Sorry if this is an amateur question, but how does a tax on corporate profit distort investment incentives?
Large companies have access to debt financing. If they make investment decisions based on some hurdle rate of return, it shouldn't really matter whether they have a lot of retained earnings to invest.
My hunch is that there's an important market failure here, which consists of CEOs overestimating their own business acumen and the prospects of their own sector. It seems as if it would be more efficient for companies to pay out most of their earnings as dividends and then go back to the capital markets when they have what they think is a worthwhile plan for expansion.
So wouldn't the ideal policy be one that allowed corporations to deduct dividend payments (as they already can for debt interest) but then taxed retained earnings at a high rate? It would counter the natural bias of managers towards using shareholder profits to build their own empires, even when that's not an optimal allocation of capital.
Yes, but it's a tax on capital that was not converted into labor or capital goods. Just about every corporate tax scheme I've run into taxes net profits, not revenue, and assumes that money spent on equipment or research is not taxed.
This is way above my pay grade but I think these models are looking at a different issue. They discuss the optimal tax rate on individuals' capital income (of all types).
What I was wondering about is the justification for taxing this specific form of capital income twice: first when the corporation earns it and then when the shareholder earns it.
Regardless of what the optimal tax rate on capital income may be, it seems as if it would be distortionary to tax *distributed* corporate earnings at a higher rate than other forms of capital income.
But I think a punitive tax on *retained* earnings might be antidistortionary, because managers have a natural tendency to increase their own power and prestige at the expense of shareholder value. I think this bias is documented, for example, by the fact that corporate mergers are often wealth-destroying.
The idea is that all companies are owned by households, so taxing their profits is the same as taxing household capital income. Obviously this is not really true, and there are reasons that corporate taxes don't work like, say, capital gains or dividend taxes. But the basic idea is there: When you tax profits, you tax the value of productive capital, so less gets accumulated over time.
That's true. There are overhead costs to accessing a bank loan or issuing a bond, and I suppose they don't rise proportionately to the sum being raised.
So I think it might be optimal to impose a heavy tax on the retained earnings of companies, but only above a certain threshold. There should be a zero bracket large enough to cover all the earnings of a small business.
(The market failure I'm talking about is a principal/agent problem, which shouldn't be a major issue for small businesses anyway.)
No shame admitting I am a very recent (very happy) reader, I followed a link here from a Paul Krugman article about the Sri Lanka currency crisis. Thanks for the excellent and thoughtful content!
Thinking back it could have been Steve Williamson who sent me to Noahpinion - was studying Macro 3 and his textbook was making life hell at the time. I remember commenting back in the day on a Noah post and Steve Williamson responded to my comment and I was like this is wild.
GPT-3 comes from OpenAI not Google, btw. (OpenAI started as an anti-evil AI thing apparently because Elon’s been hanging out with lesswrong posters, but turned into a product company part owned by Microsoft or something.)
> gptI’m in the mood to write about some econ papers today.
I think the 4 centuries of data is definitely enough to show TFP is exponential. With 4 parameters I can fit an elephant; if you need 4 different slope parameters in your piecewise-linear function, it's not looking good for your theory. I could fit better than that paper if you gave me 4 parameters too!
From what I can tell the paper's tests of functional form are dramatically biased because they don't account for the long memory of TFP. After removing the exponential trend, TFP isn't *quite* nonstationary, but it's very close to it, and needs fractional differencing to get good estimates of standard errors. You can have very, very long deviations before regression to the mean kicks in, which can give the appearance of linearity over short timespans.
I don't think this shows that 4 centuries isn't enough to get the right answer. I think it shows that if you're really inefficient with your data, it can take you more than 4 centuries to figure out the right answer.
I still think that Matt Yglesias had this right, and in fact very high tax rates on individuals' capital income and corporate profits drive companies to sink excess cash into R&D expenses that deduct from profits, rather than paying tax on the profits and then letting their investors pay even more tax on dividends.
I know those weren't your words, as you put in a graph from Cabellero, but why would 4% be an inflation "disaster"? It seems like a moderately higher, but predictable inflation rate would not reduce growth and would The Fed more maneuvering room in the case of recessions.
4% inflation is what we had _after_ the Volcker disinflation, during the "Morning In America" recovery. I am seriously baffled as to why people think it would be so terrible. After we get inflation back down to 2% for a year or so, the Fed should announce they're raising the target to 3, as well as continuing to treat it as a target, not a ceiling.
(Or, even better, they should just move to NGDP targeting, aiming for something like 5%, typically hoping for something like half growth and half inflation.)
