Expectations and the Neutrality of Interest Rates

John H. Cochrane1

Abstract

Our central banks set interest rates, and do not even pretend to control money supplies. How do interest
rates affect inflation? We finally have a complete economic theory of inflation under interest rate targets
and unconstrained liquidity. Its long-run properties mirror those of monetary theory: Inflation can be stable
and determinate under interest rate targets, including a peg, analogous to a k-percent rule.
Uncomfortably, stability means that higher interest rates eventually raise inflation, just as higher money
growth eventually raises inflation. Sticky prices generate some short-run non-neutrality: Higher nominal
interest rates can raise real rates and lower output. A model in which higher nominal interest rates tem-
porarily lower inflation, without a change in fiscal policy, is a harder task. I exhibit one such model, but it
paints a more limited picture than standard beliefs. Generically, without a change in fiscal policy, monetary
policy can only move inflation from one time to another.
The last decade has provided a near-ideal set of natural experiments to distinguish the principal theories
of inflation. Inflation did not show spirals or indeterminacies at the long zero bound. The large monetary-
fiscal expansion of the covid era produced a temporary spurt of inflation. The same money unleashed in
quantitative easing had no such effect.

1. Introduction

A half-century ago, Robert Lucas (1972a) published the watershed “Expectations and the Neutrality
of Money.” Lucas studied expectations and the neutrality—and temporary non-neutrality—of, as the title
says, money. His paper was the foundation of modern intertemporal monetary economics.
But our central banks set interest rates. The Federal Reserve does not even pretend to control money
supply, especially inside money. There are no reserve requirements. The quantity of M2 is whatever people
feel like holding in that form. Super-abundant reserves pay the same or more interest as short-term treasuries
and overnight money markets. The Fed controls interest rates by changing the interest it offers on abundant
reserves, not by rationing scarce zero-interest reserves. Other central banks follow similar policies. With
passive money supply, monetarist theory cannot determine the price level. It is a fine theory, but inconsistent
with current institutions.
We need an analogous theory of inflation under interest rate targets. Ideally, the theory should also
be based on robust and clean economics, starting with a well-described frictionless and neutral benchmark,
and adding minimal frictions to describe temporary non-neutrality. The theory should be consistent with
current institutions, including an interest rate target, ample reserves that pay market interest, no control
of inside money, and a plausible description of how central banks operate. Its basic mechanisms and signs
should be explainable to undergraduates, central bankers, and intelligent laypeople.
I offer a short stylized history of efforts to develop such a theory. I argue that by integrating interest
rate targets, rational expectations, and fiscal theory we have at last such a theory. In this theory, inflation

1 Hoover Institution, Stanford University and NBER. john.cochrane@stanford.edu, https://www.johnhcochrane.com. This

paper stems from a talk given at the Foundations of Monetary Policy Conference celebrating 50 years since the publication of
Lucas (1972a) “Expectations and the Neutrality of Money,” Federal Reserve Bank of Minneapolis, September 2022. I thank
Ed Nelson, Greg Kaplan, Loukas Karabarbounis, Christopher Sims, and anonymous referees for helpful suggestions.

Preprint submitted to Elsevier May 9, 2024

is stable and determinate under an interest rate peg, analogous to monetarist analysis of a k percent money
growth rule. The theory starts with a frictionless and neutral benchmark, and it remains approximately
long-run neutral with sticky prices.
However, stability and long-run neutrality inexorably imply that higher nominal interest rates eventually
produce higher inflation, just as higher money growth eventually produces higher inflation in a monetarist
analysis. The final piece, a theory of temporary non-neutrality, the central contribution of Lucas (1972a) for
money-based theory, is unfinished — or, if it is finished, it is not commonly accepted that the models we have
are the best we can do. With sticky prices, higher interest rates lower output. But it is not settled whether
and how the central bank can temporarily lower inflation by raising interest rates, without a contemporary
fiscal tightening.
The fiscal caveat is important. A fiscal tightening can lower inflation, and so an interest rate rise paired
with a fiscal tightening can lower inflation. Many models assume that monetary policy induces such a fiscal
tightening via “passive” fiscal policy. But central banks have no power directly to tighten fiscal policy. And
pairing higher interest rates with a fiscal tightening leaves open just how much inflation reduction comes
from higher interest rates and how much is due to the induced fiscal tightening. Is monetary policy just a
carrot in front of the fiscal horse that pulls the inflation cart?
For this reason, I focus on the question, can higher interest rates lower inflation without a change in
fiscal surpluses? We can treat this focus as simply an interesting theoretical orthogonalization, to help us to
understand how monetary policy works. We can then combine the result with fiscal shocks to understand
events in which both fiscal and monetary policy move.
In a range of contemporary models, higher interest rates do not lower inflation unless they are accom-
panied by tighter fiscal policy. The best model we have in which higher interest rates lower inflation without
tighter fiscal policy includes long-term debt, and temporarily lowers inflation only by raising later inflation.
The mechanism is a form of unpleasant arithmetic, moving inflation over time, and transferring resources
from long to short-maturity bondholders. This effect is weaker than most economists and central bankers
believe of monetary policy, and it is based on a completely different mechanism than they profess.
We also lack robust empirical understanding of the effect of interest rate changes, not accompanied by
fiscal tightening, on inflation. Since fiscal policy authorities also respond to inflation, historical episodes likely
combined fiscal and monetary policy changes. Successful historical disinflations have typically combined fiscal
and monetary policy reforms. The hoped-for negative effect of interest rates acting alone may not be there.
That monetary and fiscal policy must always be analyzed together is a truism usually acknowledged
but frequently ignored. Monetarists long warned that the central bank must be insulated against pressure
to print money to cover fiscal deficits. Sargent and Wallace (1981) “unpleasant monetarist arithmetic”
crystallized this tension in modern intertemporal terms.
Even without such direct fiscal pressure, monetary policy, now interest-rate policy, has fiscal conse-
quences: Higher real interest rates raise interest costs on the debt, and unexpectedly lower inflation is a
fiscal transfer to nominal bondholders. If higher interest rates slow economic activity, they induce deficits.
Such fiscal-monetary interactions were often glossed over as second-order effects. But that is no longer
true, if it ever was. With 100% debt to GDP ratio, a one percentage point higher real interest rate means one
percentage point higher debt service costs and one percentage point of GDP larger deficit. A one percentage
point decline in the price level is a one percentage point of GDP windfall for nominal bondholders. Each
recession sparks trillions in additional deficits.
Assuming that fiscal policy will always tighten to pay these costs is dangerous, as the assumed fiscal
tightening may not happen. In the current environment of large debts and structural deficits, fiscal authorities
may not wish to or be able to raise surpluses. Such lack of fiscal support may explain the failure of some past
monetary tightenings. (For example, Sims (2024) argues that this effect explains the resurgence of inflation
after 1975.)
Stability, neutrality, and the effect of interest rates on inflation are also crucial for policy debates. Must
central banks substantially raise nominal rates above inflation, as the Taylor Rule recommends and as we

