What’s up with the “term premium”?

Posted on December 1, 2016

If you could know exactly where short-term interest rates will be in the future, it would be easy to determine whether longer-term bonds offer attractive yields today.

But that’s not the world we live in. Investors have to live with the risk that their best guesses about future borrowing costs won’t be good enough and hope the “term premium” embedded in bond yields is high enough to compensate for the uncertainty.

This premium can’t be observed in the wild. Instead it’s estimated by subtracting your best guess of what interest rates “should be” at every point in time from whatever they really are. The challenge is disentangling actual changes in expectations about inflation and real rates from changes in the risk surrounding those expectations. (Related.)

The most popular of these models comes from researchers at the Federal Reserve Bank of New York, which we’ve quibbled with before. (Among our concerns: ultra-long-term inflation expectations apparently didn’t move between 1960 and 2010.)

A new note from rates strategists at Citi inspired us to revisit what’s going on in the NY Fed’s model.

Start with what the “term premium” is supposedly capturing. Uncertainty about the level of interest rates should rise as you gaze deeper into the future, yet almost all of the term premium seems to exist within a five-year window:

Term premium 5yr and 10yr

Historically, American business cycles only last about four to six years on average — and the recent ones have been even longer. In other words, most of the action in the New York Fed’s estimate of the “term premium” occurs over a time horizon when inflation is expected to move around a bit relative to any stable longer-term target and when real short rates will probably go up and down quite a lot thanks to changes in policy. There might be lots of volatility in inflation and real short rates in a five-year window, but not necessarily much reason to think the range of outcomes should widen or shrink dramatically within a fraction of a business cycle.

On the other hand, the narrow difference between the implied 5-year and 10-year “term premiums” suggests investors get little compensation for the danger that inflation might accelerate significantly over time. Imagine the re-pricing that would be required if the Fed responded to a future recession by moving its target to 4 per cent annually, up from 2 per cent, as some worthies have suggested. It’s also hard to square the idea that buying by central banks at the long end of the curve has had much of an impact on the term premium when most of the volatility is occurring in the first five years.

Another way to look at this is to look at the difference between 5-year and 10-year interest rates over time, as well as differences in the implied “term premiums” and the implied “risk-neutral” rates:

5s10s decomposition

In general, 10-year interest rates have been higher than 5-year interest rates, although there were long periods when investors were essentially indifferent between the two time horizons. Longer-term rates were rarely much lower than medium-term rates, and when they were, it was during the sharp and deep recessions of the 1970s and early 1980s.

Weirdly, the NY Fed model suggests this has occurred despite traders’ expectations that real rates would fall and inflation would slow down for much of the past 55 years, which is why the teal line is negative for two-thirds of its history. (The current period is an unusual exception.)

Offsetting this apparent belief that short-term interest rates would fall over time, or at least almost never go up, was a modest risk premium — a premium that stayed confined within a range of 1 percentage point between the early 1970s and the early 2000s despite massive changes in the trend of inflation and several significant changes in the policy regime.

Is it reasonable to think people in the late 1970s “expected” short-term interest rates to be lower in the future than in the present, even as the country was suffering from a worsening inflationary spiral? Is it reasonable to think people in the mid-1990s “expected” short rates to fall over time even when the economy was booming and before the productivity boom had become fully evident?

Citi makes some other interesting points. For one thing, you can generate very different results using the New York Fed’s model if you tweak two key inputs: the start date for your regressions and the expected holding period for your bonds. The chart below shows what happens when you use different time periods to tell the model about macro relationships:

Citi 10y term premium different start dates

Focusing only on data since Volcker’s departure from the Fed makes the implied “term premium” appear much less volatile and range-bound. The blue line above is towards the bottom of its history right now, but not far off from levels in the mid-1990s. From this perspective, investors in 10-year bonds are getting paid about a full percentage point above what they might expect from buying short-term debt and reinvesting maturing instruments in new ones for a full ten years.

We have no insight into whether one approach is better or worse than another, but the magnitude of difference between the blue and purple lines suggests it’s significant.

Similarly, consider what happens when you combine the previous decision to only use data since 1987 with a change to the assumption about your holding period:

Citi 10y term premium different holding periods

Again, we don’t have strong opinions on whether one approach makes more sense than another, but it’s impressive how much the picture changes when you make these two simple tweaks. The term premium today, far from being near its all-time low, now appears to be towards the upper end of its 30-year range. (We also would point out the supposedly big swings in the “term premium” in the 2000s vanish.)

Citi also used the NY Fed’s model to estimate “term premiums” for Germany and the UK and compared those to America. Intriguingly, they found that the American government supposedly pays a lot more to secure longer-term financing than the UK after accounting for inflation and real rate expectations — potentially more than 2 full percentage points depending on the specific model inputs:

Citi 10y term premium cross country

Is that plausible? On the one hand, the UK has robust demand for longer-term debt from its pensions, which also shows up in breakeven inflation rates. On the other, it seems to contradict the common narrative that America is a safe haven uniquely privileged (or burdened) with low yields thanks to foreign inflows.

One last wrinkle: Citi found that the spread between nominal 10-year Treasury yields and 10-year Bund yields tends to be inversely related to the size of the UST “term premium” implied by the NY Fed’s model. When American yields are far above German ones, the implied “term premium” appears small, but when yields are close together, the compensation for duration risk is large:

Citi 10y term premium bund spread

Citi thinks higher US yields attract foreign inflows that compress the “term premium”. This is an interesting finding although we’re not sure the mechanism is plausible.

One of the common explanations for the widening spread is the yawning divergence between the monetary policies of the European Central Bank, which is stepping up its bond-buying, and the Fed, which is itching to raise rates. Capital has flowed out of the euro area and the currency has weakened against the dollar, but it’s less clear how or why this would affect the embedded pricing of duraton risk.

If anything, the policy mix might cause American duration risk to rise relative to Germany, especially since central bankers have traditionally viewed QE as having a big impact on the term premium. At the same time, it’s reasonable to think increased ECB activism combined with Fed tightening should encourage convergence between American and Germany in longer-term expectations for inflation and real growth. That’s the opposite of Citi’s theory.

As we said the last time we covered this subject, we have great sympathy for anyone attempting to extract an estimate of the underlying duration risk embedded in long-term bond prices. We don’t have any great insights into how to do it better than the increasingly-canonical New York Fed model, but it’s worth pointing out oddities in their results, especially when they seem directly contrary to the basic concepts of expectations and risk surrounding those expectations.

Related links:
Trust, Treasuries, and Trump — FT Alphaville
Decomposing the yield curve — Cochrane and Piazzesi

Copyright The Financial Times Limited 2016. All rights reserved. You may share using our article tools. Please don’t cut articles from FT.com and redistribute by email or post to the web.

You must be logged in to post a comment Login