Recent Posts

Wednesday, June 1, 2016

Summary: Why Quantitative Easing Was Pointless Policy

This article offers a summary of why I argue that we cannot see any measurable statistical link between the amount of Federal Reserve purchases of long-dated Treasury securities ("Quantitative Easing" or "QE") and the level of 10-year Treasury yields. Although I would summarise this as "QE did not work," there are some nuances involved in my position.

(This article was triggered by an email exchange with an ex-colleague. To a certain extent, I am restating what I wrote in earlier articles, although with some details expanded. It also reflects comments on a recent paper.)

Can We Measure The Effect Of QE?

I was careful to use the phrase "measurable statistical link" between QE and the level of 10-year yields. I am not using this phrase based on regurgitated statistical tests, rather a common sense modelling perspective.

In order to predict the effect of Treasury purchases on yields, we need to be able to predict 10-year yields in the absence of said purchases. Based on my long experience of wading through economists' models of bond yields, the best we can hope for in any public domain bond model based on "fundamentals" is a prediction error of about 100 basis points. but in practice 200 basis points is more common. (Remember all those predictions of 10-year yields going to 4% after 2010? Those were based on models that were eventually all taken behind the barn and shot.)

(Note: One common error for people new to the field is to look at models which feed in other yields, or the previous values of yields, and compare them to fundamental bond models. Such models can get much better fits. Since bond yields usually move less than 10 basis points in a day, we obviously can get an error of only 10 basis points or less if we feed in yesterday's yield. However, we then need to find a fundamental explanation for yesterday's yield. Furthermore, if we want to predict the yield a year from now using such a model, we would need to predict the yield one year less a day ahead, and we are not much further ahead.)

So if our model prediction error is 100 basis points, it is clear that it cannot usefully predict effects that are much smaller than 100 basis points. In other words, we would need to be able to find an effect of QE that is greater than 100 basis points in order to say that the effect is "measurable" (no matter what statistical significance tests say). More realistically, given the quality of model fits, the effect would have to be even larger than that.

In order to justify that QE had a measurable effect, it would be necessary to show that the 10-year Treasury yield yield would be at least 100 basis points higher than the observed yield. Given that the Fed started hiking in 2015, and is hiking at pace of about 50 basis points a year, I would suggest that the rate expectations view can easily justify the observed level of yields. However, we would need to look at the proposed models to be come up with a more specific analysis.

Where "QE" Works

There are a few ways in which QE can be seen to have "worked."
  1. Based on my comment section, many people lump in the Fed purchases of risky assets (or lending against them, which is effectively the same thing) during the Financial Crisis with "Quantitative Easing." Although I would distinguish the two type of operations (lending against risky assets is a lender-of-last-resort operation), if you insist on lumping the two types of operations together as "QE", then "QE" works under your definition of "QE" (but not under the definition I prefer).
  2. Purchases of Treasurys acts as a form of signalling mechanism -- if the central bank is engaging in QE, that is a signal that rate hikes will not happen any time soon. In other words, they are adding some muscle behind "open mouth operations." This is an entirely plausible story, but I would argue that the size of purchases is largely irrelevant (so long as they are not laughably small). The implication is that if the Fed ever engages in QE again in the future, the multiplier between the magnitude of purchases and the signal for rates is going to be variable. Furthermore, as seen in the "Interest Rate Conundrum" episode (Section 4.4 of Interest Rate Cycles) central bank jawboning has only a limited effect on bond yields. When you have Fed Governors yammering on about inflation risks based on crypto-Austrian reasoning, you tend to discount policymakers' long-term forecasting skills. This limits the ability of signalling to drive bond yields.
  3. If the segment of the bond market does not have alternative sources of supply, the yield curve can be greatly distorted by supply and demand effects. The long end of the curve is particularly of interest, as the private sector cannot credibly supply long duration low risk assets. Everyone knows that any corporation is three years and one empire-building CEO away from bankruptcy. The U.K. gilt curve was a notorious horror show during the 1990s as a result of pension demand for liability-matching, and the dive in yields in Japan could reflect the lack of alternative sources of duration. In the U.S. dollar markets there are tons of alternative sources for duration that are equivalent to 10-year Treasurys, so this market segmentation effect is unlikely to apply to the 10-year Treasury yield. That said, I would be willing to accept that one could isolate an effect of QE on the 30-year Treasury yield. 

