Danish government makes its new economic model open source
github.com
github.com
The authors working for the Danish government deserve an A+ for making their code and assumptions public, along with comprehensive high-quality documentation.[b] It was a delightful surprise to click on the link and find... a carefully curated repo.
I have only one nitpicking: This looks like the kind of project that should have been -- and in fact probably should be -- implemented in Julia, instead of in a proprietary language. I suspect many heuristics and approximations could be more accurately modeled by ML components, interspersed between equations, as has long been proposed by Julia folks like Chris Rackauckas.
--
[a] https://news.ycombinator.com/item?id=30028188
[b] https://media.githubusercontent.com/media/DREAM-DK/MAKRO/mai...
The only reason I ask is because your model is written in a proprietary language. While Gurobi models are written in Python (and solved by the silver).
Just curious if you found it not effective.
When you develop models like this, what is your stance on testing? All that stuff in a software engineers toolbox, unittests, integration tests, exploratory tests, manual tests etc.
Oh god, our economy is doomed.
Econometrics gives people everything they want to hear, and with a scientific veneer as a bonus. The HN crowd should however appreciate how susceptible this is to the tyranny of metrics. Claude-Frédéric Bastiat warned of this over a century ago with his parable of the broken window: Think not only of what you can see (the glazier who is employed to repair the window), but also of what you cannot see (the alternative purchase of a suit) had the window not been broken.
Why not?
Money would majorly come from those who have the greater means to pay larger amounts into the system, and on the hypothesis that bettering the material conditions of those in the lower and middle class will necessarily lead to increasing productivity and thus (and 'thus' is a large leap) greater general income that can partly be paid back into the system in a cycle.
A reasonable overview: https://econreview.berkeley.edu/unboxing-universal-basic-inc...
How much you plan to pay in taxes to support my UBI?
So you're saying that you're a poor economist, then.
I expect to pay and continue paying a lot, yet the benefit is more than what I pay in taxes. What's better is that our current UBI is quite a bit above zero already, but the distribution is fraught with large costs in time. Re-organizing and streamlining UBI distribution would result in significant cost improvements to the current US welfare approach, and improve its impact.
Can you back your claim with facts? UBI is universal, not magic.
> I expect to pay and continue paying a lot,
Just say a number, so we will put it into formula to see is your expected taxation will match minimum requirement for UBI.
Typical wealth distribution without UBI:
^
|
|
| *
| * *
| * *
| * *
| * *
|* * * *
+------------------------------------------------>
^ - number of people, > - income
^ *
|
|
|
|
|
|
|
| *
| *
| *
| *
| * * *
+--U------------------------------------------->
^ - number of people, > - income, U - UBI level
As you see, UBI erases middle class. Everyone becomes poor, except some rich.
The US already has a non-zero UBI[1], it's simply inefficiently distributed via the negative income tax rate and various welfare efforts. That's a clear first area to fix things up. Then, if we need to increase the historically low tax rates on high income households or implement capital tax rates on the wealth, increases in taxes can be considered.
Further, from a theoretical perspective, the economy is not zero-sum. A rising tide lifts all boats; if less money has to be doled out to industries supporting enforced poverty, such as private jail systems among others, then there is more money in the economy for utilization for many things (including but not limited to UBI). General equilibrium effects often have counterintuitive effects, such as they are.
[1] https://en.wikipedia.org/wiki/Universal_basic_income_in_the_...
I answered your earlier questions already. I encourage you to read full threads.
As for your other point, I guess the issue the lack of a hypothesis regarding basic income. You can't test "basic income" as such, and with such unspecific setup you are sure to conflate rather than illuminate mechanisms of relevance. In any case, it depends on your question whether you need to take income away from anyone. Are you going to start by examining distributive effects? Really? No physisict would test their airflow experiment on a flying rocket, while also trying to measure how he exhaust affects the wind in Costa Rica next week and saying 'something' about the future climate. They would start by putting their airflow gizmo into an isolated chamber - which is, afaik, more or less where UBI experiments are.
