Tuesday, July 17, 2012

Monetary Policy: A Blunt Club or a Surgeon's Scalpel?

And a reason why NGDP futures are a critical credibility tool

Should monetary policy be viewed as a blunt club or a surgeon's scalpel? While we often discuss central banks credible actions to accurately hit a policy target, there is less discussion on whether central banks can precisely hit a target. Because while both blunt clubs and surgeons' scalpels can wound and hurt, a club "beats around the bush" while a scalpel (ideally) targets a specific, well defined area. In the language of monetary policy, we know central banks can raise the level of nominal GDP growth, but can they precisely control the magnitude of the increase? How much of a base injection does the central bank do to make the potatoes hot enough without burning our fingers? This precision problem is at least as important as the accuracy problem because it is a key reason for why explicitly declaring a five percent nominal GDP level target is not sufficient. Because there are real costs of extremely high inflation, the central bank is likely to shy away from policies that might "accidentally" lead to excessively high inflation. If economic variables become more volatile, the lack of monetary policy precision can become a major barrier to monetary policy accuracy.

The core of my argument starts from a mechanics-credibility theorem, which states that if policy does not have a concrete mechanism, it can not be credible. This theorem means raising nominal GDP can be credible. We know that central banks have an infinite number of ways to inflate: currency depreciation, asset purchases, forward guidance, nominal GDP futures targeting, among other options.

However, what does the theorem say about policies to raise nominal GDP to a specific, precise level? What do we have to guarantee this? A typical market monetarist response would be that expectations bridge the gap. Because private agents believe that the central bank will raise nominal GDP to reach trend growth, the market will do the work for the central bank. They will invest in riskier assets, hire more people, engage in more research and development projects, and other ventures up until current nominal GDP returns back to the trend line. Once we're back at trend growth, the central bank can just moderate the economy without too much work because the perfectly credible declaration will induce the market to stabilize itself. 

Although this logic is appealing, it is nonetheless circular. According to this line of argument, a precise target is credible because the central bank will inevitably hit the target, but the reason the central bank is precise is because it is credible. Before we can wave the magic wand of expectations, we need to show that there exists a vanilla, "elbow grease" monetary action that can credibly hit a target with both accuracy and precision.

In normal times, the relationship between the short term nominal interest rates is reasonably stable, and central bankers can manipulate the short term rate quite effectively to stabilize nominal GDP. This is a game that the Federal Reserve has played for the entire Great Moderation. However, once the economy is in a liquidity trap, the short term rate loses its effectiveness, and other more drastic measures are needed. Now that we're in uncharted territory, here is where the precision problem arises.

Take asset purchases for example. As noted by Miles Kimball, these asset purchases can have large effects, if implemented on a large scale. There are certain monetary frictions that allow quantitative easing to break out of the liquidity trap, but for those frictions to exert an appreciable effect quantitative easing must take place at a large magnitude. A specific friction that I see is Kiyotaki and Moore's liquidity friction, which states that the bonds central banks purchase are not as liquid as the cash that central banks dish out. Bonds cannot be sold as quickly, so the large stock of bonds represents a promise that the central bank will maintain a higher monetary base in the future. This is how the central bank gets around the Woodford, "double the monetary base in this period and all subsequent periods" credibility problem. While central bank declarations of a future elevated monetary base may not be seen as credible in light of inflation targeting, committing to large scale bond purchases commits the central bank to a brief period of above average inflation by locking in the larger monetary base. In the words of Brad DeLong:

But purchasing bonds for cash has another effect. Cash is a perfect substitute for short-term Treasury bonds now. It won't always be the case. When interest rates normalize, the price level will be roughly proportional to the high-powered money stock. Not all of today's purchases of bonds for cash will be unwound when the economy exits the zero lower bound. If we believe that the high-powered money stock will be roughly $1 trillion after exiting the zero lower bound, and if we believe that a fraction λ of marginal bond purchases won't be unwound, then an extra $100 billion of quantitative easing boosts the expected price level ten years hence by 1%--and boosts expected inflation after the next decade by an average of 0.1%/year. That is enough to spur higher spending and a more rapid and satisfactory recovery.

