Tuesday, August 19, 2014

Secular Stagnation: Useful Ideas or Hot Air?

There is an ebook available on "Secular Stagnation: Facts, Causes, and Cures," which consists of a set of papers by a group of economists, brought together in an attempt to understand secular stagnation, and how we might address the problem, if it is one. Some of the papers are summaries of published or unpublished research, while others are more speculative.

What is secular stagnation? Apparently it's frequently on the lips of some people, but according to Barry Eichengreen,
...while the term ‘secular stagnation’ was widely repeated, it was not widely understood. Secular stagnation, we have learned, is an economist’s Rorchach Test. It means different things to different people.
I think the problem here is that an idea cannot be understood if it's not an idea. The people who claim that the idea exists - principally Larry Summers and Paul Krugman - trace it to Alvin Hansen's "Economic Progress and Declining Population Growth," from 1939. Nicholas Crafts summarizes what Hansen had to say:
The first time around, ‘secular stagnation’ was a hypothesis famously articulated by Alvin Hansen ... Hansen argued that the US economy faced a crisis of underinvestment and deficient aggregate demand, since investment opportunities had significantly diminished in the face of the closing of the frontier for new waves of immigration and declining population growth. It was as if the US was faced with a lower natural rate of growth to which the rate of growth of the capital stock would adjust through a permanently lower rate of investment.

As we all know, these fears were completely without foundation – the delusions of a hypochondriac rather than the insightful diagnosis of a celebrated economist.
I think we could go even further than that. Not only were Hansen's writings the "delusions of a hypochondriac," but we would have a hard time making sense of his arguments in the context of the modern theory of economic growth, and what we know about the causes of growth and the reasons for differences in standards of living across countries. Suffice to say that a re-reading of Alvin Hansen won't enlighten us as to what we should expect from the world economy over the next twenty years or so.

But, what about the contents of this ebook? There are really two parallel notions of stagnation discussed in these contributed papers. For convenience, I'll call these growth stagnation and Keynesian stagnation. The growth stagnation idea appears to take conventional economic growth theory as a basis for how we should think about future economic growth, in the U.S. and in the world. From Hsieh and Klenow, empirical work on economic growth tells us that we can account for 10-30% in income differences across countries by differences in human capital, about 20% by differences in physical capital, and 50-70% by differences in TFP (total factor productivity). This should also apply to the time series. So if, for example, we are pessimistic about future TFP growth in the U.S., we then have a strong reason to be pessimistic about future real GDP growth in the U.S. On one side of the argument, Robert Gordon is a pessimist. Gordon points out that TFP growth was much lower after 1970 than in the period from 1920-1970, and he is confident that the average rate of TFP growth we experienced for 1970-2014 will persist for the next 25-40 years. Further, he is worried about four "headwinds," i.e. demographics, education, inequality, and government debt. Basically, a smaller fraction of the U.S. population will be working, educational attainment in the U.S. has plateaued and the quality of U.S. education may be in decline, inequality in incomes has increased and may continue to do so, and there are reasons to think that government debt could increase relative to GDP.

If reading Gordon's entry in this volume is inclined to make you depressed, Joel Mokyr's piece is a great pick-me-up. He says:
There is nothing like a recession to throw economists into a despondent mood. Much as happened in the late 1930s, many of my colleagues seem to believe that ‘sad days are here again’. Economic growth as it was experienced by the world through much of the 20th century, they tell us, was a fleeting thing. Our children will be no richer than we are. Some of the best economists of our age, including Larry Summers, Paul Krugman, and my own colleague Robert J. Gordon, are joining the chorus of the doomsayers. It is said that we are faced by headwinds that inevitably will slow down growth and perhaps condemn us to secular stagnation. There is no denying that the population of the world is getting older, and that the fraction of people working (and supporting the aged) is falling everywhere except in Africa. The ‘big pushes’ driven by millions of married women taking jobs and the huge increase in college graduates that drove post 1945 growth were one-off boons, but they are no more. Growing inequality exacerbates demography. Slow growth is here to stay, say the secular stagnationists.