Fair point. I guess you could run an experiment where money is simply taken away from people, destroyed and then look at the effect of that in isolation.
That it is harder to draw conclusions from experiments in biology than in physics doesn’t imply biology isn’t a science. Economics adds the problem that we (rightfully) don’t want to do realistic controlled experiments (would at times be hard, too. Put a few million people in a country with a given law regime, watch for a couple of decades, draw conclusions. Repeat) but again, to me, that doesn’t imply it isn’t a science.
I think it’s more part of the humanities than a science, though.
As for Econometrics, I agree that studies employing observational causal inference in statistics are much harder (and harder to swallow) than experiments. Usually.
But perhaps, if you dig a bit deeper and ask yourself things like: "what makes a hypothesis falsifyable" or, "could there be a hypothesis falsifyable by observational data?". Have your ever written down the exact exclusion condition (or exogeneity condition or whatever) you use in your experiment? You might find out something interesting.
Controlled experiments, much like any econometric (or statistic) technique, require identification assumptions. And of course, these can not be proven. Even though many people believe it is so, experiments are not really different in kind, the assumptions are just usually easier to believe. Not always, though, which can also be fun. Even in an experiment, you - in fact - observe reality only once. In other cases, however, we find out that our assumptions about the identification in the experiment were wrong, due to factors we didn't know about. This happens even in physics (shocker, I know).
Keep an open mind, would be my advise. Your critiques are well known, and in the broad sense tackling them is the business of statistics. By the way, there are indeed smart people here and there.
A good heuristic: If your find yourself rejecting entire disciplines, like econometrics, all while implictily using auction theory, regulation, pricing etc. in your amazon, ebay, cellphone, or energy grid, you are probably on the wrong peak of a Dunning Kruger curve.
The number of places for bias to creep in are way to many. I think the econometrics people have gone down a hopeless path, guided by wishful thinking and demand from people who want us to be able to make statements about things which we just cannot do.
The parts of economics that are valuable are not based on econometrics, but upon reasoning from first principles: Incentives, supply and demand, game theory etc. under which I would argue all of your examples fall.
Could you be more specific what is a lab setting for you? Natural experiments are not what I meant when I spoke of experiments, as natural experiments are observational inferential methods!
btw did they ever tell you about that physics experiment in a lab that was built just over a subway line? So much confounding!
Jokes aside, your assumptions amount to assumptions about what "controlled" and "in a lab" means. I hope you see it makes more sense to define the requirement for experimental identification mathematically, in which case you'd see that the conditions are not different in kind, only in degree. etc.
Edit:
If that counts as revealed preference data, I can assure you that econometrics are used to great effect in pricing, auctions etc - one of the reasons certain orange and blue companies hiring experts on this have become mucho rich in the past couple of years. Some oldschool econ models may not be the best to describe the altruistic, boundedly rational present-preferrer. But oh boy, can you make some serious dough if you can mechanism design the auction platforms and the best the firms can do is to try to train some machine learning models against it.
There is a substantial difference in degree when dealing with atoms, molecules, homogenous materials vs. people who think, face incentives and understand game theory.
As far as economic reasoning goes, there are some studies you can do (like optimal auction clearing) where you can hold the inputs equal and then adjust some aspects of the optimization (the objective, constraints, or decision variables) and get a very accurate picture of how market changes would impact the results. I think that is probably an exception rather than the normal case though.
In general, my thoughts during my studies has been that the economic theories of supply & demand, incentives, elasticity, theory of the firm, game theory, government intervention...etc all has more use as a generic framework of thinking (ex: if we put in a price cap, here are the expected impacts), rather than an equation I can use to get an exact answer (basically exactly what you're saying above). Doing some kind of analysis on things as complex as the US economy must have massive inaccuracies involved as there are too many factors and even behavioral economics doesn't have all the answers for why people don't always act rationally.
So how do you confirm these first principles? I'd argue that those are an even wider door to let your biases flow into the analysis. I mean this type of argument has given us trickle down economics which has significant real world evidence it doesn't work.
