October 15, 2019 / 03:59PM
/ By Gertjan Vlieghe, External MPC Member /
Header Image Credit: BoE
Given at the Money Macro & Finance Society Monetary and Financial Policy (MMF) Conference 2019, London
Today I want to discuss what I see are some key differences in how monetary policymakers think about the economy and set policy, relative to the pre-crisis period.
I want to focus on three major changes.
First, structural questions like the level of the neutral interest rate have become at least as important as cyclical questions like the size of the output gap.
Second, monetary policy has more tools than it had before. And, thinking beyond monetary policy, central bank policy has added even more tools.
Third, there currently is an asymmetry in the power of monetary policy: our ability to ease further is significantly smaller than our ability to tighten.
I will discuss these in turn, and then consider the implications for how we set monetary policy in the future. I will conclude with some remarks about the current outlook.
1. Structural vs Cyclical questions
We now spend relatively more of our time on structural questions like the level of the neutral interest rate, or r*, than on cyclical questions like the size of the output gap.
Previously, we mostly worried about the output gap. The neutral rate and productivity growth were believed to be fairly stable.
Policy roughly followed growth. With a fairly stable level of potential output growth, high growth meant the output gap was shrinking, and would lead to interest rates going up. Low growth meant the output gap was widening, and interest rates would go down.
The early phase of the crisis was still about the output gap. It was so big that interest rates could not be lowered enough.
But recent years have seen the output gap closing, and policy rates still remaining at or near the historical lows.
The discussion is now about forces that will require low rates even when the output gap is closed, or at least much smaller than before. That is a debate about the neutral rate, or r*.
This is a topic I have previously discussed on several occasions. I first emphasised several long term determinants of real rates, which I labelled the 3Ds: debt, demographics and the distribution of income. 1
More recently I showed how we can also think about low frequency changes in equilibrium real rates, from a finance and risk perspective, by linking them to precautionary savings and macro-economic tail risk. 2 These are not separate explanations. Rather, they are different ways of looking at the same problem.
The real risk-free rate is a fundamental variable in the economy that is potentially linked to every other aspect of the economy. In technical terms, it is potentially a function of all "state variables" (fundamental factors) that determine all the equilibrium outcomes. It is therefore all too easy to run into the circularity of general equilibrium when discussing what explains real rates. "Real rates are low because demand is weak" or "demand must be weak because real rates are low".
The appeal of the 3Ds is that, to different degrees, they seem to be fundamentals that determine, but are not initially determined by, real rates (this is particularly true of demographics and income inequality), and so the circularity is avoided to a large degree. This is rare in macroeconomics, and is purely the result of how persistent these factors are. This comes at a cost: because they move so slowly, it is difficult to determine empirically their impact on the economy. This is also why, until very recently, these were not variables that policymakers or monetary economists paid much attention to.
How can we map the 3Ds into the finance and risk story? To explain this it is worth remembering that the finance and risk story boils down to saying that real rates will be lower: the higher is aggregate patience (a time preference parameter); the lower is the expected consumption growth rate; and the higher is consumption risk (measured by higher volatility, fat tails and more downside skew). So, to do the mapping, we just need to explain how the 3Ds may affect one or more of these factors.
In the case of demographics, there is an increase in the supply of savings when the population ages, such that there are more individuals at a high savings stage of their life-cycle. 3 There is a further increase in the supply of savings that results from an increase in longevity without a commensurate increase in the pension age, so that more savings are required to sustain consumption in retirement. The link to risk is that pensioners, having to rely only on investment income, are typically more risk-averse than those in work, so will have a preference for safer assets. In other words, they will tolerate a lower risk-free rate (compared to those in work) to avoid the investment income risk associated with riskier assets.4
In the case of debt, I have in mind a story of credit cycles, debt overhang and fragility: during periods of high growth expectations, or financial innovation, credit growth is high and economic growth is strong. Eventually, this process leads to the economy becoming more fragile, as highly indebted firms or households are increasingly sensitive to changes in income expectations. In a downturn, the economy weakens sharply as debt needs to be repaid. 5 This pattern of long booms followed by sharp downturns implies a riskier economic environment (as measured by the fatness of tails and their downside skew) than one where the economy is less leveraged.6 Hence economies with strong credit cycles will have lower real interest rates on average, and in particular during the deleveraging phase of the cycle.
In the case of the distribution of income, a potential link to risk is as follows. As more income accrues to the top of the income distribution, it becomes concentrated among those with a low marginal propensity to consume. Interest rates need to fall just to sustain consumption at previous levels. Recent research shows that the impact is much larger if the income inequality is associated with increased individual income risk.7
The effect will be larger still if it occurs when nominal interest rates are close to their effective lower bound, because there is less scope for interest rates to offset the demand weakness, hence the economy risks falling into a prolonged period of weakness. The higher risk of that happening will itself feed back onto risk-free rates, making them lower still. Moreover, research for the US has shown that the income risk of the top of the income distribution, who disproportionally hold (and hence determine price of) assets in the economy, has been increasing in the last few decades.8
Figure 1: The Emergence of Fat Tails and Negative Skewness in Consumption Growth after 19141
In previous work 9 I have used the risk framework to explain changes in real interest rates over very long periods. I showed that we can use it to understand why real interest rates were above 3% during the 19th century gold standard era, and closer to 1% in the past 100 years. As shown in Figure 1, even though consumption growth was higher in the recent century, risk was higher as well (volatility, downside skew, fat tails) and this pushed real interest rates down. And the increase in risk can quantitatively explain the fall in average real interest rates.
Of course, the 20th century was hardly a period of constant risk. So a natural follow-up question is whether this framework is useful for explaining variation in real interest rates within the past century, rather than just explaining the average.
I think the approach is promising, and I would like to share some results with you.
Recall that the approach in our earlier paper was that we estimate, using only consumption data, the parameters that determine macroeconomic tail risk, to check if they can explain average interest rates across the two very long sub-periods. There are five key parameters: the mean, the standard deviation, the probability of a jump or 'tail event' (which captures the potential occurrence of infrequent bad outcomes), the average size of a jump, and the uncertainty about the size of a jump.10
As a simple first check, I now ask the following question. What if we just allow the probability of a bad outcome to vary over time? Would that be able to match the variation in real interest rates? The answer, shown in Figure 2, is yes. Just moving this one risk parameter around can match almost exactly the trend in real interest rates in the post-WWI sample.11 The range in which the parameter moves over the entire sample is [0, 40%], and less than 22% in the period excluding World Wars. The expected fall in consumption (jump probability multiplied by the expected jump size, which is held fixed at 3.1%) relative to the trend of 2% is between [0, -1.25%], meaning that the expected consumption growth is never lower than 0.75%. So these are not outlandish, never-before-experienced tail risks (see Appendix A.1 for details).
In other words, because there are nontrivial effects of negative skewness and fat tails on the real interest rate required by a risk averse individual, we are able to explain a drop in real rates from 4.3% before the crisis to -1% without changing trend consumption growth, and with a fall in expected consumption growth from 2% to only 1.25%.12
Kindly read the PDF speech here.