4 ms·

> Economics has a reputation as a dry, heartless subject, full of boring equations. Not if you have any training in a hard science: then you think of economics

> Economics has a reputation as a dry, heartless subject, full of boring equations. Not if you have any training in a hard science: then you think of economics as something largely devoid of equations and made up of lots of ad-hoc regression analysis and hand waving. This might not be the fault of the economists but of the difficulty of the subject matter of course. But economists really should learn to accompany their predictions with error bars as every undergraduate physicist must.

imtringued 4y ago
Economists have "hard science" models but those are empirically unreliable. Honestly the boring answer is that e.g. liquidity is the ability to change your mind instantly, which also means that any prediction can become invalid at any given moment. If you want to increase predictability, you would have to encourage people to make decisions as soon as possible and stick with those decisions for as long as possible.
fezfight 4y ago
As a mechanical engineer (not a hard science), economics has always looked more like history than science. As in, you can build models to show what happened, but not what will happen. Edit: I think Ive had my mind changed enough in this thread to edit this. It's obviously a very deep subject and I've been convinced it has predictive qualities. I definitely dont think of dry equations though, more like psychology, and politics. But I'm looking forward to learning about it and possibly having my mind changed yet further.
- ChadNauseam 4y ago
  Econ is great at predicting the future. Let’s say your uncle tells you about his new stock trading strategy that’s guaranteed to make him 10% returns per month. Will he be rich two years from now? Econ’s efficient market hypothesis predicts he will not. Or let’s say the president of Japan announces a plan to beat deflation once and for all. He says that next year, he’ll fire their central bankers and hires the ones from Venezuela, give them a 100% annual inflation target, and promise them millions of USD in bonuses if they meet or exceed that target. Even though this plan is going into effect next year, what will you see the next day? (Assuming people believe the president is serious.) Econ predicts you will start seeing inflation immediately. And we see effects like these all the time when looking at how the actions of central bankers effects the economy – if Powell says he’s going to change interest rates, there’s a reaction in the markets when he makes the announcement and not one when he makes the actual change.
  - thfuran 4y ago
    >Let’s say your uncle tells you about his new stock trading strategy that’s guaranteed to make him 10% returns per month. Will he be rich two years from now? That depends on whether the stock he bought was GME.
  - fezfight 4y ago
    So a bit of psychology in there, eh? Pretty neat field, if you ask me.
  - tsimionescu 4y ago
    These are not predictions, they are assumptions. The difference being they are in no way quantifiable. Econ has lots of equations, but it can't actually predict anything beyond some broad directions like you show, and even those are sometimes wrong. There was even a story (no idea if true, so maybe of limited relevance...) of a reverse bank run happening in Japan, with people rushing to deposit money to save a failing bank.
- colinmhayes 4y ago
  There's a big difference between micro and macro. Macro is basically predictions as you say because in the end it's largely underpinned by market expectations which are not always rational. Micro on the other hand looks at the choices of individuals, which are by definition at least somewhat rational. Individuals will always make the choice that they believe will maximize their utility, so the whole thing is really just equations whose derivative you set equal to 0 in order to maximize/minimize.
  - JumpCrisscross 4y ago
    > which are by definition at least somewhat rational Or can be demonstrably arbitraged to take advantage of the irrationality in a tangible, quantifiable and repeatable way. Put-call parity, substitution, et cetera.
influxmoment 4y ago
"error bars" are only relevant if you have a reliable model and theory. Economists buying into their own models of humans is the problem
Gimpei 4y ago
Before you lay into an entire academic discipline, it would help to understand it a little better. Fine if you’re just bagging on DSGE models, since they really are garbage (although I would argue that it is because of all the equations involved). But the vast majority of economics is concerned with causal analysis where there are no predictions. And everybody reports standard errors and has for decades.
- sudosysgen 4y ago
  Do the standard errors actually hold up empirically?
erdos4d 4y ago
This. The subject is definitely not the rigorous science that its practitioners would like you to believe. They mostly just fit some sort of regression model and analyze it, but rarely is there any serious work to demonstrate that reality actually obeys the model. I understand that proving the real world actually conforms to theory is very hard, but that is what makes science what it is, and why science is really hard.
- nequo 4y ago
  Out of curiosity, could you give a few examples of published papers that uncritically reported regression estimates?
  - erdos4d 4y ago
    No I don't want to read through that crap anymore, I had enough in grad school helping econ friends. What I'm talking about is that they always make a bunch of simplifying assumptions about the world and then fit their model and analyze. But what they almost never do is seriously justify the assumptions. One popular paper is to do "Assuming the world follows theory X, we propose this tweaked version of model Y from paper Z and show how it better predicts this dataset of 600 points from the world bank." I knew numerous people in grad school who played that game. There is no way the tiny little dataset they had could justify any of this, but that's what publishes in econ. And if you ask them about this, they will tell you that they only had this tiny little dataset because that's all that is publicly available and so on. That's fair, but also gives them a huge dodge to hide behind. That's what I saw of the field and I came away thinking that it stank.
nequo 4y ago
> But economists really should learn to accompany their predictions with error bars Virtually all published papers in economics that use regression analysis which you mention include "error bars." The discipline calls these standard errors and confidence intervals. If a paper is non-parametrically estimating functions rather than single parameters, they might also report confidence bands.[1] Standard errors may be calculated analytically (resting mostly on asymptotic theory) or numerically (using a bootstrap estimator for them). There is a large literature on the appropriate standard errors for various contexts, with clustered errors being mainstream in applications.[2][3] These are taken seriously in peer review, and published papers come with long lists of robustness checks which turn papers into book-length publications when you include the appendices. [1] https://cattaneo.princeton.edu/papers/Cattaneo-Crump-Farrell-Feng_2022_Binscatter.pdf https://cattaneo.princeton.edu/papers/Cattaneo-Crump-Farrell... is one example. [2] See https://economics.mit.edu/files/13927 https://economics.mit.edu/files/13927 for a recent discussion. [3] See https://gregoryeady.com/ResearchMethodsCourse/assets/readings/Cameron,%20A.%20Colin%20et%20al.%20-%202008%20-%20Bootstrap-Based%20Improvement%20for%20Inference%20with%20Clustered%20Errors.pdf https://gregoryeady.com/ResearchMethodsCourse/assets/reading... for bootstrap-based methods.
- kwhitefoot 4y ago
  I stand corrected. Thank you.

