Sunday, January 31, 2016

Central Banking by Process of Elimination

Here's a remark I've heard more than once from macroeconomists who are old enough to remember the 1970s: If you could go back to 1979 and tell people that the big problem facing central banks in 2016 would be getting the inflation rate up to 2%, they would all have fits of laughter leading to cardiac arrest. Case in point is the Bank of Japan, which has tried keeping its policy interest rate close to zero for twenty years, and has engaged in a massive QQE (qualitative and quantitative easing) program since April 2013, with the result that inflation is still well short of their 2% target. On Friday, the BOJ announced a new program - "Quantitative and Qualitative Monetary Easing with a Negative Interest Rate," which involves a continuation of the QQE program, along with a reduction in the interest rate on reserves held by banks with the BOJ.

As described here, the plan for paying interest on reserves is somewhat complicated, with three tiers of reserves and different interest rates applied to each. Paraphrasing, interest will be paid at 0% on required reserves, at 0.1% on a reference level of reserves (a bank's average excess reserve balance during 2015), and at -0.1% on anything in excess of the sum of those two quantities. The effect of this will depend on how reserves are distributed in the financial system. For some banks the marginal unit of reserves will earn -0.1%, and for some banks the marginal unit will earn zero or 0.1%.

Why can nominal interest rates be negative? In most mainstream monetary models there is in fact a lower bound of zero on the nominal interest rate. Why? For example, in a standard New Keynesian (NK) "cashless" model, people hold zero money balances in equilibrium, and all other safe assets, including government debt, bear interest. If the nominal interest rate on all of those safe assets were negative, then people could borrow at the safe rate of interest (in baseline NK models there is no default), hold cash, and make a profit. Arbitrage then dictates that the nominal interest rate cannot fall below zero. In practice, the nominal interest rate on currency is always zero, so if it were costless to hold currency and to convert it into reserves and vice-versa, then if banks faced a negative interest rate on reserve balances they would always prefer to hold currency rather than reserve balances. The reason that the interest rate on reserves can be negative is that it is not costless to hold currency, nor is it costless to convert reserves into currency, or vice-versa. Currency can be stolen, and it takes up physical space, which makes it extremely inconvenient for large interbank transactions. For a bank in the United States, converting currency into reserve balances means putting the currency in a truck and taking it to the local Federal Reserve Bank (or one of its branches), which of course uses up time and resources.

There is nothing new about negative interest on reserve balances - some central banks, including the Swiss National Bank, the Central Bank of Denmark, the Swedish Central Bank, and the ECB, have been doing this for some time. Further, negative nominal interest rates have been discussed extensively by economists. Some people like the idea, including Ken Rogoff and Miles Kimball, and have devised schemes that would in principle allow central banks to violate the zero lower bound in a big way. That is, at some sufficiently low negative nominal interest rate, arbitrage profits must overcome the costs of holding currency, and the nominal interest rate can't go any lower. But what if central banks did away with currency, or imposed extra costs on holding it? That would give them plenty of leeway on the south side of zero.

So, central banks can set interest rates on reserves below zero, and in so doing can put overnight market interest rates in negative territory as well. But why do they contemplate such a thing, and implement negative interest rate policies? Stefan Ingves, the Governor of the Swedish Riksbank, explains it here:
GDP growth is fairly good and the labour market is strengthening gradually, but inflation is too low in relation to the Riksbank's target for inflation. In recent months, inflation has begun to rise and the repo rate needs to be this low so we can be sure that inflation will continue to rise and attain the target.
A repo rate just below zero gives a further boost to household consumption and company investment. Together with the repo-rate cuts we have made earlier, this gives higher inflation in the long run and higher interest rates, which will also benefit all savers. It will take us back to a more normal situation in the Swedish economy and this will be good for us all.
He's saying that, in terms of the Riksbank's goals, things are going well, except that inflation is running below the Riksbank's 2% inflation target. The Governor hopes for a future world in which inflation and nominal interest rates are higher. So, you might ask, why doesn't he just increase nominal interest rates, and allow the Fisher effect to produce higher inflation? No, he says, the way to do it is - here we have to read between the lines - to reduce the nominal interest rate below zero, stimulate real activity, and then the Phillips curve will produce the inflation that the Governor can then tamp down by increasing the nominal interest rate. Then we're back to normal.

