Tuesday, October 18, 2011

Matt Rognlie Needs to Learn Some Monetary Economics

Matt thinks that monetary frictions don't matter. Mike Woodford made the same mistake, and set a large fraction of macroeconomists off to work on New Keynesian models. Then, the financial crisis hit, and central banks started to engage in some unprecedented interventions, about which standard New Keynesian models had nothing to say.

In a basic Woodford model (see for example Mike's book, Interest and Prices), all monetary frictions are stripped away. In a Woodford world, the only friction comes about because of nominal price stickiness (and perhaps wage stickiness too), which leads to relative price distortions. What does a central bank do in a Woodford world? It sets the price of a bond which is a claim to "money" in the future, but this money is not actually held by anyone, in spite of the fact that all prices and wages are denominated in units of the stuff.

Why does central banking matter? It matters because a central bank can engage in intermediation activities that are not replicated in the private sector. A typical central bank has a monopoly on the issue of currency (in the US this is implicit), and on the large-value payments system. Thus, currency and bank reserves are liabilities that cannot be issued by private financial institutions. When the central bank issues its liabilities in order to buy assets, this in general matters. In particular, asset prices move. To understand how that process works, one has to model the frictions that make private intermediation useful, and the frictions that make assets of all kinds (including the ones conventionally called "money") useful in exchange. By "exchange," I mean exchange of all kinds, including retail exchange, and exchange among financial institutions.

Matt, for starters, you can read this blog post, this piece with Randy Wright, our chapter in the Handbook of Monetary Economics, and this forthcoming AER paper. The latter shows you why you need monetary frictions to understand the financial crisis and unconventional monetary policy.

25 comments:

  1. isn't it possible to think money matters without thinking that monetary frictions matter? one is a general concept about the economy and the other is a particular mathematical and verbal assumption made to generate a phenomenon resembling that concept within preexisting models.

    ReplyDelete
  2. Geeze, it's as if you are groping to reinvent the monetary economics of Mises and Hayek.

    OK, I'm well aware, you have no knowledge of their seminal work on the topic, extending the marginalist revolution to money, finance & macroeconomics.

    But just sayin'

    ReplyDelete
  3. I tend to be on your side on this question. But in Matt's defence, he did say he was talking about *conventional* monetary policy.

    To my mind, the biggest problem in Mike Woodford's model is this: start in full equilibrium, with expected inflation equal to zero, and the central bank setting a nominal interest rate equal to the natural rate. Now suppose the central bank just flips out and raises the interest rate. The result is supposed to be a recession. Output supposedly drops below even the inefficient (because of imperfect competition) natural level of output.

    But if there were no monetary frictions, why don't the firms barter their way back to the natural level of output? Hell, why stop there; why don't the n imperfectly competitive firms do a big n-sided barter exchange that gets them all the way to the efficient level of output? Or n(n-1)/2 two-sided barter exchanges could do the same thing.

    There has to be some implicit monetary friction in that model, or the "equilibrium" doesn't make any sense. But it's implicit.

    ReplyDelete
  4. In other words: If you allow barter in Woodford's model, even if all barter exchanges are done at the sticky prices, the only inefficiencies, *even if the interest rate were far too high*, would be minor relative output distortions because the Calvo fairy visits different firms at different times. You would never see *all* firms producing less than the efficient level of output.

    ReplyDelete
  5. "...in Matt's defence, he did say he was talking about *conventional* monetary policy."

    Unconventional policy highlights the problem, but you need to model the frictions to understand the effects of "conventional" policy too.

    ReplyDelete
  6. Stephen is spot on that monetary frictions do matter, once we look at all levels of exchange. Matt seems to focus on just base money exchange and trivializes all other exchanges.

    Nick point is on the money (pun intended). If money frictions don't matter, then why use money at all?

    ReplyDelete
  7. Nick,

    The equilibrium concept in a NK model is coherent. To make it simple, suppose there is no capital. At equilibrium relative prices, each firm produces a differentiated product, pays the representative consumer a wage payment in goods, and also distributes profits to the representative consumer in goods. Then the representative consumer consumes the lot. At equilibrium prices, the consumer is optimizing given his or her budget constraint, and the firm is also optimizing, given the rules of the game. We could have standard monopolistic competition in a real model, without all the extra Woodford stuff, in which case you get some suboptimality from the monopolistic competition. Woodford adds a story so that the relative prices are distorted in other ways. That's about it. Why anyone finds that a useful way to think about monetary policy is beyond me.