2
observed in the early 1980s, in order to control inflation? Or can inflation ease without such high nominal
interest rates, as stability implies, and as we observed in 2022-2023? If the central bank raises interest rates,
and fiscal policy does not or cannot tighten to pay higher interest costs on the debt, does the interest rate
rise still lower inflation? Does the zero bound represent a dangerous cliff, with spiraling deflation on the
other side, or a benign stable and quiet inflation?
That such important foundational questions are unsettled is great news for researchers. The 1970s were
a golden decade for macroeconomic research, as much as they were a miserable decade for the economy. The
2020s may well repeat both features.
I trace the evolution of an interest-rate-based theory of inflation via a stylized history. ISLM-AS/AD
old-Keynesian style modeling mixes interest rate targets with adaptive expectations. This view describes
common policy beliefs, and allows interest rates higher than inflation to reduce future inflation. It predicts
that inflation is unstable under an interest rate peg, leading to inflationary or deflationary “spirals.” The
central bank can stabilize inflation by moving interest rates more than one-for-one with lagged inflation.
This theory places irrational behavior and model-inconsistent beliefs as an unavoidable necessary ingredient
to understand monetary policy.
New-Keynesian theory uses rational expectations with an interest rate target. In this view, inflation is
stable at an interest rate peg, but indeterminate, with multiple equilibria, leading to undesirable volatility.
The central bank removes multiple equilibria by deliberately destabilizing the economy, threatening hyper-
inflation or deflation to select equilibria. Our central banks don’t do that, so this model also is inconsistent
with current institutions.
Adding fiscal theory of the price level to an interest rate target with rational expectations, we obtain a
fully economic model that is at least not glaringly inconsistent with current institutions. Inflation is stable
and determinate under an interest rate peg.
Leaving aside theoretical issues, the last few decades have provided nearly ideal natural experiments
for distinguishing the four great theories of inflation.
Monetarism and fiscal theory agree that a large money-financed deficit, which people do not expect
to be repaid, causes a transitory burst of inflation as we saw in 2021-2023. Monetarism predicts however,
that an open market operation, an exchange of debt for new reserves with no additional deficit, has the
same effect on inflation. The comparison of quantitative easing with the pandemic and post-pandemic fiscal
expansion is a nearly ideal experiment. QE was even bigger, but had no detectable effect on inflation.
The decade of the zero bound in the US and Europe, and three decades in Japan, offer a nearly ideal
experiment to distinguish old-Keynesian, new-Keynesian, and fiscal theory models of interest rate targets.
In the zero-bound era, there were no spirals or sunspot volatility. Inflation was if anything quieter than when
central banks actively moved interest rates. There was no spiral again in 2021-2022 when central banks kept
interest rates at zero despite large inflation. These are central robust stability and determinacy properties
of the models, in as clear experiments as we could ask for.
This paper is a synthetic review, a stylized history of thought, and a primer. The contemporary
monetary economics literature is full of models that add ingredients. My aim here is to simplify and to see
which ingredients are really necessary to the basic properties of monetary policy. Some of these points are
also made at longer length with more complex models in Cochrane (2023) and Cochrane (2022b).

2. Theories of Inflation Under Interest Rate Targets

What is the dynamic effect of interest rates—not money supply—on inflation? I use a very simple
standard model to think about this question, and the historical development of our answers to it,

xt = Et xt+1 − σ(it − πte ) (1)

πt = πte + κxt , (2)

3
where x = output gap, π = inflation, π e = expected inflation, and i = interest rate. Variables are all
deviations from steady state. The central bank sets the interest rate. I include policy rules such as a Taylor
rule below.
Equation (1) is the first-order condition for consumption, in an economy without capital so that output
equals consumption. Higher real interest rates encourage saving, so depress current consumption relative to
future consumption. Equation (2) is the Phillips curve. High output coincides with high inflation, relative
to expected inflation.
Lucas specified that expectations enter the Phillips curve as πte = Et−1 πt . Expected inflation has no
output effect, as Friedman (1968) famously explained. But unexpected inflation still causes output to rise.
Lucas paired that Phillips curve with, essentially, M V = P Y and constant V , which determines the price
level. Lucas already had in hand a theory of the price level, and needed only to extend that theory to
describe non-neutrality. We have to work to get to Lucas’s launch pad.
I hesitate to write down such a model without preferences, technology, market structure, definition of
equilibrium, and recursive statement. Lucas’ most important contribution may have been methodological, to
express monetary economics with a completely articulated general equilibrium model. But this is well-trod
ground and it is well known how to provide those foundations. See, for example, Woodford (2003) and Galı́
(2009).
I simplify further by dropping Et xt+1 on the right hand side of (1), leaving a simple statement that
higher real interest rates depress output,

xt = −σ(it − πte ). (3)

This simplification turns out not to make any difference for the points I make, and leaving it out allows me
to do everything with transparent algebra. Equation (1) iterates forward to
∞
X
e
xt = −σEt (it+j − πt+j ) + lim Et xt+T ,
T →∞
j=0

so when output reverts to its mean, both static and dynamic versions give lower output in response to a
higher real interest rate. The interpretation of σ changes.
Substituting output out of (2)-(3), we obtain a relationship between interest rates and inflation:

πt = (1 + σκ)πte − σκit . (4)

The dynamic response of inflation to interest rates now depends on how expectations are formed.

2.1. Expectations, Stability, Determinacy, and Neutrality, and Sign

Table 1 summarizes the steady forward march of expectations in the Phillips curve. Each equation is
simplified to be emblematic of an era. Actual Phillips curves also include disturbances.

Author Phillips curve Expectations

Phillips (1958) πt = π0 + κxt Absent
Dynamic empirical (1960s) πt = απt−1 + κxt , α < 1 Adaptive
Friedman (1968); ISLM AS/AD (1970s) πt = πt−1 + κxt Adaptive
Lucas (1972) πt = Et−1 πt + κxt Rational
Calvo (1983), Rotemberg (1982); NK (1990s) πt = βEt πt+1 + κxt Rational

Table 1: The forward march of expectations in the Phillips curve

Phillips didn’t have any expectations or other variables to shift the Phillips curve, nor did the Keynesian
advocates of inflation in the early 1960s such as Samuelson and Solow (1960). (See Nelson (2020) Ch. 13.)

4
 Dynamic estimates of the Phillips curve in the 1960s and 1970s added lags of unemployment or inflation,
depending which variable one put on the right-hand side of the regression. These specifications retained a
long-run inflation-output tradeoff, α < 1 here, and when thinking theoretically, adaptive expectations.2

2.1.1. Adaptive Expectations

Friedman’s (1968) address is fundamentally about neutrality. He proclaimed two things that monetary
policy cannot do: It cannot permanently lower unemployment and it cannot peg an interest rate. On the
first point, he proclaimed that the Phillips curve would shift once people come to expect inflation, so the
long-run Phillips curve is vertical.
On the way there, Friedman describes explicitly adaptive expectations: “This price expectation effect is
slow to develop and also slow to disappear.” Phelps (1967) also writes “a sort of Phillips Curve . . . that shifts
one-for-one with variations in the expected rate of inflation; . . . the expected inflation rate adjusts gradually
over time to the actual inflation rate.”
As a simple model of adaptive expectations, suppose expected inflation is just last period’s inflation,

πte = πt−1 .