What is wrong with this story? The one word answer is ‘technology’. The responsibility of economic historians is to remind the world what things were like before 1800. Growth was imperceptibly slow, and the vast bulk of the population was so poor that any disruption in food supply caused by a harvest failure could kill millions. Almost half the babies born died before reaching the age of five, and those who made it to adulthood were often stunted, ill, and illiterate. What changed this world was growth driven by technological progress. Starting in the late 18th century, innovations and advances in what was then called ‘the useful arts’ slowly began improving life, first in Britain, then in the rest of Europe, and eventually in much of the rest of the world. The story has been told many times over, but as Nobelist Robert Lucas once wrote, once you start thinking about it, it’s hard to think of anything else.
So, we should stop moaning, and recognize that life is pretty good and likely to get much better.

Mokyr's point is that there is much important scientific advance happening right under our noses, and that this new science will be applied in many ways that we might find hard to imagine from our 2014 viewpoint, just as Alexander Graham Bell would have a hard time imagining an I-phone. Further, the effects of current and future innovations on economic welfare may not be measured well. For example, information has become much more accessible in myriad ways that make us better off, but not all of that is captured in GDP.

Technological change does create economic problems that we need to deal with, though. It is now well-understood that an important factor in the increase in the dispersion in income in the U.S. in the last 30 years or more has been technological change. This change can bring huge rewards to innovators while depreciating particular types of human capital. For example, David Autor has written about the hollowing out of the skill-distribution because information technology makes middle-level skills obsolete. This then becomes a challenge for U.S. education. Indeed, better access to public education at all levels is a possible remedy for the income-distribution problem that Gordon seems to be concerned with.

In terms of the growth stagnation story, there is nothing in this volume that sheds new light on the growth process, and would permit us to confidently project stagnation in the medium to long term, in the U.S., or in the world. We know a lot about how TFP, human capital accumulation, and physical capital accumulation, work to produce growth in per capita incomes, but we perhaps know little about the actual process of innovation, and how to predict it.

The idea that seems to have spurred the publication of this volume, however, is not growth stagnation, but Keynesian stagnation. But Keynesian stagnation does not appear - at least to me - to be in the Keynesian tradition. Keynesians have been quite comfortable with the idea that mainstream growth theory could guide our thinking about "long-run" issues, while sticky-price and sticky-wage economics could guide our thinking about "short-run" issues. For example, Robert Solow provided us with the foundation for modern growth theory, but also wrote (with Paul Samuelson) a classic paper on how to exploit the Phillips curve tradeoff. Similarly, Mike Woodford took a several-generations-later version of Solow's growth model (with credit to Cass, Koopmans, Brock, Mirman, Kydland, and Prescott, along the way), put in some sticky prices and monetary policy, and convinced central bankers that it would be a good idea to use such a model to think about short-run monetary policy. New Keynesian models have the property that monetary policy is non-neutral in the short run, but the mechanics of the basic growth model take over in the long run.

This is definitely not what Larry Summers and Paul Krugman have in mind. Here's the basic hypothesis, as stated by Summers:
Unfortunately, almost all work in both the New Classical and New Keynesian traditions has focused on the second moment (the variance) of output and employment. This thinking presumes that, with or without policy intervention, the workings of the market will eventually restore full employment and eliminate output gaps. The only questions are about the volatility of output and employment around their normal levels. What has happened in the last few years suggests that the second moment is second-order relative to the first moment – the average level of output and employment through time.
I'll take "New Classical and New Keynesian traditions" to be represented by the ideas of Mike Woodford - prices and/or wages are sticky in the short run, and in the long run the world works according to the Solow/Cass/Koopmans/Brock/Mirman/Kydland/Prescott growth model. But what Summers sees is a world that is not at "full employment" even in the long run. He doesn't say this, but presumably he thinks that price rigidity and/or wage rigidity can persist indefinitely.

So what's the problem? Summers points out that the real rate of interest has declined over time, and argues - in typical New Keynesian fashion - that this has created a zero-lower-bound problem. The actual real rate has fallen, but it should be lower than it is (the "natural rate" is lower than the actual real rate), but monetary policy cannot lower the real rate further, because the short-term nominal interest rate is at its lower bound of zero. Summers argues that this problem could persist long into the future. Stagnation is then essentially a nagging output gap, that monetary policy cannot correct in the "usual" fashion. Paul Krugman is basically on the same wavelength.

