Reconciling Macroeconomics And Finance From The US Corporate Sector, 1929–2022

>> John Taylor: Let's begin. We're very happy to have Andy Atkinson here with his colleagues, Jonathan Fabrizio, where is Fabrizio? Sorry, Fabrizio. And Jonathan is up on the screen somewhere, okay. And it's welcome to be here at Stanford. You've been here a few times before, and it's great to have you back.

The title of this talk is Reconciling Macro and Finance, the US Corporate Sector, 1929 to 2020. What, 22, it says here.

>> Andrew Atkeson: It's through 22.

>> John Taylor: 22, Okay, through 22. But welcome back. And you can speak as long as you want.

>> Andrew Atkeson: All right, thank you very much.

I had a busy weekend, if you read the headline over there, I had to break four minutes in the mile before coming here.

>> Andrew Atkeson: Running for Yale, which I did 35, 40 years ago. But anyway, should we go to the slides? So there we are. So this is a paper joint with Jonathan and Fabrizio.

And lemme just get into it right away. Obviously, first question is, what needs to be reconciled? And in a broad sense, macroeconomists looking at NIPA data or fixed asset tables will see that macro aggregates tend to be very smooth, and the amount of capital in the economy tends to be very smooth.

And yet, if you use public firm data, crisp compustat, you see very volatile market valuations of claims on corporations and also very volatile returns. And so there's a big agenda, so how should we reconcile those two sets of observations? And a lot of discussion is this because there's, in the background fluctuations in expected returns or fluctuations in expected growth of payouts or maybe some other factor like bubbles.

So the thing we aim to do in this paper, I mean, the paper has three parts, I would say probably the most important goal, and if we get through that, I'll be happy. Is to advertise that I think these new integrated macroeconomic accounts that have been created by the BEA and the Federal Reserve in a joint project.

Actually might serve as a very useful, unified data source for macro and finance, because they've basically put together the flow and stock data with the market valuations in a way that they've done a lot of interpolations, but I think works pretty well. And so the first thing you want to demonstrate is just the properties of returns shown from that data look a lot like crisp returns.

So that maybe in the future, people doing macro finance can talk to each other, because they might be referring then to a common dataset. I hope to get through that part and convince you of something there. The next parts might not go so well. We're gonna try to use that data to revisit some classic issues.

And so the first one is just like, well, what do you get out of Campbell Schiller regressions that when you see a high valuation relative to current cash flows, is that predictive of future returns or future payout growth? And so it turns out you get different results in the IMA data than you get in the public firm data.

And we'll try to talk a little bit about why. So that's in some sense model free. And then we're gonna try to do like a minimal extension of a standard stochastic growth model that we're gonna call an accounting model. And the purpose of the model is to recognize that a portion of the value of firms is coming from the physical capital that they have installed, that we see measured in the fixed asset tables, and a portion is coming from something else.

And so we're gonna call that something else in this particular model, factorless income. So the accounting model is to divide the firm valuation we see into those two components, the physical capital and the rest, and the cash flows into those two components. The cash flows to owners of physical capital and then the factorless income.

And cuz the point of that is that we think that a macro-finance reconciliation should account not just for returns and valuations of equity of US corporations, but what are the returns on the physical capital stock? We should also be getting that. And in this simple accounting model, we will do that decomposition.

And we're gonna say there's a puzzle that remains about the returns on the physical capital stock that I'll try to show you, we'll see if we get there. Obviously, the puzzle could be a failure of our model, could be a failure of the data, but we think that that's an interesting area for research going forward, is to try to complete that reconciliation there.

So that's the agenda. So leme get started. There's a bunch of literature. John Cochrane has done quite a bit of this literature, will be closely related to something by Lorraine and Yogo from 2008 and papers, I'm sorry, Hanno is at a ski conference, it's difficult to be a business school professor, I guess.

But there's literature that there might be shocks to the share of value-added that's going to owners of corporations, and that'll be important in our story. And then, of course, a huge literature in macro-finance looking at both of the issues that I mentioned what are the returns to physical capital, and then about the valuation of firms.

So that's an incomplete survey of the literature, but lemme get into it. Some of you may be familiar, but starting in 2008, there was an international agreement that every country should create these things called integrated macroeconomic accounts. The US is probably the only one that has really implemented that to a large extent.

And so that actually divides the economy up into nine sectors and tries to do consolidated cash flows, fixed assets. And then financial balance sheets with financial flows where everything fits together in the sense that you would expect for a corporation to have an income statement, a cash flow statement, and a balance sheet, and a market valuation that all fit together.

So we're gonna focus on the two sectors, five and six, that form the corporate sector in these accounts. And so the object we're looking at, or the set of firms we're looking at, is a little different than what you see in public firm data. It includes both public and private firms.

And the BA and the fed have to do an imputation of valuations to the private firms, and they have a methodology for doing that. And then there's this concept of a US resident corporation, which is a little bit of an odd thing. It is the US subsidiaries of US multinationals, the US subsidiaries of foreign multinationals, it does not include the foreign subsidiaries Of US multinationals, but everybody's nationally product accounts are constructed on this resident basis.

And so, a lot of the work that goes into constructing this integrated macroeconomic accounts is to make adjustments to public firm data and use tax data, which is a different source. And put valuation and balance sheets on a consistent basis with this notion of resident corporations. And that's why I think it's a lot of work so, Bob Hall has done a lot of this type of work in the past, but now the BEA and the flow of funds are doing it for us in a consistent way.

You have a question?

>> Speaker 6: I just don't understand the data. If I'm a US firm that operates internationally as a single entity, how is that handled in the data?

>> Andrew Atkeson: So, what goes on is, in your public financial statements, your 10Ks, you do a consolidated reporting, but for corporate tax purposes, you file tax returns in every country that you operate.

And the subsidiaries that you have in each of these countries is considered a corporation in that locale, so, like Toyota of America is a US corporation.

>> Speaker 6: I get that, and you explained that earlier, but let's suppose-

>> Andrew Atkeson: The data source is then a corporate tax data.

>> Speaker 6: So if I'm Apple, and I have expenses in Thailand, but no income, and I don't have a company, let's suppose in Thailand, what happens to Thailand in my US accounts, in your data?

 

>> Andrew Atkeson: I believe nothing I mean, in other words, Apple's gonna submit a US corporate tax return, and a bunch of the data is gonna to come off of that. And so, there's gonna be flaws in the data and there's gonna be imputations. Ellen is a data expert, and I think that what I wanna stress is that this is an ongoing project.

So, I mean, in our interactions with people with the flow of funds because I have a student there, they'll change things if people think that there's a better way of doing it. So the way I see this is not a final data set, it's gonna have a lot of problems.

But if we kind of agree that this is something we could focus on, we could collectively improve that.

>> Speaker 7: If you can ask in reply to Tara's question-

>> Andrew Atkeson: I wanna go back there to Luigi, I think you said it first.

>> Luigi: Yeah, sorry, the big issue for Apple is Ireland and not Thailand or places like-

 

>> Andrew Atkeson: No, no, no-

>> Luigi: How do you consider, all that income disappears from here, right? Because it's considered an income.

>> Andrew Atkeson: So we'll have we'll have discussions, and we'll kinda get to that. I agree, all these issues are there, and the BA has to wrestle with it, yes.

 

>> Luigi: Reconcile any of the tech sector. This is huge.