So, how has that strategy been working? This is the overnight repo rate in Sweden, along with the Riksbank's forecast of where it's going:
And this is the Swedish CPI inflation rate:
So, inflation has been close to zero for three years, in spite of (if you're a Keynesian), or because of (if you're a neo-Fisherian) the Bank's lower interest rate policy. But the Riksbank is confidently predicting that it will be back to its 2% inflation target in a year, and will overshoot it by a wide margin in two years. It's interesting to see what they were predicting two years ago:
Same thing basically. The point estimate at the beginning of 2014 was that inflation would currently be in the neighborhood of 3%, but actually inflation is currently at about 0%, outside the 90% confidence interval in the 2014 forecast. That is, what actually happened was supposed to be highly unlikely.

Some people have made a big deal of what they see as a policy mistake by the Swedish central bank - the interest rate hikes you see in the first chart beginning in 2010 which were subsequently undone. We could quibble about whether the Bank did the appropriate thing in 2010 - though I tend to agree with Marvin Goodfriend and Mervyn King that the Riksbank's 2010 tightening was well within the bounds of normal central bank behavior. However, what has happened post-2013 seems like a more obvious failure - the Riksbank has been pursuing a policy which isn't paying off as they say it will. Inflation remains persistently below the 2% target.

What about the other countries with central banks that have gone negative? Here's the overnight interest rate and the CPI inflation rate in Switzerland:
Here's the ECB - inflation rate followed by the overnight interest rate:
Finally, Denmark - again, inflation rate followed by the interest rate on 7-day deposits with the central bank:
So, that tells a consistent story. Central banks that have slipped into negative nominal interest rate territory don't tend to get any closer to their inflation targets. Sometimes they get further away. But that shouldn't be surprising, as that's what standard monetary theory predicts. As I show here, lowering the nominal interest rate below zero just makes inflation lower.

So, if the theory - neo-Fisherian theory - is so obvious, and given the experience we have already had with negative nominal rates, why does the Bank of Japan think that its new program will increase inflation? For a very long time, a Phillips curve theory of inflation was what central bankers believed, and in some countries those central bankers were successful in keeping inflation low, but not too low (assuming a 2% inflation target, say). This worked for the Fed, for example, during the Great Moderation period - parts of that episode look like the Fed had adopted a 2% inflation target, and was succeeding. Central bankers would cut rates in a recession, and then increase rates in the recovery. Even if inflation was not actually increasing in the recovery phase, central bankers could argue that a tighter labor market would surely increase inflation in the future. A typical argument was that, in the recovery phase, tightening needed to occur sooner rather than later, particularly given "long and variable lags," etc. As John Cochrane says, they could then pat themselves on the back for controlling the inflation which they had themselves created.

But that doesn't work if central bankers don't get the timing right. If they wait too long to increase interest rates in the recovery phase, low inflation sets in, along with doubts about the wisdom of "tightening." Doubts give way to panic, which leads to lower and lower interest rates, and various experiments with unconventional polices - QE and nominal rates below zero. The Phillips curve was sometimes a convenient fiction, but in this instance it's a curse - believe in it and you're in trouble.

So, what to do? Central banks, and many economists, won't declare the Phillips curve dead. The parrot is motionless at the bottom of the cage, but perhaps he's only pining for the fjords? Some, like Olivier Blanchard, are at least willing to concede that the parrot appears quite lethargic. But, religious conviction is very hard to overcome. One approach is to put religion aside, and show people that, even in models they like to use - models including Phillips curves - neo-Fisherian results hold, i.e. increases in nominal interest rates by the central bank make inflation go up. So, people can abide by their religious beliefs, but we are all neo-Fisherians. Mike Woodford is a neo-Fisherian, Knut Wicksell (and by implication Paul Krugman) was a neo-Fisherian. So, is Narayana Kocherlakota, though he's trying hard to deny it.

So, for central bankers currently seeing low inflation and wondering what to do, take the express route. There's no need to resort to a process of elimination. Most of the supposed solutions for low inflation - QE, negative nominal interest interest rates - have been tried, and haven't worked. Be neo-Fisherites. You'll sleep much better.


  1. Very interesting. Is there a single example in an advanced economy where Zirp & QE have actually worked and increased inflation?

    Wondering how long mainstream economic orthodoxy will hold if inflation stays near zero?