If you are a young macroeconomist, you might be thinking of Summers and Krugman as some creaky dinosaurs blowing hot air. Where is your model, Summers and Krugman, you might say. Well, Eggertsson and Mehrotra have fleshed out a theory that they think captures what Summers and Krugman are trying to get at. The Eggertsson and Mehrotra chapter in this volume is a summary of a formal academic paper that I discussed in this post. The gist of that blog post is that Eggertsson and Mehrotra - as with Eggertsson/Krugman, which is closely related - focus on the wrong problem. The key inefficiency in their model arises from a credit friction, but they are focusing their attention on the secondary zero-lower-bound inefficiency that the credit friction creates. Basically, the problem is insufficient government debt, and the solution is straightforward.

Now we are getting somewhere. The contribution in this volume from Cabellero and Farhi gives a nice synopsis of safe asset shortages and why such shortages produce the low real interest rates we have been observing. Before the financial crisis, high savings in the world combined with financial innovation created a high demand for safe assets - as stores of wealth, as collateral, and for exchange in asset markets. Governments can supply safe assets, but the private sector can also do it. So, if governments do not increase their outstanding debt in the face of an increased demand for safe assets, then the price of safe assets rises and real interest rates fall. This creates a profit opportunity for the creation of safe private assets. Indeed, asset-backed securities could perform such a role. But the financial crisis showed us that, in the face of poor regulation, the capacity of the private sector to produce safe assets can be limited. Further, when the private sector builds up a stock of "safe" assets which proves not be safe, the ensuing destruction and loss of trust can result in persistent inefficiencies.

The private sector is rebuilding its capacity to produce safe assets, but changes in private sector regulation are also serving to increase the demand for safe assets. For example, the liquidity coverage ratio included in Basel III banking regulation will create an additional demand for safe assets by commercial banks. Though there are things that central banks can do in the face of safe asset shortages (as I show here and here), a safe asset shortage is basically a fiscal problem. The safe asset shortage is reflected in binding financial constraints that imply the economy is non-Ricardian. Government debt matters, and an expansion in the stock of government debt can be welfare improving. Presumably this also implies a lower net cost of financing government projects, meaning that a safe asset shortage provides an opportunity for the government to finance education and infrastructure on the cheap. Note that we can come to that conclusion without ever invoking stickiness, multipliers, fallacy-of-this, or fallacy-of-that.

An interesting feature of this paper is that it captures some of Larry Summers's concerns about monetary policy and financial stability. In the model, when there is a shortage of safe assets, low real interest rates can create incentive problems in asset markets. Basically, creating safe private assets is profitable when the real interest rate is low, but misrepresenting unsafe assets as safe ones is potentially even more profitable. Conventional monetary easing acts to reduce the real interest rate, and therefore aggravates incentive problems. Indeed, if incentive problems are severe enough, a safe asset shortage induces a situation in which the central bank should not push the nominal interest rate to zero.

The editors of this volume, Coen Teulings and Richard Baldwin, seem convinced that the potential for secular stagnation, whatever it may be, requires some radical rethinking of policy approaches. I don't think so. While there is much we don't know about how economies work, and we continue to learn, normal economics is certainly not at a loss in dealing with the problems we face, or will face.

Monday, August 4, 2014

The ON RRP Facility and Post-Liftoff Fed Policy

[See some related ideas in a post by John Cochrane.]

When the FOMC ultimately decides that short-term nominal interest rates in U.S. financial markets should rise, a lot will have changed since pre-financial crisis times. Principally, the Fed's balance sheet is about 5 times larger, total reserves have increased from about $10 billion on average in 2007 to about $2.8 trillion currently, and the Fed now pays interest on reserves.

Any central bank with a significant quantity of excess reserves outstanding overnight operates within a floor system, in which the interest rate on overnight liabilities of the central bank determines the overnight interest rate - in a financial system with no significant frictions. In the case of the U.S. Federal Reserve System, the frictions make all the difference, and create a thorny monetary policy problem for liftoff from essentially-zero overnight nominal interest rates.

