Slashdot Mirror
Machine Learning Could Solve Economists' Math Problem
An anonymous reader writes: Noah Smith argues that the field of economics frequently uses math in an unhealthy way. He says many economists don't use math as a tool to describe reality, but rather as an abstract foundation for whatever theory they've come up with. A possible solution to this, he says, is machine learning: "In other words, econ is now a rogue branch of applied math. Developed without access to good data, it evolved different scientific values and conventions. But this is changing fast, as information technology and the computer revolution have furnished economists with mountains of data. As a result, empirical analysis is coming to dominate econ. ... [Two economists pushing this change] stated that machine learning techniques emphasized causality less than traditional economic statistical techniques, or what's usually known as econometrics. In other words, machine learning is more about forecasting than about understanding the effects of policy. That would make the techniques less interesting to many economists, who are usually more concerned about giving policy recommendations than in making forecasts."
>> the field of economics frequently uses math in an unhealthy way
Yikes, if you think that's bad, you'd better not look at the "social" sciences then.
Many aspects of economics, at least "real world" economics, are not based on logic but on emotions like greed or fear. As a result, they are difficult to describe with rigorous mathematical models. Machine learning may in fact be a more accurate way to predict such things, but as the old saying goes past performance may not guarantee future results, and since all machine learning algorithms have to go on are past performance, they too will be prone to surprises and the quirks of human nature.
No, because most of economics was not scientific. Economists would get a "feeling" of how it should be, then fudge the numbers to prove their theories.
I'm a good cook. I'm a fantastic eater. - Steven Brust
who are usually more concerned about giving policy recommendations than in making forecasts
What? Is this implying that they want to make suggestions about what to do in the present and future to change the future without being the least bit concerned with forecasting the future? I don't think I would listen to anyone who wants to make important changes/suggestions without them being very concerned with attempting to predict the future of the situation at hand.
Also,
the field of economics frequently uses math in an unhealthy way
As an EE having taken many econ classes, I can wholeheartedly agree with this statement.
timeo Danaos, et dona ferentis
Let's be clear: how you think economics is defined by your ideology, and most economics is bad math with unfounded assumptions arriving at un-supportable conclusions.
So, if you're the Chinese government and think you can manipulate markets to suit your beliefs, you'll be horribly mistaken. Likewise, if you subscribe to the ridiculous Austrian School of economics (which refuses to have empirical evidence), then you likewise believe your theory is so perfect it doesn't need to be validated.
Nobody has ever had any proof for "trickle down economics" other than they think it should work and it suits their ideology, but 30 years of actual real world data mostly shows it's utterly failed to work as planned.
Economics is useful to look at what came before, and understand some limited problems ... but in general many people believe once you try to use that to predict things, or influence outcomes you get into a level of complete bullshit and voodoo. Time and time again when people try to take action or set policy based on economics, it fails utterly.
And until economics is based on anything other than sketchy math and ideology, it can never be a real science or have much more meaning than something people use to defend their ideology. But since people never look at economics separated from their ideology, it will never happen.
Economics is mostly a tool to make it look like the things you believe should happen, based on how you want the system to behave, have any actual relationship with the outcomes you expect to achieve with policy. The problem is that is a lie.
But it sure as hell can't be called an objective science. First you have to believe in the ideology and then you believe in the methodology.
The problem is people like to believe that the ideology is objective reality, and that their observations are in fact rules. And that simply isn't true.
Lost at C:>. Found at C.
machine learning is more about forecasting than about understanding the effects of policy
And forecasting depends on some underlying behavioral model. Problem is, people keep changing the model. They try to anticipate or short markets to gain an advantage and very rapidly, this new response to market conditions replaces the previous behavior, invalidating previous relationships.
Have gnu, will travel.
My understanding of the difference is that this produces somewhat testable results WITHOUT requiring a theory of how and why those effects occur.
