Sunday, October 6, 2013

Dupor and Li on the Missing Inflation in the New-Keynesian Stimulus

Bill Dupor and Rong Li have a very nice new paper on fiscal stimulus: "The 2009 Recovery Act and the Expected Inflation Channel of Government Spending" available here.

New-Keynesian models are really utterly different from Old-Keynesian stories. In the old-Keynesian account, more government spending raises income directly (Y=C+I+G); income Y then raises consumption, so you get a second round of income increases.

New-Keynesian models act entirely through the real interest rate.  Higher government spending means more inflation. More inflation reduces real interest rates when the nominal rate is stuck at zero, or when the Fed chooses not to respond with higher nominal rates. A higher real interest rate depresses consumption and output today relative to the future, when they are expected to return to trend. Making the economy deliberately more inefficient also raises inflation, lowers the real rate and stimulates output today. (Bill and Rong's introduction gives a better explanation, recommended.)

So, the key proposition of new-Keynesian multipliers is that they work by increasing expected inflation. Bill and Rong look at that mechanism: did the ARRA stimulus in 2009 increase inflation or expected inflation?  Their answer: No.


This is a quantitative question. How much do the large-multiplier models say the ARRA should have increased inflation? Their answer: 4.6%. Where is it?

We know, of course, that inflation (especially core inflation) basically did nothing during the period of the ARRA, and Bill and Rong have some nice graphs. Defenders might say, aha, but except for the stimulus, we would have had a catastrophic deflation spiral. Critics might reply, that's what George Washington's doctors said while they were bleeding him. As always, teasing out cause and effect is hard.

Bill and Rong have a range of interesting facts that address this question. Here are two that I thought particularly clever. First, they look at the survey of professional forecasters, and examined how the forecasters changed inflation forecasts along with their changes in government spending forecasts, i.e. when they figured out a big stimulus is coming. I plotted the data from Bill and Rong's Table 2

Dupor and Li Table 2
As you can see, in 2008Q4 and 2009Q1, many forecasters updated their views on government spending, a few by a lot.  However, there is next to no correlation between learning of a big stimulus and increases in expected inflation, especially among the forecasters who strongly update their stimulus forecasts.

Bill and Rong's interpretation is that the stimulus failed to increase expected inflation. The main defense I can think of is to say that this evidence tells us about professional forecaster's model, not about true inflation expectations. Professional forecasters are a bunch of old-Keynesians, not properly enlightened new-Keynesians; they don't realize that stimulus works through inflation, they're still thinking about a pre-Friedman consumption function. That's probably true. But if so, it's hard to think that everyone else in the economy does understand the new truth, and changed their inflation forecasts dramatically when they learned of the stimulus.

Another nice piece of evidence: The US had much bigger government spending stimulus than the UK. The behavior of expected inflation revealed in the real vs. nominal treasury spread was almost exactly the same. (Yes, Bill and Rong delve into the TIPS pricing in the crisis.)

Source: Dupor and Li

Finally, a key point missing in most of the stimulus debate. These models predict big multipliers not just at the zero bound, but anytime that interest rates don't respond to inflation. We don't have to just rely on theory, there is some experience. New-Keynesians since at least Clarida Gali and Gertler's famous regressions have said that the Fed was not increasing interest rates fast enough in the 1970s, and the 1930s and interest-rate peg of the late 40s and early 50s are another testing ground. Using standard measures of exogenous spending increases, Bill and Rong find no impact of government spending on inflation in any of these periods.

New Keynesian stimulus analysis has been particularly slippery, on the difference between the models and the words, and on advocating the policy answers without checking or believing the mechanisms. The models are Ricardian: the same stimulus happens whether paid for by taxes or borrowing. The opeds scream that the government must borrow. The models say totally useless spending stimulates. The opeds are full of infrastructure, and roads and bridges. (At least, the "sprawl"  complaint is temporarily quiet.) The models say that spending works by creating inflation, not through a consumption function. Inflation being totally flat, and the counterfactual argument weak, you don't hear much about that in the opeds.  The models say we should be in a huge deflation with strong expected output growth. The facts are protracted stagnation. (More in my last stimulus post.) The models are models, worthy of careful examination and empirical testing. All I ask is that their proponents take them seriously, and not as holy water for a completely different old-Keynesian agenda.

29 comments:

  1. I'm not sure the professional forecasts are a reasonable metric. Remember that in 2009 the BEA made almost unprecedentedly large revisions to its previous GDP figures, showing that in fact the recession had been far worse than anyone, including the forecasters, thought. So the failure of forecasters to update their forecasts in the way Dupor and Li expected could easily be explained by the coincidental revisions to official GDP figures.