The Fed's problem would be easy in a financial system like that in Canada, for example. In normal times, the Bank of Canada operates under a channel system. It targets an overnight interest rate, which is subject to an upper bound, which is the rate at which the Bank lends to financial institutions (the "Bank rate"), and a lower bound, which is the deposit rate for financial institutions at the Bank - the counterpart of the interest rate on reserves, or IOER in the U.S. Typically, the Bank rate is set at 0.25% above the target, and the deposit rate at 0.25% below the target. The idea is that no financial institution would borrow from another one at a rate above the Bank rate, nor would it lend to another financial institution at a rate below the Bank's deposit rate, so arbitrage will keep the target rate between the upper and lower bounds. Here's a chart showing the actual overnight rate in Canada:
Typically, the Bank pretty much nails its target. For example, in the chart, the target since mid-2010 has been 1%, and there has been little variability in the actual overnight rate around the target. An important point to note here is that lending and borrowing at the overnight rate is secured - this is a repo market.

Though normally the Bank of Canada operates a channel system, from April 21, 2009, to June 1, 2010, there was a floor system in Canada. Over this period, the Bank's deposit rate and the target rate were both set at 0.25%, with the Bank rate at 0.50%. Commensurate with that configuration of policy rates, the Bank also kept a positive quantity of reserves in the financial system overnight:
In Canada, there are no reserve requirements on banks. Overnight reserves would be zero, but for some slippage due principally to bad timing. For example, a financial institution could receive a large payment near the end of day, in which case it would be convenient to hold this as reserves overnight (no idea what was going on in early 2012). During the floor system period, you can see that the Bank of Canada targeted the quantity of overnight reserves at $3 billion - what the bank thought was sufficient reserves so that the floor of 0.25% would bind. With a floor system in place, there are sufficient reserves overnight such that arbitrage dictates that the overnight interest rate is equal to the interest rate on reserves.

But, you might say, that's not what has been happening in the United States since the financial crisis. There is a very large quantity of excess reserves in the U.S. financial system, and the IOER has been set at 0.25% since late 2008, but the fed funds rate - the overnight rate the FOMC currently focuses on - looks like this:
So, typically, fed funds have been trading below the IOER, and the interest rate differential has tended to increase over time. Over the last year, fed funds have typically been trading at 15 basis points or more lower than the IOER. Clearly, something is inhibiting arbitrage. What is it?

When Congress amended the Federal Reserve Act to permit payment of interest on reserves by the Fed, it specified that government-sponsored enterprises (GSEs) could not receive these payments. The key GSEs that matter in this respect are Fannie Mae, Freddie Mac, and the Federal Home Loan Banks (FHLBs). Most people know what Fannie Mae and Freddie Mac are up to, but most economists I have run into had no idea until recently what a FHLB is. The FHLB system was set up during the Great Depression as a housing finance vehicle. There are twelve FHLBs, each with a district (much like Fed districts, though FHLB districts are not identical to Federal Reserve districts) and member financial institutions. The members each hold nontradeable stock in the FHLB, and the primary activity of a FHLB is issuing tradeable securities to finance lending to member institutions. This lending is typically secured by mortgages, and the intention seems to be to direct financing toward low-income mortgage borrowers. FHLBs also hold a relatively small portfolio of mortgages, purchased outright. As is the case with Fannie Mae and Freddie Mac, FHLBs have reserve accounts with the Fed and, since GSEs receive zero interest on those reserve balances while commercial banks and other financial institutions currently receive interest at the current IOER rate of 0.25%, there would appear to be gains from trade.

GSEs which would otherwise hold reserve balances overnight could lend overnight on the fed funds market to commercial banks, for example. Those commercial banks could then hold the funds as reserves overnight, earning 0.25%, pay the GSEs x%, where 0 < x < 0.25, make a profit, and the GSEs and commercial banks would be better off as a result. Indeed, frictionless arbitrage would dictate that the GSEs would get all of the gains from trade, with x = 0.25. Why is this arbitrage activity, between GSEs and financial institutions with interest-bearing reserves important? Given that the U.S. financial system is awash in reserves, we might predict that there would be little reason for any financial institution to borrow on the fed funds market overnight, so that most fed funds activity should currently just be arbitrage. Indeed, Afonso et al. estimate that about 75% of lending on the fed funds market was accounted for by FHLBs, by the end of 2012.