    ReplyDelete
  8. Steve: "Woodford adds a story so that the relative prices are distorted in other ways. That's about it."

    We have argued about this before.

    What you say makes perfect sense to me. I can imagine a coherent model that gives exactly the results you say it does. But those aren't Woodford's results. And I'm saying Woodford's results don't follow coherently from his model.

    Take an extreme example. Suppose the Calvo fairy visited all the firms at the same time. So there can't be any relative price distortions. Start in full equilibrium, now suppose the central bank flips out and increases the interest rate (God only knows how, but let's forget that). Does this cause any additional inefficiences?

    You (I think) would say "no". But Woodford and all the NK's would say "yes, it causes output to drop".

    Your answer makes sense. Woodford's answer *would* make sense (roughly) if there really were monetary exchange in the model, so that firms couldn't barter their way back to the original level of output. But is there really money in the model? No.

    ReplyDelete
  9. Nick,

    I think the first part of what you are saying is correct. Suppose a different sort of Calvo fairy. The fairy either lets everyone change their price, or no one can do it. Say prices change with probability p. If we are in a period when prices can change, and there is a change in policy, then I think that you are correct that this is neutral, barring some unusual second-order effect that might be in there. If you put money in the model, then you'll get the usual nonneutralities - not Keynesian ones of course. For example, i.i.d. money growth shocks won't produce a change in output in the periods when prices are flexible, but anticipated money growth will matter.

    ReplyDelete
  10. Steve: I'm in full agreement with your 9:41.

    Let's simplify, and assume the *level* of money follows a random walk. (expected money growth is always zero). p% of the time (when prices can change) money is neutral. (1-p)% of the time, money is non-neutral (for 1/(1-p) periods, on average). If we added a bond market, the real interest rate on bonds would also be affected by money shocks in those periods when money is non-neutral. There will be a negative correlation between real interest rates and real output.

    What Woodford did was say "OK, if we keep bonds in the model, but delete money from the model, we still get the same results".

    Which I don't think is right.

    ReplyDelete
  11. Woodford uses a limiting argument to show why he thinks it's useful think in terms of his "cashless" economy. The argument is a bit slippery - depends on how money enters the model.

    ReplyDelete
  12. Yep. There are two questions: how big a stock of cash do people hold; do they use cash to buy and sell everything else. I'm less worried about the first question. It's the second question that matters, I think.

    ReplyDelete
  13. There is a nice summary about the difference between Woodford style frictions and Williamson type frictions linked below. From what I see the Woodford type frictions make a bit more sense in practice.

    http://www.newyorkfed.org/research/economists/eggertsson/079274D1-DDF1-7867-958FA949F866FBCB.pdf

    ReplyDelete
  14. In his conclusion, Gauti says:

    "Instead, it seems to me that the greatest potential is in exploring
    in more detail "inside money", i.e., the is the evolution of various claims of
    individuals on one another which I guess could be characterized as the nitty
    gritty details of Narayana’s "memory ledger"."

    I'm one step (or two or three) ahead of him:

    http://www.artsci.wustl.edu/~swilliam/papers/web%20page%20paper.pdf

    ReplyDelete
  15. By the way: For what it's worth, I wrote a follow-up post where I explain my position. As I tried to emphasize in the first post as well, I'm not saying that monetary frictions in the broadest sense "don't matter"; rather, I think that the distinction between base money and other forms of liquidity doesn't matter much at the margin in alleviating monetary frictions.

    I'm confident that this is an empirically true statement, and it opens up the question of how conventional monetary policy can possibly have an effect when it merely exchanges base money for other highly liquid instruments. This is a question that I think can only be answered by New Keynesian models.

    ReplyDelete
  16. Matt,

    "...it opens up the question of how conventional monetary policy can possibly have an effect when it merely exchanges base money for other highly liquid instruments. This is a question that I think can only be answered by New Keynesian models."

    That is a very confused statement. You're saying that monetary policy during "normal" times, i.e. swaps of reserves for T-bills, essentially, can't matter much. But that has to be true in the New Keynesian world too. In a New Keynesian model, we just say that the central bank can move the short-term nominal interest rate at will, without modeling how the underlying asset swaps make that happen. You want to answer the fundamental question: "Does monetary policy matter and why?" In order to answer the question you need to write down a model which captures the roles of "base money and other highly liquid instruments." I'm afraid that New Keynesian models don't cut it in that respect.

