## Speech

### 1: Introduction

At its meeting last month, the Monetary Policy Committee came together to deliver on its pledge that it would react forcefully in response to more persistent inflationary pressures. A large majority agreed that this condition had been met and that we were indeed seeing persistence so the hike by a half percentage point was the appropriate course of action at that meeting. This sends a strong signal that the MPC is committed to leaning against domestic inflation pressures becoming further entrenched.

Now, some commentators have pointed out that the MPC’s central forecast projects a pro-longed
recession over the next quarters, and that this slack would yield inflation falling below the target
in year three.^{footnote [1]} Therefore, they claim inflation to
be vanquished already and indeed that monetary policy already has become contractionary. Further,
since the shocks hitting the UK economy have come mostly from external sources – supply chain
frictions, energy prices, war – they claim that the MPC need not tighten further, but rather can
‘look through’ these shocks, since they will mean-revert.

In this speech I’m going to argue that, while some elements of this line of argument might hold in normal times, it is based on an incomplete view of the inflation process and of the channels through which monetary policy can achieve our remit. Specifically, in today’s environment, inflation expectations take a central role, alongside the standard channel of aggregate demand and slack. Looking through the lens of inflation expectations, achieving the remit depends on ensuring that inflation expectations in the short-term do not become adaptive, and that medium-term inflation expectations do not drift, so that long-term expectations remain anchored.

What most people seem to have in mind when thinking about the determinants of inflation is a variant of the Phillips curve (Phillips, 1958), which relates inflation to some measure of slack in the economy. In the earliest version of this model, inflation today is determined by economic outcomes yesterday – inflation therefore will adjust slowly and with considerable lag. To bring inflation down, when the MPC raises interest rates, it increases borrowing costs for firms and households causing them to spend less, unemployment eventually rises and slack opens up. Only then do companies adjust prices in response to slowing demand for their goods and services. In Friedman’s (1961) story of “monetary policy having long and variable lags”, this is where the “long” comes from. Importantly, since lagged inflation is pre-determined, the only way to get inflation down from high levels is to depress aggregate demand for an extended period of time. In this formulation, disinflation is costly and necessarily so.

I do not dispute that this mechanism exists, but I hope that the MPC will not have to depend on it alone to bring inflation down to target. In the more nuanced formulation of the Phillips curve discussed below, inflation today does not simply depend on past inflation but depends as well on markets, firms, and household’s expectations, and crucially, how these expectations react to each other, are formed over time, and interact with our and others’ policy choices. The MPCs’ evaluation of inflation expectations therefore should take a central role in monetary policy decisions.

In this more complex and arguably more realistic and relevant version of the inflation model, a fast and forceful monetary tightening, potentially followed by a hold or reversal, is superior to the gradualist approach because doing so is more likely to promote the role that inflation expectations can play in bringing inflation back sustainably to 2% over the medium term. This policy strategy would reduce the risks of a more extended and costly tightening cycle later that depends primarily on shrinking aggregate demand.

### 2: Micro-foundations and expectations in the Phillips curve

The first expectations-augmented Phillips curve was a crucial innovation in understanding the
inflation process. In this formulation, inflation expectations are not just backward-looking, but
agents in the model are allowed to anticipate how policy choices today may shape macroeconomic
outcomes tomorrow.^{footnote [2]} If we allow for both backward- and
forward-looking expectations formation, we arrive at the hybrid Phillips curve (Clarida et al., 1999
and Gali & Gertler, 1999). Here, the non-slack term in the equation becomes a weighted average
of past and expected future inflation:

\[\pi_{t}=\left ( 1-\beta\right)\pi_{t-1}+\beta \pi_{t+1}^{e}+\lambda\bar{y}_{t}\](1)

Depending on the value of β, this model encompasses the accelerationist Phillips curve (as β → 0) but importantly also encompasses the so-called New Keynesian Phillips curve (as β → 1). In the latter model, inflation is entirely forward-looking and determined only by expectations of future inflation, which is a function of expected future slack or more precisely, to the discounted path of expected future real marginal costs. To see why that is interesting and potentially relevant to the current conjuncture, let me briefly walk through the foundations of the New Keynesian Phillips curve, which makes the bridge between microeconomic fundamentals and macroeconomic outcomes.