For example people use says law to argue saving does not create unemployment but says law can also be interpreted as saving is impossible over the long run and therefore unemployment shouldn't happen either.
Would you critique astronomy the same ways as economics? What is the critical difference?
[1] Yes, there was a lot of simultaneous work in other areas as well.
Economics is a dismal science and models rarely predict real word outcomes correctly. It’s more akin to astrology than astronomy.
Whether the stuff you’re observing can read and understand your observations, and then choose to do the exact opposite.
- Astronomers cannot influence the system they are describing; there's no feedback. In economics there is.
- Astronomy describes systems with a few, simple particles whose inner state is not relevant to the behavior at large. Economics describes (tries to..) systems with many complex 'particles' whose inner state is essential to the behavior at large.
Not saying that this is what's fundamentally separates astronomy from economics, just pointing out some potential differences.
One of the observation of the author is that models work better if the human beings accept to restrict their behavior and follow simple rules. Another observation is that in social science, it is hard to design models that do a prediction significantly better than "tomorrow will behave the same as yesterday".
[0]: http://www.ens-lyon.fr/en/article/research/your-life-numbers...
Astronomy doesn't fight back, economics does.
* Observers are also actors. They can and will significantly influence the observed.
* Once observations are revealed at large it'll have short/long term consequences. There's a good reason fed gives multiple indications before changing any parameters such as interest rates, tapering etc.,
* Humans don't necessarily act like rational selfish agents (as economists expect) all the time.
* Economics is often entangled with other aspects such as trade agreements, geopolitics some of which take years and decades to manifest.So when we criticize "economists" are we criticizing the scientists like Galileo and Copernicus that were right and got their books banned, or are we criticizing the ones that supported the political power of the day?
Note that some of those scientist would have genuinely believed their 'church-approved' theories. Just their opponents couldn't challenge them openly without incurring wrath.
And would generic criticism of "scientists" have benefitted science or the established political power at that time?
It's not that economics can't be a science like physics. It's that (most) economists don't treat it like a science.
Contrast that to physics, where we have been very successful with predicting the behavior of masses both on small and on large scale, with relatively few exceptions.
The seen/measurable: employment in the production of windows. The unseen/unmeasurable: missing employment in the production of suits.
The even worse part is that some assumptions don't even work in once you look at the math of e.g. accounting.
Keynes and Silvio Gesell and Wolfgang Stützel are much closer to how our accounting system known as money works even though both of them lived during a gold standard where it was normal to bend some accounting rules like digging out new gold to create more money or just bend them outright by printing money without any asset backing it.
Assumptions don't need to be absolutely true, they have to be good enough to have predictive power.
>Keynes and Silvio Gesell and Wolfgang Stützel are much closer to how our accounting system known as money works
I doubt.
From Popper’s The Open Society and its Enemies
[1] https://en.wikipedia.org/wiki/General_Algebraic_Modeling_Sys...
The Federal Reserve has two models ( https://www.federalreserve.gov/econres/us-models-about.htm and https://www.federalreserve.gov/econres/edo-models-about.htm ), written in EViews and Dynare, respectively.
[1] https://www.dst.dk/en/Statistik/nyheder-analyser-publ/Publik... [2] https://www.dst.dk/da/Statistik/ADAM/Modellen-ADAM/Download
It is a step forward in transparency (even though 90% of danes probably don't understand a thing of what was published) which is something no one I know is opposed to.
Hope this sets a precedent for other countries to follow!
But far beyond 'Black Swan' events it's hard to account for innovations, geopolitical shifts, things that change the dynamic.
Plastics, the dishwasher, women in Engineering and the workplace, the birth-control pill and Rock and Roll combined to shift things in ways we could not put down in math, at least beforehand.
But the biggest 'black hole' is that it does not account for consumer surplus: we only measure what is bought and sold, at that price. We don't measure the leverage gained by consumers for cheap fashion, the benefits of more variety in goods, more vacation, shifting to bikes from cars etc..
