Was talking with an Econ PhD student recently, and it sounds like there's still a gap between economic training and modern statistical methods (for most students). I wonder if a large portion of econ PhD's opting for tech rather than academia will change that (quicker).
Can you elaborate on that? If you mean that ML stuff isn't taught much in coursework, that's probably true --- but getting that stuff to work for economic questions isn't exactly trivial (e.g. we only have observational data but want to estimate the effects of hypothetical policy interventions).
If you mean something else by "modern statistical methods" I'd be curious to know what you mean.
Well, we talked about using cross-validation and the bootstrap/bagging methods. I was surprised that he hadn't heard of these ideas, much less using.
Deep learning variants are definitely farther out; I think the current lack of interpretability makes for a real strenous case wrt economic applications.
These methods are not usually taught in economics courses.
Why?
Not enough data.
In most cases economists are dealing with survey data, time series data and panel data. The benefits of cross validation and bootstrap/bagging grow with data size. When you're dealing with minimal amounts of messy/misbehaving data these methods lose their power.
Other methods become more important ie: Instrumental Variable estimation, Probit and Tobit models, Vector Autoregressions, Vector Error Correction models. Im sure your econ PhD student friend would know what these are.
"Not enough time" and "harder to do inference" are bigger reasons. It's hard to see why cross validation would do worse than the AIC or BIC for lag length selection in VARs, the bootstrap is widely used for inference for all of the models you mentioned, IV isn't known for its exceptional small sample properties, etc. People are working on this stuff, but it takes a little while to get it to work well for Econ research, and it takes some clear new empirical findings before it becomes mainstream enough to teach it in classes. Everyone recognizes that there's a lot of promise, though.
" Asymptotically, minimizing the AIC is equivalent to minimizing the CV value. This is true for any model (Stone 1977), not just linear models. It is this property that makes the AIC so useful in model selection when the purpose is prediction."
Thanks, that's a bit surprising, but definitely helped me make up my mind on what I'll teach in the last few weeks of my class next year. The bootstrap is a weird omission, but economists usually only see it for variance estimation and inference, not in other contexts.
Interesting, I think bootstrapping is overused in financial modeling, because most long term financial time series don't obey the assumptions, covariance stationarity, normal distribution of residuals etc. A lot of regime changes, fat tails, latent predictors that you don't realize are important until they break your model.
disclaimer: hand-waving
OTOH there are a lot of situations where you need to model something with a lot of potential predictors and limited data. e.g. testing a macro model with 100s of potential predictors and 100 years of relatively poor macro data.
ML might find interesting relationships in those cases.
Traditional statistics has a strong theoretical foundation, you assume a bunch of things about the shape of the data, and you can prove your estimator is best and what the error looks like based on the amount of data. It makes heroic assumptions about underlying data that we know don't apply.
So it often doesn't work well but we know why.
ML just wants to find things that work well in cross-validation without worrying too much about proofs... ML is a little like QE ... it works but we don't really know why.
Anyway, any sufficiently complex ecosystem is a market design, makes sense that if you have a sufficiently valuable ecosystem you would want some people who study markets.
I had always presumed the vast majority of economists working in tech were simply doing transfer pricing... shuffling profits overseas to minimize taxes.
Some, yes. However, a lot of economist work goes into real-time ad auction platforms, marketing, fraud detection, and many other places where human interaction and incentives meet automated systems.
As an economist in the technology/data science space, I've found that the training and methodology brings a lot to table, such as the feature engineering side of ML or investigating business processes for incentive misalignment.
I studied econ undergrad but don't see much reason to get a PhD outside of academia or leveraging a public sector job such as a Fed job into a private sector finance role. I can see some of the technical skills such as survey construction/analysis, bayesian analysis/higher level statistics being somewhat useful, but a lot of that is also very role specific and can be learned on the job, no? Thinking with a game-theory/incentive-centered mindset is something that does not require too much formal econ knowledge.
