julia at all), and found an answer in a previous article: > Julia has two main advantages from our perspective. First, as free software, Julia is more accessible to users from academic institutions or organizations without the resources for purchasing a license. Now anyone, from Kathmandu to Timbuktu, can run our code at no cost. Second, as the models that we use for forecasting and policy analysis grow more complicated, we need a language that can perform computations at a high speed. Julia boasts performance as fast as that of languages like C or Fortran, and is still simple to learn. http://libertystreeteconomics.newyorkfed.org/2015/12/the-frbny-dsge-model-meets-julia.html http://libertystreeteconomics.newyorkfed.org/2015/12/the-frb...
- ced 9y ago
  Julia boasts performance as fast as that of languages like C or Fortran, and is still simple to learn. I think the greatest benefit is that Julia code is both high-performance and (mostly) high-level, which makes it easy to change. I don't mind implementing a completely-specified algorithm in C or Fortran, but making significant changes to these code bases is simply much more work than in languages like Python or Julia.
  - ska 9y ago
    see also lisp.
    - ced 9y ago
      Yes; Julia isn't unique in being a modern high-level high-performance language. But it's the only one I know that explicitly targets numerical computing, and it chooses its trade-offs accordingly.
      - philip1209 9y ago
        Native matrix syntax and operations is a huge plus, IMHO.
one-more-minute 9y ago
See also the case study on the JC website, which includes a talk from one of the developers explaining some of the rationale: https://juliacomputing.com/case-studies/ny-fed.html https://juliacomputing.com/case-studies/ny-fed.html

68 points by matthjensen 9y ago

Most policy decisions these days are heavily influenced by proprietary forecasting models. Just look at the fuss that was made because the House passed a health bill without a Congressional Budget Office score. The CBO score will certainly play a large part in the Senate's rewrite. The problem is that the CBO and other organizations like it are quite secretive about their modeling. Most of the time they only produce point estimates, and they don't publish many of the assumptions behind their modeling. When there is a bill that contains both taxes and spending, the Joint Committee on Taxation models the tax part, the CBO models the spending part, and they just smash the results together because even those two organizations aren't willing to share and integrate their models. The NY Fed is serving as an important leader in this field. Policy analysis should be transparent and scientific, and that's what the NY Fed is moving the field towards.

matthjensen 9y ago
The Open Source Policy Center, where I work, is focussed on this issue. www.ospc.org Most of OSPC's work is focussed on fiscal policy rather than monetary policy.
- Gtifn 9y ago
  What languages do you use?
  - matthjensen 9y ago
    Primarily Python. See, for example, https://github.com/open-source-economics/tax-calculator https://github.com/open-source-economics/tax-calculator. We also make many of the models available through webapps like https://www.ospc.org/taxbrain https://www.ospc.org/taxbrain and https://www.ospc.org/ccc https://www.ospc.org/ccc. The source code for those is available at https://www.github.com/opensourcepolicycenter/webapp-public https://www.github.com/opensourcepolicycenter/webapp-public

Phithagoras 9y ago

code at https://github.com/QuantEcon/RBA_RBNZ_Workshops https://github.com/QuantEcon/RBA_RBNZ_Workshops

matthjensen 9y ago
That's for workshop materials. Here is the model itself https://github.com/FRBNY-DSGE/DSGE.jl https://github.com/FRBNY-DSGE/DSGE.jl

morley 9y ago

I was wondering why they specifically chose Julia for this (since I know little about Julia at all), and found an answer in a previous article: > Julia has two main advantages from our perspective. First, as free software, Julia is more accessible to users from academic institutions or organizations without the resources for purchasing a license. Now anyone, from Kathmandu to Timbuktu, can run our code at no cost. Second, as the models that we use for forecasting and policy analysis grow more complicated, we need a language that can perform computations at a high speed. Julia boasts performance as fast as that of languages like C or Fortran, and is still simple to learn. http://libertystreeteconomics.newyorkfed.org/2015/12/the-frbny-dsge-model-meets-julia.html http://libertystreeteconomics.newyorkfed.org/2015/12/the-frb...

ced 9y ago
Julia boasts performance as fast as that of languages like C or Fortran, and is still simple to learn. I think the greatest benefit is that Julia code is both high-performance and (mostly) high-level, which makes it easy to change. I don't mind implementing a completely-specified algorithm in C or Fortran, but making significant changes to these code bases is simply much more work than in languages like Python or Julia.
- ska 9y ago
  see also lisp.
  - ced 9y ago
    Yes; Julia isn't unique in being a modern high-level high-performance language. But it's the only one I know that explicitly targets numerical computing, and it chooses its trade-offs accordingly.
    - philip1209 9y ago
      Native matrix syntax and operations is a huge plus, IMHO.

one-more-minute 9y ago

See also the case study on the JC website, which includes a talk from one of the developers explaining some of the rationale: https://juliacomputing.com/case-studies/ny-fed.html https://juliacomputing.com/case-studies/ny-fed.html