imtringued 4y ago

Economists have "hard science" models but those are empirically unreliable. Honestly the boring answer is that e.g. liquidity is the ability to change your mind instantly, which also means that any prediction can become invalid at any given moment. If you want to increase predictability, you would have to encourage people to make decisions as soon as possible and stick with those decisions for as long as possible.

fezfight 4y ago

As a mechanical engineer (not a hard science), economics has always looked more like history than science. As in, you can build models to show what happened, but not what will happen. Edit: I think Ive had my mind changed enough in this thread to edit this. It's obviously a very deep subject and I've been convinced it has predictive qualities. I definitely dont think of dry equations though, more like psychology, and politics. But I'm looking forward to learning about it and possibly having my mind changed yet further.

ChadNauseam 4y ago
Econ is great at predicting the future. Let’s say your uncle tells you about his new stock trading strategy that’s guaranteed to make him 10% returns per month. Will he be rich two years from now? Econ’s efficient market hypothesis predicts he will not. Or let’s say the president of Japan announces a plan to beat deflation once and for all. He says that next year, he’ll fire their central bankers and hires the ones from Venezuela, give them a 100% annual inflation target, and promise them millions of USD in bonuses if they meet or exceed that target. Even though this plan is going into effect next year, what will you see the next day? (Assuming people believe the president is serious.) Econ predicts you will start seeing inflation immediately. And we see effects like these all the time when looking at how the actions of central bankers effects the economy – if Powell says he’s going to change interest rates, there’s a reaction in the markets when he makes the announcement and not one when he makes the actual change.
- thfuran 4y ago
  >Let’s say your uncle tells you about his new stock trading strategy that’s guaranteed to make him 10% returns per month. Will he be rich two years from now? That depends on whether the stock he bought was GME.
- fezfight 4y ago
  So a bit of psychology in there, eh? Pretty neat field, if you ask me.
- tsimionescu 4y ago
  These are not predictions, they are assumptions. The difference being they are in no way quantifiable. Econ has lots of equations, but it can't actually predict anything beyond some broad directions like you show, and even those are sometimes wrong. There was even a story (no idea if true, so maybe of limited relevance...) of a reverse bank run happening in Japan, with people rushing to deposit money to save a failing bank.
colinmhayes 4y ago
There's a big difference between micro and macro. Macro is basically predictions as you say because in the end it's largely underpinned by market expectations which are not always rational. Micro on the other hand looks at the choices of individuals, which are by definition at least somewhat rational. Individuals will always make the choice that they believe will maximize their utility, so the whole thing is really just equations whose derivative you set equal to 0 in order to maximize/minimize.
- JumpCrisscross 4y ago
  > which are by definition at least somewhat rational Or can be demonstrably arbitraged to take advantage of the irrationality in a tangible, quantifiable and repeatable way. Put-call parity, substitution, et cetera.