Then from (4) inflation and interest rates are related by

πt = (1 + σκ)πt−1 − σκit . (5)

Steady states have higher inflation with higher interest rates, and no change in output, employment, or
real interest rate. In this sense the model displays long-run neutrality. The term (1 + σκ) in front of lagged
inflation is greater than one. Inflation is unstable. Thus, Friedman proclaimed a key doctrine: The central
bank cannot peg the nominal interest rate. If it tries to do so, the economy will spiral off to hyperinflation
or hyperdeflation. In Friedman’s description, the central bank needs to print more and more money to keep
the interest rate down. In the non-monetary ISLM AS/AD tradition, a too-low nominal interest rate lowers
the real rate, which boosts demand, which boosts inflation, and around we go. The left panel of Figure 1
illustrates instability and the inflation or deflation spirals that break out under an interest rate peg.

Figure 1: Adaptive (left) vs. Rational Expectations (right) Dynamics

Thus began the tradition in which an interest rate target is a fundamentally incomplete price-level
anchor. Friedman’ answer, of course, was that the Fed should control the money supply rather than interest
rates.
The last term of Equation (5) is negative: Higher interest rates lower inflation. Indeed, higher interest
rates set off an unstable deflationary spiral. Friedman said so (in the opposite, inflationary, direction),

2 Among many others, Lipsey (1960), Gordon (1970). Gordon (1976) p. 192–193 provides a nice summary of the era. Sargent
(1971) and Lucas (1972b) also summarize insightfully.

5
though quickly adding that the central bank would soon give up the peg and lower money growth to stop
the spiraling inflation.
The Taylor rule repairs Friedman’s critique of interest rate targets.3 Let the central bank systematically
respond to inflation with higher interest rates,

it = φπt + ut (6)

with φ > 1. Substituting for it in (5), inflation dynamics become

1 + σκ σκ
πt = πt−1 − ut . (7)
1 + σκφ 1 + σκφ
The central bank’s interest rate policies stabilize the economy. The coefficient on lagged inflation is less
than one. It appears the central bank can target interest rates after all. But the economy itself remains
fundamentally unstable. Instability only tamed if the central bank reacts swiftly and more than one for one
to inflation, like a seal balancing a ball on its nose. Where I discuss stability and determinacy at a peg as
an interesting conceptual experiment, the real dividing line is a less than one for one φ < 1 vs. greater than
one for one φ > 1 reaction.
Higher interest rates still lower inflation. The second term in (7) is negative, so a positive monetary
shock lowers inflation. The positive monetary policy shock raises interest rates, so we also see higher interest
rates with lower inflation. This isn’t obvious. In some models, a positive monetary shock lowers equilibrium
interest rates. In the period 1 of the shock we have π1 = −σκ/(1 + σκφ)u1 and i1 = 1/(1 + σκφ)u1 so
π1 = −σκi1 . The higher interest rate sets off a deflation spiral, but the interest rate soon chases inflation
downwards and stops the spiral.

2.1.2. Rational Expectations and New-Keynesian Models

New-Keynesian models use rational expectations,

πte = Et πt+1

and consciously play by the Lucas rules of how to write and solve intertemporal general equilibrium macroe-
conomic models. (The standard new-Keynesian model writes the Phillips curve πt = βEt πt+1 + κxt ; Calvo
(1983), Rotemberg (1982). I use β = 1 for simplicity. I show below that β < 1 makes no important difference
to my points.) Now from (4) the dynamic response of inflation to interest rates is
1 σκ
Et πt+1 = πt + it . (8)
1 + σκ 1 + σκ

This model starts to provide a fully economic model of inflation under nominal interest rate targets.
Inflation is still long-run neutral. Inflation is now stable since 1/(1 + σκ) < 1. Inflation eventually converges
to the nominal interest rate, just as higher money growth eventually leads to proportionally higher inflation.
This model also has a sensible frictionless κ → ∞ limit and limit point,

Et πt+1 = it . (9)

Higher interest rates lead to instantly higher inflation, as with flexible prices higher money growth instantly
leads to higher inflation. Price stickiness draws out dynamics and adds output effects. People are rational,
markets clear.
But this model so far only ties down expected inflation. There are multiple equilibria. Unexpected
inflation πt+1 − Et πt+1 can be anything, or wander up and down following sunspots. The right hand panel

3 McCallum (1981) is the first formal statement that the φ > 1 principle resolves indeterminacy with rational expectations.

Taylor (1999) shows how a Taylor rule resolves instability with adaptive expectations. Taylor (1993) is the most influential
statement of the rule and its practical implementation. Wicksell (1898), (1965) proposed verbally that a central bank could
stabilize the price level by raising and lowering an interest rate target. In the 19th century, the Bank of England systematically
raised and lowered the discount rate to defend the gold standard, in part in response to inflation and vice versa.

6
of Figure 1 illustrates, with the question mark indicating all the many equilibria that could break out at
that point. In a related model, Sargent and Wallace (1975) modified Friedman’s doctrine: Inflation is
indeterminate under an interest rate target. Central banks should target money.
Sargent and Wallace’s stable and indeterminate is different from Friedman’s unstable (and determinate).
Both models suggest volatile inflation, and a deep problem with interest rate targets. But spiraling away on
a determinate path is different from batting up and down unpredictably.
The sign on the last term in (8) has also changed. Now higher nominal interest rates raise expected
inflation, as also in the flexible price model (9), in both the short and long run. We will return to look for
additional frictions that can reverse this conclusion. Sticky prices alone are not sufficient.
One often thinks of rational vs. adaptive expectations as a small issue, and adaptive expectations as
a small friction that can easily deliver somewhat more realistic dynamics of a baseline model, icing on the
cake. That is not true here. Rational vs. adaptive expectations fundamentally change the stability and
determinacy properties of the model, and change the sign by which interest rates affect inflation. This is the
cake.
New-Keynesian modelers resolve indeterminacy with a novel application of the Taylor principle. If we
add it = φπt + ut in this case, inflation dynamics (8) become
1 + σκφ σκ
Et πt+1 = πt + ut . (10)
1 + σκ 1 + σκ
With φ > 1, dynamics are now unstable. Adding a rule against nominal explosions, new-Keynesian modelers
can now choose the unique initial value of inflation that precludes an explosion, and thus produce determinate
inflation; πt = 0 with no shocks, and
∞  j
σκ X 1 + σκ
πt = − Et ut+j
1 + σκφ j=0 1 + σκφ