    ReplyDelete
  17. It is a completely coherent statement. I am aware that most of the time, New Keynesian models do not explicitly outline the channel through which asset swaps affect the short-term nominal interest rate. In fact, that's the point: New Keynesian models demonstrate that these swaps are important because the short-term nominal interest rate (and therefore any asset swap that affects this rate) has such a powerful influence. No explicit model of the swap itself is necessary—and we have many other reasons to think that the details of the interbank reserves market are only minimally important compared to the ultimate effect on the interest rate.

    You also haven't responded to my criticism, except by providing more of the same rhetoric about the pressing need for microfoundations. (I agree in principle, but it is impossible to include microfoundations for every element of a model---we will inevitably make choices about what to include and what not to include. I am providing arguments for why it is strategically sound and quantitatively reasonable to exclude the details of monetary policy transmission and currency's role as a medium of exchange.)

    In your model, monetary policy matters because it drives a wedge between the yield on money and the rate of time preference; a larger wedge affects the real economy by making the decentralized market less attractive. I am arguing that this does not map onto the real world in a quantitatively significant way, particularly when we are talking about currency: there are very few transactions (aside, perhaps, from drug deals) that will not take place because there is a annual wedge of 4% between holding $100 in your pocket and holding it in the asset markets. Have you ever made significant adjustments to your consumption because the annual loss of a few dollars in foregone yield made it undesirable to use cash in a situation where cash was necessary?

    ReplyDelete
  18. "No explicit model of the swap itself is necessary—and we have many other reasons to think that the details of the interbank reserves market are only minimally important compared to the ultimate effect on the interest rate."

    That's where you're wrong. How can you just assert that the central bank moves the short rate without showing how it happens?

    "You also haven't responded to my criticism..."

    That's the response. In your "reply" you're just repeating what you said in the first post. In your last paragraph above, you make clear that you don't understand my model. Go read the paper again - carefully.

    ReplyDelete
  19. Of course, I'm sure you also object to an MIU formulation in the first place---shouldn't we be modelling money explicitly, rather than just sticking it into the utility function ad-hoc? Again, I think that while more microfoundations are always better in principle, we have to make strategic choices about what elements of the world we include. More importantly, even though the dominant welfare effect of conventional Fed policy in your model comes from the wedge between the real return on paper currency and the rate of time preference, I don't think that any paper in the existing academic literature microfounds the use of currency in a way that's consistent with empirical reality.

    Most paper currency is simply not used for legitimate transactional purposes; instead, it is held outside the US, hoarded to evade taxes or for idiosyncratic reasons, and used for illegal transactions. If you believe that microfoundations are crucial to addressing monetary matters, why aren't you including these phenomena in your models? You mention them briefly in your paper, but you don't really elaborate. Do you dispute that these phenomena account for the majority of currency demand? If so, they seem just as important as the classic model of currency as a medium of exchange.

    And the dynamics of your baseline model of frictions can nevertheless be embedded in a reduced-form utility function u(c,m), or maybe u(c,m_1,m_2) if we're looking at the model that also includes outside money backed by government debt. (If you don't think that this is possible, I am happy to provide a derivation.) So when monetary economists study functions of the form u(c,m) and find that the complementarity between c and m is of minimal importance in practice, have they been making some terrible mistake? Is there some kind of bad functional form assumption that's ruining their results? If the phenomena you're studying are really important, it seems like they should show up when we look at u(c,m). And that doesn't seem to happen in practice.

    ReplyDelete
  20. First part of the above comment (somehow it didn't go through the first time---sorry that everything is out of order):

    "That's where you're wrong. How can you just assert that the central bank moves the short rate without showing how it happens?"

    It's easy to include the open-market operation mechanism in a New Keynesian model: simply put money in the utility function and you'll obtain a money demand equation that includes both m and i. By assuming that the central bank is able to manipulate m, we can also say that it is able to manipulate i. Now, in some circumstances leaving m out of the model entirely and assuming that the central bank directly controls i will miss important effects: for instance, if there is substantial complementarity in the utility function between c and m (which would be true in a very reduced-form version of your model), then obviously a New Keynesian model that excludes m will be unsatisfactory. But there is a lot of evidence to suggest that the role of "m" in the utility function is indeed unimportant---New Keynesians haven't just carelessly ignored it. Woodford does a back-of-the-envelope calculation on page 118 of his book and comes up with an estimate of 0.01 for the elasticity of marginal utility of consumption with respect to real money balances. And I'd argue that even this number is overstated!