By virtue of being “micro-founded”, the New Keynesian model can connect all its macroeconomic equations to some microeconomic (i.e. firm- or household-level) optimisation problem. The key bridge between macro measures of inflation and the micro-foundations is the price-setting problem of the firm. To presage the conclusion: In their pricing strategy, firms face adjustment costs, consider trend inflation and competitor prices, and evidence downward wage and price rigidity. Collectively these yield a non-linear and potentially shifting Phillips Curve, with important implications for the role for monetary policy to influence expectations.

The first crucial point for the inflation model is that prices typically are not infinitely flexible.
Firms will only choose to bear costs of changing prices once the adjustment is sufficiently
large^{footnote [3]} or because prices are fixed
through contractual arrangements (e.g. rent, utilities)^{footnote [4]}.
Calvo (1983) formulates these various types of stickiness by assuming that a certain proportion of
firms is allowed to reset its prices while the rest remain at last period’s. As shown in
**Chart 1** (adapted from Werning, 2022), if the firm anticipates that prices will go
up, it will optimally overshoot the current price level to ensure that it hits the correct price on
average. Uncertainty about how long they will stay at the new price and how fast the frictionless
price level is rising will create incentives to overshoot by even more to insure against
exceptionally long spells or more rapid rises. Thus, the adjusting prices have momentum above, and
can pull up, trend inflation.

Further, when we take seriously the way firms behave in the real world, there may be price
coordination between firms. When a firm sees a competitor raise their price, they might be
emboldened to do the same even if their marginal costs have not actually moved. If a series of large
and salient shocks increase costs for some but not others, this “me-too-ism” would show up as a
persistent rise in desired mark-ups, driving inflation upwards with seemingly no connection to cost
or demand conditions.^{footnote [5]}

Various firm-level behaviors affect the shape of the Phillips Curve. The optimal reset-price behavior along with me-too-ism induce an upward bias in inflation in the short term. On the disinflationary side, while firms are quick to raise prices when they can, prices rarely outright fall in aggregate.

### Chart 1: Firm-price overshooting in sticky price models

#### Footnotes

- Source: Werning (2022).
- Notes: Adapted from Ivan Werning’s slidepack presented at the NBER Summer Institute 2022.

This downward nominal rigidity can be seen in deep recessions: Even during the Great Financial Crisis
or the Covid lockdowns, consumer prices in the UK never actually fell in year-on-year terms. Firms
facing bankruptcy may raise prices to generate revenues to pay debtors, even if reducing prices
might make more sense in theory (Gilchrist et al., 2017). Nominal wage rigidity is well known and
since labour is an important cost this will bolster the downward rigidity of prices (Daly &
Hobijn, 2014). All told, for a given change in economic slack, prices will rise more in expansions
than they would fall in contractions.^{footnote [6]} The Phillips curve is non-linear.

A second innovation, particularly important now, is that inflation expectations may change over time. In theoretical work, the change between the backward-looking, accelerationist and the forward-looking views often is modelled as a largely exogenous regime shift. However, it is plausible that there is a smoother transition between states, endogenously shaped by macroeconomic outcomes. For example, contemporaneous research from the 1970s suggested that the time away from target was an important determinant of the degree of backward-lookingness: Robert Gordon in 1970 rejected that the US Phillips curve was accelerationist. By 1977 his updated estimates suggested inflation expectations were fully backward-looking.

One plausible mechanism for such endogenous regime shift – proposed by Cornea-Madeira et al.
(2019)^{footnote [7]} – arises if we think of two types
of firms and households in the economy: forward-looking “fundamentalists” and backward-looking
“random-walkers”. These have different forecasting rules, and firms and households will switch
between them when the one rule outperforms the other. If inflation varies modestly around its
target, the fundamentalists dominate. This is akin to Alan Greenspan’s desired Central Bank outcome
where firms and households ‘ignore’ inflation when making their decisions. The literature formalizes
this behavior as ‘rational inattention’.^{footnote [8]}

But, when shocks drive inflation away from target for extended periods and the fundamental rule produces larger forecast errors, more firms and households use the backward-looking rule to form adaptive expectations. In this case, experiences of past high inflation can become embedded in firms’ and households’ price-setting decisions and the aggregate Phillips curve becomes more accelerationist over time.