>> Andrew Atkeson: Yeah, so we'll have pictures. And I agree, I mean, Monday evening-

>> Speaker 10: Pharmaceuticals, a lot of other industries, too.

>> Andrew Atkeson: Yeah, yeah, no, so Monday evening, I met your colleague Uran Ma. And I was at a social dinner, and she whipped out her laptop and said, let's do this in Compustat.

And so maybe the next iteration of this paper will have a much more careful reconciliation. She does magic that I wouldn't know how to do in Compustat. All right, but the point is that-

>> Speaker 7: In response to Daryl's question, but wouldn't the answer be, well, it is an input purchase, so-

 

>> Andrew Atkeson: No, we'll run into it in slides. We're gonna come back to this in slides. So the NIPA, of course, is the stuff in the cash flows from operations, we'll call it. The stuff that macroeconomists are familiar with from the tables for the corporate sector in NIPA and fixed assets.

But then the balance sheets and flows are organized nicely in that cash comes out of those operations. And you also start with a starting balance sheet. And then you have transactions, you have actually property income goes out. But transactions and asset liabilities, revaluations of those things, there are statistical discrepancies, and then you have an ending balance sheet.

So what we're gonna do with that is try to construct two measures of the value of US corporations. I'm gonna start with the one we think is most closely connected to macro, which is a cash flow measure we call free cash flow from operations, and a valuation measure we call enterprise value.

So if you think about a standard growth model in which a firm all equity financed and has no financial assets or liabilities, the cash flow or the dividend should be gross value added. You pay some taxes of various kinds, you pay your labor, and you invest in physical capital, and that's the cash flow left over to distribute to owners of all kinds in the corporation.

And the corresponding valuation measure we call enterprise value, you take the market value of the equity plus the liabilities and subtract off the financial assets to get what we call that. And so when we construct a measure of returns in the annual data, we're just gonna take balance sheet stuff is at the end of the period.

So Vt would be end of period enterprise value, and then next period, there's some free cash flow and enterprise value. So that'll be the notion of returns that we're using.

>> Luigi: And where is excess cash? Because a bunch of tech firms have billions and billions on the balance-

 

>> Andrew Atkeson: Okay, does this is work? The vision that we have is this is what a private equity guy would do to unlever the firm and strip it of its assets. So, we're heavy reliance on a Modigliani–Miller theorem, that if they were to change their payout policy or their financial policy, it wouldn't affect value, it wouldn't affect returns on the enterprise value.

It would affect what you see in the data. So we are gonna use the second measure that I discussed.

>> Luigi: And many times, they take net debt, right? So it's whatever debt you have, leverage you have, minus the excess cash.

>> Andrew Atkeson: Yeah, that's what we're basically doing, is net debt.

 

>> Luigi: Okay, that was all

>> John Taylor: The answer to this question should have been, it's the last item in our list.

>> Andrew Atkeson: Yes, I had to understand what he was asking. Okay, so what I just wanna show you is that since 1929, maybe, I didn't realize we'd be so far away from the screen.

The thing on your left is this enterprise value over the gross value added annually in the non-financial, I mean, I'm sorry, in the corporate sector as a whole. And I just wanna emphasize it's quite volatile. It goes up and down a lot. The thing on your right is this free cash flow measure as proportion of gross value added.

And you can't really see, at least I have to put my glasses on. Those of you who have good vision can see that before World War II, the free cash flow was about 14% of gross value added. It falls to about 6 for 50 years, and then it's gone up quite a bit since then.

So it's got this big U-shaped pattern. So the left-hand side then is the corresponding dividend price ratio using these concepts, which you see doesn't really show much of a trend over time. And on the right side, we just overlaid the The free cash flow with a valuation multiple of 31.25, just because it made the axes work.

And so you can kind of see at the low frequency with enterprise value, you can see at the low frequencies that the movements in cash flows and the movements in value roughly have something to do with each other. And so if the line on the left had been completely flat, we would go, we're done, it's all cash flows, but obviously it's not flat.

So there remains this issue of what is driving the movements on the left hand side, and we'll come to that. But just kind of as a robustness.

>> John Taylor: Question over here.

>> Speaker 9: I've seen that you shape a lot, like on income inequality, immigration, stuff that lines up to there.

And in the back of your mind, think about whether there could be any possible link, cuz there.

>> Andrew Atkeson: We'll come back to that. So the high free cash flow in the beginning is gonna turn out to be very different the source than the high free cash flow at the end, so we'll talk about that.

Okay, which could be a data error, I mean, whatever, we'll discuss that. Another concept you can use, and one advantage of the richness of this data is that we can also look at a return to equity using a more traditional way of thinking of things. Where every year there's a certain amount of dividends that are paid, and at the end of the previous year there's a certain amount of equity, market value of equity.

But in the accounts they have the capital gain on the outstanding equity. So you can construct a straightforward equity return, which would be the sum of the capital gain plus the dividends that you got, as if you just bought the equity and held it for a year.

>> Speaker 10: Could you take a minute and just describe how they're coming up with market values of non-traded corporations?

 

>> Andrew Atkeson: Okay, we link to a webpage in the paper. It's basically by a method of comparables, I mean, so they're gonna take various valuation ratios by industry and do stuff. The issue Luigi raises about what do you do with Toyota America, for example, on another paper we have, this is a big issue.

They made a choice to use us valuation ratios to value Toyota America, you could conceivably use Japanese valuation ratios to value Toyota America. For the finance crowd, I would say there's a nice thesis to be written in doing some kind of factor analysis to say, which one should you be using?

And the BA and the FED might change their methods, in the face of that research.

>> Speaker 11: Domestic firms are different, they present a different set of issues, they're much less capital physical capital intensive and much less human capital intensive than.

>> Andrew Atkeson: Yeah, no, no. So the.

>> Speaker 11: Evaluation model is not obvious how to.

 

>> Andrew Atkeson: Right, okay. So the nice thing about this is that there's several hundred people who are working and constructing this data. I agree we should ask questions, but to force macroeconomists and finance people together, we're gonna have to accept that there's some sausage going on. And what I wanna show.

 

>> Speaker 12: The question is these facts are dependent on the private part or if I just did them with publicly traded.

>> Andrew Atkeson: Well, I'll show you. Okay, so just out of curiosity, what's the relationship between enterprise value and this equity valuation? So I have to describe on the left hand side, we've taken ratios to gross value added.

What you cannot see, but I wish you could, on the axis, is that enterprise value is just consistently 50 percentage points of gross value added more than equity value. So it's like over the century, the sector as a whole has just had a constant amount of net leverage.

The right hand side is the free cash flow over gross value added and the dividend concept over gross value added. So you can see that the dividends is the smoother line, there is smoothing of dividends through the cycle, as you might imagine, and so they're using financial assets and liabilities to achieve that smoothing.

Okay, but this is the graph? Yes.

>> Ellen: Corps in there, right?

>> Andrew Atkeson: Yes, we do. So this is, we just wanna just literally take, I mean, Ellen raised a point about S corps that I'll come back to, so can I just hold that one? Yeah, okay. So we just, let me say, let's just take these annual returns that we get from the IMA and just compare them to the crisp valuated returns.

In each graph, there's a 45 degree line, and you see the unconditional statistics in the bottom. So the first one has the crisp valuated return on the x-axis on the left, and then the enterprise value return on the right. And since we saw that equity is kind of a levered version of the free cash flow, the line's a little bit flatter.