with shocks ut .
The central bank is imagined to deliberately destabilize an economy which is already stable on its
own, exactly the opposite of the adaptive expectations economy in which the Taylor rule stabilizes an
otherwise unstable economy. The central bank threatens hyperinflation or hyperdeflation in order to select or
“coordinate expectations” on the equilibrium it likes. The Taylor principle is an equilibrium-selection policy
not a stabilization policy. Indeed, the central bank may simply announce its inflation target, announce this
threat, and inflation jumps to whatever value the central bank desires.
These statements are easiest to see in the κ = ∞ case of flexible prices, in which the interest rate sets
expected inflation directly via (9), it = Et πt+1 . They are also easiest to see if we rewrite the policy rule
equivalently as
it = φπt + ut = i∗t + φ(πt − πt∗ ) = Et πt+1
∗
+ φ(πt − πt∗ ). (11)
The second and third equalities define i∗t and translate between the ut and i∗t , πt∗ notation for the monetary
policy disturbance. (King (2000) invented this clever notation.) Eliminating it from (9)-(11), the model’s
equilibrium condition is
∗
Et (πt+1 − πt+1 ) = φ(πt − πt∗ ). (12)
With φ > 1, the unique bounded equilibrium is πt = πt∗ , it = i∗t .
In this notation, we may interpret {πt∗ } as the central bank’s stochastic inflation target, the value of
inflation it wishes to produce in each date and state. The central bank implements that target with an interest
rate policy i∗t , which sets the equilibrium interest rate which we observe, and a separate equilibrium-selection
policy φ(πt − πt∗ ), which threatens to produce an expected hyperinflation or deflation should unexpected
inflation come out against the bank’s desires. The interest rate policy determines expected inflation, the
equilibrium-selection policy determines unexpected inflation. For example, if πt∗ is i.i.d., observed interest
rates never move. The central bank simply announces each period what inflation it would like to see, and
that inflation occurs.

7
 Don’t we know from regressions that φ > 1, at least away from the zero bound? No. Since in equilibrium
πt = πt∗ and it = i∗t , we never observe the φ(πt − πt∗ ) reaction that generates determinacy. The last example
with constant interest rate and i.i.d. inflation gives a vivid illustration: The regression coefficient of it on
πt is zero in that case. Estimated Taylor rules measure a correlation between equilibrium interest rates and
inflation, not how off-equilibrium interest rates react to off-equilibrium inflation, which selects equilibria.
For example, suppose the central bank chooses an inflation target πt∗ with Et πt+1 ∗
= θπt∗ + ut . Then it
∗ ∗ ∗ ∗
would pick it = θπt + ut and again threaten it = it + φ(πt − πt ). The equilibrium condition is still
∗
Et (πt+1 − πt+1 ) = φ(πt − πt∗ ). We observe it = θπt + ut ; we measure θ not φ. The equilibrium-selection
parameter φ does not appear in observed, equilibrium quantities. (For a fuller treatment of these points, see
Cochrane (2011).)
What matters to equilibrium selection is that the central bank makes inflation rise more than one for
one with past inflation and hence explode for all but one value; that φ > 1 in (12). Whether interest rates
rise more than one for one with inflation is not directly important. In this model, interest rates that rise
more than one for one with inflation generate inflation that rises more than one for one with inflation because
higher interest rates raise inflation. If higher interest rates lowered inflation, following standard intuition, a
Taylor principle would no longer select equilibria. In larger models, interest rates that rise more than one
for one with inflation are often neither necessary nor sufficient for the requisite stable eigenvalue (less than
one) to become unstable.
In all these ways, regression evidence and historical episodes in which (by definition observed, equilib-
rium) interest rates rise more than one for one with inflation are not central evidence for the new-Keynesian
equilibrium-selection evidence.
The habit of writing it = φπt + vt and restricting vt to follow an AR(1) has led to a lot of confusion on
these points. That superficially plausible specification involves implicit identifying assumptions, in particular
that the off-equilibrium threat φ is the same as the on-equilibrium relation between interest rate and inflation
targets θ. But there is little reason for such an assumption. Historically, authors naturally started with
it = φπt + ut and AR(1) ut , and only much later discovered the i∗t , πt∗ notation that makes the implicit
identifying assumptions clear.
To verify the same points in our little sticky price model, again write the policy rule as

it = φπt + ut = i∗t + φ(πt − πt∗ ), (13)

where now
1 + σκ 1 ∗
i∗t ≡ ∗
Et πt+1 − π . (14)
σκ σκ t
∗
Equation (14) generalizes i∗t = Et πt+1 , applying the equilibrium condition (8) to the starred variables. If we
want to think in terms of an interest rate target and an inflation target, those targets must be compatible
with private sector equilibrium conditions. With this policy rule, the equilibrium condition (8) becomes

∗ 1 + φσκ
Et (πt+1 − πt+1 )= (πt − πt∗ ). (15)
1 + σκ
Again, with φ > 1 the central bank destabilizes an economy that is stable on its own. By doing so, it repairs
indeterminacy: πt = πt∗ and it = i∗t is the unique locally bounded equilibrium.
New-Keynesian modelers can produce lower inflation when interest rates rise, overcoming the positive
signs in (8), (9), and (10), by such equilibrium-selection choices. Suppose that coincident with an unexpected
rise in interest rates at time t, the central bank also announces a lower stochastic inflation target πt∗ . Now
current inflation πt jumps down. Translate back to {ut } and you have the conventional new-Keynesian
model. However, as you can see by this rewriting, the higher interest rate has nothing to do with the lower
initial inflation. The central bank could just as easily lower initial inflation without a higher interest rate
by equilibrium-selection policy alone. Indeed, the higher interest rate, by raising future inflation, requires a
larger downward equilibrium-selection jump for the model to produce a protracted disinflation.
The trouble with this approach is that central banks do not have equilibrium-selection policies. They
do not threaten hyperinflation or deflation if inflation comes out against their desires. Such threats being

8
contrary to their objectives, nobody would believe them if they tried. Central banks do not intentionally
de-stabilize economies that are stable on their own. Ask central banks. Look at central bank websites. They
loudly announce that they stabilize economies; no matter what inflation does, they promise to act resolutely
to bring it back.
As I rejected the beautiful M V = P Y because central banks set interest rates and do not limit money
supply, I argue that we should also reject the new-Keynesian equilibrium-selection approach, because our
monetary institutions simply do not remotely behave as this model specifies. The new-Keynesian theory
also requires an extra rule against non locally bounded equilibria, not a part of the standard definition of
Walrasian equilibrium. And yet we see hyperinflations. Again, empirical performance rather than theoretical
beauty is the most important test, and the quiet (low volatility) of inflation in the zero bound era is also
vital evidence.
For decades, economists worked to develop a theory of inflation under interest rate targets. So far in
my story, they did not succeed. But central banks target interest rates, inflation is something, and indeed
was rather successfully controlled by the mid 1990s. How?