The problem with this approach is that it is not robust to shifts in expectations. Working with Brad's terminology, λ is positively correlated with the scale of bond purchases. As a result, expected inflation is non-linear with respect to bond purchases. Initial bond purchases may not raise inflation expectations by much, but past a tipping point at which λ start rising quickly, inflation expectations could drastically shift. Even if such bond purchases occur at a steady rate so as to avoid sudden changes, expectations can still suddenly shift if an unforecasted shock occurs. If key economic parameters, such as money velocity, become increasingly unstable, expectations fragility will only grow worse. An example of this kind of "expectation fragility" is with the Swiss National Bank's currency floor. While the fundamental value is below the floor, we get suppressed volatility. Once the fundamental value rises, even if it's purely because of static, there can be lots of chaos as a previously fixed price is now allowed to move. In the context of monetary policy, given the large stock of currency reserves, a slight unforecasted rise in inflation expectations can be the trigger for a massive investment reallocation.

Lars Svensson, in his paper on a "foolproof" way to leave a liquidity traps, also highlights this uncertainty:

It is difficult to determine how large an open-market operation would be needed to reduce the long interest rate, because of difficulties in estimating the determinants of the term premium of interest rates (that is, the difference between long and short interest rates and its dependence on the degree of substitutability between short and long bonds). However, Bernanke (2002) has proposed an elegant operational solution to this problem. The central bank simply announces a low (possibly zero) interest-rate ceiling for government bonds up to a particular maturity, and makes a commitment to buy an unlimited volume of those bonds (that is, potentially the whole outstanding volume) at that interest rate. This commitment by the central bank is readily verifiable – since everyone can verify that the central bank actually buys at the announced interest rates – and achieves the desired impact on the long interest rate, without a need to specify the precise magnitude of the open-market operation required. The central bank may have to buy the whole outstanding issue of the long bond, though.  

In the absence of an expectations mechanic, there is a wide range of uncertainty on the policy response. Other forms of unconventional monetary policy, including currency depreciation or forward guidance, all suffer from this "how much is too much?" problem. Once you inject a few non-linearities into the stories, and have market expectations change at unknown threshold, you realize that prospects for precision are grim.

When the monetary policy result is uncertain, central banks are less likely to use the tools to raise nominal GDP. They may fear excessively high inflation or other side-effects of over-expansionary monetary policy, and choose to be "cautious" and live with high levels of unemployment instead. This is the reason why limits in precision can become limits in the accuracy or effectiveness of monetary policy.

A historical counter-example to this theory of uncertainty would be FDR's dollar devaluation program, which Scott and Marcus Nunes have shown to be very effective in restoring both the price level and output during the Great Depression. Yet I  don't find their example to  contradict my argument about the imprecision of monetary policy at the zero lower bound. Perhaps FDR got lucky. Perhaps the price level was already so depressed that you would have needed hyperinflaiton. Perhaps, back then, finance was not as tightly coupled and the investment effects of getting out of treasuries wouldn't have been as strong. Nonetheless, in our faster, more leveraged, more volatile world, this granularity and imprecision of monetary policy is a bigger impediment.

Scott Sumner often uses Australia as another example, pointing out that their monetary base to nominal GDP ratio is much lower than that in the United States. This shows that Australia has not had to inject as much money into their banking system to stabilize nominal GDP, demonstrating that the stabilizing nominal GDP growth should not be very difficult. But Australia is unique in that it never was at risk of the zero lower bound. While part of this may have just been good monetary policy, it still leaves open the possibility of a large, unforecasted shock that puts a county at the zero lower bound.