    ReplyDelete
  2. Will higher levels of G really increase inflation or expectations of inflation if the economy has slack labour markets and overcapacity?

    Higher inflation will send up borrowing rates because lenders don't want to lose out. The real rate of interest on base will go down if it stays at some nominal number while inflation goes up but its irrelevant to the broader economy because the broader economy doesn't borrow base money directly from the fed.

    "stimulus works through inflation"

    Doesn't stimulus work through higher Y. If my inflation increases Im not going to borrow more or invest more or feel better about my economy. If my income rises thats a different story.

    ReplyDelete
    Replies
    1. The issue is how new Keynesian models work, and whether their predictions are correct. The issue is not how you think the economy works, or old Keynesian thinking, or anything else. Good economics tests the actual models at hand, and that's what this paper and this post are about.

      Delete
    2. Exactly. So the model must be incorrect if it is incorrectly constructed. The model seems to imply that higher inflation wont affect market borrowing rates. Of course real rates on CB lending will be lower with higher inflation. But like I said the economy doesn't borrow from the CB. Lenders will lose out lending into higher inflation economy if they don't increase rates. Do you disagree?

      Delete
  3. Professor Cochrane,

    Your final empirical point is that, in a variety of scenarios without monetary offset, an increase in government expenditures does not lead to higher inflation. Can you explain how this fits into your view of the macroeconomy? Even in a world with perfect Ricardian equivalence, the demand effect of higher G is offset, but aggregate supply is pushed down (assuming the government is less efficient than the private sector). This should drive up inflation as NGDP stays stable but inflation replaces RGDP.

    ReplyDelete
  4. The two graphs showing the increase in real government expenditure and expected inflation, at least for the US, looks like there is the increase in expected inflation post the stimulus. Again, for the US, it looks like a drop in expected inflation in 2010 when the stimulus wanes.

    I know you are using the graphs for a different purpose, but if isolating the performance of US inflation expectations, doesn't it show the NK prediction of more inflation with stimulus is valid?

    ReplyDelete
  5. I have to agree with Matthew--now I'm not a professional economist, but it seems like Dupor and Li are tracking inflation by using economists' expectations for inflation. However, the finger pointing at the moon is not the moon; have they compared gov't. expenditure with price changes, the CPI (chained and unchained), or something else?

    ReplyDelete
    Replies
    1. The variable of interest in their paper is expected inflation, not actual inflation. Forecasts are the relevant indicator for that, not CPI.

      Delete
    2. Please help me understand this--again, I'm not an economist. Why would we need to look at expected inflation more than actual inflation?

      Delete
    3. The real interest effect. Increases in expected inflation will ex ante decrease the real interest rate. The real interest rate thought of today going into tomorrow is nominal interest rate minus EXPECTED inflation. For it to actually be a stimulus today real rates must be lower between today and tomorrow, which would happen if people expect higher inflation.

      If you thought people were good forecastors you could look at ex post inflation, and treat it as the prior expectation, but I doubt that would actually help the NK models in this case.

      Delete
  6. The key issue is your assumption that the counterfactual about deflation is weak. What is this assumption based on? if we look at asset price deflation at the height of the crisis ( plunge in values of various financial assets ) - this would be quite strong empirical evidence for the counterfactual

    ReplyDelete
  7. Not to be nitpicky, but I think Old Keynesian models also predict higher inflation for an upward-sloping aggregate supply curve. Yes, that is a by-product of raising income directly instead of being the channel of transmission as in the New Keynesian story, but it looks like the paper's results are also bad news for someone thinking about fiscal policy in terms of the Old Keynesian model.

    ReplyDelete
  8. "New-Keynesian models act entirely through the real interest rate. Higher government spending means more inflation."

    Maybe this is true of some New Keynesians, but not the ones I mostly read. For example, Simon Wren Lewis:

    "...as Professor Michael Woodford has recently shown, we get exactly the same multiplier of one if consumers are much more sophisticated, and look at their entire lifetime income when planning their consumption. The basic intuition here is that any temporary tax increase gets smoothed over their lifetime, so the impact on current consumption is small. As the simple Keynesian case shows, any short term impact there is will be offset by higher incomes generated by higher government spending.

    This all assumes an unchanged level of real interest rates."

    (Emphasis mine.)

    http://mainlymacro.blogspot.ie/2012/02/balanced-budget-expansion-debate-begins.html

    ReplyDelete
    Replies
    1. '...if consumers are much more sophisticated, and look at their lifetime income when planning their consumption..."