Does Ngdp targeting ignore inflation or try to do something specific about it? I mean there is the standard formula real equals nominal minus prices or inflation. This formula can be appplied to gdp amongst other things.
I'm not sure whether this link works if you're not a Slow Boring subscriber, but Matt Yglesias has a fairly good summary of the case for NGDP targeting:
Noah might know some other economists who've described this.
But basically yes, the idea is that when the Fed decides whether they need to tighten or loosen monetary policy, they should look at the growth rate of nominal GDP (i.e. how much money is circulating through the economy) rather than the rate of change in prices. At a basic level this is an easier quantity to measure (you don't have to worry about differences between "core" or "expectations-based" inflation versus volatile commodities, and hedonic adjustments, and so on), and in some sense it bears a more direct relationship to monetary policy. (You can go back to Milton Friedman's famous "PQ=MV" formula.) And if your NGDP growth target is, say, 5%, then basically it's the Fed's job to try to keep that on track, and then leave it up to private actors, and fiscal policy, to determine whether that ends up being 2% real growth and 3% inflation, or 1% real growth and 4% inflation... It also leaves room for the Fed to not freak out and apply thumbscrews to the economy if we have a burst of minimally inflationary growth.
"Vollrath shows that if you assume that TFP is equal to the natural log of the number of ideas, and if ideas give birth to each other at a constant rate, you get perfectly linear growth. Cool result!"
Somehow this one paragraph did more to undermine my confidence in you, Noah, than all the mistakes you discussed in your recent roundup. That's not a result, it's the definition of logarithm -- and of course "natural log" is irrelevant, any logarithm would work here. Maybe it's just inartfully worded, but MAN, how inartfully!
I think a significant reason we don't have exponential growth, or at least higher order than linear, is that we (humanity) does not have a way to take advantage of the non-rival quality of intellectual capital. I have an economic model I wrote down with coupled differential equations in which intellectual capital is, as is the natural model, non-rival. This is a physics based model where we only consider the real entities: labor, capital, production, consumption. I break this down into physical and intellectual components, so for instance new physical capital is produced by combining labor and intellectual capital and consuming some physical capital. Because it is a mean field model new intellectual capital is instantly fed back into the equations for producing new physical and new intellectual capital.
But in real life we artificially make intellectual capital restricted, i.e. it doesn't feed back uniformly across all entities, through the use of patents and trade secrets and non-compete clauses etc. Basic research science which I did for a career does meet the non-rival status, your only product is the papers you write and the talks you give. But as soon as you get away from basic science people try to capture intellectual property and not distribute this with zero friction. People and companies often find it in their financial interest to restrict the use of intellectual capital.
As an egregious anecdotal example when I was working at a National Lab which had a very large super computer as well many smaller clusters all running Linux. SCO, a failed Unix company, sued the Lab claiming it held patents on some underlying code (which they hoped to find). In the discovery phase we had to submit every piece of code running on open source code bases to a repository. Technically if for instance you published a paper doing analysis with R, Perl, Gnu Fortran, Postgresql, etc. every line of code you wrote had to be submitted so they could search through it and try and find instances where someplace in the stack where they thought they could find a legal violation.
As a mean field physics problem productivity can easily expand above linear. As a human problem perhaps not.
300 years is not a long time in the span of history. I've seen estimations of the growth in ancient Rome that asked similar questions and it seems to me that if we broaden our timespan nations and history exhibit rapid growth when successful until they reach a point of peak hubris and arrogance at which point they collapse. Surrounding civilizations take the good ideas and the cycle repeats itself. I'd rather call it hog cycles personally versus strait line or exponential. Maybe a fibonacci or golden ratio built into the natural environment.
An interesting line of research these days is chaos theory and fractals. As you approach a finer and finer stasis "order / strait line" the change to destabilize that stasis becomes smaller and smaller as it approaches zero until finally it is destabilized. Once it is destabilized the direction of correction is a fractal relationship of the entire system not just the parts closest to the destabilization and so the consequence can only be predicted as a factor of the entire system. this being the case we see in history violent swings from what appears to be very stable and people never predict the collapse of civilization or world wars until it is already to late. Well... That is at least my hypothesis obviously.
Has anyone applied this kind of revision-of-exponentials thinking to Moore's Law? When the surface temperature of a microprocessor projected for a few years out started to exceed the surface temperature of the sun, it became clear even to silly physicists and certainly to reasonable engineers that something had to change. And the slope of the curve of log (memory capacity) vs time has slowed down considerably from the original estimate of 18 months. So the next hope for exponential growth is to plot capability against total product shipped, integrated since the dawn of time. This may be important for batteries, which have a history going back to the 1800s, starting with voltaic, then lead-acid, and now lithium-based technologies. In the previous century, the doubling time when you plot capacity or energy density vs years was about ten years. Once the electric car industry started building battery factories, it has apparently speeded up.