I find this framework useful because it is more nuanced than the discrete switching between “expectations are forever anchored” and “expectations are suddenly unanchored”. Even if long-term expectations remain stable and the central bank’s target is credible, such a model can generate different degrees of backward-lookingness and intrinsic inflation persistence. From the theoretical perspective, this time varying expectations formulation also represents a departure from the strict rational expectations formulation of the canonical New Keynesian Phillips curve. In this more complex formulation, monetary policy, by having an expected effect on macroeconomic outcomes, can feed back to affect the inflation expectations process and therefore current inflation outcomes.

To summarize the importance of this world for monetary policy: Monetary policy does not just focus on
demand management, but also on coordinating firms and households to agree on some fundamental
equilibrium,^{footnote [9]} in which prices and wages rise at
target-consistent rates and the economy is growing at a sustainable pace and any shocks are expected
to wash out quickly.

From the standpoint of monetary policy, these variations yield different implications for channels
through which monetary policy can affect inflation. In the old, backward-looking, accelerationist
version, inflation always *follows *economic slack. In the rational expectations model, when
expectations are forward-looking, inflation can move *even before* changes in the monetary
policy stance have affected real economic activity. And in the time-varying model, monetary policy
can influence aggregate inflation expectations via the share of fundamentalists versus random
walkers. It follows that, if monetary policy can act on inflation expectations, then the dependence
on the aggregate demand channel to discipline firms’ price setting is reduced.

Collecting up the theories points to a Phillips Curve that is both non-linear and can shift.
(**Chart 2**). When we think about it, the Phillips curve is actually a set of
isoquants, which trace out all attainable combinations of inflation and slack given a value for
expected inflation. When expectations change, so does the location of the curve.

### Chart 2: Shifting Phillips curves due to changing inflation expectations

Modest shift ^{(a)} and acceleration ^{(b)}

Here, let the blue line be the Phillips curve of the economy at the outset of any shocks or firm reactions. Given initial conditions for inflation expectations as well as desired mark-ups, there is a level of slack y* consistent with achieving the inflation target π* at the intersection of the horizontal dotted line and the blue curve. Small disturbances will push the economy away from that intersection but only by modest amounts and we can always – along the curve – travel back to the same equilibrium.

Now consider the case in which inflation expectations drift up: the newly attainable combinations of
inflation and slack are shown by the red curve. Keeping inflation at target requires more slack in
the economy, more unemployment, and lower growth (the intersection of the horizontal dotted line
with the red curve). On the flipside, holding the level of slack constant at the old y* will lead to
higher inflation (the intersection of the vertical dotted line with the red curve).^{footnote [10]}

If we could be confident that any shift was short-lived and that the red curve would shift back in line with the blue curve by itself, this might be a situation in which MPC could choose to “look through” the shock and tolerate a temporary inflation overshoot. Our remit explicitly allows for this as long as medium-term price stability is not threatened.

If, however, there is a risk of further acceleration of inflation and further drift of expectations,
to ‘look through’ the inflation changes, we have to be very confident that inflation does not
becoming embedded in expectations and outcomes. If there are such shifts in inflation expectations,
more economic pain is required to bring inflation back to target (red dashed curve on the right-hand
side of **Chart 2**). Which situation are we in now? What do we know from the data
about the slope and potential shifting of the Phillips Curve?

### 3: Taking these models to the data

A first observation is that the Phillips curve set ups above are not the same as the correlation between inflation and an output gap measure or unemployment (sometimes dubbed the “Phillips correlation”). Instead, they describe structural, causal relationships between inflation yesterday, today, and tomorrow along with associated economic conditions. This relationship may not be immediately visible from a simple scatterplot. Just as correlation does not imply causation, causation does not imply correlation.

In fact, if one were to simply fit a line through the scatter in **Chart 3**, which
plots UK unemployment against CPI inflation in quarterly frequency, there would be no detectable
relationship. However, if we trace out the time series underneath the point cloud, we can see some
interesting patterns:

The first thing that jumps out is of course the high-inflation period of the 1970s and the subsequent
moderation of the 80s (in light blue). This moderation in inflation did not come painlessly but
instead was only achieved alongside high rates of unemployment. Then, the thirty years thereafter
trace out a continuous downward trend in both inflation and unemployment and finally, variation
becomes so small that it is barely visible compared to the first twenty years. Finally, compared to
the recent history, the Covid period does not look like a particularly egregious outlier.^{footnote [11]}

### Chart 3: Unemployment rate ^{(a)} and CPI
inflation ^{(b)} in the UK since 1971

Percent of labour force ^{(a)} and year-on-year percentage
changes ^{(b)}

#### Footnotes

- Source: Office for National Statistics.
- Notes: The chart shows combinations of CPI inflation and the unemployment in the UK at quarterly frequency. Inflation before 1989 is based on the ONS’ historical series for UK CPI available at ‘Modelling a Back Series for the Consumer Price Index, 1950 – 2011’.

