January 16, 2013 at 6:40 pm #382
Of the interests and the limits of dynamics
“SFC models are intrinsically dynamic. Because of its consistency, every monetary flow is recorded as a payment for one sector and a receipt for another sector. These flows will impact end-of period
stocks levels, which will then generate next period flows, which then update the stocks, and so on. The long-run dynamics of SFC models are thus composed of a path of short-run period interconnected with each other via the stocks.
The difficulties arising from the models’ convergence properties force a groping methodology in order to calibrate each model. Or rather, a “guesstimation” of the parameters, so that the model presents some characteristics such as smoothness, stability or desired behaviors. This arbitrary choice often prevents a deeper reflexion on each parameter’s impact on the economy at hand. In this perspective, one might find an answer via analytical solutions of static and dynamic equilibriums.
However, this implies an increasing complexity and ever-rising computational time. This highlights the need of a strong methodological framework.
What are the software to use? Eviews allowed the large diffusion of SFC models thanks to its flexibility and thanks to the website sfc-model.net. Indeed, the website offers all the models from the reference textbook of Goldey and Lavoie (2007), as well as from numerous papers. However, because Eviews is not a programming software and because it is not freely available, some limits to its use are appearing. What other tools could provide a better solution? Matlab and Mathematica are not open-source, R does not have a numerical solver,..?”January 16, 2013 at 7:31 pm #394
Thank you very much for this very exciting proposal. I propose some preliminary comments on this proposal.
First, I don’t see why EViews is not a programming software. I think you should precise what you heard by “programming software”. EViews has a very convenient Windows interface but you can choose to not use it. As witnessed by the Cambridge Alphametrics Model (CAM) of Francis Cripps, EViews can be adapted to solve very large complex empirical models in which ceilings and floors can be introduced. Besides, in their reference textbook, Wynne Godley and Marc Lavoie had used the MODLER package.
Second, I don’t see any reference to supply and to behavioral expectations In most cases, SFC models doesn’t include any supply constraint. Even if most of them strongly reject Say’s law, I think that these kind of modelling could be not adapted in some cases and especially in the case of the euro crisis. We should introduce some supply constraints in order to be more realistic in our policy prescriptions.
Anticipations could be also problematic. I think this point is crucial in the perspective of challenging mainstream economics. Even if this point is already partially analyzed in the Godley-Lavoie textbook, I think we need to going further (…together ?) on this point to propose a fully-fledged alternative to rational expectations.
I hope that these remarks will stimulate further discussions !
Novice in SFC modellingJanuary 16, 2013 at 7:39 pm #395
We have discussed the software issue somewhat at the Inet conference in Waterloo, since some of us were pressed to support the Minsky initiative started by Steve Keen.
My personal view is that we need translators from one platform to another, say from Mathematica to Eviews etc., rather than trying to persuade a whole community to adopt the same software.
We do need some simple software for teaching, and possibly the Minsky software could be a solution (I have not tried it yet!) and we need an open-source solution for complex models, and R seems a good candidate.January 16, 2013 at 8:10 pm #397
You will find my response to Jamel hereunder in order to clarify the bad wording we chose, but let me first answer to Gennaro.
The first point I want to make when discussion the choice of a software is that it should be open-source or at least free for download. The cost of software such as EViews but above all of Mathematica is clearly a disadvantage. But I also believe that having open-source software allows for much more flexibility if we feel that a plug-in is missing.
This is why I also thing that R is a good candidate. Indeed, it has a wide community of developers that could help us, if needed. Furthermore, R, like EViews, was designed to handle data and thus already has lots of nice functions and plug-ins. What is missing, to my understanding, is a solver for large system of non-linear equations.
Finally, I also agree that trying to convince everyone to use the same software is a loss of time. Also because everyone has its own habits and tricks when developing a model… If you allow me to point out the biggest advantage of Mathematica, to my view: analytical resolution. The thing allows you to obtain the analytical solution of some (not all) non-linear systems, which comes very handy when playing with the model.
As promised, my answer to Jamel:
Thank you for your e-mail.
Let me first have a general response. The proposal you all received is certainly not an extensive list of all the topics open to debate and we welcome any new ones. We probably need to find a better place to discuss these, let me come back to you when I found a technical solution such as a forum or a blog (done!).
Regarding EViews, the wording should have been chosen more carefully. Our goal was to raise questions and certainly not to be dismissive. Of course, we are well aware, as stated in the proposal, that EViews allows to solve large models and is a very useful tool. My understanding, and it is limited as I have have not used EViews much, is that EViews doesn’t offer the flexibility that other softwares such as Matlab, Mathematica or R offer. For example, the error handling seems rather crude. Furthermore, linking this point with your second point, one way to include supply constraints would be to write the model using Karush-Kuhn-Tucker conditions between price and quantity. Does EViews allows such a formulation? I know Mathematica doesn’t and I had some trouble when trying to model such a feature. However, since Mathematica allows for very flexible coding, I could circumvent these limitations. That said, I’ll gladly accept any argument showing that my concerns regarding EViews are not founded.January 21, 2013 at 3:47 pm #399
Thanks for your interesting replies.
I have some further remarks and suggestions.
Firstly, I think that error handling on EViews is quite complete and acceptable relatively to other software as STATA or GAUSS.
Secondly, in works of Jean-Louis Brillet, you can find supply constraints in a macroeconomic model solved with the EViews package.
Nevertheless, according to me the most promising way in SFC modelling is agent-based modelling. I think that EViews (or even maybe R) could not be fitted to solve this kind of models. We could use programming language as Java.
The MINSKY initiative started by Steve Keen seems to be very attractive as it underlines the role of the banking system in macroeconomic modelling.
Find here (http://www.economist.com/news/finance-and-economics/21569752-efforts-are-under-way-improve-macroeconomic-models-new-model-army), an article of The Economist in which alternatives to DSGE modelling are quoted (including the MINSKY initiative).
JamelJanuary 30, 2013 at 12:23 am #403
I agree that proprietary software is a significant barrier to the diffusion of this approach. E-views is bad enough but my god, MODLER! (there has to be a story there). R is a good candidate, especially since many learnt it in stats. Octave is an open source clone of matlab … but I don’t like it. Neither of these are particularly ‘nice’ languages. I’d suggest looking at Python. It’s open source and far more intuitive than the other two. InsightMaker is a cute program I once built SIM on. Its far to limited to do anything else, but it might have some pedagogical use. Translators are an interesting idea but may or may not violate terms of service.
You must be logged in to reply to this topic.