    Now, another potential problem is that by modeling injection of money as a "helicopter drop" (a lump-sum transfer), we ignore the reality that an asset swap of currency for government bonds both (1) increases the supply of currency and (2) decreases the supply of government bonds, which are useful assets to support the creation of inside money. Thus a conventional open market asset purchase results in both (1) a lower nominal interest rate on government debt and (2) a larger yield spread between government debt and other, non-liquid securities; only channel (1) is addressed in a New Keynesian model. While this is not an ideal state of affairs, I don't think it's very bad, because in practice effect (1) of an open-market operation is vastly larger than effect (2). The stock of government debt, and other assets suitable for transformation into outside money, is vastly larger than the stock of currency, and an increase of X in the supply of currency will cause a vastly larger change in the federal funds rate than a similar increase in the supply of government bonds will cause for the liquidity premium on government debt. Moreover, in an equilibrium where part of the supply of base money is held by banks an interest-paying reserves (as in your model), and the Fed adjusts nominal interest rates by changing the rate on reserves, effect (2) does not exist. Thus I believe that focusing on (1) a pretty decent approximation.

    Don't get me wrong: I think that studying the liquidity characteristics of non-currency assets like government debt is a perfectly worthwhile and valuable endeavor. The extent to which this matters for the effects of conventional monetary policy, however (rather than, say, activist fiscal policy or banking regulation), seems to be minimal. Do you disagree because you think that effect (2) actually is important?

    ReplyDelete
  21. That's the response. In your "reply" you're just repeating what you said in the first post. In your last paragraph above, you make clear that you don't understand my model. Go read the paper again - carefully.

    I think I understand the model perfectly well---and if I don't, I would appreciate being informed about what my error actually is.

    In your model, welfare depends on the total surplus in matches in the decentralized market. There are two cases: (1) nonmonitored bilateral meetings that require currency, and (2) monitored meetings where bank debt is possible. For the sake of understanding conventional monetary policy, I have been concentrating on (1) because I think that the dominant effect of an open market purchase (in these kinds of models) is its expansion of the stock of currency. (Alternatively, in a world where the Fed conducts monetary policy by adjusting the interest rate on reserves, (1) is the only case affected by the policy rate, making this focus even more reasonable.)

    When we're concerned about (1), the key question in your model is the spread between the real rate of return on money (the inverse of the inflation rate) and the rate of time preference. (You even say this in the optimality section of your paper, where you point out that as long as mu > beta we will not have optimal x_n.) Assuming that conventional monetary policy has little effect on the spread between the real interest rate on government debt and the rate of time preference, the change in the spread that interests us is equal to the change in the nominal interest rate. And that was the premise of my post: if you look at a given change in the nominal interest rate, it is extremely unlikely that the "New Monetarist" effect is even close in magnitude "New Keynesian" one over a business cycle timeframe.

    ReplyDelete
  22. "...I have been concentrating on (1) because I think that the dominant effect of an open market purchase (in these kinds of models) is its expansion of the stock of currency."

    No, in my model, an open market purchase can be non-neutral precisely because it is working on the other margin of liquidity. An open market purchase works to reduce the real quantity of the second class of liquid assets - interest-bearing government debt and privately-created liquidity - used in financial trade. This class of liquid assets becomes more scarce and the real interest rate falls, giving you more private lending as the private sector attempts to quell the shortage. This is very important for understanding the financial crisis and how it differs from the Great Depression, for example. In the Great Depression (and earlier banking panics), there was a currency shortage - a particular kind of liquidity shortage. The financial crisis is a about a shortage of the second class of liquid assets. This is very important for understanding why the nature of the liquidity trap we are in differs from the conventional one.

    ReplyDelete
  23. Congratulations on the World Series.

    "No, in my model, an open market purchase can be non-neutral precisely because it is working on the other margin of liquidity. An open market purchase works to reduce the real quantity of the second class of liquid assets - interest-bearing government debt and privately-created liquidity - used in financial trade."