2.1.3. Fiscal Theory with Rational Expectations

The fiscal theory of the price level adds an equilibrium condition, or rather recognizes one that was
there all along and has been left out of the analysis so far. The log real value of nominal government debt
vt evolves as
ρvt+1 = vt + it − πt+1 − s̃t+1 (16)
where ρ = e−r < 1 is the steady state discount factor and s̃t+1 is the real primary surplus scaled by the
steady state value of debt. Higher real interest rates raise debt, higher primary surpluses lower debt. I use
here the simple case of one-period debt and no economic growth; I generalize to long-term debt below. This
equation is also linearized. See Cochrane (2023) Ch. 3.5 for a derivation. The consumer’s transversality
condition also requires
lim Et ρT vT = 0. (17)
T →∞

We can add these conditions to the VAR(1) statement of the model. But in this simple case, we can
solve the model analytically by iterating (16) forward to
∞
X
vt = E t ρj [s̃t+1+j − (it+j − πt+1+j )]. (18)
j=0

The real value of debt is the discounted present value of future surpluses. Since (18) holds with Et+1 as well,
so does (18), and we can take innovations ∆Et+1 ≡ Et+1 − Et of both sides of (18), to obtain.
∞
X ∞
X
∆Et+1 πt+1 = −∆Et+1 ρj s̃t+1+j + ∆Et+1 ρj (it+j − πt+j+1 ). (19)
j=0 j=1

Unexpected inflation devalues outstanding debt. Thus, unexpected inflation corresponds to the revision in
the present value of future primary surpluses. In the second term, higher discount rates lower the value of
debt and cause inflation. Equivalently, if interest costs on the debt rise, but current or future surpluses do
not rise to pay them, then the resources must come by inflating away outstanding bonds.
The rational expectations model left an indeterminacy indexed by unexpected inflation ∆Et+1 πt+1 .
Condition (19) restores that determinacy, in place of central bank equilibrium-selection rules. This point is
also easiest to see in the simplest case of flexible prices. Then, (8) and (19) boil down to

it = Et πt+1 (20)
∞
X
∆Et+1 πt+1 = −∆Et+1 ρj s̃t+1+j . (21)
j=0

9
The interest rate target sets expected inflation but leaves multiple equilibria. Fiscal policy determines
unexpected inflation and picks one equilibrium.
With sticky prices, inflation dynamics follow (8), which I repeat for convenience,
1 σκ
Et πt+1 = πt + it , (22)
1 + σκ 1 + σκ
along with (19). Equation (22) describes a set of stable paths for expected inflation, and fiscal policy in (19)
selects one of the many possible equilibria. It’s straightforward to add policy rules in which surpluses and
interest rates respond to inflation and output, and all the other complications of DSGE models.
In sum, with (19) to choose unexpected inflation, inflation is stable and determinate at an interest rate
peg, or more generally Taylor coefficients φ < 1, overcoming Sargent and Wallace’s contrary doctrine. The
peg can include zero. The economy can follow the Friedman (1969) optimal quantity of money rule and leave
us satiated in liquidity, without worry over spirals and sunspots. (Leeper (1991) is the watershed first article
that described fiscal theory with interest rate targets. Woodford (1994), Woodford (1995) showed that an
interest rate peg is determinate and stable with fiscal theory, including a peg.)
The volatility of inflation depends on the size of fiscal and discount rate shocks. But economics picks
one value for inflation. In addition, we now have a model of the price level, a nominal anchor, rather than
simply a model of the inflation rate with a remaining indeterminacy about the price level. In (18), vt is the
ratio of nominal debt to the price level. The price level adjusts so that (18) holds.
Monetarists, ISLM-AS/AD or adaptive-expectations Keynesians, Sargent and Wallace, and new Key-
nesians make explicit assumptions to wipe out fiscal theory. Careful versions of these theories include the
government debt equilibrium condition, but they assume that fiscal authorities “passively” adjust surpluses
as needed to validate inflation determined elsewhere. The central bank picks ∆Et+1 πt+1 , and then fiscal
authorities supply whatever surpluses s̃t+1 are necessary to satisfy (18) or (19), often via lump-sum taxes.
(Leeper (1991), Woodford (2003) section 4.4.)
Since the equilibrium conditions are the same, the fiscal and new-Keynesian ways of completing the
sticky price model make predictions for equilibrium time series that are formally observationally equivalent,
at least without further assumptions. (See Cochrane (1998) and Cochrane (2023) Ch. 17, 22.) Thus, for the
purposes of everything that follows, you may think in terms of the new-Keynesian rather than fiscal-theory
version of equilibrium formation, if you disagree with my argument.
However, new-Keynesian modelers typically do not examine what the required surpluses are. For a full
evaluation of the theory, we must examine them. Observational equivalence includes model’s predictions for
“passive” fiscal policies. For example, in the standard new-Keynesian policy experiment, a monetary policy
shock that lowers inflation comes with a “passive” fiscal tightening to pay higher interest costs on the debt
and a windfall to bondholders. Fiscal theory calls this rise a “fiscal shock” coincident with the interest-rate
increase. New-Keynesian theory calls it an induced fiscal policy response to the equilibrium-selection part
of the central bank’s monetary policy. By either name, however, surpluses must rise, or inflation cannot
fall. Standard new-Keynesian analysis doesn’t look. So, you can interpret my equations as exploring fiscal
underpinnings of conventional new-Keynesian models.
We should also not overstate the consequences of observational equivalence. Formal tests on equilibrium
time series are not the only way to distinguish theories. For example, in the fiscal theory model, a fiscal
shock (a decline in the present value of surpluses) results in inflation to devalue debt that the central bank
cannot completely avoid by any path of interest rates. In the new-Keynesian model, the central bank fully
controls inflation, expected and unexpected. Fiscal shocks are ruled out: If the central bank does not
choose inflation, fiscal authorities repay all deficits in full. Whether there are inflationary fiscal shocks,
and whether the central bank alone can or cannot fully control inflation are pretty important policy and
doctrinal issues, which historical analysis can illuminate. Did we suffer inflation in 2021-2022 because of a
fiscal shock, or because central banks failed to credibly commit to appropriate equilibrium-selection threat to
support a rock-steady 2% inflation target, inducing hyperinflation for anything greater? Knowledge of how
our institutions work, evidence in their words rather than equilibrium time series that central banks simply

10
do not make explosive equilibrium-selection threats, is information that can help to distinguish theories.
And history speaks on the central stability and determinacy issues, for example stability at the zero bound.
While observational equivalence tells us that there is a contrary story, Occam’s razor also applies.
Fiscal theory does not have to be married to with rational expectations. I investigate fiscal foundations
of adaptive expectations models below. If the rational expectations Calvo pricing structure turns out to be
a dead end, fiscal theory can apply to any macroeconomic theory in which government debt is not a free
lunch.

2.2. Lucas’s Phillips Curve

In my little Phillips curve history (Table 1), I skipped over Lucas. Lucas (1972a) first made Phillips-
curve expectations rational. His Phillips curve relates output to unexpected inflation only, first moving
forward the time subscript in the Phillips curve, from πte = πt−1 to πte = Et−1 πt . In the spirit of rational
expectations, it makes most sense to pair Lucas’ Phillips curve with rational expectations in the bond market
and consumption. So let’s use Lucas’ Phillips curve in an interest-rate model by writing

xt = −σ(it − Et πt+1 ) (23)

πt = Et−1 πt + κxt . (24)

Eliminating xt , inflation dynamics (4) are now

1
Et πt+1 = it + (πt − Et−1 πt ). (25)
σκ
Further in the future,
Et πt+j+1 = Et it+j ; j ≥ 1.
Lucas’ specification of the rational expectations Phillips curve, along with our IS curve, passive fiscal policy,
and an interest rate target, leads to an economy that is stable and indeterminate, like the new-Keynesian
model. Relative to the flexible-price model it = Et πt+1 , Lucas’s Phillips curve gives one period of additional
inflation after a shock, which then reverts to the frictionless value. It gives a one-period output effect.
Adding fiscal theory to this model we again restore determinacy, and name the unexpected inflation shock.
One could alternatively add new-Keynesian equilibrium selection and policy rules.