And the IMA equity and the crisp equity one line up much more closely, as you would expect they would. So at the end of the day, with all these interpolations that they do to get private firm values and the rest, they end up with something that looks a lot like crisp data.

 

>> Speaker 10: And it's hard to know what to make of that because maybe that you're just recovering, covering the models that they're using to place values on the traded firms. I mean, the Lorraine and the Lorraine paper that you cite and Yoko, they literally took the crisp, the ratio of crisp market value to book value multiplied that by.

 

>> Andrew Atkeson: Yes, so they wrote before the IMA, so the difference between us and them is we can blame it on the BEA and the fed, rather they did it themselves. But our interpretation of this, is that if you work with public firm data, you have this problem that it doesn't match NIPA cuz it's on a different basis and it doesn't necessarily go back that far.

And if you work with NIPA data for the macro side, it doesn't match public firm data. So we kind of see that the opportunity here is to get macroeconomists and finance it, people to talk to each other. But somebody's got to bridge the two data universes and we're feeling that they're doing that.

So, John.

>> John: IMA is not a miracle cuz they are using values as an input, unlike lots of previous macro things where they wanna use cost or accounting values or something.

>> Andrew Atkeson: Yes.

>> John: So what you're just saying is you take market value as an input, you extend it to a bunch of private firms and other things that aren't measured.

And those things you get, something goes up and down with market value.

>> Andrew Atkeson: So, yes, I'm saying I agree it's good, but there's a, it's a little.

>> Andrew Atkeson: No, no, it's not a test of anything, this is what I'm saying, this is an invitation for people who work in macro finance to say, you know, if I'm gonna put forward a.

Model and it doesn't have capital in it. I'll say you got a non starter. You should compete with a macroeconomist. And if you're gonna have a macroeconomist, who's gonna do a valuation model, this is your data target. So I'm just saying this is a data target.

>> Speaker 10: If you were the type of person, none of them in this room, of course, who were to say, stock markets are all completely crazy, let's ignore their information.

Good old fashioned book value is the right, is the fundamentals, not the market value, then say, well, they've just completely polluted the numbers by putting these crazy market values into things.

>> Andrew Atkeson: Yeah, but I'm not that type of person.

>> Speaker 15: When private guys go public, isn't there any checks?

If you're a super micro iss person, to match them up and see.

>> Speaker 16: No, that's not what they're doing.

>> Andrew Atkeson: I want to keep the pace going.

>> Speaker 16: Keep going.

>> Speaker 17: I had a very low level question, which is if the origin, I mean, the source of the data is tax based.

And I'm curious about business expenses that take place in a locality, where no income is being generated. I thought your answer would have been, it's gonna show up as net business income to the extent that the income that is generated by the business purchase is taxed in the United States.

Is that the right answer?

>> Andrew Atkeson: No.

>> Speaker 17: What's the answer?

>> Andrew Atkeson: If you look at table S5 or S6, you'll start off with these concepts of getting down to free cash flow, and then they're gonna have property income, which is the money coming in from the foreign subsidiaries, largely.

And then they're gonna have property outflow, which is dividend and interest and rent payments. And so the interaction with the Apple in Ireland, is the way we do accounting for FDI income, is that there's a property income entry. And then if Apple doesn't bring the money back, there's a flip side, reinvestment in Ireland entry.

 

>> Speaker 17: And some are the business purchase, not the input purchase in Thailand, not the Ireland.

>> Andrew Atkeson: No, that stuff.

>> Speaker 17: How does that show up?

>> Andrew Atkeson: It doesn't, because it's not a us resident corporation.

>> Speaker 17: The net income that is being taxed is.

>> Andrew Atkeson: No, they're not taxing that income.

That Nipah is done for US corporations and a US corporation is a US object.

>> Speaker 17: And your initial piece of data, is there any?

>> Andrew Atkeson: Mmm?

>> Speaker 17: And your additional data is tax based? I missed the beginning.

>> Andrew Atkeson: No, but let's.

>> Speaker 17: Nothing.

>> Andrew Atkeson: It's a mess of, it's a, nip a Dave's a sausage making exercise.

So let's hold off. We could spend the entire day on it, and I don't think it did. So we want to do now is we have a couple of alternative data series, of returns and cash flows. And we had this issue that in each of these series, we have what I'll call a dividend price ratio or a cash flow to current value ratio that's going up and down.

And there's this classic issue of, why is it going up and down? Is it because people expect that returns going forward are gonna be higher or lower, or because cash flow growth going forward is gonna be higher and lower? And a very popular tool for getting at that question, is to run these things called Campbell Schiller regressions.

And as John articulated beautifully, you can set up three linked regressions, so you're always using as an explanatory variable the current dividend price ratio. And then you're trying to forecast cumulative returns, cumulative cash flow growth, and a terminal dividend price ratio over a certain horizon. And if you do a log linearization, there's a relationship among the coefficients that must be satisfied.

So you kind of set up this, maybe you call it a trichotomy between is it mostly returns, is it mostly cash flow growth or is it mostly this tail term that I might wonder about? So we just ran those regressions, we should have done all three. And the regressions are known to have bad sampling problems and out of sample forecasting problems.

But a regularity in the crisp data that John articulated, and John Campbell articulated in his textbook, is that for the public firm data, there's very little evidence of forecast ability of cash flow growth. From the IMA data, when we run these regressions, what we have in the first column is that first regression, look what happens to accumulated returns over 15-years.

The second regression, what happens to cash flow growth using the different measures of cash flow. The third one, what's happening to the tail coefficient, the parameters across a row, the first one minus the second one plus the third one, add to one. And usually in the first row, you should get a very small number.

This one's actually bigger than normal on the cash flow growth, so the second item in the first row tends to be small. In contrast, in the IMA data, you see that 0.71, you get a much larger coefficient on cash flow growth, okay? Yeah,

>> Speaker 18: no concern about comparing an average of ratios to a ratio of totals.

That is the Jensen effect of averaging across many firms and then comparing the two numbers,

>> Andrew Atkeson: you say, ask if I have a concern or the people

>> Andrew Atkeson: who run these regressions have a concern

>> Speaker 18: whether there's likely to be a significant bias in the difference between zero and the average of ratios.

 

>> Andrew Atkeson: So the literature that we're reacting to is also using indices and so.

>> Speaker 18: They're totaling up all the firms before they do the division of the numerator and the denominator.

>> Andrew Atkeson: Yeah.

>> Speaker 18: Evaluated portfolio.

>> Speaker 19: It's not firm level data.

>> Speaker 20: No, no. There's other regressions that do for level data.

 

>> Andrew Atkeson: This is not, no. We know,

>> Speaker 21: good question. You also add up to the total portfolio of firms, or is this an average of ratios?

>> Speaker 20: No, it's got the aggregate data.

>> Andrew Atkeson: We just got the aggregate data. Yeah, but, okay.

>> Rob: What about the standard errors on these things?

 

>> Andrew Atkeson: Gigantic.

>> Rob: Yeah, so, I mean, there's.

>> Rob: You have 94 years.

>> Andrew Atkeson: No.

>> Rob: There's six non overlapping observations. And, I mean, if you take a year off and add a year, these things can change drastically.

>> Andrew Atkeson: Okay, so slow down.

>> Rob: Six times 15 is 90.

 

>> Andrew Atkeson: So, Rob, this is my view of what's going on. It is well known that these things have horrible sampling properties. I have Eva Welsh is a colleague, so he says they're out of sample, forecasting is also BS. But, John made an important point, that you have this restriction across the parameters.