      I don't understand this. It appears to be one whale of an assumption. Isn't it possible to arrive at any conclusion given an inappropriate "helpful" assumption (such as this one appears to be)? Is Wren-Lewis looking at the world as it is, or a hypothetical world that does not exist? Isn't the problem of any model the failure to make realistic assumptions?

      If consumers were actually that sophisticated, why are consumers, on average, woefully short on retirement savings?

      Delete
    2. The quote, and that part of Woodford's "simple analytics" is about the background 1 multiplier. The issue is how to get multipliers greater than 1, which is about intertemporal substitution in the NK model.

      Vivian illustrates the paradox of Keynesian economics. We are in trouble because, apparently, a glut of savings driving the "natural rate" below zero, and this terrible saving must be stopped and consumption increased. Yet somehow everyone hits retirement with no savings.

      Delete
    3. If "The issue [for you] is how to get multipliers greater than 1", fair enough, you're entitled to select your own issues. "New-Keynesian models act entirely through the real interest rate" is another thing altogether. If dY/dG = 1 even with no movement in the interest rate then that claim is false.

      Delete
    4. That's not correct that 'everyone hits retirement with no savings'. A handful of people end up in retirement with huge amounts of money, the vast majority do not. The people that had lots of money have an 'excess' of savings. The people that didn't, still don't. There is no contradiction here. Surely you need to look at the aggregate savings.

      Delete
    5. So then why are there aggregate lower levels of investment. If savings are not well distributed, in fact they belong mostly to just a few, then we should have even more investment spending because rich people invest at much higher rates than the middle class.

      Delete
    6. in fact, most people being short on savings in retirement presents a much bigger problem for classical economics than Keynesian. Keynesian assumes a temporary situation in which economy is in a bad equilibrium. Classical assumes that people are rational and markets are efficient. Surely, rational actors in efficient markets would rationally over their lifetimes save enough money to have sufficient savings, especially given pretty high level of GDP per capita in the U.S. Yet this is not what happens. h

      Delete
    7. KyleN: Again, there is no contradiction here. Aggregate investment is low because aggregate demand is low and the economy has over capacity. Rather than invest in new capacity, the people with savings park them in US Treasuries. As near as I can tell, all Dupor and Li and Cochrane have shown is old Keynesian models work better than new Keynesian models.

      Delete
    8. Re: Retirement savings - Don't you think that social security and medicare (the whole panoply of social welfare programs) has had an impact on the savings of the typical U.S household? When we socialize retirement savings we eliminate an important source of diversification across individuals both cross-sectionally and over time and are simply left with the systemic risk. In the end, social security and medicare are more risky for each individual than if each were solely responsible for there own retirement.

      Delete
  9. "The models say totally useless spending stimulates. The opeds are full of infrastructure, and roads and bridges"
    To play Devil's Advocate, it could be that useless spending is good, but useful spending is even better, so why not go for useful?

    ReplyDelete
    Replies
    1. In my view I think there is a multiplier for some sorts of infrastructure spending. After all, it is undeniable that commerce increases when a new modern road is built between two cities. But that return on investment is only noticeable over a long period of time. Much too long to be an answer to a recession.

      Delete
  10. Dear Professor,

    I find it somewhat amusing that in your prior post (and your related new paper) you were asking "where is the deflation?" predicted by NK models (given the large negative shock to the "natural rate of interest"), while now you are asking "where is the inflation?" predicted by the same model (given the fiscal and monetary stimulus ). So which one is it that would contradict the NK model? Maybe the US government (and the FED) just managed to exactly offset the negative shock by their expansionary fiscal and non-conventional monetary policies.

    In fact, under the optimal commitment plan (in the same NK model) inflation barely moves throughout a liquidity trap episode. It's virtually constant. That's because there's no need, in fact it's harmful, to promise much future inflation given that it is just as costly as the deflation you want to prevent. So instead you lower the ex-ante long term *real* interest rate by promising a zero *nominal* interest rate beyond what you would do under discretion but not so long as to actually generate inflation at the end. Thus, after being stuck at zero for just the right amount of time, the nominal interest rate jumps up discretely (according to Werning and my own work) so as to prevent any inflation post-liquidity trap.

    Of course the authorities must be real masters of the universe to be able to get the timing just right. All I'm saying here is that it is theoretically possible, and that it is the optimal prescription under commitment.

    ReplyDelete
  11. Apart from Dupor and Li’s evidence that New Keynsianism doesn’t work, NK strikes me as being a THEORETICAL non-starter.

    If I’ve got it right, more government spending is supposed to raise expected inflation. But what possible transmission mechanism do the “expectors” have in mind?