2.3. Anchored Expectations

The 1970s adaptive expectations model as I have presented it doesn’t fully capture the views of today’s
policy consensus, as reflected for example in central bank analysis and communications. We can capture
those views a bit better by writing expectations as
∗

e
πt = P∞ π if anchored
j=0 aj π t−j if not

along with the model (2)-(3),

xt = −σ(it − πte ) + σux,t (26)

πt = πte + κxt + uπ,t . (27)

I add here IS (“demand”) and Phillips (“supply,” “cost”) shocks, and use u rather than ε to emphasize
that such shocks are persistent. I multiply the demand shock by σ so it has rate of return units. Inflation
dynamics are then
πt = (1 + σκ)πte − σκit + σκux,t + uπ,t . (28)

Inflation expectations underlying consumption-saving and price-setting behavior are usually “anchored”
at the bank’s inflation target, π ∗ = 2%. Those expectations remain anchored even if a rational or rational-
expectations (using the model) forecast of inflation temporarily deviates from 2%, for example in response
to a u shock. The anchor comes from a belief that eventually the central bank will do something to control

11
inflation, even if inflation breaks out in the meantime. That belief is malleable by speeches, policy “strate-
gies,” forward guidance, and other bank communication. If the belief is broken, then expectations become
“unanchored,” but remain slowly adaptive rather than forward-looking and model-consistent or “rational.”
Something like a Markov switching process with probabilities that depend on observed inflation, interest
rates, and bank communication describes a shift from anchored to unanchored.
In (28), while expectations remain “anchored” with π e = π ∗ , a shock to inflation eventually goes away
on its own without aggressive bank action. Central banks are still important. Higher interest rates can lower
inflation and thereby partially offset u shocks elsewhere in the economy. However, such higher rates lower
output, see (26) with π e = π ∗ . While expectations are anchored, central banks face a basically static Phillips
policy tradeoff between inflation and employment.
Inflation is stable and determinate as long as expectations remain anchored. We might call it “condi-
tionally stable.” Unlike the inflation spiral of the adaptive expectations model, a slow reaction to inflation
such as 2021-2022, while harmful, does not threaten to spiral out of control and indeed will slowly go away
on its own. The damage of delay, while perhaps sorting out just what shocks did hit the economy and what
the data mean, is limited.
But as inflation rises, perhaps due to u shocks, with no central bank action, the chance of de-anchoring
rises, and then an inflation spiral breaks out. So raising rates eventually is doubly, and perhaps mostly,
important via a signaling channel to reanchor expectations. Indeed, if expectations have become a bit
unanchored, πte > π ∗ , then a rate rise that signals a firmer commitment to doing whatever it takes can
quickly lower inflation πt by snapping back expectations, with no Phillips curve output effect, and no period
of interest rates above current and past inflation as required by the adaptive expectations model. Thus, the
story goes, the beginning of rate rises in 2022 brought inflation down quickly and with no recession, even
though interest rates stayed below inflation for a year.
Once expectations become unanchored, the policy model reverts to adaptive expectations, but with
expectations that adjust slowly to experience, say
∞
X
πte = (1 − a) aj πt−1−j
j=0

with a large value of a. Now inflation dynamics, suppressing the u disturbances for simplicity, are
∞
X
πt = (1 + σκ)(1 − a) aj πt−1−j − σκit . (29)
j=0

We can express this result as4

πt = [1 + σκ(1 − a)]πt−1 − σκ(it − ait−1 ).

For 0 ≤ a < 1, inflation is still unstable, 1 + σκ(1 − a) > 1. But more slowly adapting expectations, larger a,
mean a slower explosion. Slowly adaptive expectations also mean that it takes the central bank longer and
with a deeper recession to bring inflation under control.
This scenario answers the nagging question of “anchored” expectations, “anchored by what?” The
answer is, the central bank will repeat the adaptive-expectations story of 1980, lowering inflation via the
−σκit term at the cost of severe recession.

4
 
1−a
1 − (1 + σκ) L πt = −σκit
1 − aL
 
1 − [1 + σκ(1 − a)]L
πt = −σκit
1 − aL

12
 The policy model has some features of ancient fixed-expectations analysis, while expectations are “an-
chored.” But it also takes some of the lessons of the rational-expectations revolution, that expectations are
not fixed functions of past experience, and depend on policy “regimes.” It takes expectations as something
of a free variable, malleable by announcements, but also dependent on the Fed’s reputation, and ability, to
eventually do what it takes to constrain inflation by higher interest rates. But expectations are still not
model-consistent; the expectations of the model are not the expectations in the model. Perhaps most deeply,
in both adaptive and policy models, expectations are not reactive. In the rational expectations model,
expectations react, immediately, to a change in interest rate or other shock. Here, they mostly stay put.
Most policy discussion also sees “long and variable lags” in the response of the real economy to the
interest rate (26), which also induces such lags in inflation dynamics (28). Therefore, higher interest rates
do not lower current inflation as in my (28), but rather slowly lower future inflation. But our purpose is to
understand basic properties with simple economic models, not to match dynamic beliefs with ad-hoc lags,
so I move on.

3. Long-Run Properties

Combining fiscal theory with rational expectations, then, we have, finally, a complete economic (rational
agents, Walrasian equilibrium) theory of inflation under interest rate targets, compatible with what we know
about how central banks behave. In this theory, and like monetarist theory, the economy is determinate,
stable, starts from a neutral flexible-price benchmark, displays (potentially approximate) long-run neutrality,
and offers a foundation for models with frictions that create additional short-run non-neutralities.
These properties are on full display with flexible prices,