So the observation, if you're arguing whether the forecasting of returns is meaningful, the observation that the coefficient on dividends is informative, for that, it's a. Close to zero is informative for that, that's the dog not barking, so, all we're saying here is in the IMA data, you no longer have the cash flow growth having a zero coefficient.

 

>> Rob: Unless the standard error is one point.

>> Andrew Atkeson: Well, it could be zero, it could, right, but-

>> Speaker 23: But you cannot reject it, it's not any number-

>> Andrew Atkeson: So, this is the picture-

>> Rob: You're making a statement that this number is bigger than this other number, and I'm saying that that's a testable hypothesis, and would you reject that hypothesis?

 

>> Andrew Atkeson: John wrote a beautiful paper called how big is the random mark in GDP, I view this as the same issue.

>> Rob: Well, the point estimates are really, when the point estimates are zero and one and they change from one to zero, that's interesting.

>> Andrew Atkeson: This is the picture.

 

>> Rob: 1.7.

>> Andrew Atkeson: Okay, but this is the picture I'm interested in discussing and getting reaction to, and this is where the whole thing might break down. Okay, so what I'm showing in this graph, there's a U-shaped line that is blue, that goes from 1929 to the present, which is the nip of dividends over gross value added.

There's a red line that comes down initially and then stays down it's actually the index of crisp dividends, and we just set the first number to 0.16 to set the level. So, you see that for a while, the first half of the sample, they kind of looks, go in the same place, and then the second half of the sample, they dramatically diverge.

Ellen's gonna say S corporations but let me keep going, so then there's two other lines on this figure.

>> Ellen: Andy, if I can, since we're having graduate school deja vu, certainly the proportion of cash paid to shareholders, that is, dividends versus repurchases, has changed very dramatically in that.

 

>> Andrew Atkeson: So now let's look at the other two lines on the graph, okay? So, there's an article, Zeng and Luk, who work for S&P Global, and they conveniently for us, organized measures of S&P 1500 payouts in table four of their article. So, we were able to copy the dollar numbers down and divide them by gross value added.

So, there's a yellow line, which is that measure of S&P 1500 dividends from that table relative to gross value, which seems to match the crisp data quite well, as you would expect. Because the S&P 1500 is so, and then there's a purple line that's kind of jaggedy, that is the sum of dividends, buybacks and acquisitions, okay?

So, there's a cryptic sentence in John Campbell's textbook on page 141 about how if a low frequency growth in dividends per share that their regressions might not pick that up. And so we conjecture that that's what the source of the difference of the results is. But the thing we wanna emphasize is it looks as well from public firm data that there's been a dramatic increase in the share of corporate gross bio added that's available to be paid to owners of firms.

That there's been all this literature and macro about rising monopoly power, whatever you wanna call it. It does seem to be showing up in cash, which was new to us in terms of looking at, so both in the NIPA data and in the public firm data. I will say for Alan's benefit, the fact that the wiggly purple line and the blue line lie close to each other is a pure coincidence, there's two offsetting effects that just happen to cancel.

So, one effect is that more and more us corporations or S corporations and S corporations are required to pay out everything as dividends. So, they're a huge portion of NIPA dividends without being necessarily a huge portion of gross added. So that's pushing the NIPA dividend line up at the same time for the public firms as many have observed, much of their income is coming from abroad.

So, on a consolidated basis, you would think public firms should be a subset of public plus private, but when you include the international, that's getting the purple line up. So the fact that the two lines land on top of each other, I don't want you to take anything away from that other than pure coincidence.

I just want you to take away that whether you, I guess this is a way of putting it, some people would say this increase in corporate profitability is purely because of this offshore stuff. And we wanna say no, it's also showing up in the NIPA stuff, whether you use free cash flow or the dividends, so it is occurring for us resident corporations, it's not just Ireland and NIP.

 

>> Speaker 24: One aspect that might be part of this is that the US firms are getting older, they were until the pandemic.

>> Andrew Atkeson: I'm getting older, my cash flow's not going up.

>> Speaker 24: Initial period of loss leadership, trying to establish a market hold and so on is diminished as a phenomenon in the US at least from 2000 onwards.

So, that might be, I'm not saying that's the whole story here, but that would work in the direction of more cash flow.

>> Andrew Atkeson: I'm going way back, sorry, so that U-shape of free cash flow over gross value added, you notice that there's a 50-year period from World War II to 2000 where it's about 6%.

And if you think of a standard growth model, and you think in the data, the capital output ratio for the corporate sector is about two. And if you think R minus G is 3%, if you're teaching macro, you'd go, well, my model says free cash flow should be 6% of corporate gross value added.

So, somebody who was doing macro in that 50-year period would say, we're pretty much at a competitive benchmark, the standard growth model gets me the value of firms I can ignore finance. You see, obviously, prior to that, you're more than double that benchmark, and after that, you're more than double that benchmark.

So, we think that there's something to be this to be discussed, okay? So, let's get back to where I was, all right, so we got through that, so the baseline accounting model, what do we wanna do? So, it's gonna be clear in the data that the value of US corporations is something different than just their installed physical capital, as you would have in the standard growth model, so we have to do something.

So, we're gonna to put in the simplest modification, or a simple modification, that there's just a time varying wedge between revenue and cost that's gonna lead to what Carica, Bonus and Niemann called factorless income. And that will mean that the corporation's value will come from two components. It'll come from their installed capital, and it'll come from the discounted present value of their expected factorless income, and of course- Correspondingly, the cash flows that are there will be something we call a free cash flow to physical capital.

So, what you would get in a standard growth model would be capital rentals minus investment would be that free cash flow, and then the rest is factorless income. But the discipline we have is that everything has to add up to NIPA when we do this decomposition. So that's it, yeah?

 

>> Speaker 25: In your regressions, you boosted up the cash flow forecasting, but mostly at the expense of the long-term dividend yield coefficient. It sort of looked like the Campbell story, that they were just smoothing the dividends of the cash flow, but the return coefficient was the same. So, of the very volatility of price whatever ratios, you're still getting the same amount from time-varying returns as you had before, it's all coming out of that terminal condition.

So, when you're building a model now with constant expected returns.

>> Andrew Atkeson: Not yet.

>> Speaker 25: You're trying to account for fluctuation in value with some factorless income, as opposed to just a time varying interest rate or risk framework.

>> Andrew Atkeson: Well, okay, so there's a number of valuation ratios that people like to look at.

We were using this free cash flow enterprise value. There's price dividend ratios coming from S&P or CRISP. There's price earnings ratios that you can do equivalents of in NIPA and in CRISPR S&P data. And there's Tobin's q. No, I don't think it's in the paper, it's not in the slides.

If you take the log of Tobin's q, the log of Shiller CAPE, and the log of our, what I call the Buffett ratio. Andrea, I asked me, what's the Buffet ratio? I said, how can you be a finance professor and not know? If you take the log of those three things, they're all the same.

You take the means out. And so-

>> Speaker 25: Market prices.

>> Andrew Atkeson: Right, basically, but it's kind of saying earnings to gross value added, the capital stock to gross value added are super stable, and just somehow, the stock market's going up and down. Right now, we don't wanna plant our flag on it's not expected returns.

So I'm sympathetic to what you say, that the regressions are consistent with some of it being expected returns, some of it being cash flows. I talked to Jen Yang Zhang about doing an estimation exercise that we're not capable of doing, but I'll point to it when we get there.