Before I move to Caballero, please help me understand this:
“If people tend to judge their situation by comparing themselves to their parents, then linear growth should still satisfy them, because — barring negative shocks — everything still keeps getting better and better.”
I get the feeling millennials do not think they are better off than Boomers (their parents). What metrics say they are? Yes, technology & information exchange is faster but are projects of wealth?
The biggest problem with the corporate tax cut thesis is that it presumes that the money is burnt, and not used to improve productivity in non-market sectors (such as the education of labor) and maybe even the improvements to TFP.
It's that sort of blinkering propaganda that is intended to prevent consideration of solutions that don't just concentrate wealth.
On the psychological impact of slower overall growth...
I don't think people really compare themselves to their parents. The technological gap is just too big. Smartphones are 'recent' yet they are so ubiquitous, life doesn't make sense without them anymore. "The past is a different country" as they say.
People compare themselves to their neighbours and, to a lesser extent, to what they see on TV.
If you like getting answers straight from economic research, check out the new product we are launching: https://consensus.app/. Our mission is to make research more accessible for everyone.
Sorry if this is an amateur question, but how does a tax on corporate profit distort investment incentives?
Large companies have access to debt financing. If they make investment decisions based on some hurdle rate of return, it shouldn't really matter whether they have a lot of retained earnings to invest.
My hunch is that there's an important market failure here, which consists of CEOs overestimating their own business acumen and the prospects of their own sector. It seems as if it would be more efficient for companies to pay out most of their earnings as dividends and then go back to the capital markets when they have what they think is a worthwhile plan for expansion.
So wouldn't the ideal policy be one that allowed corporations to deduct dividend payments (as they already can for debt interest) but then taxed retained earnings at a high rate? It would counter the natural bias of managers towards using shareholder profits to build their own empires, even when that's not an optimal allocation of capital.
Here's a summary of the most famous paper that made an extreme version of this argument.
https://en.wikipedia.org/wiki/Optimal_capital_income_taxation#The_Chamley%E2%80%93Judd_zero_capital_income_tax_result
The basic idea is that corporate tax is a tax on capital, which is something that gets accumulated over time, so that taxes compound.
Yes, but it's a tax on capital that was not converted into labor or capital goods. Just about every corporate tax scheme I've run into taxes net profits, not revenue, and assumes that money spent on equipment or research is not taxed.
This is way above my pay grade but I think these models are looking at a different issue. They discuss the optimal tax rate on individuals' capital income (of all types).
What I was wondering about is the justification for taxing this specific form of capital income twice: first when the corporation earns it and then when the shareholder earns it.
Regardless of what the optimal tax rate on capital income may be, it seems as if it would be distortionary to tax *distributed* corporate earnings at a higher rate than other forms of capital income.
But I think a punitive tax on *retained* earnings might be antidistortionary, because managers have a natural tendency to increase their own power and prestige at the expense of shareholder value. I think this bias is documented, for example, by the fact that corporate mergers are often wealth-destroying.
The idea is that all companies are owned by households, so taxing their profits is the same as taxing household capital income. Obviously this is not really true, and there are reasons that corporate taxes don't work like, say, capital gains or dividend taxes. But the basic idea is there: When you tax profits, you tax the value of productive capital, so less gets accumulated over time.
That's true. There are overhead costs to accessing a bank loan or issuing a bond, and I suppose they don't rise proportionately to the sum being raised.
So I think it might be optimal to impose a heavy tax on the retained earnings of companies, but only above a certain threshold. There should be a zero bracket large enough to cover all the earnings of a small business.
(The market failure I'm talking about is a principal/agent problem, which shouldn't be a major issue for small businesses anyway.)
When y'all start reading Noah? I started at uni so around 2007-2010. Noah is my longest relationship.
I started my blog in 2010! :-)
No shame admitting I am a very recent (very happy) reader, I followed a link here from a Paul Krugman article about the Sri Lanka currency crisis. Thanks for the excellent and thoughtful content!
I'm pretty sure I found Noahpinion for the first time from a Krugman hat tip.
I forget who referred me, but yeah, definitely some other blogger -- most likely Krugman, DeLong, or Yglesias.