To give an extremely simplified example, assume that a certain coin is flipped every day. For the past 20,000 days, it has always come up heads. (Obviously not a fair coin). The machine will predict that it will probably come up heads tomorrow. Traditional economic theory will try to understand WHY it keeps coming up heads before making predictions. That's the first difference.
The requirement for a theory that explains how and why economic effects occur also means that the theory is subject to subject to be supported or decried based on political considerations or other irrelevant factors. A system which accurately predicts what will happen without comment on politically sensitive policy questions may be useful.
economists don't use math as a tool to describe reality
They can't. economic realities predicated upon hysteresis or long term analytical trends are constructed with touchstones of bogus equations founded on wild conjecture. a sizeable component of many modern economies are unquantifiable, having been intentionally obscured by human interest. Why did the fed hold reserve rates? what is the nature of a cyclical economy when infused with more than one trillion dollars of emergency funding to prevent an uncodifiable collapse? what factor or theorem transmogrifies and determines the value of liabilites as assets? So much of what economics seeks to do is undermined by the fact that most economies work by the rules of human nature. the Federal reserve practically has its own language to communicate change or lack thereof, and it is intentionally designed to avoid concrete resolution or effable transmission of meaningful information.
Marx as an economist is one of the few to provide a very broad overview of the concept as it applies to modern capitalism. he describes the capitalist economy as a cycle similar to a bird with a broken wing. it builds, booms, collapses and in turn affords a widely acceptable cycle of observed failure. From the great depression to the lincoln savings and loan scandal to the United States Failed leveraged buyout of United Airlines to the dotcom bubble and finally the great recession of 2007 we've all grown to accept this cycle as normative.
Good people go to bed earlier.
That's because economics is a blend of math and psychology. The math assumes a rational actor with all the necessary information. The psychology is rarely rational and involved decision making influenced by the decisions of others, highly varied interpretations of historical events which preclude deterministic mechanisms, and imperfect information viewed through personal biases and strengths. Inaccuracy results from improperly weighting the relative value of these two in economic outcomes and from difficulty in modeling the psychological elements. Bad math is the least of the challenges facing economics.
> Isn't this the same as having economists doing the work, just faster?
Of course not! A machine is a cold, emotionless, insensitive, empathy-deprived entity, while an economist...
Hmm, it really is the same...
Isn't this the same as having economists doing the work, just faster? You are still using past data to predict the future
Yes and no. In a sense, letting AI learn the salient traits of the available data just saves a human from needing to do it; but, you can do something with an algorithm that you can't reliably do with a human - You can model the existing system, then test billions of hypothetical situations to see how they respond.
Humans work amazingly well at pattern matching, even in the absence of understanding of "why". We can even get good at predicting what will likely happen if we change a few inputs to a system. But we don't do so well at figuring out what will happen if we tweak a large number of inputs simultaneously.
Think of this as nothing more than finally making batch hypothesis testing possible in an objective way, in a field where a persuasive argument matters more than facts and where a real experiment can take a few generations to fully show its outcome.
To illustrate, the summary could easily be restated this way:
"...field of [data science] frequently uses [machine learning] in an unhealthy way. Many [data scientists] don't use [machine learning] as a tool to describe reality, but rather as an abstract foundation for whatever theory they've come up with."
Replacing "math" with "machine learning" isn't going to make a difference if the practitioners don't understand how to use it properly. Machine learning models are much more subtle and complex than simple mathematical models and very easy to misuse. To use them properly, you really need a much stronger understanding of the math behind them than most people have.
See the entire field of psychology and most GWAS studies for an example of where over reliance on (simple) models can get you into a lot of trouble.
-Chris
I think you mean the other Social Sciences since Economics is one. It's the main reason the Left likes to trash the discipline as unscientific (whilst championing popular figures like Stiglitz, Krugman and Piketty). In truth they're all scientific: in part. They're equally comprised of philosophy.
By which you mean un-knowable magic.
There is no such thing, never has been. Someone is always lying and cheating, cartels will form, consumers do not have perfect information and make irrational choices. A free market is a complete lie, never has existed and never will.