So we're gonna do something much simpler. But I could see down the road, it could be of interest to someone who's very good at estimating essentially affine pricing kernel models. No, I said it could be of interest, it could be of interest to somebody to do that and to try to do that type of decomposition.

But I think it's gonna be very difficult no matter what you do, because of this fact that everything's persistent relative to your sample size. So you're always gonna have giant standard errors.

>> Speaker 25: You're building a model here with constant expected returns in-

>> Andrew Atkeson: Not yet.

>> Speaker 25: Okay, well, then I will be quiet until you build your model.

 

>> Andrew Atkeson: So the first statement about dividing enterprise value into the two components, because we have no adjustment costs, can be done without choosing any model parameters or saying anything about expected returns. Cuz of this, we know Tobin's q holds there cuz we have data on the capital stock.

The valuing, the factorless income part is where variation in expected returns would come in. And that'll be something where we would have to take a stand. Okay, so the equations of the model would be as follows. There's gross value added, it's just the standard thing, capital and labor with technology being z, the capital is accumulated with investment and depreciation that can vary over time with no adjustment costs.

Firms have to pay a bunch of taxes that we're all gonna just lump in as output taxes. Most of the taxes they pay are indirect business taxes, but they also pay some corporate profits taxes, but we're just gonna lump them in as an output tax. So why here Yt is gonna stand for after tax corporate gross value added, and costs would be capital rentals plus compensation of labor.

And that mu t is a wedge between revenue and after tax revenue and cost.

>> Ellen: Andy, can I have a quick question? Are you going to distinguish owned and used capital?

>> Andrew Atkeson: No.

>> Ellen: But you're gonna bring in a market value of owned?

>> Andrew Atkeson: The BA does that for you in the reproduction value of the capital stock.

 

>> Ellen: Of owned.

>> Andrew Atkeson: We're gonna take the thing-

>> Ellen: The question I have is there's what we stick in the production function and then there's what we stick in the valuation, those are different.

>> Andrew Atkeson: Yeah, so you raised this last time I presented.

>> Ellen: The rules have not changed.

 

>> Andrew Atkeson: Okay, so the rules of the exercise so far is we're copying the numbers off of S5 and S6. And Ellen has an opinion that the capital number on S5 and S6 is not correct. And I advocate, we work on a memo to my student at the flow of funds and see if we can get to change those numbers.

And then I'll show the different results.

>> Speaker 26: Whether Tobin's q is one or some number larger or smaller than one?

>> Andrew Atkeson: Well, the fact that Tobin's hue can differ from one here is gonna come from that time bearing wedge mu that there's this factorless income that could potentially be positive or negative.

That's how we're gonna have deviations from Tobin's q.

>> Speaker 26: That's a separate. I thought we were just talking about a measurement issue.

>> Andrew Atkeson: No, no, I think what Ellen's concerned about is that much of the capital used in the corporate sector is actually owned by people who are not in the corporate sector cuz they lease it like I'm leasing a commercial real estate.

Or vice versa, the corporate sector might own a bunch of capital that's used by somebody in some other sector. And so Ellen's concerned with, there's something that's on a balance sheet table and there's something that goes into production function. And in a sectoral analysis, those two things can differ.

And I totally accept that statement. And Ellen has proposals for how they would change the data to make those things fit each other.

>> Ellen: No I keep the data the same, change what you do.

>> Andrew Atkeson: Go ahead.

>> Speaker 27: And, and Bob, the thing you call k, is that going to be the thing you measure as the value of the capital?

 

>> Andrew Atkeson: No, okay, so I dropped this equation, but I see that was a mistake in the slides. There's a thing that they call the reproduction value of the capital stock, and there's a thing called gross fixed capital formation, which These are both dollar numbers and they have a revaluation term for the capital stock, which is supposed to be the change in the price of capital.

If you were thinking of an adjustment cost model, whatever, that would be your q. Okay, now you can ask me what is that in the data? Mostly it's the difference between the investment price deflator and the GDP deflator. So in other words, they're not using-

>> Speaker 28: They're violating the capital.

 

>> Speaker 29: Tobin's ideas about Tobin's Q are not reflected.

>> Andrew Atkeson: Yes, that is correct. The BEA doesn't, so there's a question here. Yes.

>> Speaker 30: So when I look at the capital, and maybe this is something that you've already thought about and is not relevant here, but there's growth options and assets in place and the cost of capital and what the investors perceive about their riskiness could be different here.

We are just ignoring all of that.

>> Andrew Atkeson: Yes.

>> Speaker 30: And that is the, it gets at the whole debate of intangibles. And-

>> Andrew Atkeson: No, the intangibles is something else. Effectively everything is being pushed into this time bearing wedge.

>> Speaker 30: Okay, so what is different from what Barkai did five years, six years ago, except that you call it factorless income rather than profit?

 

>> Andrew Atkeson: Okay, so the difference with Barkai is that he attempted, he didn't use valuation data. He instead said, I think I have a measure of the cost of capital. And from the capital oiler equation I can infer rental rate with a user cost of capital.

>> Speaker 30: Yeah, exactly.

That's exactly what you do, that you subtract some cost of capital, that is cost of rental, and attribute the rest to profits.

>> Andrew Atkeson: No, but we're gonna hold off, you'll see in a second. We're gonna use the valuation data directly. In other words, we're never gonna bring in the capital Euler equation to estimate an arcade.

So with Barkai, what he's doing is he's saying, look, riskless interest rates have fallen a lot. I am going to have some opinion of what happened to the equity premium. And based on that, I'm gonna infer what's happened to that RK, and then I'm gonna say, I'm gonna compute the mu as I see the data on WL, I have my imputed RK, I have my K and I have my Y, and that gives me the mu.

So nowhere in that calculation did he use the valuation itself.

>> Speaker 30: But using valuation is tricky when there might be monopoly power. Think about, but the medallion of the taxi.

>> Andrew Atkeson: Yeah, so this is, I'll get there in the next slide. So the it just means mechanically, that after tax, gross value added is gonna be divided into three shares.

There's gonna be like a pure profit share. There's gonna be a labor share and implicitly a capital share.

>> Speaker 30: That's exactly what Barkai did.

>> Andrew Atkeson: Yeah, yeah, in terms of this part of the structure, yes. So-

>> Speaker 31: Who gets this? I mean, everything's paid out. I thought the equity holders were the owners of the capital.

So it's not.

>> Andrew Atkeson: They're also owners of the factorless income. So-

>> Luigi: The managers get a big chunk of it.

>> Andrew Atkeson: No, no, but okay, so now let's go to Luigi. So something that's there in the data is how would you get the value of firms worth less than the installed physical capital?

Okay, so there has to be a possibility for this wedge to result in negative factorless income. And how would you interpret that? And so we think of it as for a firm that owns its own capital stock, that rental is an implicit cost. It's not a cash flow cost.

And so if your managers are not so good or they're taking some of the money. The revenue that you earn from your installed capital and your labor might not cover the implicit cost of the capital. And that would leave you with a tobin's q less than one. Of course, if you have monopoly power, it could go the other way.

So we have a line here. Maybe this was for you, Luigi, in some long run sense that you should have these competitive pressures in the product market and in the market for control of firms that would try to push that factless income towards zero. But that might take a long time and may never happen.

But that's how we're thinking of it. But this would be our explanation for how you could have Tobin skew less than one is that either management's incompetent or they're taking the money or something like that. But it's not a-

>> Luigi: That's mutual funds who are trading a discount for a long time, right?