    Right. I explicitly recognized this above when I wrote:

    Now, another potential problem is that by modeling injection of money as a "helicopter drop" (a lump-sum transfer), we ignore the reality that an asset swap of currency for government bonds both (1) increases the supply of currency and (2) decreases the supply of government bonds, which are useful assets to support the creation of inside money. Thus a conventional open market asset purchase results in both (1) a lower nominal interest rate on government debt and (2) a larger yield spread between government debt and other, non-liquid securities.

    But ultimately I have severe doubts that this channel makes much of a quantitative difference. When the Fed adjusts policy through open market operations, over the short to medium term it's making purchases in the tens of billions of dollars; maybe $100 billion at the very most. Meanwhile, the MZM money stock is $10 trillion, and that's an underestimate of the true size of the universe of liquid assets. Fiscal shocks happen all the time that adjust the quantity of liquid government debt by much more than Fed operations normally do; if you're positing that this an important channel for the effects of Fed policy, it follows that the Fed is at most a minor sideshow next to the Treasury. That doesn't ring empirically true to me.

    And we can presumably test this hypothesis. Do expansionary surprises to monetary policy lead to a measurable increase in the spread between Treasuries and less liquid instruments? My guess is that there is no sign whatsoever of this effect in the data; it is simply too small, and liable to be overwhelmed by noise and various secondary effects. Indeed, even vastly larger fiscal shocks don't seem to have a conspicuous effect on the spread between Treasuries and less liquid assets; my guess is that the supply of Treasuries and comparable forms of liquidity is simply too deep for short -term shocks from the Fed or Treasury to make too much of a difference.

    ReplyDelete
  24. The financial crisis is a about a shortage of the second class of liquid assets. This is very important for understanding why the nature of the liquidity trap we are in differs from the conventional one.

    I am certainly sympathetic to the claim that the financial crisis was about a shortage of liquid assets; I wrote an entire post about it back in the day. But after that point, my interpretation is very different: I think that the shortage of liquid assets was destructive because the increase in the liquidity premium and the zero lower bound on interest rates led the effective real rate (the one on less liquid assets) to rise far above its natural level, producing a recession.

    Your interpretation, on the other hand, seems to be that liquidity plays an economically important role in transactions, and that an increase in the liquidity premium made many kinds of economic activity more expensive, leading to a contraction. No doubt that this is true to an extent, but I'm skeptical of its comparative importance.

    In fact, I think we can make a solid argument that at best, your proposed channel cannot be more important than the New Keynesian channel I'm proposing. Once we've reached the zero lower bound, each 1% increase in the liquidity premium leads directly to a 1% increase in the real interest rate on non-liquid assets. Effectively, this adjusts the rate governing consumers' intertemporal Euler equations by 1% (since the liquidity premium should not appear there). This makes all consumption today 1% more expensive relative to tomorrow.

    Now, in your model a 1% increase in the liquidity premium makes monitored consumption in the decentralized market 1% more expensive. But assuming that the decentralized market accounts for only a fraction of overall consumption, this effect must be a lot smaller than the New Keynesian effect, where the 1% matters for all consumption.

    Now, since the Fed was able to bring interest rates down from 2% to 0% when the financial crisis hit, the "New Keynesian" effect on the real rate appearing in the intertemporal EE was a few percentage points smaller than the change in the liquidity premium. But in practice this is presumably swamped by the fact that the DM only accounts for a fraction of consumption.

    In a sense we're comparing apples to oranges here, but I don't think that this observation is necessarily favorable to your point of view. Since the New Keynesian effect works through intertemporal substitution, it depends on exactly how long the real interest rate on nonliquid assets will remain elevated. If it will last a year, then directly comparing percentage points like I did above is appropriate. If it will last two years (before an expected "return to normal"), then the New Keynesian effect should be multiplied by 2: at the zero lower bound, a 1% increase in the liquidity premium that lasts for 2 years will make consumption today 2% more expensive relative to "business as usual" two years from now, while the New Monetarist effect is still only 1%. Now, if the expected duration of the high liquidity premium is less than a year, then this goes the other way---but was that really the case?

    As always, my fundamental point is this: while the New Monetarist effect is qualititatively valid, it is hard to slice the numbers in a way that makes it come near to the quantitative importance of the New Keynesian effect.

    ReplyDelete
  25. I would of course like to take credit for the fine performance of the Cards. The truth is that I did not even watch the game. If the Blues ever made it to a Stanley Cup final, I would watch that.

    I moved this discussion to another blog post, in hopes we get some other people to join in. Hopefully not silly people.

    ReplyDelete