Now consider the time from 2021 Q4 onwards. The configuration of the most recent data (the final red
dot) with July inflation at double digits and unemployment at under 4% looks more like 1973 than any
other time, as indeed does the importance of energy prices in both periods.^{footnote [12]} Of course, many things have
changed since then – one of which is the recognition of the power of an independent central bank –
but this configuration of the data should cause concern.

Strong, independent, and credible central banks may have contributed to why the Phillips correlation
has ‘disappeared’ in recent periods across the world. Although, perhaps the Phillips curve never
actually went away and it probably always had been relatively flat (Barnichon & Meesters, 2021).
It just became more difficult to identify because monetary policy became more systematic and
predictable. As McLeay and Tenreyro (2020) explain, if monetary policy were able to perfectly offset
demand shocks, there would be no visible correlation in the data. It may even be the other way
around as the central bank would seek to increase inflation in times of deficient demand and vice
versa.^{footnote [13]}

Additionally, to underline the importance of expectations in the inflation process, much of the apparent flattening in more recent samples can be attributed to a better anchoring in inflation expectations (Hazell et al., 2022). But it is probably best if Emi Nakamura explained this all herself tomorrow.

Arguably, this anchoring is what differentiates the 1970s and 80s from the rest of the data in
**Chart 3**: Back then, shifting expectations made it impossible to achieve low
inflation without generating a high degree of economic slack. Policymakers were too confident that
inflation would eventually return back to target, for example due to the endogenous demand
destruction caused by high oil prices. This miscalculation allowed inflation expectations to ratchet
up to high levels. The required economic slack can be seen by the long and painful path from the top
left to the bottom right in the chart. Between 1980 and 1985, inflation in the UK fell by about 10
percentage points but at the cost of a more-than doubling in the unemployment rate from 5 to nearly
12%.

This is of course put millions of people out of a job with millions of livelihoods ruined – a
prospect we would very much like to avoid. There is another way: When inflation expectations are
time-varying monetary policy can affect them directly, shifting the curve back towards the
fundamental equilibrium π^{*} and y* via both movements along the curve but also shifting
it.

What data help us to evaluate how successful we are in keeping inflation expectations consistent with the fundamental equilibrium? I will argue first: Long-term inflation expectations remain relatively well anchored and the MPC has credibility to be able to fulfil its remit. This is unequivocally a good thing but survey measures do not tell us how we actually get to the objective of 2% – through aggregate demand compression or other ways. Second: Short-term inflation expectations are worryingly elevated but mostly reflect past inflation and are likely not a good guide for policy choices or outcomes in real time. Third, of significant concern: Medium-term measures of inflation expectations have drifted up alongside realized inflation, albeit not by as much. This may indicate a worrying shift in trend inflation, e.g. a shift in the Phillips Curve. A drift in medium-term inflation expectations is the development that monetary policy needs to firmly lean against and it should be a key yardstick for whether the MPC’s decisions are effective.

Going to the data on expectations. In general, we can partition different measures of inflation expectations as derived from surveys or financial market pricing in accordance with the distance to some shock that moves inflation off the target. In the very short run, roughly up to the 1-year horizon, they mostly reflect the direct impact of the shock on the economy. As people (and policymakers) observe macroeconomic outcomes in real time, they learn about the nature of the shock and form expectations about how it may play out in the immediate future.

Short-term expectations are, therefore, quite correlated with recent inflation outcomes as can be
seen, for example, from price expectations of firms in the Decision Maker Panel. As **Chart
4** shows, expectations of price changes over the next 12 months have increased by
similar amounts as price changes over the past 12 months. This is certainly not good news and points
to a more protracted inflation “hump” than would be implied by one-off shocks to the price level
(Mann, 2022a) but it is also could be consistent with a relatively swift return to target
thereafter, albeit tempered by downward stickiness.