The manager was stealing some of the money.

>> Andrew Atkeson: Yeah, no, okay, so because we have no adjustment costs in fiscal capital. The value of the capital stock is always the reproduction value at the end of the period which we're gonna take off the data. And so we have enterprise value is the V without markings on it.

So the value of factorless income is just the residual. So without choosing parameters other than the no adjustment cost parameter, We get to see those two components. So that's what I mean by no parameters needed to measure this. So this is a graph of the two components. And this is what I was saying about the pre World War II data look a little different than afterwards.

So the left panel shows a red line, which is the reproduction value of the capital stock that's in the data overgrowth value added to the corporate sector. The blue line is the same old enterprise value. And then the right panel is enterprise value and this inferred value of factorless income where the axes are different because the capital output ratio is about two.

So what I want you to see is that in this great depression to World War II period, it's almost like the story you would expect that the reason firms got more or less valuable is the installed capital relative to output went up or down. So it's like a straightforward story as to where that came from, why they did that is a different question.

But that would be a story that would be straightforward. After that, it's like it's all Tobin's Q. It's all fluctuations in this thing other than the capital stock. And so that's what we'll kind of focus on afterwards. Okay, so wait, did I lose you? Okay, so now we wanna divide the cash flows and look at those.

And so we're gonna have to add one parameter. So we'll the heroic assumption that the production function is fixed over the time period. And so we pick the parameter alpha. We can look at the labor share and infer what fraction of output is factless income. So I admit it's heroic, but let's look at what we do.

 

>> Speaker 32: Period is constant.

>> Andrew Atkeson: 29 to the present, so that's a long time, or to 2022. So that's a long time, so but I just wanna show you what happens when you do this. And so what it says is, you take the free cash flow that you see out of the data, and you're just dividing it into a component that you're saying is rents, and a component that you're saying is free cash flow to capital.

Based on this heroic assumption, and a single choice of the parameter alpha, in this graph on the left side, we see the labor share, and we know it all, everybody knows it declined, and so that's data. And so, if you follow this identifying assumption of a constant production function and you choose a particular alpha, you're forced to say, factorless income went like that.

So it's just kind of the mirror image of it, the level of factless income goes up and down if you change alpha and so, with this particular choice of alpha, the average value of it over the sample is 1%. Okay, now, I said the interest.

>> Speaker 33: You used the phrase rent for the first time just now, and there's a competitive rent, which you can measure by assuming an adjustment cost structure that seems to be missing here.

 

>> Andrew Atkeson: It is, we have no adjustment costs. So I should say at the start that one can give different accountings with different models, so one could do a model with adjustment costs. One could do a model with intangible capital, so we're doing this as kind of a proof of concept with these data, why this might be useful to do, and so we're not trying to say that the other model, the one you used, is the wrong model.

We're just,

>> Speaker 33: I used it and concluded that the amount of it was way too great to be adjustment cost. So, in some sense, I'm on.

>> Andrew Atkeson: Yeah, I think so. I mean, yes, but plus, you unfortunately wrote your paper before the massive boom in cash flows occurred.

I mean, you wrote it when there's a massive boom in valuations, in the tech boom, but then you were asking, like, where are the cash flows we just had? If you'd waited 20 years, you would have seen them.

>> Speaker 33: Valerie's fault. She was the editor at the AER who enthusiastically accepted the paper.

 

>> Andrew Atkeson: So the lines here that I wanna stress are, so this is using so, Luigi, relative to Barkai, this is the top line is arcade. And so Simcha, Simca. Simca would have a different line, but. So, I mean, that's essentially the difference, there's the line. But what we find interesting is the red line subtracts off from those capital rentals investment, and you see this huge secular decline in free cash flow to physical capital.

This is gonna be kind of at the root of our puzzle, and Abel Mankiw Sommers Zechauser wrote a famous paper about dynamic efficiency, where free cash flow to physical capital was their measure of, is private investment dynamically efficient? They said yes, but with more updated data, it seems like were getting dangerously close to the answer being no.

And this could be a problem with our model, but it, it's something that's coming out of this accounting framework.

>> Luigi: I also had problem with the definition of capital, because if you invest in software, then it's not investment.

>> Andrew Atkeson: It is, it's in there.

>> Luigi: How do you measure it?

 

>> Andrew Atkeson: I didn't do it, the BA does. They added software, artistic I love it that artistic originals, which is the output of LA, is in there and R and D is in there.

>> Ellen: But you need to be careful about the k value. They have the investment, but they have zero information about the.

 

>> Andrew Atkeson: Yeah, yeah, but we're not using the stock of capital here, this is the cash flows that would be left over.

>> Ellen: What's the denominator?

>> Andrew Atkeson: Gross value added, the corporate sector.

>> Luigi: So the investment, adding the cost of software, etc, is still going down tremendously

>> Andrew Atkeson: well.

What's basically going down is investment is going up, but the capital rentals are not really going down very much, and Simca thinks they're actually going down more than we have there. So when you subtract investment off of your capital rentals, you say, what cash do I have left over, as the owner of capital?

You would say none, so this is not a return, this is a. This is just what cash is available to owners of this capital.

>> Speaker 17: Can I ask you, sorry, but I don't want to just throw down, I was trying to understand about the alpha as opposed to having an output cost wedge that was factor specific.

Is the alpha doing that? Yep, the mu t, that is just a scaling factor for the total cost, you could have a mu t, Mu k times k. You know, that's what the alpha, I miss the Alphabet.

>> Andrew Atkeson: No, the alpha is just saying that the alpha being constant, is saying the share of physical capital in factor costs is constant, that's what we're using.

 

>> Speaker 17: And towards if it is not throwing you off on the measurement structure.

>> Andrew Atkeson: Basically, we're just saying if you assume this Cobb Douglas production function with this wedge on the outside between revenue and factor costs, if you assume that the alpha is constant, you can just look at the labor share and you see what that wedge is

 

>> Speaker 17: Again is a proof of concept, the alpha constant is not natural to me, it should be constant.

>> Andrew Atkeson: Well, I mean this is an exercise that says this would you have to do, everybody has to do some division of free cash flow into the two components. We thought it was a simple division, but it's leading, leading to this thing which is little startling does.

 

>> Speaker 34: Can you remind me why that hints at dynamic inefficiency again?

>> Andrew Atkeson: Okay, so basically think of

>> Speaker 34: like r minus g stochastic version being negative.

>> Andrew Atkeson: No, but think about it, just think of yourself on some balanced growth trend. And if the investment you have to do to stay on the balance growth exceeds your capital rentals, then you're like what are you doing, what business are you in?

So their papers about proving that even in a stochastic model that on an average basis or a trend basis this thing being positive or negative is the thing you wanna look at

>> Speaker 35: why things fishery. These firms could have been investing enormous amounts for future cash flow not past pre 2000.

 

>> Andrew Atkeson: Right? This will get my shot at intangible capital, another explanation as opposed to practice income is that there was massive installment of physical capital. Intangible, I'm sorry, intangible capital. And so there's two questions in my mind that I won't speak for my co authors that come up with regard to that.

One is and Ellen will jump in, I'm sure. One is that Carol Carrado and co authors have done a big project to try to measure this stuff and assume dependent. Depreciation rates and construct capital stocks. And they have a nice Journal of Economic Perspectives article from a year ago and they linked their most recent data.