it = Et πt+1
∞
X
∆Et+1 πt+1 = −∆Et+1 ρj s̃t+1+j .
j=0

There is only one equilibrium value of inflation. The central bank can follow a peg, analogous to a k% rule,
and inflation will neither spiral away nor suffer multiple-equilibrium volatility. (More generally, it may follow
a Taylor rule that reacts less than one for one to inflation, φ < 1.) Like a k% rule, a peg may not be optimal,
but it is possible. A one percentage point higher interest rate with no change in subsequent fiscal policy
produces a one percentage point higher inflation, with no change in real rate or output. That’s neutrality.
The sticky-price version of the model (22)-(19) also offers a first step of explicitly modeled short-run non-
neutrality, keeping the frictionless model’s properties in the long run, again similarly to standard monetary
theory. The model remains stable and determinate: Inflation slowly converges to the interest rate. Output
and real interest rates vary, but that variation dies away in the long run. A one percentage point higher
interest rate leads eventually to one percentage point higher inflation, but not right away.
Higher interest rates raise inflation? Even if qualified by “in the long run,” this “Fisherian” property
is so contrary to standard doctrine that one looks up in disbelief from the screaming implications of the
equations. Really?
Stability is the central property behind the Fisherian result. If inflation is stable under an interest
rate peg, it follows that raising the peg must eventually raise inflation. The adaptive expectations model is
long-run neutral, in that steady states with higher inflation have higher interest rates. But it is unstable, so
raising the interest rate from such a steady state lowers inflation.
How do higher interest rates produce higher inflation? The intuition is not as easy as simply printing
up money and handing it out. What about standard intuition that higher interest rates depress aggregate
demand, so should reduce inflation?
First, consider the full consumer first-order condition xt = Et xt+1 − σ(it − Et πt+1 ), with no pricing
frictions. Raise the nominal interest rate it . Before prices change, a higher nominal interest rate is a higher
real rate, and induces people to demand less today xt and more next period xt+1 . That change in demand

13
pushes down the price level today pt and hence current inflation πt = pt −pt−1 , and it pushes up the expected
price level next period pt+1 and thus expected future inflation πt+1 = pt+1 − pt . So, standard intuition is
correct, and refers to a force that can lower current πt inflation as it lowers the current price level relative
to the future. Fisherian intuition is correct too, and refers to a natural force that can raise expected future
inflation Et πt+1 , as it raises the future price level relative to the current one.
But which is it, lower pt or higher pt+1 ? This consumer first-order condition, capturing an “intertem-
poral substitution” effect, or telling us the slope of the price level over time, cannot tell us. Unexpected
inflation and the overall price level is determined by a “wealth effect,” by the overall level of prices, by the
“nominal anchor.” Handing money or government debt out for free with no change in fiscal policy generates
such a wealth effect. Hence, if we pair the higher interest rate with no change in surpluses and thus no
wealth effect, whether “actively” or “passively” produced, then the initial price level pt does not change and
the entire effect of higher interest rates is a rise in pt+1 . If instead there is a positive “wealth effect,” a
concurrent rise in the present value of expected surpluses, then we have a lower price level pt and less current
inflation πt . Intuition that higher interest rates lower inflation implicitly assumes that fiscal policy tightens
in concert with monetary policy.
Second, consider the Phillips curve. Suppose a higher nominal rate means a higher real rate in the IS
curve and pushes output down, as it does in classic static ISLM thinking and in my simplified model. Lower
output xt in the Phillips curve,
πt = πte + κxt
means lower inflation πt , yes, but relative to expected inflation. With adaptive expectations, πte = πt−1 ,
lower inflation πt means that inflation decreases over time. But with rational expectations, πte = Et πt+1 ,
lower inflation means that inflation increases over time. So we might again be talking to cross-purposes,
confusing current inflation that is lower relative to expected inflation with inflation that decreases over time.
And lower inflation relative to expected inflation may not mean lower inflation, as expected inflation may
rise so much that both current and expected inflation are higher than otherwise, as happens in this case.
Third, verbal intuition may also confuse adjustment to equilibrium with the movement of inflation over
time in equilibrium. Hold fixed expected inflation, i.e. assume it is “anchored.” Then higher real interest
rates and lower output lower inflation πt . But that means lower inflation today, right away, than otherwise
would have occurred. It does not mean that higher interest rates today provoke lower output in the near
future, and slowly drive inflation down in the further future, as most policy discussion describes. These
equations do not have “long and variable lags.” So part of the contrast with standard intuition may come
from confusing the process of adjustment to equilibrium right now (t) with the evolution of equilibrium over
time.
(Since Ball (1994), the fact that the standard new-Keynesian Phillips curve relates high output to
decreasing inflation has been a persistent concern, attracting a wide variety of additional ingredients or a
return to adaptive expectations. The point here is not to pursue this issue, but to understand the intuition
of the new-Keynesian result.)
Stability means that higher interest rates must eventually lead to higher inflation. Stability is a natural
and robust consequence of forward-looking or rational expectations, while instability is a natural consequence
of backward-looking expectations. If you drive a car looking through the rear view mirror, you will veer off
the road. If you drive looking forward the car will stably follow the road.
If intuition still rebels, keep in mind that stability and a Fisherian response may take a long time to take
hold, and keep in mind the experiment: permanently higher interest rates with no change in fiscal policy.
Central bankers seldom abstain from moving interest rates for very long, monetary and fiscal policy both
respond to events, monetary policy induces fiscal responses, and lies in the shadow of fiscal events which
push inflation around. We do not obviously have a lot of experience that bears on the Fisherian proposition.
Exchange rate pegs and purchasing power parity are a good analogy. If a country wishes a lower
nominal exchange rate, a less valuable currency, it can do so by pegging at that rate and waiting long
enough. Any peg also requires fiscal backing, the ability to get as much foreign exchange or the will to
print as much domestic currency or debt as needed to enforce the peg. Pegs fail when governments give

14
up on these fiscal requirements. One may have to wait a very long time for purchasing power parity and a
long slow inflation to kick in. Nonetheless, the proposition that an exchange rate peg is eventually neutral,
that relative price levels eventually adjust, if a government can stick to that peg long enough, is intuitively
clear, as are the fiscal, foreign exchange, and time-consistency limitations of that proposition for policy and
historical analysis. The proposition that higher interest rates lead eventually to higher inflation is analogous.
One pegs a cross-sectional nominal price and waits for real prices to return to normal. The other pegs an
intertemporal nominal price and waits for real prices to return to normal. Standard intuition holds that
an exchange rate peg with adequate fiscal backing is stable, just as it holds that a money growth peg with
adequate fiscal backing is stable. We just extend those ideas to a nominal interest rate peg.
In sum, stability and consequent view that higher interest rates eventually raise inflation are logically
linked, explainable by simple intuition, and a robust result of forward-looking expectations. They are hard
to avoid.