influxmoment 4y ago

"error bars" are only relevant if you have a reliable model and theory. Economists buying into their own models of humans is the problem

Gimpei 4y ago

Before you lay into an entire academic discipline, it would help to understand it a little better. Fine if you’re just bagging on DSGE models, since they really are garbage (although I would argue that it is because of all the equations involved). But the vast majority of economics is concerned with causal analysis where there are no predictions. And everybody reports standard errors and has for decades.

sudosysgen 4y ago
Do the standard errors actually hold up empirically?

erdos4d 4y ago

This. The subject is definitely not the rigorous science that its practitioners would like you to believe. They mostly just fit some sort of regression model and analyze it, but rarely is there any serious work to demonstrate that reality actually obeys the model. I understand that proving the real world actually conforms to theory is very hard, but that is what makes science what it is, and why science is really hard.

nequo 4y ago
Out of curiosity, could you give a few examples of published papers that uncritically reported regression estimates?
- erdos4d 4y ago
  No I don't want to read through that crap anymore, I had enough in grad school helping econ friends. What I'm talking about is that they always make a bunch of simplifying assumptions about the world and then fit their model and analyze. But what they almost never do is seriously justify the assumptions. One popular paper is to do "Assuming the world follows theory X, we propose this tweaked version of model Y from paper Z and show how it better predicts this dataset of 600 points from the world bank." I knew numerous people in grad school who played that game. There is no way the tiny little dataset they had could justify any of this, but that's what publishes in econ. And if you ask them about this, they will tell you that they only had this tiny little dataset because that's all that is publicly available and so on. That's fair, but also gives them a huge dodge to hide behind. That's what I saw of the field and I came away thinking that it stank.

nequo 4y ago

> But economists really should learn to accompany their predictions with error bars Virtually all published papers in economics that use regression analysis which you mention include "error bars." The discipline calls these standard errors and confidence intervals. If a paper is non-parametrically estimating functions rather than single parameters, they might also report confidence bands.[1] Standard errors may be calculated analytically (resting mostly on asymptotic theory) or numerically (using a bootstrap estimator for them). There is a large literature on the appropriate standard errors for various contexts, with clustered errors being mainstream in applications.[2][3] These are taken seriously in peer review, and published papers come with long lists of robustness checks which turn papers into book-length publications when you include the appendices. [1] https://cattaneo.princeton.edu/papers/Cattaneo-Crump-Farrell-Feng_2022_Binscatter.pdf https://cattaneo.princeton.edu/papers/Cattaneo-Crump-Farrell... is one example. [2] See https://economics.mit.edu/files/13927 https://economics.mit.edu/files/13927 for a recent discussion. [3] See https://gregoryeady.com/ResearchMethodsCourse/assets/readings/Cameron,%20A.%20Colin%20et%20al.%20-%202008%20-%20Bootstrap-Based%20Improvement%20for%20Inference%20with%20Clustered%20Errors.pdf https://gregoryeady.com/ResearchMethodsCourse/assets/reading... for bootstrap-based methods.

kwhitefoot 4y ago
I stand corrected. Thank you.