You download it and you see massive investment. But if you look at what has happened to what they think is the stock of this stuff that's unmeasured, it's about 20% of value added and hasn't moved in two decades.

>> Speaker 36: Just a second, there's an objectionable feature in that work, at least as it was done a while ago, which was that almost all the intangible capital is taken to be advertising.

Yeah.

>> Andrew Atkeson: No, no, so I'm just saying if you follow a strategy, and I'm challenging Andre Isfeld and Dimitri Spafnik Lau on this, if you follow a strategy of saying, I think I see what the investment is, and they do it with SG&A and stuff. If I assume a depreciation rate, I construct a capital stock with a perpetual inventory method.

And can I explain a boom in valuation of 100%, 200%, 300% of value-added based on accumulation of this capital of the same magnitude, of the exercises of this kind of, which I'm aware of, I don't think that's what comes out. I don't think anything close to that comes out.

 

>> Ellen: Because they don't see the capital stocks.

>> Andrew Atkeson: Okay, so another approach Bob took it and Ellen takes is to use the Tobin's q to infer how much of that capital is there. So implicitly you have to be saying, when we see this enterprise value going like this, you know, 100%, 200%, 300% of gross biotech, is that over the century, there's massive swings in the capital-output ratio for intangible capital.

She's shaking her head no. She's gonna say taxes on dividends. Okay, that's coming next.

>> Ellen: Next week.

>> Andrew Atkeson: No, so the Daryl, the question is if the firms were doing some investment, like paying a bunch of workers to do some R&D or do something that wasn't being measured by the BEA, then that would be lowering their free cash flow somewhere.

And so we should see somewhere periods of low free cash flow to compensate for this.

>> Speaker 35: We're seeing low cash flow around 2000.

>> Andrew Atkeson: No, no, but if we go back, we had that 50-year period where free cash flow was like super stable and we still had the market going up and down a bunch.

So if you wanted to do intangible capital for the 50s, 60s, 70s and 80s, you'd be sitting there looking at, Ellen's gonna say it was taxes. You'd be saying like they're somehow managing to pull off huge variations in the intangible investment rate without it affecting their free cash flow, which is an interesting trick to pull off.

 

>> John Taylor: I'm gonna make sure you're on time here.

>> Andrew Atkeson: Yeah, okay, well, I can finish any time, cuz these are issues.

>> Andrew Atkeson: I mean, this is the one that I'll probably lose people on. So this is the one we'll probably lose people on, which is to just ask, and unfortunately, John has gone to teach, Shiller, planted a flag and said there's just too much volatility in valuations relative to cash flows, this excess volatility calculation.

So we're just gonna redo an excess volatility calculation, but we have to do it a certain way. So what I'm gonna try to do, we have this measure for this kappa t, which is the fraction of gross value added that is factorless income. And in valuing it, I feel I suffered doing this in an appendix.

You can price it kind of like zero coupon bonds the same way you do for the term structure of interest rates, but you start by pricing a claim to aggregate output, and then factorless income is a, is a claim to a fraction of aggregate output. So they can be positive or negative.

So in general, you could posit a pricing kernel and work out formulas for this. And in an appendix, we did it for essentially affine pricing kernel, where kappa can be positive and negative. So I've never seen this formula before, but it's kinda trivial and I think people must know it, but I'm gonna skip it in the interest of time.

But, so, if you want to estimate this whole damn thing, it's one can do it. But what we're gonna do is assume, when we say that expected returns are constant, we're just gonna assume that the price to a claim of aggregate output is constant. And the covariance, the additive covariance term that comes into the valuation exercise is also constant over time.

And so when we do that, we have to say, well, we're gonna posit a stochastic process for the factorless income share and call out two numbers, I'll show you what the numbers are. What is the value of a claim to aggregate output relative to aggregate output? And either three numbers, I guess the persistence of Kappa and the f.

But we're gonna make one change, most people do VARs with a fixed endpoint, so long-run expectations never move. We're goNNA do AR(1) with a time-bearing endpoint. So it's basically what you have to think of is this x shock is investors are thinking in the long run, the share of factorless income is gonna go to this number versus that number.

And why is that important? Well, it's gonna affect the valuation of factorless income. You see a coefficient VY/Y in terms of how big is a claim to aggregate output or the price dividend ratio for that relative to the price. Anyway, I'll show you why that's important. So this is a baseline result that we get with a price dividend ratio for aggregate output of.

 

>> Speaker 17: XT is backed out residually, then?

>> Andrew Atkeson: Mmh?

>> Speaker 17: XT is backed out.

>> Andrew Atkeson: It's gonna be backed out, if you picked a VY/Y, Xt is the only thing you don't see. So the blue line in the figure is the kappa t we had before. The red line is where investors at each date are thinking it's gonna go in the long run.

That's backed out from the valuation data. And a key number is we're using by VY/Y of 50. It does also true that that red Xt forecasts where kappa is going in the future. Not very strongly, but at least it's there. But the question should go through your head is, okay, is that too volatile?

And the answer to that question, we wanna say is really critical on what you think the price of a claim to aggregate output is, or the price dividend ratio. So we did sensitivity saying, let's make that 25 and let's make that 100. If you make it 25, the red line on the left starts getting super volatile.

And then I would be sympathetic with saying that if I discount the long run at something close to the discount rate for equity, I'm gonna fail an excess sensitivity exercise. But if I think that a price to claim to aggregate output, the price dividend ratio is super high because aggregate output's not that risky a claim.

I'm gonna say, no, I have no problem with X sensitivity. And Olivier Blanchard effectively argued And Stavros does this, too, that the price-dividend ratio for a claim to output is infinite. And then you get volatility as much as you want. And so we don't wanna say that returns are constant, but we wanna say excess volatility exercises that have been done are using the wrong discount rate.

And so they're arriving at the conclusion that there's excess volatility, as we have in the left graph. And if you use the right discount rate, it's a much more open question as to whether there's excess volatility.

>> Speaker 37: You're kind of flying the last minute.

>> Andrew Atkeson: Yes, I agree.

 

>> Speaker 37: Can you back up like a slide and just give us maybe a helicopter view? I thought I got lost. Hate to tell you on that last, this x how you used it. A summary again.

>> Andrew Atkeson: Okay, I'll try to summarize it this way. So the logic of the original excess volatility exercise, and then corrected with Campbell Schiller 87, is that I see a return on equity that's pretty high.

So an r minus g for equity is a pretty high number, like maybe six, seven, whatever. And so, and then I have a VAR of what I think the cash flows are gonna be.

>> Speaker 37: AR, VAR, VAR.

>> Andrew Atkeson: Well, they use the VAR.

>> Speaker 37: Okay.

>> Andrew Atkeson: Yeah, but a key thing about that is if you have a fixed endpoint, your expectation of what cash flows are gonna be in the long run never moves.

And so you've got two things going on.

>> Speaker 37: What is the fixed endpoint? Could it the world stationary or something or?

>> Andrew Atkeson: Yes.

>> Speaker 37: We can't be-

>> Andrew Atkeson: Well, they did it in price-dividend ratios, and they want those to be stationary.

>> Speaker 37: Okay.

>> Andrew Atkeson: So you have two forces going on that you say, my discount rate is super high, so expectations of the long run have trouble influencing today much.

But plus, I force people to have time in varying expectations of the long run. And what we're saying is open up the possibility that they have changing expectations of the long run and then recognize that what we're pricing shouldn't be discounted at the r for equity. It should be discounted at the r for a claim to aggregate output because the risk terms are additive, that's for the people who care about this.