However, one form of nominal GDP targeting seems to sidestep these problems: nominal GDP futures targeting. This would allow market participants to instantly improve estimates of future inflation by bidding on futures contracts. Their bidding one way or another would immediately translate into changes in central bank open market operations such that nominal GDP always stays on track. This approach sidesteps the non-linearity of expectations because it allows the market to aggregate all the necessary information and automatically has the central bank adapt to the new found information. Even if expectations did shift in response to unforecasted shocks, the policy response would be immediate and taken in decentralized steps as individual investors bid on futures contracts. In this case, mechanics-credibility theorem is satisfied because the mechanic by which the Fed earns its nominal GDP credibility directly interacts with market expectations while avoiding the circularity problem. Market expectations of nominal GDP feed into futures market volumes, which directly changes the monetary base. The market answers the questions of "how much" with the level it thinks is "just right".

This is one of the key advantages of an nominal GDP futures targeting regime relative to a conventional "wait-and-see" regime. It cements in credibility, and rolls with the waves of external volatility. In a sense, it floats like a butterfly and stings like a bee. It takes monetary policy from the world of "Bernanke Smash" to "Sumner Slice", and allows for greater accuracy and precision in the control of a central nominal aggregate: nominal GDP.


  1. I don't understand.

    The Nominal GDP level target provides expectations effects on the up side.

    Suppose nominal GDP rises above target. Then we know that the central bank will tighten to slow nominal GDP growth (perhaps to negative growth rates) to get it back to target. All the heavly lifting to return to Nominal GDP to the target level is done by the market.

    To me, much of the new Keynesian literature that concerns you is based upon inflation targeting.

    Nominal GDP level targeting does mean you don't worry about the price level or the inflation rate over any period.

    If, instead of nominal GDP level trageting, you go with nominal GDP growth rate targeting then perhaps the same problems as with inflation targeting would develop.

    We must increase base money so much that people will believe it is impossible to reverse, so inflation will be higher and real interest rates lower.

    Well, that isn't the goal at all.

    I do agree, however, that if nominal GDP is on target and expected to be on target, and the market clearing interest rate on everything the central bank could buy is still more negative than the cost of storing currency, then quantitative easing won't work.

    I think some of the rational expecations models (well, maybe all of them) require that this is the situation we are in. Otherwise, we would never have been in a recession.

    1. Bill,

      I agree with your points on growth rate and inflation targeting, so let's just focus on the level targeting discussion.

      I think, at this point, declaring a 5% trend level target that fills in some fraction of the output gap is good going. It would likely lift the economy out of the gutter, and likely with minimal "monetary policy bluntness"

      What I'm more concerned about is a hypothetical of what happens at the zero lower bound with limited credibility. Policy takes time to work, and it would be foolish to expect monetary policy to be perfectly tuned to get the exact amount of increased nominal GDP. So the question at hand is how much variance should we expect around the level target?

      While you have faith that the market will undoubtedly restore the trend, I'm a little less certain. If the Fed has policy lags, would there be potential for momentum in nominal GDP? Note that nominal GDP data only comes in quarterly, subject to large revisions, so how does the market know when the Fed will tighten to bring trend growth back down? Even if the Fed responds, will the faster NGDP growth at that point be baked into certain expectations, ala Mankiw and his asymmetric information-Philips curve argument? And if the Fed is tested and forced to take action to bring down growth, how do they guarantee they won't overshoot? There is no guarantee, so there's little reason for the market to find the Fed's claim to limit variance about the trendlien to be credible.

      We can follow this credibility chain. The Fed commits to 5% NGDPLT. Markets find this weird, they don't think it's credible. Fed does massive balance sheet operations, proves its credibility. Fed knocked to zero-lower bound again, and, due to imperfect credibility, people don't know what it will do. Fed commits to large balance sheet actions to re-establish credibility, but exogenous shock in re-inflaiton process knocks the Fed above its target and it scrambles to re-establish trend. But in the meantime, there's already action going on that expects the above trend NGDP for a period as a result of momentum/higher NGDP growth being baked in. I'm asking for hard mechanisms that the Fed can use to quickly and efficiently restore imbalances in NGDP (which futures targeting does). Only if these are sufficient will market expectations make the job easier.