Econ's undergrad degree is remarkably unrepresentative of Econ grad school. But every phd program in any subject is training you to do research first and foremost, so don't do it for technical training.
The level of math. An Econ/math double major is about the right background for an Econ phd program. (I was a math major in undergrad and didn't feel like a lack of much exposure to Econ hurt me much in grad school, but it would have given me better perspective on the field.) Many Econ undergrads want a "businessey" degree and don't want to take math beyond calculus, so trying to teach "real" macro (which would require coursework equivalent to differential equations) or "real" micro (similar to but less fussy than real analysis) or even empirical research (statistical tools that build on regression) would be rough going. So we teach "intuition" and watered down versions of the results and punt on justification. Plus the "core" undergrad courses have a large service component --- we need good business students to be able to take intermediate micro, for example, so unilaterally ramping up the required smath wouldn't be feasible even if some of the students wanted it.
tl;dr phd programs can spend a lot of time teaching students how to use and extend mathematical tools and can cherry pick students with the right backgrounds. Undergraduate programs have a big service component, have to assume a range of mathematical preparation, and have to actually teach "economics" in their core courses -- game theory needs to teach game theory, not how to be a game theorist.
Econometrics is essentially a field of applied math. Statistics is also pretty closely related to econ, though it sounds like theres a wide range of statistics backgrounds between programs. I had a roommate that studied econ/statistics and always thought it was pretty neat some modern algorithms didn't come from computer scientists, but from statisticians (particularly Random Forests).
i used to work at an institutional investment firm. a huge one. (say it like trump) huuuuuge. we had lots of economists. i chilled with them. i partied with them. and when we partied, when the booze flowed and we all shared sordid stories about how our jobs really went down, they basically said the work they did was, meh, adding very little, and at times no, marginal value and that most of the time the traders did their own thing regardless.
also, lets not forget what happened to south america when the chicago school of economics prescribed its bitter medicine. they're still recovering/reeling both economically and culturally.
so, i shudder when i read that some of the most influential and active corporations, those determining the way of our society/culture and economy, are adding all these PhDs from a discipline whose track record is absolutely crap from both a social/cultural and an economic angle. (full disclosure i have a b.s. in economics, so, yes, i understand some of what an economist does)
The kinds of economists hired by these places are basically just applied statisticians with more practical training in causal inference. Microsoft/Google/Amazon aren't hiring macroeconomists to forecast interest rates or whatever; they're (mostly) hiring microeconomists who can estimate causal effects and design experiments.
Also, any applied economist at a top tech firm knows undergrad level applied ML.
How many economists are there at Google, compared to other non-engineering professions at Google, say chefs, cleaners, doctors, physiotherapists, accountants, secretaries, barbers, lifeguards etc.?
"Athey says Microsoft is another company with “at least a dozen” economists on staff. She knows of nearly 100 economists employed by tech firms, the big majority joining after 2010."
A few dozen economists for a company the size of Microsoft or Google, and 100 across all the major firms seems few when you consider the myriad and quantity of other non-economic non-engineering staff these technology companies hire.
I think the point is not the absolute size of the number (would you expect PhD economists to be comparable in number to most other professions you mentioned?), but the recent hiring trend. I think it's interesting for a few reasons. One is that tech companies have a bit of a historical tendency to make economics decisions without much input from economists. Another is the notable big exceptions such as Varian at Google and the generalized second price ad auction. A final one is the interaction between econ and CS in terms of methods, viewpoints, priorities, decisions. (Disclaimer: I've collaborated with economists and computer scientists at both companies you mention.)
The most amazing example of such cooperation is Yanis Varoufakis, who was working for Valve on virtual currencies and economy of Steam Market before becoming Minister of Finance in Greece.
In his own words: http://blogs.valvesoftware.com/economics/it-all-began-with-a...
I saw Gabe make a presentation at the University of Texas a few years ago, and Yanis was sitting in the front row. Not being an economist myself, I had no good questions for them. And I gather that Gabe was a little annoyed at all the questions around upcoming releases and not the topic of his speech.