Thinking back it could have been Steve Williamson who sent me to Noahpinion - was studying Macro 3 and his textbook was making life hell at the time. I remember commenting back in the day on a Noah post and Steve Williamson responded to my comment and I was like this is wild.
I know I started reading it by August 2011. Amd I'm sure I read back to the first post, mining it for good criticisms of libertarian arguments.
Sorry that should read only, not longest.
GPT-3 comes from OpenAI not Google, btw. (OpenAI started as an anti-evil AI thing apparently because Elon’s been hanging out with lesswrong posters, but turned into a product company part owned by Microsoft or something.)
> gptI’m in the mood to write about some econ papers today.
Also, stray GPT here.
Also, https://www.bart.gov/news/anime.
I think the 4 centuries of data is definitely enough to show TFP is exponential. With 4 parameters I can fit an elephant; if you need 4 different slope parameters in your piecewise-linear function, it's not looking good for your theory. I could fit better than that paper if you gave me 4 parameters too!
From what I can tell the paper's tests of functional form are dramatically biased because they don't account for the long memory of TFP. After removing the exponential trend, TFP isn't *quite* nonstationary, but it's very close to it, and needs fractional differencing to get good estimates of standard errors. You can have very, very long deviations before regression to the mean kicks in, which can give the appearance of linearity over short timespans.
I don't think this shows that 4 centuries isn't enough to get the right answer. I think it shows that if you're really inefficient with your data, it can take you more than 4 centuries to figure out the right answer.
I still think that Matt Yglesias had this right, and in fact very high tax rates on individuals' capital income and corporate profits drive companies to sink excess cash into R&D expenses that deduct from profits, rather than paying tax on the profits and then letting their investors pay even more tax on dividends.
https://slate.com/business/2012/07/xerox-parc-and-bell-labs-brought-to-you-by-high-taxes.html
I agree. High taxes and antitrust enforcement are why we had our golden age of research labs like Bell Labs, IBM Watson and Xerox PARC.
I know those weren't your words, as you put in a graph from Cabellero, but why would 4% be an inflation "disaster"? It seems like a moderately higher, but predictable inflation rate would not reduce growth and would The Fed more maneuvering room in the case of recessions.
4% inflation is what we had _after_ the Volcker disinflation, during the "Morning In America" recovery. I am seriously baffled as to why people think it would be so terrible. After we get inflation back down to 2% for a year or so, the Fed should announce they're raising the target to 3, as well as continuing to treat it as a target, not a ceiling.
(Or, even better, they should just move to NGDP targeting, aiming for something like 5%, typically hoping for something like half growth and half inflation.)
Does Ngdp targeting ignore inflation or try to do something specific about it? I mean there is the standard formula real equals nominal minus prices or inflation. This formula can be appplied to gdp amongst other things.
I'm not sure whether this link works if you're not a Slow Boring subscriber, but Matt Yglesias has a fairly good summary of the case for NGDP targeting:
https://www.slowboring.com/p/the-case-for-ngdp-targeting
Noah might know some other economists who've described this.
But basically yes, the idea is that when the Fed decides whether they need to tighten or loosen monetary policy, they should look at the growth rate of nominal GDP (i.e. how much money is circulating through the economy) rather than the rate of change in prices. At a basic level this is an easier quantity to measure (you don't have to worry about differences between "core" or "expectations-based" inflation versus volatile commodities, and hedonic adjustments, and so on), and in some sense it bears a more direct relationship to monetary policy. (You can go back to Milton Friedman's famous "PQ=MV" formula.) And if your NGDP growth target is, say, 5%, then basically it's the Fed's job to try to keep that on track, and then leave it up to private actors, and fiscal policy, to determine whether that ends up being 2% real growth and 3% inflation, or 1% real growth and 4% inflation... It also leaves room for the Fed to not freak out and apply thumbscrews to the economy if we have a burst of minimally inflationary growth.
"Vollrath shows that if you assume that TFP is equal to the natural log of the number of ideas, and if ideas give birth to each other at a constant rate, you get perfectly linear growth. Cool result!"
Somehow this one paragraph did more to undermine my confidence in you, Noah, than all the mistakes you discussed in your recent roundup. That's not a result, it's the definition of logarithm -- and of course "natural log" is irrelevant, any logarithm would work here. Maybe it's just inartfully worded, but MAN, how inartfully!
I think a significant reason we don't have exponential growth, or at least higher order than linear, is that we (humanity) does not have a way to take advantage of the non-rival quality of intellectual capital. I have an economic model I wrote down with coupled differential equations in which intellectual capital is, as is the natural model, non-rival. This is a physics based model where we only consider the real entities: labor, capital, production, consumption. I break this down into physical and intellectual components, so for instance new physical capital is produced by combining labor and intellectual capital and consuming some physical capital. Because it is a mean field model new intellectual capital is instantly fed back into the equations for producing new physical and new intellectual capital.