Because you say so without proof or evidence. Society cannot exist without maintaining the structures which allow society to exist, and you can't pretend there is a self-regulating human condition which exists independent of society without being delusional.
As usual, Roman, you've brought your own brand of crazy to this conversation. You spout things as if they are objective natural facts, instead of bullshit premises you take to be true and expect the world to accept because you say it loudly and with bluster.
And as usual you utterly fail to introduce facts or evidence. Because you believe your magic bullshit is so self evident as to not need facts or evidence.
It's still not true.
And this is the fundamental flaw in your argument.
Nothing is "mis" allocated, it is allocated where the people who control them put them.
That this fails to match up with your perfect theory isn't a problem with the random, selfish, and stupid shit people do. It's a problem with your model which says people will achieve the perfect outcome your flawed model predicts.
What you're saying is, if people will only behave according to how you believe they should, there would be perfect outcomes. And I hate to tell you, but that will never happen.
I say there is no such thing as perfect outcomes, there is no such thing as rational actors, and there is no collective goal or "proper" allocation of anything. Merely a bunch of greedy, selfish, irrational actors doing whatever the hell makes sense at the time, or what they've been hoodwinked into believing will yield perfect outcomes. Many of whom will utterly fail to play by any rules or with anything other than pure, shortsighted greed.
If your theory can't account for the randomness of the human animal, it's your theory which is defective. Because the human animal will simply NEVER do anything according to someone's theory which has perfect outcomes.
The notion of "they're doing it wrong because it doesn't match my ideology" is the problem here.
Lost at C:>. Found at C.
It takes no account of the irrational human animal.
Quite the contrary; there are a few economic think tanks around that take plenty account of that and "tailor" (is "completely fabricate" too judgemental?) data to totally agree with the political stance of their primary benefactors.
Economics is not scientific in the mathematical sense. It takes no account of the irrational human animal.
That's entirely and demonstrably untrue. In 2002 the Nobel prize was awarded to Daniel Kahneman "for having integrated insights from psychological research into economic science, especially concerning human judgment and decision-making under uncertainty". Much recent economic study has been explicitly examining the irrationality of people and their economic decisions. Economics is a science but it is one where it is challenging to design experiments and so much of the data comes from empirical analysis.
That is why it is more like the weather than mathematics.
Mathematics is a language to describe things. Economics isn't a branch of mathematics any more or less than any other science. It is however a bit like studying the weather in that the forecasting models are trying to make predictions about a very complex and chaotic system. The models get their data from historical real world events but perfect accuracy in the models is nigh unacheivable.
Yeah, Economists (and I'm an avid amateur) use a language that is not precise to describe behavior that is not able to be precisely measured. Take the word "rational" for instance: Rationality in psychological terms implies something on the order of optimization in general, while rational in economic terms is optimization for individual behavior. Yes, an individual may change to the faster moving line or lane, but that only has a cumulative effect on the mass of behavior, and does not add to strategy as a whole. Individual behavior does not advance knowledge of the subject; it is only the accumulated effects of individual behaviors that reveal Economic behavior and may reveal Economic principles.
The closest thing we have in in tech terms is "emergent systems" and economics actually closely resembles an experiment in complex systems.
The other imprecision is to generalize about Economists. Some Economists are idealists or philosophers and deal in theory only, trying to determine "the way things should be." Some Economists are technologists and are mostly concerned about using the information to manipulate outcomes. (Many negative effects are a result of Economists who do both of these!) Somewhere in between these two types is the Economic Scientist whose main concern is finding out what is going on, proving that it is valid, and who can use the information to actually predict Economic behavior. It seems to me that machine learning actually serves the Economic Scientist.
"The mind works quicker than you think!"
Putting that stuff near "science" or "maths" is an insult to those fields of endeeavour.
Not any more than meteorology or ecology or geology or any other field that gets its data from complex and chaotic empirical sources.
Which economists predicted 2007/2008?