But why is the arbitrage not perfect, or even close to it? Financial institutions that receive interest on reserves also face balance sheet costs associated with borrowing on the fed funds market. For example, deposit insurance premia depend on total assets, and there are capital requirements and other restrictions on commercial banks. As a result, the cost of fed funds borrowing is idiosyncratic - it will depend on a bank's size, and on the composition of its portfolio, for example. Of particular note is that branches of foreign banks in the U.S., because they do not have domestic retail deposits, do not pay deposit insurance premia, so they have a cost advantage over domestic banks in borrowing on the fed funds market. What would we predict then? Branches of foreign banks should be borrowing fed funds from FHLBs. Again, Afonso et al. estimate that, by the end of 2012, about 60% of borrowing on the fed funds market was being done by foreign-owned banks.

Once liftoff occurs, presumably with a stock of reserves in the financial system that is on the order of what it is currently, what would happen if the Fed adhered to its existing operating strategy? One might expect that, as the IOER increased, the fed funds rate would follow. Perhaps the current margin of about 15 basis points between the IOER and the fed funds rate would be maintained. Perhaps that margin would increase. It's hard to say, given our imperfect understanding of what explains the margin in the first place. One might also ask: who cares about the margin anyway? Surely the overnight rate that is critical to most financial market participants is the IOER? Maybe in this idiosyncratic floor system we should be ignoring fed funds market activity, as that is just about arbitrage between GSEs and foreign-owned banks?

There are other alternatives, though. The New York Fed has been conducting experiments using an overnight reverse repurchase agreement (or ON RRP) facility. A reverse repo for the Fed consists of a loan, typically overnight, to the Fed, secured by collateral (typically Treasury securities) from the Fed's asset portfolio. But why would the Fed need to post collateral on a loan? Surely the Fed is good for it? But, if the Fed were to borrow unsecured, we would call that reserves. If the Fed borrows by way of reverse repos, then it is permitted to pay interest to anyone, including GSEs, and it can also borrow from financial institutions that do not hold reserve accounts. Indeed, to enlarge the set of financial institutions that can hold interest-bearing Fed liabilities, the New York Fed has approved an expanded list of counterparties which includes domestic commercial banks, foreign-owned banks, GSEs (including some FHLBs) and money market mutual funds (which do not have reserve accounts). ON RRP activity by the Fed is shown in the next chart:
We should then think of ON RRPs simply as reserves by another name. Relative to regular reserves, the only difference is that reserves are unsecured while ON RRPs are secured, but since the Fed is the borrower, that's irrelevant. The only relevant difference between reserves and ON RRPs is that, while some financial institutions (e.g. commercial banks) can hold both reserves and ON RRPs and earn interest on both, some financial institutions (GSEs) hold reserves at 0% interest and ON RRPs with positive interest, and some other financial institutions (e.g. money market mutual funds) cannot hold reserves at all, but can hold ON RRPs.

But, given that the ON RRP facility now exists, what to do with it? To figure this out, it helps to understand how policy would work in the liftoff phase in the absence of ON RRPs. Given the size of the Fed's portfolio, total liabilities of the Fed - in this case currency and reserves - is essentially fixed in nominal terms. For a particular IOER, asset prices, the prices of goods and services, and quantities, including the fraction of Fed liabilities held as reserves, adjust so that consumers and financial institutions are willing to hold the total stock of Fed liabilities. If the Fed sets a higher IOER, then everything adjusts, and presumably the fraction of Fed liabilities consisting of reserves will increase.

Now, throw ON RRPs into the mix. There are different ways in which the Fed could intervene in the repo market. For example, the Fed could decide on a quantity of borrowing on a particular day, and then conduct an auction among its counterparties. However, what the Fed seems to ultimately envision is a "fixed rate, full allotment" allocation mechanism, according to which a rate is set, and the fed accepts whatever is forthcoming at that rate. So far, full allotment appears not to have been attempted, as the ON RRP experiments conducted by the New York Fed, beginning in January of this year involved putting per-counterparty caps on takeup, with a fixed rate of .05%. Caps were increased over time, to $10 billion per counterparty in April 2014. As you can see in the last chart, outstanding ON RRPs in the last few months have come in mostly between $200 billion and $300 billion - on average somewhat less than 10% of total interest-bearing Fed liabilities.