John Taylor Article

The article "New Test Finds No Impact of QE on Long-Term Interest Rate" by John Taylor was the initial point of discussion in my email conversation. He quickly noted an earlier papers that suggested that QE was effective (and was based on an event study), and then discussed results of Ansgar Belke, Daniel Gros, and Thomas Osowski, which asserted that if we look at the international bond market, QE had no effect.

Although I agree that QE had no effect on 10-year yields, I have reservations about the use of correlation analysis in this case.

The Abuse Of Event Studies

Event studies (the analysis of yield movements around specific events) have its place in fixed income analysis. In fact, today is June 1st, which is an event that has considerable impact in the Canadian bond market (the other big day is December 1st). (Other markets have more sensible bond index structures, and those markets have auction/month end effects.) However, all an event study can do is look at what happens around the events in question, and tell us nothing about the overall level of yields.

For example, one could construct the "I Hate Mondays" yield series. This is a yield series that is defined by:
  • On Mondays, the value is the 10-year Treasury yield plus 7.5 basis points.
  • On Tuesdays, the value is equal to the 10-year yield.
  • On Wednesday, Thursday, and Friday, the yield is the 10-year yield less 2.5 basis points.
Any properly applied statistical test will tell us that the "Monday Event" is associated with a 10 basis point rise in the yield of this series. This would suggest an obvious trading strategy, although there would still be residual risks involved.

A person who does not understand how to apply event studies could then argue that since the yield is rising by 10 basis points (on average) each Monday, the yield will rise by at least 500 basis points a year.

Of course, since the average of the "I Hate Mondays" yield series is equal to the average of the 10-year Treasury yield (when taken over a full set of weeks, without holidays), the yield series is actually falling (since 1982, at least).

In summary, you cannot use event studies to make statements about the average level of yields. In particular, they cannot be used to justify the statement that QE lowered bond yields.

Correlation Not Just "Eyeball Econometrics"

Although I like the conclusions of the paper by Ansgar Belke, Daniel Gros, and Thomas Osowski, I have reservations about the methodology. However, it points out the obvious flaw in previous QE analysis. If we think supply and demand matter, we cannot isolate just the Fed's purchases, we need to look at all the factors affecting duration supply and demand. And yes, the developed interest rate markets are heavily integrated, so we would need to look at external supply and demand factors. Once we do this, the phrase "drop in the bucket" comes to mind when we think about Fed purchases.

That said, I disagree with John Taylor's suggestion that the correlation between markets is just "eyeball econometrics." Such a correlation is exactly what is predicted by rate expectations theory.

If we believe any of the following (and I would argue that all are true):
  • the business cycle is now global,
  • financial crises are now mainly global,
  • central banks now look at the spread of their policy rate versus their peers when setting the policy rate (for example, as the result of a belief that the level of the currency is affected by policy rate spreads),
the policy rates driving bond yields are now correlated. (The conditions listed above were less applicable in earlier eras.) As a result, we expect to see cross market yield spreads to show some mean reversion properties. And mean reversion implies correlation.

Since we expect bond yields across markets to be correlated, we will have a difficult time using cross-market studies to disentangle the causes behind yield changes. 

(c) Brian Romanchuk 2016


  1. Crystal clear.

    Your open reminds me of Willem Buiter's strongly-worded endorsement of the heli drop, which he claims any "minimally intelligent" set of policy makers could easily pull off. He mentions in notes that he wrote the paper on Larry Summers' urging that he express his case briefly. That was about 18 months ago. So far, Summers has no taste for the heli drop. Not that he is necessarily right, although he is in this case. Your open reminded me of that, is all.

    1. Thanks. Don't think I saw that Buiter article.

  2. There is one important side effect of Fed purchases under QE which might be of importance: It exchanges a fixed coupon bond with reserves at the Fed paying a variable interest rate equal to Interest on Reserves. In effect it resembles the floating leg of an interest rate swap with the Fed assuming "interest rate risk". The more reserves are created, the less is the interest rate risk for the banking system as a whole since it can hedge this risk through Interest on Reserves.