    Presumably the solution, to get inflation to target along with raising interest rates is looser fiscal policy. Lower or even Negative income tax rates & higher spending at least temporarily?

    How would neo-Fisherite ideas be implemented in practise?

    Would the Fed just hike regardless of the stage of cycle or wait until the next early stage growth upswing?

    1. "Is there a single example in an advanced economy where Zirp & QE have actually worked and increased inflation?"

      Not to my knowledge.

      "...mainstream economic orthodoxy..."

      In fact, one can readily understand what is going on with orthodox economics. It's orthodox central banking practice that goes awry.

      "How would neo-Fisherite ideas be implemented in practise?"

      I think most macroeconomists would agree that there is a countercyclical role for monetary policy - the central bank should lower nominal interest rates in a downturn. But, for inflation control, it's very important to be mindful of the Fisher effect, under which lower nominal interest rates cause lower inflation. And the effects of monetary policy on real activity - employment and real GDP - are temporary, but the Fisher effect is permanent. In current circumstances, for many central banks in the world, the only solution to the low inflation problem is to increase nominal interest rates.

    2. "I think most macroeconomists would agree that there is a countercyclical role for monetary policy - the central bank should lower nominal interest rates in a downturn."

      How does this work with the fisher effect? Wouldn't lowering rates cause lower inflation expectations which means shifting less of the consumption from the future to present?

    3. The Fisher effect isn't everything. There's also a short-run liquidity effect. When the nominal rate goes down, in the short run the real rate goes down too. So, in the short run, the Fisher effect and liquidity effect work in opposite directions with regard to inflation. But you would need a very large liquidity effect - more than one-for-one - for the inflation rate to go up when the nominal interest rate goes down.

  2. Seems to me that the emperor is wearing no clothes. Certainly lowering interest rates has not worked, but is there any reason to believe raising the interest rates (per the not-so-implicit neo-Fisherian model) will work? If so (examples, please? Line forms on the right...) why no description of the model and how it would affect the economy?

    The actual solution, per the denigrated Keynesian theories, is that lowering the interest rate is a non-preferred solution: the actual preferred and prescribed solution is to increase government spending. Politically, however, that doesn't seem to be possible. Indeed, government spending, particularly in the U.S., has shrunk as a fraction of the economy (popular belief to the contrary).

    1. "...but is there any reason to believe raising the interest rates (per the not-so-implicit neo-Fisherian model) will work? If so (examples, please? Line forms on the right...) why no description of the model and how it would affect the economy?"

      You're not paying attention. Follow the links in the post, and if that doesn't satisfy you, search my archive.

      "...the actual preferred and prescribed solution is to increase government spending."

      If the problem is that inflation is too low, we usually think it's the central bank's job to increase it.

    2. I should add again that there is no single "neo-Fisherian model." As far as I've been able to determine, all standard macro models are neo-Fisherian.

  3. "As far as I've been able to determine, all standard macro models are neo-Fisherian."

    What about Garcia-Smchmidt and Woodford NBER's paper ( The Neo-fisherian results in the New Keynesian model are a pure disturbance due to rational expectations.

    If you assume that the real interest rate is fixed, then your model must be Neo-fisherian. But in this case, why would you care about monetary policy anyway?

    1. "What about Garcia-Smchmidt and Woodford..."

      I said "standard" macro models. That's got a learning mechanism that kills some equilibria. But it's also true in that model that, if the central banker increases the inflation target, that will increase the nominal interest rate one-for-one in all states of the world. Hard to get away from the Fisher effect, don't you think?

      "If you assume that the real interest rate is fixed..."

      Who is assuming that? I said "standard," which includes models in which the real interest rate is not fixed.

  4. Well, you're doing comparative statics between different equilibria. In the long run, if the central bank increases the inflation target, then yes the nominal interest rate will be higher in the long run. This may well be true in the short run.

    However, does this mean that the reverse causality holds? No. It is true in real models but as Kocherlakota showed, those models are not robust. It is also true in NK models if agents have rational expectations and if they all expect the central bank to peg the nominal interest rate forever. As Garcia-Schmidt Woodford showed, this is not robust either. Furthermore do you really believe that agents would expect the central bank to peg the nominal interest rate forever? Is this what central banks are doing?

    Everyone agrees that the fisher effect has to hold in the long run, but you're making claims about the short run dynamics.