A more nuanced view on inflation expectations comes from looking at expectations for the year after
next; that is price changes between 12 and 24 months ahead (what in financial markets would be
called 1-year-1-year inflation). At this would be a horizon, we would plausibly expect inflation to
reflect policy choices. **Chart 5a** shows the time series of expectations from the
Bank’s Inflation Attitudes Survey as well as financial market pricing of inflation at that 2-year
mark.

### Chart 4: Firms’ own-price outcomes and short-term expectations

Year-on-year percentage changes

#### Footnotes

- Source: Decision Maker Panel and Bunn et al. (2022). Latest observation: August 2022.

### Chart 5: Intermediate-term inflation expectations and
financial market pricingYear-on-year percentage changes ^{(a)} and share of respondents
^{(b)}

#### Footnotes

- Source: Bloomberg and Bank of England/Ipsos Inflation Attitudes Survey.
- Notes: Red line on the left-hand side shows monthly averages of financial market pricing for inflation extracted from inflation-linked swaps. Yellow line shows the median expectation for inflation 2 years ahead from the household survey, the right-hand side chart snapshots of its distribution. Latest observation: August 2022 for financial market pricing, May 2022 for survey expectations (August data for the BoE/Ipsos survey will become available on 9 September 2022).

The two series’ levels are not directly comparable since the financial market instruments behind the red line are based on RPI and likely incorporate risk premia, i.e. they are not a clean measure of the underlying inflation expectation. Nonetheless, comparing each series to its recent history is still instructive. Both measures are somewhat elevated currently, but while household expectations have been on the rise for four quarters in a row now, financial market measures have come off their peak, albeit still holding well above historical average levels. What might be the reason for this difference in evolution of expectations 2 years out? Financial markets, being more forward looking, anticipate a worsening macro environment resulting from the massive energy price squeeze and also the tightening of financial conditions that they, to a degree, are responsible for.

While household price expectations do not reflect these macroeconomic prospects, we can observe a
worrying increase in the risk of sustained inflation well in excess of target. Compare the red bars
on the right-hand side of **Chart 5** with the blue, especially in the
larger-than-five-percent bracket. In the latest survey round, more than a fifth of all respondents
said that inflation in two years’ time would be higher than 5 percent, and another fifth see
inflation higher 3%.

Moving now to the solidly medium term, roughly expectations of inflation between 3 and 5 years ahead. At that horizon, direct effects of shocks ought to have already played out. Therefore, expectations at that point mainly reflect possible second- and third-round effects as well as, importantly, the effects of current and anticipated policy choices. We need to pay significant attention to these measures since they can tell us something about the adequacy of our policy strategy.

Of concern is that the measures that we have for this horizon have been drifting up and have remained
at levels inconsistent with the target. For example, the DMP introduced a new question in May of
this year directly asking for firms’ expectations of aggregate consumer price inflation.^{footnote [14]} While their backward-looking
perception of consumer price inflation has generally been in line with measured CPI on average (e.g.
9.6 percent for July 2022 over July 2021 in the survey compared with 10.1 percent in the official
data), their 3-year ahead expectation has held firmly at 4.2 percent, despite prospects for a
significant slowdown in economic activity.

**Chart 6** shows the distribution of responses to those questions. Backward-looking
perceptions on the left-hand side have, as expected, moved rightward with actual inflation. However,
the distribution of 3-year ahead inflation on the right is nearly indistinguishable from what it was
in May.

### Chart 6: DMP inflation perceptions ^{(a)} and
medium-term expectations^{ (b)}

Density

#### Footnotes

- Source: Decision Maker Panel and Bank calculations.
- Notes: The charts show kernel density estimates of the distribution of responses about CPI inflation over the past year (on the left-hand side) and about CPI inflation 3 years ahead (on the right-hand side).

Recall that those months saw a marked deterioration in consumer sentiment indicators, talk of an imminent recession in much of the Western world, and robust monetary tightening by many central banks. Typically, we would expect such an outlook to dampen measured price inflation which should be reflected in the DMP. However, these measures show no such move, neither in the central tendency nor in the tail. If anything, the right tail has even fattened in August – a worrying sign of embeddedness beyond the short term. Further, although we do not have a long time series of the question of 3-year ahead CPI expectations, that there is no apparent sensitivity to expected macroeconomic conditions should give us pause.