    By lowering interest rate risk and as a result, the risk of capital losses for bondholders (from an increase in yields) QE might be able to move long-term interest rate expectations.

    Just a thought.

    1. It certainly affects the interest rate risk ("duration") for Treasurys outstanding. The question is whether that affects pricing. There are other factors affecting duration, such as the fact that the Treasury extended the duration of bond issuance during that period. Plus, there is private sector supply, foreign currency duration, etc. Why focus on Fed purchases when there are these other factors as well?

      The biggest problem facing supply and demand stories is that supply spikes in a recession - when rates are the lowest.

    2. I am not talking about supply and demand or duration but rather about the fact that the Fed exchanges a fixed interest rate instrument with a floating interest rate instrument and assumes 'interest risk' itself. The banking system as a whole probably has lower interest rate risk since it holds several $tr of bank reserves paying Interest on Reserves (which can be regarded as a very large Interest Rate Swap).

    3. "Duration" is shorthand for interest rate risk. (Technically, you need to use a measure like DV01 - Dollar Value of a Basis Point - but less people are familiar with the concept.) So "reducing duration" (or the "supply of duration") is equivalent to saying "reducing interest rate risk". Fed purchases are only one factor affecting that interest rate risk; if you look at the history of the Flow of Funds, there are huge changes in the flow of fixed coupon bonds without there being a discernible effect on yields.

      If you want to isolate banks, there might be some effects. However, banks are supposed to keep their net interest rate risk near zero; they make money off of spreads, not by duration mismatches. (This is the response to the way S&L's blew up in the Volcker era.) They will adjust other instrument holdings to achieve their net interest rate risk target. Banks' liquidity positions are presumably improved. However, by reducing the odds of a financial crisis, I would argue that rate expectations should be higher, not lower.

    4. My point is that by providing the private sector with several $tr of reserves paying Interest on Reserves the Fed has effectively moved interest rate risk to the government balance sheet and created corresponding 'space' on the private balance sheet. That allows for easier hedging and might make it easier to move 'sticky' long-term interest rate expectations.

    5. I agree that the interest rate risk for the private sector has been reduced. The issue is whether this change in isolation is going to have any measurable effect on yields.

  3. In the United States, when the central bank (Fed) makes a loan at the discount window, or otherwise lends money to a bank, this results in an increase in bank reserve balances but no corresponding increase in the transaction account (checking deposit) levels of nonbanks. I think the monetary statisticians refer to reserves created by this mechanism as "borrowed reserves." By contrast, if Fed purchases a security from a nonbank counterparty in the open market (or does a certain type of repo agreement) then it results in an increase in bank reserves and a corresponding increase in checking deposits when the banks clear the transaction between Fed and a nonbank. The reserves created in this manner are called non-borrowed reserves. During the financial crisis Fed initially attempted to provide banks with non-borrowed reserves but these reserves had to be "sterilized" by the sale of Treasuries to retain control over the fed funds rate. Fed sought authorization to pay interest on reserves and instituted securities purchases to provide a large volume of nonborrowed reserves to avoid the problems associated with controlling interest rates while providing a large volume of borrowed reserves. The interest rate transmission mechanism is not well understood so we are stuck with "correlation is not causation" models where Fed adjusts its rates and markets adjust their portfolio decisions in a complex way that defies a clear recognition of cause and effect. Most people who discuss QE (asset purchase from nonbanks) do not explain how it provides liquidity for banks and nonbanks, that FDIC increases insurance on bank deposits, and so banks and nonbanks both gained liquid fixed net asset value liabilities of the central bank and Treasury while swapping out floating net asset value assets from their portfolios. The concept of "printing money" means an floating NAV is swapped for a fixed NAV and this helps stop the run on both bank and non-bank financial intermediary liabilities in a financial panic. Later one can criticize Fed/Treasury/FDIC for saving the world and speculate about the impact on interest rates when the main reason is to stabilize an unstable market structure and fight deflation.


Note: Posts may be moderated, and there may be a considerable delay before they appear.

Although I welcome people who disagree with me, please be civil.

Please note that my spam comment filter appears to dislike long "anonymous" posts. I get no warning about this, and only go through my "spambox" infrequently. The best bet it to keep comments short, and if you think the spam filter struck, let me know with a short comment.