Given a setting for IOER, there will in general be some critical value for the spread between the IOER and the ON RRP rate such that takeup is zero below that rate, and positive above it. From the New York Fed's experiments, it seems we are safe in assuming that this critical spread is at least 20 basis points. Further, suppose the spread is lower than the critical value, and the ON RRP rate increases with IOER held fixed. What happens? Fed liabilities will in general be more attractive, and ON RRPs will be more attractive relative to either reserves or currency. We would expect that short-term market interest rates would increase (not IOER of course), and more Fed liabilities would be held in the form of ON RRPs, with less held as reserves and currency. For example, perhaps some FHLBs which would formerly have been lenders on the fed funds market would now hold ON RRPs. This would reduce reserves and activity on the fed funds market.

Clearly, given the IOER, the ON RRP rate can be sufficiently high that reserves go to zero. That is, we know that if the ON RRP rate exceeds IOER, then no financial institution would wish to hold reserves. But, it's possible that reserves could go to zero with the ON RRP rate less than IOER, if there is sufficient demand for ON RRPs from the GSEs and money market funds. Further, with sufficient interest-bearing Fed liabilities in the system, it is possible that the ON RRP rate could be high enough that fed funds market activity could dwindle essentially to zero. We know that much of current fed funds market activity is just the result of arbitrage between GSEs and banks that earn interest on reserves, and this arbitrage activity would go away with a sufficiently high ON RRP rate.

So, now that we have thought through how a system with three primary Fed liabilities - currency, reserves, ON RRPs - might work, what potential concerns might there be with such a framework?

1. There is potentially a lot going on here. How will the FOMC communicate monetary policy actions to the public in a simple and clear fashion? This was one of the subjects of discussion at June FOMC meeting (see the minutes):
Most participants agreed that adjustments in the rate of interest on excess reserves (IOER) should play a central role during the normalization process. It was generally agreed that an ON RRP facility with an interest rate set below the IOER rate could play a useful supporting role by helping to firm the floor under money market interest rates. One participant thought that the ON RRP rate would be the more effective policy tool during normalization in light of the wider variety of counterparties eligible to participate in ON RRP operations. The appropriate size of the spread between the IOER and ON RRP rates was discussed, with many participants judging that a relatively wide spread--perhaps near or above the current level of 20 basis points--would support trading in the federal funds market and provide adequate control over market interest rates. Several participants noted that the spread might be adjusted during the normalization process. A couple of participants suggested that adequate control of short-term rates might be accomplished with a very wide spread or even without an ON RRP facility. A few participants commented that the Committee should also be prepared to use its other policy tools, including term deposits and term reverse repurchase agreements, if necessary. Most participants thought that the federal funds rate should continue to play a role in the Committee's operating framework and communications during normalization, with many of them indicating a preference for continuing to announce a target range. However, a few participants thought that, given the degree of uncertainty about the effects of the Committee's tools on market rates, it might be preferable to focus on an administered rate in communicating the stance of policy during the normalization period. In addition, participants examined possibilities for changing the calculation of the effective federal funds rate in order to obtain a more robust measure of overnight bank funding rates and to apply lessons from international efforts to develop improved standards for benchmark interest rates.
To clarify what is going on here, if the Fed continues to announce policy in terms the fed funds market, this would require that the ON RRP rate be set sufficiently low relative to the IOER so that the fed funds market remains active. Use of the ON RRP facility has some advantages in this context. It works against segmentation in financial markets and thus gives interest-bearing Fed liabilities a broader reach, and it also puts a floor under the fed funds rate. The higher the ON RRP rate relative to the IOER, though, the less fed funds market activity there would be, and the greater the fraction of the interest-bearing Fed liabilities consisting of ON RRPs. Indeed, if the ON RRP rate were equal to the IOER, this might imply that most of the stock of Fed interest-bearing liabilities would be in the form of ON RRPs, as it is less costly for the GSEs and money market funds - which cannot earn interest on reserves - to intermediate ON RRPs than it is for commercial banks to intermediate reserves. Presumably the money market funds would expand and the commercial banks would contract.