The observation of a firming drift in medium-term inflation expectations is consistent with Bank
staff research of underlying trend inflation. One such piece of analysis, the Underlying Inflation
Measure of Lam, Potjagailo, and Wanengkirtyo (2022), uses a dynamic factor model of item-level CPI
data to extract a common factor of broad-based inflation. In **Chart 7** I plot this
measure alongside actual CPI inflation and the volatility-based inflation measure from my speech at
the start of the year (Mann, 2022a).

### Chart 7: CPI inflation and measures of trend inflation

Year-on-year percentage changes

#### Footnotes

- Source: Office for National Statistics, Lam, Potjagailo & Wanengkirtyo (2022), and Bank calculations.
- Notes: The volatility-based measure is the average of inflation rates of the least volatile 20% of CPI components. See Mann (2022a) and Mann & Brandt (2022) for more details. The underlying inflation measure (UIM) is based on a dynamic factor model that statistically captures broad-based joint co-movement across 438 CPI item series. For more details on its construction, see Lam, Potjagailo & Wanengkirtyo (2022, forthcoming). Latest observation: July 2022.

Both series correlate quite well, indicating that they are measuring the same object: the underlying rate of inflation common to the entire basket. Over the past, this rate has been quite persistent so it could plausibly remain elevated even as headline inflation comes down.

I am worried about the drift in this component since, theoretically, this would be the eventual resting point of inflation once shocks have washed out. As I explained in that earlier speech, the ‘underlying’ rate does not have to equal 2 percent to be consistent with achieving the inflation target. But, if not, there need to be other persistent influences that keep headline inflation on target. Before the GFC, those might have been intensifying globalisation and falling goods prices, the question is what will it be now?

A tighter monetary policy stance, on average, would be one such factor, at least for as long as medium-term inflation expectations and measures of trend inflation are elevated. The financial markets’ evolving yield curve is one metric of how much they think the MPC will have to tighten.

### Chart 8: Long-term inflation expectations and financial market pricing

Year-on-year percentage changes ^{(a)} and share of
respondents ^{(b)}

#### Footnotes

- Source: Bloomberg and Bank of England/Ipsos Inflation Attitudes Survey.
- Notes: Blue line on the left-hand side shows monthly averages of financial market pricing for inflation extracted from inflation-linked gilts. Yellow line shows the median expectation for long-term inflation from the household survey, the right-hand side chart snapshots of its distribution. Latest observation: August 2022 for financial market pricing, May 2022 for survey expectations (August data for the BoE/Ipsos survey will become available on 9 September 2022).

What about long-term inflation expectations? **Chart 8** shows these measures for
average inflation starting five years ahead. The news is mixed. On the one hand, these measures are
in line with their history, consistent with inflation being anchored, and consistent with
credibility of the target.

However, the survey measure on the left-hand side masks underlying movements that warrant attention,
particularly for households. Households matter because their buying habits either will discipline
firms’ prices or will allow the me-too-ism that underlies aggregate inflation. For example, since
mid-2020 we have seen a steady increase in respondents that say inflation in the long run will be 5%
or more on average and a steady decrease in those that expect below-target inflation. Therefore, we
need to be vigilant that these long-term expectations do not keep moving higher. Looking over the
time-series, aggregate long-term expectations remain below where they were in 2019 and in 2014. So
despite the increase in respondents in the larger-than-five-percent bucket, as of now, unanchoring
does not appear in the median.^{footnote [15]}

In addition to households, evidence from the corporate sector (specifically professional
forecasters), corroborates this assessment: Longer-term expectations remain on target but there is
some movement in the medium term. **Chart 9** shows average expectations from the
Bank’s Survey of External Forecasters which usually do not exhibit large time-variation making the
uptick in the 2-year-ahead measure even more noteworthy. However, the 3-year ahead measure, which
could be interpreted as the most fundamentalist forecast available, shows no such movement. This is
consistent with the Consensus panel of professional forecasters: There also, long-term expectations
are consistent with achieving the inflation target in the long run while those at medium-term
horizons have drifted.

### Chart 9: External forecasters’ average expectations

Year-on-year percentage changes