    1. "In the long run, if the central bank increases the inflation target..."
      No, take the model, increase the inflation target, and I think it's correct to say that the nominal interest rate will be higher in all states of the world in their model. That's the sense in which their model is Fisherian.

      The experiment in which the nominal interest rate goes up once and for all is just that - an experiment to illustrate an idea. The fact that Woodford can construct an example in which this can't be an equilibrium doesn't tell me much. How come central banks manage to peg nominal interest rates for extended periods of time with no problem?

      "It is true in real models but as Kocherlakota showed, those models are not robust."

      You bought that idea?

      "you're making claims about the short run dynamics..."

      Supported by standard models and data.

  5. Stephen, do you have an inflation model that can produce forecasts with those confidence interval bands? I'd love to see how you do against the Swedish forecasts (and against reality of course).

    1. "I'd love to see how you do against the Swedish forecasts (and against reality of course)."

      If I couldn't do that, my econometrics teachers would be very disappointed. These models are boilerplate. Everything reverts to trend, with a bit of Phillips curve thrown in (and the Phillips curve is part of what's throwing them off). The confidence intervals you can't take seriously.

    2. I'd be interesting to see the results (fit & forecasts) from your favorite neo-Fisher model. Would your favorite neo-Fisher model also model changes of the trend vs time? "Econometrics" and "revert to trend" makes me think of a VAR model.

      For example, here's a simple low order trend model (theory here) showing its fit to old data and its forecast as compared to the NY Fed DSGE model forecast.

    3. "your favorite neo-Fisher model"

      Same confusion as eyesoars. All standard models that macroeconomists work with have neo-Fisherian properties. But a lot of central bank models don't have those properties. In some cases - for example the FRB/US model - they're some variant of large-scale econometric models developed in the 1960s. Most of those don't even have some standard properties we require of our theory - e.g. the neutrality of money in the long run. Further, when used for forecasting, the model isn't doing much work other than making sure accounting identities hold - the add factors of the forecasters are doing the heavy lifting.

      Now, back to forecasting. Forecasting inflation is notoriously difficult. You might call this cheating, as I'm doing this ex post, but a naive forecast that says inflation at every future date is the same as it is today would have been quite accurate for Sweden - forecasting inflation today as of early 2014 - than the Riksbank's forecast.

    4. Thanks for the information Stephen.

      "inflation at every future date is the same as it is today would have been quite accurate for Sweden"

      Lol, yes. In fact the author of that trend model above compares his forecasts to several others, including just a horizontal line (which does pretty well in the near term). He also does several other countries. It might take a decade or more to see a significant difference, but if he calibrates his ~3 parameters (or whatever it is: it's not many) to data, say from 1960 to 1990, the out of sample results from 1990 till today fit amazingly well.

      I'd love to find more people doing the same kind of thing, especially over several decades. For the sake of comparison.

    5. Turns out he just updated Switzerland, which you cover here.

  6. Stephen, check out this plot of the many failed Riksbank forecasts (source: Nordea Markets and Macrobond):
    Why doesn't some notable economist out there put together an accurate forecast and rub the Riksbanks' repeated forecasting failures in their face? Even your simple forecast you describe above (a flat line) might do the trick! Once they rack up another 50 complete failures won't somebody start asking questions?

    It seems to me the forecasting "market" is DYING for a better product.

    And here's another dozen embarrassing failures for the BOE (source BOE), and another three sets from three more central banks.

    1. The problem isn't a standard one of lack of forecasting ability. In this case, the institution is forecasting something that it is supposed to have control over. The Riksbank has a 2% inflation target (though it's clearly concerned with other things too), so it always predicts that inflation will go back to the target (sometimes with some overshooting). Therefore, the conclusion is that the institution doesn't know how to control the thing it's supposed to control.

    2. Well, just in case, I thought I should go on record now.

    3. Maybe they should have a separate "red team" that gets to see everything the blue team is doing (policy, internal forecasts), but it's the red team that publishes official forecasts to the public. Red team is evaluated on accuracy only. And they share their models with blue team (which may have it's own models). Blue team is graded on effectiveness in hitting their targets only.

      I don't know: I'm just making stuff up. Seems like an institutional problem though. If blue team keeps failing, but red team shines, they might want to start adopting red team's models. If blue team does well, but red team's forecasts suck, then red team might want to adopt blue team's models.

      Of course both teams could suck.