In hindsight, I would have asked Yanis if he was overwhelmed by the detail of the data that Valve collects on their market - both the Steam store and the in-game markets. Conventional public economic data would seem to have to be used with caution, as there are people at all levels looking to sway the numbers in different directions (ahh .. politics!), but the Valve data would have been base numbers without any "processing". And there's a lot of it.
27 comments
[ 3.2 ms ] story [ 72.1 ms ] threadIf you mean something else by "modern statistical methods" I'd be curious to know what you mean.
Deep learning variants are definitely farther out; I think the current lack of interpretability makes for a real strenous case wrt economic applications.
I don't really want to draw conclusions from n=1 convos but a couple other links had piqued my curiosity towards this relationship: https://www.quora.com/How-will-Machine-Learning-affect-econo..., https://news.ycombinator.com/item?id=11460412
Other methods become more important ie: Instrumental Variable estimation, Probit and Tobit models, Vector Autoregressions, Vector Error Correction models. Im sure your econ PhD student friend would know what these are.
A different tool-kit to solve different problems.
" Asymptotically, minimizing the AIC is equivalent to minimizing the CV value. This is true for any model (Stone 1977), not just linear models. It is this property that makes the AIC so useful in model selection when the purpose is prediction."
disclaimer: hand-waving
OTOH there are a lot of situations where you need to model something with a lot of potential predictors and limited data. e.g. testing a macro model with 100s of potential predictors and 100 years of relatively poor macro data.
ML might find interesting relationships in those cases.
Traditional statistics has a strong theoretical foundation, you assume a bunch of things about the shape of the data, and you can prove your estimator is best and what the error looks like based on the amount of data. It makes heroic assumptions about underlying data that we know don't apply.
So it often doesn't work well but we know why.
ML just wants to find things that work well in cross-validation without worrying too much about proofs... ML is a little like QE ... it works but we don't really know why.
Anyway, any sufficiently complex ecosystem is a market design, makes sense that if you have a sufficiently valuable ecosystem you would want some people who study markets.
As an economist in the technology/data science space, I've found that the training and methodology brings a lot to table, such as the feature engineering side of ML or investigating business processes for incentive misalignment.
tl;dr phd programs can spend a lot of time teaching students how to use and extend mathematical tools and can cherry pick students with the right backgrounds. Undergraduate programs have a big service component, have to assume a range of mathematical preparation, and have to actually teach "economics" in their core courses -- game theory needs to teach game theory, not how to be a game theorist.
Apparently there's a lot of mathematical overlap between grad school econ and fields in Big Data: computer vision, machine learning, etc.
also, lets not forget what happened to south america when the chicago school of economics prescribed its bitter medicine. they're still recovering/reeling both economically and culturally.
so, i shudder when i read that some of the most influential and active corporations, those determining the way of our society/culture and economy, are adding all these PhDs from a discipline whose track record is absolutely crap from both a social/cultural and an economic angle. (full disclosure i have a b.s. in economics, so, yes, i understand some of what an economist does)
the suits are taking over.
Pay a bit more attention, or be a bit more honest.
Also, any applied economist at a top tech firm knows undergrad level applied ML.
"Athey says Microsoft is another company with “at least a dozen” economists on staff. She knows of nearly 100 economists employed by tech firms, the big majority joining after 2010."
A few dozen economists for a company the size of Microsoft or Google, and 100 across all the major firms seems few when you consider the myriad and quantity of other non-economic non-engineering staff these technology companies hire.
In hindsight, I would have asked Yanis if he was overwhelmed by the detail of the data that Valve collects on their market - both the Steam store and the in-game markets. Conventional public economic data would seem to have to be used with caution, as there are people at all levels looking to sway the numbers in different directions (ahh .. politics!), but the Valve data would have been base numbers without any "processing". And there's a lot of it.