    If government raises spending by just enough to bring the economy up to capacity, there won’t be much additional inflation. So there’s not much room for the NK effect.

    On the other hand if government raises spending too far, then significant inflation ensues, and the NK effect becomes operative: it will augment an already elevated rate of inflation.

    Conclusion: assuming government acts in a competent manner, there’ll be little by way of an NK effect, so the NK effect is near irrelevant.

    So what next? I think I’ll try popularising the idea that inflation is related to the number of people who think the Moon is made of cheese: I’ll call it the “cheese expectations” theory. If I succeeded, I’m sure hundreds of economists would then burn up taxpayers’ money looking for empirical evidence to back or not back “cheese expectations”. After all: they’ve got to find some way of keeping themselves employed.


    ReplyDelete
  12. John,

    "A higher real interest rate depresses consumption and output today relative to the future, when they are expected to return to trend."

    Careful here.

    First, a higher real interest rate is a result of policy, not a policy in and of itself. The federal reserve sets a nominal rate of interest, the real rate of interest is measured after the fact. A higher real interest rate is a result of depressed consumption and / or increased production not a cause of it.

    Second, you have to assume a closed economy - no external financing, no trade imbalances and you have to eliminate equity financing to reach the conclusion that higher real interest rates become a positive feedback loop - higher real interest rates cause more people to save instead of spend which drives real interest rates even higher.

    ReplyDelete
  13. John Cochrane:
    “New-Keynesian models act entirely through the real interest rate. Higher government spending means more inflation.”

    I think Cochrane expressed himself somewhat unclearly. What he meant to say is:

    New-Keynesian models generate fiscal multipliers *greater than one* at the zero lower bound in interest rates through the impact of higher government spending on inflation expectations. In such a context higher inflation expectations means lower expected real interest rates, which causes households and firms to bring consumption and investment forward due to intertemporal substitution. This is precisely how New-Keynesian models produce estimates of fiscal multipliers greater than one (i.e. why the change in aggregate expenditures is greater than just the increase in government spending).

    Is this true of all New-Keynesian models?

    Well it’s true of:

    The Zero-Bound, Zero-Inflation Targeting, and
    Output Collapse
    Lawrence J. Christiano
    March 2004

    When is the Government Spending Multiplier Large?
    Lawrence Christiano, Martin Eichenbaum, and Sergio Rebelo
    December 2010

    Optimal Monetary and Fiscal Policy in a Liquidity Trap
    Gauti B. Eggertsson and Michael Woodford
    September 2006

    Is There a Fiscal Free Lunch in a Liquidity Trap?
    Christopher J. Erceg and Jesper Lindé
    July 2010

    Simple Analytics of the Government Expenditure Multiplier
    Michael Woodford
    June 2010

    Just for starters.

    So it may not be true of all New-Keynesian models that produce such estimates, just all of the most important ones.

    I normally disagree with Cochrane when he argues that demand plays a limited role in business cycles.

    However in this particular instance Cochrane deserves to be applauded for pointing out the fact that New-Keynesian estimates of large fiscal multipliers are the result of the implausible assumption that the only channel of the Monetary Transmission Mechanism is the Traditional Interest Rate Channel.

    And based on the evidence presented by Dupory and Rong Liz (the subject of this post) if this assumption is true then the fiscal multiplier of ARRA is very unlikely to have been greater than one.

    P.S. Of course, since the assumption that the only channel of the Monetary Transmission Mechanism is the Traditional Interest Rate Channel is contrary to standard undergraduate textbook Monetary Economics (e.g. Mishkin), the fiscal multiplier is *zero* given an independent central bank will always and everywhere choose to offset fiscal policy.

    ReplyDelete
  14. "New-Keynesian models generate fiscal multipliers *greater than one* at the zero lower bound in interest rates through the impact of higher government spending on inflation expectations. In such a context higher inflation expectations means lower expected real interest rates, which causes households and firms to bring consumption and investment forward due to intertemporal substitution"

    Problem is inflation expectations will make lenders increase rates so the people wont experience lower real rates. CB rate will be lower in real terms but its irrelevant.

    ReplyDelete
  15. C+I+G has lost its meaning with G(including healthcare) more than half of I+ C and so intricately intertwined it is all one big ball of spaghetti.

    ReplyDelete

Comments are welcome. Keep it short, polite, and on topic.

Thanks to a few abusers I am now moderating comments. I welcome thoughtful disagreement. I will block comments with insulting or abusive language. I'm also blocking totally inane comments. Try to make some sense. I am much more likely to allow critical comments if you have the honesty and courage to use your real name.