Great analysis. Economics seems like untapped territory for ML models and the like. As a data scientist I'm looking forward to seeing what other applications you find!
Create a clear "Consensus"? Even when a proponent like you who can't describe it in many paragraphs. Two-consecutive-significant-declines-in-GDP may miss a few mild recessions (which are arguably indistinguishable from brief STAGNATIONS or substantial regional- and sectoral-confined downturns). But since RECESSION (which used to be labeled DEPRESSION) rightfully needs NO ECONOMIC DEFINITION because IT'S NOT AN ECONOMIC CONCEPT. No two business cycles are alike, nor are they cyclic (the worst ones were a "double-dip" in early 1980s and combined with a global FINANCIAL PANIC and real estate meltdown in the later 2000s, while 1970s unique STAGFLATION needed that new term; each has a different set of "causes." Fiscal and Monetary policy denies this bivariate distinction from CONTINUOUS measures!
I think the main point I'm getting at here is that if we are going to use "recession" as a term, then it would be important for us to know what this actually means. If we want to use my ML model as the definition (not saying I recommend this, but just for the sake of argument), then at least this would be a consistent definition that we could easily point to. On the other hand, if recession does literally have no meaning, then we should stop classifying any period as recessionary. Which is to say if you're right, good point. If you're wrong and we want to have a way to understand broadly whether our economy is doing poorly and want to label it as recessionary, we should at least have a clear avenue to doing this. Even if no two recessions are alike I can easily turn around and say no two snowflakes are alike but it doesn't take a genius to be able to tell whether or not something is snow.
Everything else in the science of Economics doesn't impose a step function (in-or-out of recession) doesn't have a base point for high or adequate growth, acceptable trade deficits, or fair enough income distribution. Those are each NORMATIVE societal standards which diverge by political parties and news network editorial policy. Even "FULL EMPLOYMENT" and the Fed's "TARGET INTEREST or INFLATION" rates for an "independent" Fed is heavily divided by political party values and current economic conditions (data and "POSITIVE" economics); the FULL employment objective, for example, varied from 5.5% by the GOP (lower for Dems) during the 1970s and 80s but in this century has been 3.5 or 4% for both parties. As a result, Dem presidents post-WW2 achieved much lower unemployment but GOP policy cares most about growth and low inflation (until Trump). No need to ever DEFINE recession. But press, media, and politicians NEED and use a crude, fixed, common sense Recession definition. Letting AI tools "validate it" does nothing for conspiracy-loving American voters!
The social scientist economists don't get it and may never. All attempts should be made to craft econ into a hard (or less soft) science. Machine Learning's ability to handle multivariate analysis will change this field forever. Never stop.
Uh, I came from quantum mechanics research programming simulations of molecular wave functions. Other economists today don't even do policy analysis at Policy Centers anymore. It's all mathematicians applying math badly to econ modeling without ever designing PRIMARY data collection (it's always tested with SECONDARY DATA collected by others for other uses) and rely on "new" Classical Economics from the 19th Century to validate IDEOLOGY. When their models yield policy prescriptions that consistently fail to improve the real world or even forecast it correctly, they blame reality (government the cause of all evil)!
Great analysis. Economics seems like untapped territory for ML models and the like. As a data scientist I'm looking forward to seeing what other applications you find!
Create a clear "Consensus"? Even when a proponent like you who can't describe it in many paragraphs. Two-consecutive-significant-declines-in-GDP may miss a few mild recessions (which are arguably indistinguishable from brief STAGNATIONS or substantial regional- and sectoral-confined downturns). But since RECESSION (which used to be labeled DEPRESSION) rightfully needs NO ECONOMIC DEFINITION because IT'S NOT AN ECONOMIC CONCEPT. No two business cycles are alike, nor are they cyclic (the worst ones were a "double-dip" in early 1980s and combined with a global FINANCIAL PANIC and real estate meltdown in the later 2000s, while 1970s unique STAGFLATION needed that new term; each has a different set of "causes." Fiscal and Monetary policy denies this bivariate distinction from CONTINUOUS measures!
I think the main point I'm getting at here is that if we are going to use "recession" as a term, then it would be important for us to know what this actually means. If we want to use my ML model as the definition (not saying I recommend this, but just for the sake of argument), then at least this would be a consistent definition that we could easily point to. On the other hand, if recession does literally have no meaning, then we should stop classifying any period as recessionary. Which is to say if you're right, good point. If you're wrong and we want to have a way to understand broadly whether our economy is doing poorly and want to label it as recessionary, we should at least have a clear avenue to doing this. Even if no two recessions are alike I can easily turn around and say no two snowflakes are alike but it doesn't take a genius to be able to tell whether or not something is snow.
Everything else in the science of Economics doesn't impose a step function (in-or-out of recession) doesn't have a base point for high or adequate growth, acceptable trade deficits, or fair enough income distribution. Those are each NORMATIVE societal standards which diverge by political parties and news network editorial policy. Even "FULL EMPLOYMENT" and the Fed's "TARGET INTEREST or INFLATION" rates for an "independent" Fed is heavily divided by political party values and current economic conditions (data and "POSITIVE" economics); the FULL employment objective, for example, varied from 5.5% by the GOP (lower for Dems) during the 1970s and 80s but in this century has been 3.5 or 4% for both parties. As a result, Dem presidents post-WW2 achieved much lower unemployment but GOP policy cares most about growth and low inflation (until Trump). No need to ever DEFINE recession. But press, media, and politicians NEED and use a crude, fixed, common sense Recession definition. Letting AI tools "validate it" does nothing for conspiracy-loving American voters!
This isn't an AI model. It's machine learning, which economists use any time they do a linear regression.
The social scientist economists don't get it and may never. All attempts should be made to craft econ into a hard (or less soft) science. Machine Learning's ability to handle multivariate analysis will change this field forever. Never stop.
Uh, I came from quantum mechanics research programming simulations of molecular wave functions. Other economists today don't even do policy analysis at Policy Centers anymore. It's all mathematicians applying math badly to econ modeling without ever designing PRIMARY data collection (it's always tested with SECONDARY DATA collected by others for other uses) and rely on "new" Classical Economics from the 19th Century to validate IDEOLOGY. When their models yield policy prescriptions that consistently fail to improve the real world or even forecast it correctly, they blame reality (government the cause of all evil)!
Thanks for sharing! As you know, I've been talking about this for a while but didn't have the data to support my hypothesis. Well done!