And in that case, in macro, it's a matter of huge debate as to what the value of a claim to aggregate output is relative to output.

>> Speaker 38: This is called the Economic Policy Working Group, but I didn't hear any policy.

>> Andrew Atkeson: I'll give you policy right now.

>> Speaker 38: Hold on, I have a question.

 

>> After each one of these seminars, I get a call from the White House and the Congress, and they say, wow, this was really an important seminar. Now, what can I take away from this to make the country work better.

>> Andrew Atkeson: Can I answer that?

>> Speaker 38: Well, that's exactly, yeah, I wanna know.

 

>> Andrew Atkeson: Right now.

>> Speaker 38: Yeah, right now, yeah, I wanna know what's going to make the country work better.

>> Andrew Atkeson: This is exactly the same problem as asking, how much do we have to change primary surpluses to pay off the outstanding government debt? Okay, so the argument as to-

 

>> Speaker 38: Wait a minute, no, firms can go under, governments don't.

>> Speaker 39: [LAUGH.

>> Andrew Atkeson: No, slow down, think of it this way, the primary surplus is a fraction of aggregate output. So when we look at the government debt and it goes up or down, we say, how does the value of a fraction of aggregate output change when we raise it, the primary surplus from 1% to 2%?

Okay, if VY/Y is 50, that's 50% points of GDP. If it's 100, it's 100% points of GDP, if it's infinite, is an infinite amount. So if you think VY/Y is high, if you're living a Blanchard, you basically say we don't have to make any fiscal adjustment to pay up our rising debts.

And if you use VY/Y is 25, you're going to say, we have to make a punishing fiscal adjustment to pay off, say, a debt to GDP going up by another 100% points. So this issue of valuing firms and valuing fiscal surpluses, we wanna argue, is the same issue.

And then this debate as to what this number is-

>> Speaker 38: Any fiscal surpluses ever?

>> Speaker 38: No, I mean, I don't mean to push this point, but I'm asking about reality.

>> Andrew Atkeson: So in reality, you are exactly right, and that's why Olivier Blanchard wrote his paper saying, VY/Y is infinite, r minus g is negative, in which case, you don't need fiscal surpluses to pay back the debt.

You're in the same case. In other words, so the two issues are closely linked.

>> Speaker 38: How much time do you spend in Washington?

>> Andrew Atkeson: I grew up there.

>> Speaker 38: No, no, no.

>> Speaker 38: Since you've been doing this stuff, since you've been a professor?

>> Andrew Atkeson: They're misguided, look, debt to GDP has been falling for the last year and a half despite massive surpluses.

How did we pull that off? That's r minus g being negative. Somehow the fed convinced people to leave interest rates low while nominal g has been very high. This is the essence of the fiscal debate, it's the same debate. Well, no, I'll do one more thing, what's the puzzle about capital?

Okay, so the puzzle about capital is we get this one parameter that allocates free cash flow between the two components. We see returns on enterprise value, we saw they look like CRISP returns, so we know they're high. Is the return to physical capital high? I'll do quickly, depends on what you choose, alpha.

What we find most difficult is that it's the dynamics of excess returns on capital. There's a period from World War II to the end of the 1970s where it appears that the realized excess returns on physical capital are gigantic. And then there's a period from the early 1980s to COVID where they look like, maybe they're not so high.

And so then we regard this as a puzzle. Why would the returns to capital be so different in these two different time periods? And taxes could be an answer, a change in the production function. But we see this as something, even as we use all this valuation data, this is a question we wanna answer.

 

>> Speaker 10: Maybe I missed it, but it'd be good to see, especially for the last part of your argument, you explicitly make the case that we're better off trying to do these volatility calculations at a macro level with highly imperfect data. Than just taking the stock market data the way Campbell and Schiller did, where we've at least got market-based valuation measures.

So it's not obvious to me, even if you're ultimately interested in the Blanchard question, which approach is better.

>> Andrew Atkeson: But we showed you that the returns in the valuation are the same in the two datasets.

>> Speaker 10: Yeah, but then I have to rely on the whatever Whatever these guys did in the sausage making, and I don't know what they did, and I know it's a hard problem what they're trying to do.

 

>> Andrew Atkeson: Yeah.

>> Speaker 10: But I don't know what they did.

>> Andrew Atkeson: So as I said, are you moved permanently? Are you still in Chicago?

>> Speaker 10: I'm here permanently.

>> Andrew Atkeson: Okay.

>> Speaker 10: I'm so happy.

>> Andrew Atkeson: Your former colleague Uran Ma. So we would hope to investigate the public firm data as much as we can and say what are the equivalent concepts in compustat that we can bring to the table.

We're gonna be limited to talking since like the mid eighties to now, but so we can investigate doing what you said, and we're totally open to that. I mean, we see the point of this as an ongoing research agenda to say macro and finance guys should try to coordinate on a common dataset, there's gonna be work to improve the measurement, I'm sure.

 

>> Speaker 10: Yeah, that requires economists digging in to what the people at the BA are actually doing, which most economists shy away from that kind of work, which.

>> Ellen: Okay, so I just want to stick up for the government statistical agencies because I was on FEESAC, federal economic statistics advisory committee for years and they do a really good job.

And there's a reason that they have academic economists come in several times a year and they call them, even in between meetings and really do try to do the best possible job. So they do get a lot of input from academic economists.

>> Speaker 10: They also make mistakes.

>> Andrew Atkeson: No, they change with input.

 

>> Speaker 10: Yes, they do. I mean, look, I'm very sympathetic to your, the line you pushed early on that we ought to engage with the statistical authorities to help them do a better job. Because what they try to do is, in this case in particular, is actually quite difficult and we have a lot of input to provide, so.

 

>> Ellen: Yes,

>> Speaker 10: and that said, I can quote examples outside where I think they blunder.

>> Speaker 40: So Andy, the factorless income was a new concept to me, so I went back to look at the caribbean paper and they lay out three sources of why it exists the first is monopoly profits, which they say they don't think it is.

The second is mismeasurement of capital, which I would think there's definitely some of that going on here. And then the other thing is mismeasurements and required rates of return that the firms are using as their discount rates, and making investment projects versus what you are measuring in the model.

Which of those three do you think is the most important? Or if you had to allocate the.

>> Andrew Atkeson: Okay, so a variant you have a highly volatile mu in the data, and what is the source of that volatility and. Okay, so the literature, Barki came up and there's a number of papers Caricaboose and Niemann comes up, there's Farigurio, and there's Cruzette and Amberly.

And Faragurio and Crissette, Nebuli and us do something so the first two, Kirkbus and Niemann and Barkai, are trying to use the user cost of capital and for rental rate and come up with. But they never use valuation data and so far, Gurio, Cruzette, Nemberle and us are trying to use valuation data, and then we're not using the user cost of capital.

So there's like, in a beautiful world, all of those would light up and so the last graph, in some sense, was, what does the return to capital that we would need to justify our valuations have to be? And we say there's a puzzle that on capital, it's like super high for three decades, and then it's, like, not so high.

And so we're posing that is, it's like there's a puzzle we have all these pieces that are supposed to fit, and this is like a piece that doesn't seem to be fitting. And so we're not saying it's done in terms of the project.

>> John Taylor: Thank you.

>> Andrew Atkeson: Thank you very much.