Quite a few of them. Some didn't get their timing right but I can introduce you to economists and financial analysts that I know personally that were warning about a likely crash in the housing market and knock on effects as far back as 2003. They obviously couldn't predict the exact outcome because that is basically impossible in a large chaotic system. (especially when you cannot perfectly model the initial state)
People have this naive idea that economists ought to be able to predict the future perfectly or it isn't a science. Predicting the recession of 2008 was something akin to a geologist trying to predict exactly when and where an earthquake will hit. There are too many unknowns to make anything better than a probabilistic analysis. They can tell you there is an X% chance of an event happening within time period Y. Asking for something more accurate than that is simply unrealistic expectations.
*Mathematics* isn't science. It is more properly a branch of philosophy that happens to be really, really useful for the sciences. The difference is that sciences are empirical---ideally, scientists observe the world, form explanations of their observations, then test those explanations with further observations. Mathematics is not empirical---mathematicians start from a set of fundamental assumptions, then use logic to deduce the consequences of those assumptions.
From the summary, it seems that the criticism is that economists are behaving more like mathematicians than like scientists---that is, they are making assumptions about how the world works, then using logic to determine the consequences of those assumptions. Instead, they should be making observations, then using the tools of mathematics analyze data taken from the real world and test their explanations.
Rhapsody in Numbers
The difference seems to be that while physicists are aware that their models are incomplete, economists (or, more likely, journalists, politicians, and the people who actually apply these models) etc. disregard these caveats and claim that this model describes the entire financial system in a few differential equations.
Economists don't disregard the limitations of their models at all. If you would spend some time speaking with actual economists (I have) you'd quickly find out that they are exquisitely aware of the limitations of their models.
Where things tend to go off the rails is when financial analysts with a profit motive try to stretch the economists models beyond what they can actually explain. A great example of this is Long Term Capital Management which was described in the book When Genius Failed. They took some models with a long list of assumptions and limitations and tried to apply the models to areas well beyond the limitations of the model. Early success begat hubris which led to greed and ultimately their downfall.
Actually the broken clock is the correct analogy here. Austrian economics predicts a crash not quite constantly, but under a certain set of conditions which very commonly exists - including a lead-up to the '08 recession, and right now. Their predictions that economies are sure to melt down "any minute now" are rarely correct.
"When information is power, privacy is freedom" - Jah-Wren Ryel
What do you mean? Keynes modeled stagflation; he didn't use that term, but it's clear that Keynes, and those who studied his work, were aware of the effect of a supply shock on an economy.
The issue with Keynesian policy and stagflation is, given two problems with conflicting resolutions, how do you address both of them?
We now know that tackling them one at a time works. First you address inflation, then you address stagnation. This isn't a weakness of Keynesian theory -- it's validation.
"Trolls they were, but filled with the evil will of their master: a fell race..." -- J.R.R. Tolkien on Olog-hai
Economics is not a science, no matter how many times you and your finance buddies tell yourselves that it is.
A science is a formal method of examining and describing the world around us. This method requires hypothesis regarding empirical data which are testable and repeatable. This process is called the scientific method. Economics follows the scientific method and any field of study that follows the scientific method can accurately be called a science. Ergo economics is a science by definition. Some sciences are easier to study than others but that has no bearing on their validity as a science.
That's not a real Nobel prize either, it's a self-aggrandizing fraud that was purchased by the banking industry in an attempt to convince the public that their twisted methods were legitimate and not just common crime.
Since the Nobel organization recognizes the Economics prize it is as real as any other Nobel prize. Your opinion on the matter is meaningless.
What do you mean "most"? I put Economists in the same fraudster category as Psychics and Clergy and Life Coaches (and somewhat worse than Psychologists). Lets not do anything that might pretend to legitimize these can artists.
Despite the examples set by the economists visible in government roles, private economists are actually proving to be valuable in certain sectors; specifically, MMORPGs. Bringing in an economist to help set up, monitor, and maintain the currencies, item prices, trade values, treasure drops, etc., can make the difference between a fun game and an endless grind.
John