2. A large ON RRP facility could eliminate fed funds market activity. Some people might ask why we should care. Borrowing on the fed funds market is unsecured, so each fed funds contract reflects idiosyncratic risk. In general, the "fed funds rate" is not a rate - it's a distribution. And in times of financial stress, the dispersion in that distribution can be substantial. You can see that in Figure 3 of this paper by Afonso et al. To get some idea of how risky the average fed funds market trade is, consider the margin between the fed funds rate and the 1-month T-bill rate in the next chart:
You can see that, even in normal times, there is a subsantial amount of risk in the fed funds market, and risk went up substantially during the financial crisis. We can certainly make a case that a central bank should be targeting an essentially risk-free overnight rate. Other central banks, including the Bank of Canada and Bank of England, see fit to target a repo rate.

In addition to the fact that the fed funds market is risky, a second problem is that the measured effective fed funds rate does not include all fed funds market trade, but only exchange through brokers. A large fraction of fed funds market activity is over-the-counter, and therefore goes unmeasured. There are indirect ways of measuring activity on the fed funds market, but these are problematic.

A third problem with the fed funds market is that central bank intervention in this market is unwieldy. In pre-financial crisis times, the New York Fed would attempt to hit a given fed funds rate target by predicting, on a given day, the demand for reserves, and then supplying the quantity of reserves that would satisfy demand at the target interest rate. This has lead to substantial fluctuations in the effective fed funds rate around the target, particularly during times of high financial market volatility. During the financial crisis, in addition to the problem that the fed funds rate was a questionable measure of the tightness of monetary policy (due to risk), the New York Fed seems to have had significant difficulty hitting the target.

All of these problems highlight the advantages of the ON RRP facility, and of the ON RRP rate as a permanent target for the FOMC. ON RRP activity is secured, so the ON RRP rate is essentially risk-free, and hitting a given ON RRP rate target is trivial using fixed-rate full-allotment. The Fed would simply fix a rate, and then accommodate forthcoming demand at that rate.

3. A dual system, with IOER above the ON RRP rate, would be more costly than it needs to be. Because holding interest-bearing Fed liabilities is more costly for domestic commercial banks than for foreign owned banks, GSEs, and money market funds, the Fed has to pay a premium to get banks to hold reserves. Thus, the Fed could achieve a given level of monetary accommodation, at lower cost, if the IOER is equal to the RRP rate, than if there were no ON RRP facility. This is roughly what Jeremy Stein is getting at:
You’re saving the taxpayer a little bit of money. You might say one job you give to the Fed is to fund its balance sheet as cheaply as possible.

4. A large ON RRP facility could make the financial system less stable. Sheila Bair, for example, has argued that a fixed rate full allotment ON RRP facility would create a kind of escape route for the liability-holders of financial intermediaries to run to in the event of perceived financial distress.
Even a relatively minor market event could encourage a massive flow of funds to the Fed while contributing to a flow away from other short-term borrowers.
Students of money and banking history might find that argument curious. To quote from the original Federal Reserve Act, Congress wanted to
... provide for the establishment of Federal reserve banks, to furnish an elastic currency, to afford means of rediscounting commercial paper, to establish a more effective supervision of banking in the United States, and for other purposes.
The idea behind "furnishing an elastic currency" was to provide an escape route. The recurrent banking panics during the National Banking era (1863-1913) were essentially shortages of retail media of exchange. The ability of the banking system to supply media of exchange was impaired during panics, and this disrupted payments and aggregate economic activity. Unfortunately, the National banking system was not designed to take up the slack. With the Federal Reserve System in place, however, the Fed could act to make up for the financial disruption during a panic by supplying more currency, either through discount window lending or open market purchases. Disruption associated with repo runs is not so different. Repos are liquid assets - media of exchange - which are provided by private financial intermediaries. In times of financial stress, as during the recent financial crisis, the repo market can be disrupted. In such circumstances, it is the job of the central bank to take up the slack. One way to do this would be through a ON RRP facility. The private sector could be having a difficult time finding liquid assets, and the Fed could supply them through the ON RRP facility.

So, in conclusion, the ON RRP facility seems neither mysterious nor scary, and could play an important role in U.S. monetary policy in the future.