      I feel this concern about the variance of NGDP around trend has serious policy implications. For as much as the Fed wants 5% NGDP, people might feel that a burst of 4, 5% inflation might be unacceptable politically, and therefore not credible. This precision analysis interests me because I'm scared by what happens to NGDP targeting if the economy is subject to more and more shocks. It'll probably perform better than inflation targeting, but the rising variance would only highlight any existing fragilities.

  2. I really like your posts. This might just be my computer, but I can't click your links in either Firefox or Chrome (but can in IE).

    1. I'm looking through my old posts via chrome, and I am not having a problem with it. If other people are having similar issues, let me know and I'll work on it.

  3. I've always wondered what this NGDP targeting stuff makes of the fact that the inflation generated might not be general. With unemployment high and labour unions weak, there is every chance that a price inflation -- say, if it was undertaken through depreciation which is the only credible channel -- might outstrip wages.

    In such a circumstance -- which, if you look at the data on wages, looks very likely -- the inflation generated by NGDP targeting would actually reduce aggregate demand as less income chased more expensive goods and services. The net effect would be to lower living standards and probably induce stagflation.

  4. Just thought I'd follow that last comment up with some empirical evidence:


    That whole mess is being floated on fiscal deficits -- i.e. the demand that is replacing full-time real wage growth is being replaced by demand flowing from the government fiscal deficits. Remember the Kalecki equation in this regard:


    The profits are being driven by fiscal deficits.

    Now, that's fine -- to an extent, although the deficits are clearly covering over some very murky macro imbalances. But now add inflation increases to this equation...

    What happens? Well, it seems obvious. Real wages will decline as inflation rises. And if the government deficits do not bridge the demand gap even further than they already are, unemployment will rise.

    Market monetarists really need to take a broader and more empirical view of the economy. What they are advocating seems very nice from an empirical point-of-view, but its pretty toxic when you look at the empirics.

    1. TheIllusionist, I think we need to be clear on the difference between the real standard of living and aggregate demand. Aggregate demand can be really high even while real living standards are low; just look at Zimbabwe. But given the short term issue isto stabilize the level of nominal spending, I don't think there's a really large issue with market monetarist analysis.

  5. This is coming from more of a lurker of this blog who is deciding to become more of a commentator on this blog...

    Once again, another great post on nominal GDP targeting, Yi-Chuan Wang! Keep up the great work! If you're looking into scholarship in the long term, you may want to get a postgraduate degree in mathematics, as a mathematician's perspective can help you understand how maths ought to be applied and how it can be misused in economics. Keynes and Mandelbrot knew these lessons all too well.

    P.S. Did you get my e-mail?

  6. Yichuan this is my first time reading your post but I really like it-I've of course read you over at Sumner.

    "While you have faith that the market will undoubtedly restore the trend, I'm a little less certain. If the Fed has policy lags, would there be potential for momentum in nominal GDP?"

    Understand I'm not picking sides here but it seems to me that one thing Sumner is always saying is that monetary policy gives us "variable leads" rather than "variable lags."

    I'm not saying this is my view necesarily but that's his claim

    1. Evilsax,

      I've considered the "long and variable leads" and the "long and variable lags" arguments, and I think they best describe two sides of the same phenomenon. Long and variable leads makes a lot of sense for financial markets, where trading is quick and nominal stickiness is less of an issue. However, long and variable lags makes more sense for goods markets where there's a lot of price/wage stickiness. Notably, if Mankiw and Reis' sticky information Philips curve has any truth, I don't see how targeting nominal GDP levels would remove that stickiness to prevent policy from lagging.

      However, to give NGDPLT some credence, we've never really been in an environment when the central bank is level targeting, so we would need to look at the microfoundations to see if information lags, even when the market expects level targeting, make policy lag. It's tough to evaluate the autocorrelation data on prices and GDP because those coefficients are all regime dependent. Because IT doesn't go back to correct its mistakes, it's understandable why the first or second quarter autocorrelations are positive. But would the same hold under a price targeting regime? I'm not sure, but I think it's an important area of analysis.