But in real life we artificially make intellectual capital restricted, i.e. it doesn't feed back uniformly across all entities, through the use of patents and trade secrets and non-compete clauses etc. Basic research science which I did for a career does meet the non-rival status, your only product is the papers you write and the talks you give. But as soon as you get away from basic science people try to capture intellectual property and not distribute this with zero friction. People and companies often find it in their financial interest to restrict the use of intellectual capital.
As an egregious anecdotal example when I was working at a National Lab which had a very large super computer as well many smaller clusters all running Linux. SCO, a failed Unix company, sued the Lab claiming it held patents on some underlying code (which they hoped to find). In the discovery phase we had to submit every piece of code running on open source code bases to a repository. Technically if for instance you published a paper doing analysis with R, Perl, Gnu Fortran, Postgresql, etc. every line of code you wrote had to be submitted so they could search through it and try and find instances where someplace in the stack where they thought they could find a legal violation.
As a mean field physics problem productivity can easily expand above linear. As a human problem perhaps not.
300 years is not a long time in the span of history. I've seen estimations of the growth in ancient Rome that asked similar questions and it seems to me that if we broaden our timespan nations and history exhibit rapid growth when successful until they reach a point of peak hubris and arrogance at which point they collapse. Surrounding civilizations take the good ideas and the cycle repeats itself. I'd rather call it hog cycles personally versus strait line or exponential. Maybe a fibonacci or golden ratio built into the natural environment.
An interesting line of research these days is chaos theory and fractals. As you approach a finer and finer stasis "order / strait line" the change to destabilize that stasis becomes smaller and smaller as it approaches zero until finally it is destabilized. Once it is destabilized the direction of correction is a fractal relationship of the entire system not just the parts closest to the destabilization and so the consequence can only be predicted as a factor of the entire system. this being the case we see in history violent swings from what appears to be very stable and people never predict the collapse of civilization or world wars until it is already to late. Well... That is at least my hypothesis obviously.
Has anyone applied this kind of revision-of-exponentials thinking to Moore's Law? When the surface temperature of a microprocessor projected for a few years out started to exceed the surface temperature of the sun, it became clear even to silly physicists and certainly to reasonable engineers that something had to change. And the slope of the curve of log (memory capacity) vs time has slowed down considerably from the original estimate of 18 months. So the next hope for exponential growth is to plot capability against total product shipped, integrated since the dawn of time. This may be important for batteries, which have a history going back to the 1800s, starting with voltaic, then lead-acid, and now lithium-based technologies. In the previous century, the doubling time when you plot capacity or energy density vs years was about ten years. Once the electric car industry started building battery factories, it has apparently speeded up.
Before I move to Caballero, please help me understand this:
“If people tend to judge their situation by comparing themselves to their parents, then linear growth should still satisfy them, because — barring negative shocks — everything still keeps getting better and better.”
I get the feeling millennials do not think they are better off than Boomers (their parents). What metrics say they are? Yes, technology & information exchange is faster but are projects of wealth?
The biggest problem with the corporate tax cut thesis is that it presumes that the money is burnt, and not used to improve productivity in non-market sectors (such as the education of labor) and maybe even the improvements to TFP.
It's that sort of blinkering propaganda that is intended to prevent consideration of solutions that don't just concentrate wealth.
Quick note: that inflation paper (and the slides) are by Ricardo Reis, not Caballero!
On the psychological impact of slower overall growth...
I don't think people really compare themselves to their parents. The technological gap is just too big. Smartphones are 'recent' yet they are so ubiquitous, life doesn't make sense without them anymore. "The past is a different country" as they say.
People compare themselves to their neighbours and, to a lesser extent, to what they see on TV.
My experience is that kids do compare themselves to parents; My kids (30 & 33 yo) definitely compare their economic situation to mine.
This is missing a discussion of what holds back the technological growth rate of the Trisolarans.
If you like getting answers straight from economic research, check out the new product we are launching: https://consensus.app/. Our mission is to make research more accessible for everyone.
Here's an example of what the results might look like?: https://consensus.app/results/?q=immigration%20effects%20on%20innovation
Also Noah liked my comment on twitter so it looks like he supports our mission: https://twitter.com/c_salem2/status/1559977087666118658?s=20&t=e6sbby3ED1JYu7GgDLocAg