The data is still a beneficial area of study however.
At a high level, I, like you, am fairly persuaded by the Hayekian point that its impossible to collect enough information or otherwise capture the necessary complexity to reliably model even a moderately complex economy with enough fidelity to allow central planning. But this is far short of that. And there does seem to be a need for general models to allow policymakers to make informed decisions. The only alternative would seem to be to make no decisions at all which a) is itself a decision and 2) does not seem to be an option that many people seriously advance.
I guess your position could be that these efforts are doomed to failure--i.e., no model will provide useful results. But that position--that no model will provide any amount of useful information, even at very low levels of granularity, is a very strong position--much stronger than Hayek's argument against planned economies--that I'd have to see some serious support for. And it certainly doesn't seem like a position that people should accept without a very convincing showing of its futility, given the significant costs of making policy decisions without even attempting to model them, or of making no policy interventions at all.
I'm not qualified to make any, but are you expecting/hoping for/accepting pull requests?
As the saying goes "any metric is bad once you optimize for it", so maybe that would lead to terrible politics.
That makes me worry if large actors can use the model to make any self-serving policy suggestion "look" like a good idea for the economy, by combining it with precise information on what the national economics model is sensitive to or isn't.
Yes, you tweak variables and look at the impulse-response functions. That's what those models are built for.
Also, is anyone aware if such large scale automatic optimisation solvers are used in other departments, such as crime regulation, health care, military?
I mean, I guess the introduction is technically correct in stating that these textbooks are not sufficient to develop a microfounded model of banking. But then, one might suggest the author to perhaps look a bit further than Mankiw's 101 book? After all, textbooks treating the subject in more detail than even the present paper do exist, even from heterodox sources, if the mainstream is unacceptable.
I studied economics, and I have some doubts about whether this kind of modelling works well enough to be useful. There's statistical issues like whether you have enough data to bring in your error bars, and theoretical issues like whether the thing you're studying moved below you because it's being studied. Neither of which invalidate the model entirely, but both of which make it a heck of a hard task.
And in the end, what matters is whether it works or not. The Lagrangian for quantum physics is some huge formula that gets pushed as meme now and again but it works. Likewise f = ma works perfectly well for your high school rolling ball experiment.
(Reference to the recent news item that Erdogan made a public statement in which he renamed Turkey to Turkiye in English.)
I always assumed those names originated externally.
https://en.m.wikipedia.org/wiki/Name_of_the_Czech_Republic#A...
Nobody gets mad that Spaniards say "Londres" instead of London. It's funny how it's mostly despotic regimes that get so twisted up about these foreign linguistic matters.
Nice! Kudos to DREAM!
https://www.quora.com/What-is-it-about-Denmark-that-many-wel...
I much prefer a predictable policy in response to the economy.
Do you know any other git hosts that have persistently proven themselves trustworthy and have superior financial incentives? Otherwise it might be better to focus on the meat of this post
git clone
Source (from 2015): https://www.dst.dk/da/informationsservice/blog/2015/12/narko...
As for tourism, it’s just one of a long list of tourist attractions in and around the city.
It would not make a huge difference if Christiania disappeared tomorrow (although Christianshavn would be a much nicer place).
Extending the scope of AI decision-making outside of the realm of economics seems much harder, though, as there's much less training data for questions like how to educate students in a pandemic, or whether to loosen planning rules for building wind turbines.
I suspect that before AI becomes capable of providing good answers to those questions, it will already have changed the world beyond recognition, and probably not in a good way. The best we can hope for is mass unemployment and a UBI, which will also make all economic models obsolete too.
But how would you decide how much to give through UBI next year? How inflation will change in response? How many people will still be needed in jobs?
You still need a model, because the world is changing, and a model helps you understand how.
But sure, it might completely miss some novel phenomenon. It is up to the humans running the model to be aware of its limitations and accuracy.
In any case, the more heads thinking about the problem the better, and the model is an approximation of a lot of heads. And the equations can show why it thinks what it thinks.