3.1. Neutrality
Stability is the central question for the possibility of an interest-rate peg and the sign of inflation’s
long-run response to interest rates. Exact long-run neutrality is not important. If a one percentage point
interest rate rise corresponds to 0.9 or 1.1 percentage point higher inflation rather than 1.0 higher inflation,
no important conclusion has changed. If that interest rate rise corresponds to an 0.1 percentage point
permanent output rise or fall, less has changed. It would take a mighty non-neutrality for steady state
inflation to decrease when steady state interest rates rise – real rates would have to rise more than nominal
rates. Similarly, standard monetary theory is also little affected by small non-neutralities, say that inflation
goes up 0.9 percentage points when money growth rises one percentage point. It would take a huge non-
neutrality for higher money growth to lower inflation.
With a standard new-Keynesian Phillips curve,

πt = βEt πt+1 + κxt ,

the sticky-price model is not exactly long-run neutral. Real variables are eventually independent of the price
level, which is one definition of neutrality, but steady states

x = (1 − β)/κπ, (30)

with a higher inflation rate have higher output, which violates a more stringent definition of long-run neu-
trality. Since β is close to one, however, this is a small effect.
This effect is contentious, with several attempts to fix it and critiques of those attempts. Its presence
is not a central part of new-Keynesian doctrine, which unlike 1960s Keynesianism does not argue for higher
steady state inflation to produce higher steady state employment. Moreover, it is only one of many pecu-
liarities of the Calvo (1983) and related pricing models that are not meant to be taken to extremes. (See
Auclert et al. (2022) for theory and Steinsson and Nakamura (2013) for microeconomic evidence).
Even with β < 1 and a small output effect, higher nominal interest rates correspond one for one
with higher inflation in the standard new-Keynesian model. And the relationship between interest rates
and inflation could also display small non-neutralities without much affecting the analysis. Since nominal
interest is taxed, at least one of pre-tax or after-tax nominal interest rates does not move exactly one for
one with steady-state inflation. Even steady inflation generates distortions via the tax code, and changes
the distribution of prices among the monopolistically competitive producers of new-Keynesian models. But
these considerations do not affect the stability, determinacy or the sign of the long-run response of inflation
to interest rates. It would take a mighty non-neutrality for higher steady state nominal interest rates to
correspond to lower steady state inflation.
I use the words “approximate long-run neutrality” to reflect these considerations.

15
4. Continuous Time

Here I develop the simple model in continuous time. This framework shows many ideas and issues
more clearly still. Some of the timing conventions that obscure the analysis vanish. Continuous time allows
inflation to jump without a price level jump, clarifying how the economy reacts to a shock. In the “primer”
function of this essay, expressing the ideas in continuous time opens a range of useful tools. I use continuous
time in several calculations below, as they give much simpler formulas.
Write the standard model (1)-(2)

Et (xt+∆ − xt ) = σ(it − Et πt+∆ )∆ (31)

Et (πt+∆ − πt ) = −κxt ∆. (32)

This standard model in continuous time is thus

Et dxt = σ(it − πt )dt (33)

Et dπt = −κxt dt. (34)

Normally a term −ρπt dt appears on the right of (34). As I simplified the discrete time Phillips curve
πt = βEt πt+1 + κxt with β = 1, I simplify here with ρ = 0. Nothing important hinges on this simplification.
With any amount of stickiness, the price level is continuous, and cannot jump or diffuse. In an instant
dt only a fraction λdt of producers may change prices. The inflation rate may have jumps or diffusions. But
Et πt+∆ − πt is still of order ∆, so the relevant inflation in the consumer’s first order condition (33) is πt .
The issue whether inflation in that condition should be rationally anticipated or adaptive disappears.
Equations (32) and (34) express the standard rational-expectations Phillips curve. The adaptive-
expectations analogue is

πt − πt−∆ = κxt ∆ (35)

dπt = κxt dt. (36)

Thus, adaptive and rational expectations differ by whether higher output corresponds to increasing (36) or
decreasing (34) inflation. Essentially, they differ by the sign of κ. Adaptive expectations also produce a
differentiable inflation, with neither jumps nor diffusion terms.
Again I simplify the model so we can see the main points without algebra, by using a static version of
the consumption equation,
xt = −σ(it − πt ). (37)
Eliminating output from the Phillips curve, we have the dynamic relation between interest rates and inflation.
With rational expectations
Et dπt = −σκπt dt + σκit dt, (38)
while with adaptive expectations
dπt = σκπt dt − σκit dt. (39)

With rational expectations, inflation is stable but indeterminate. With adaptive expectations, inflation
is unstable but determinate. “Stable” means that the coefficient in front of πt on the right hand side is
negative. “Indeterminate” means that the model as specified so far, with an exogenous interest rate, does
not fully determine inflation.
I simplify here by analyzing impulse-response functions, equivalently how the economy responds to a
single unexpected “MIT shock” at time 0. It is straightforward to generalize to responses to jump-diffusion
shocks. For rational expectations, such solutions to (38) are
Z t
πt = σκ e−σκτ it−τ dτ + e−σκt π0 (40)
τ =0

16
with a single undetermined initial condition π0 . In this context, “stability” means that the influence of past
interest rates disappears over time, while “indeterminacy” means that there are multiple equilibria.
For adaptive expectations, “unstable” means that the coefficient in front of πt on the right hand side
of (39) is negative. It is “determinate” since dπt not Et dπt appears on the left. The solutions of (39) are
Z t
πt = σκ eσκτ it−τ dτ + eσκt π0 .
τ =0

In this context, “unstable” means that interest rates and initial conditions further in the past have larger
effects today. Despite the positive sign σκπt dt on the right hand side of (39), we solve the model backward,
because there is no jump or diffusion in inflation. If we try to solve forward,
Z ∞
πt = σκ e−σκτ it+τ dτ,
τ =0

the right hand side can require an inflation jump that the model rules out. You don’t always solve unstable
roots forward and stable roots backward.
A Taylor rule stabilizes the unstable adaptive expectations model. Adding

it = φπt + ui,t ,

the adaptive-expectations dynamics (39) become

dπt
= −σκ(φ − 1)πt − σκui,t
dt
With φ > 1, dynamics are now stable and determinate. A monetary policy shock ui,t raises the interest rate
and lowers inflation.
In the rational expectations model, adding a policy rule following the Taylor principle in the new-
Keynesian tradition, dynamics (38) become

Et dπt = σκ(φ − 1)πt dt + σκui,t dt.

Now φ > 1 induces instability. We can solve the integral forward, and with a rule against nominal explosions
recover determinacy, Z ∞
πt = −σκEt e−σκ(φ−1)τ ui,t+τ dτ.
τ =0

Fiscal theory offers an alternative route to determinacy in the rational expectations model. Include the
linearized evolution of real government debt.

Et dvt = (rvt + it − πt − s̃t )dt. (41)

This version specifies instantaneous debt, i.e. a floating rate. Integrating forward and imposing the transver-
sality condition, the real value of debt equals the present value of surpluses.
Z ∞
vt = E t e−rτ [s̃t+τ − (it+τ − πt+τ )]dτ. (42)
τ =0

Again, I treat here only responses to an unexpected time-0 shock. Then, corresponding to (19),
Z ∞
dp0 = e−rt [−s̃t + (it − πt )]dt. (43)
t=0

The left hand side dp0 is a potential jump in the price level at time 0 (or diffusion in a full model). Price
stickiness rules out such jumps. Then, of multiple equilibria, we pick the inflation path in which the discount
rate/interest cost effect, the second term, exactly balances any change in surplus, the first term. In the
absence of a surplus change, we pick the inflation path so that the response of total interest costs is zero.

17
5. Can Higher Interest Rates Temporarily Lower Inflation?

Experience suggests, and it is widely believed, that higher interest rates can lower inflation. Stability,
determinacy, and long-run neutrality mean that higher interest rates eventually raise inflation. But there
could well be a temporary negative sign.
If there is a temporary negative effect, we could also reconcile stability and long-run neutrality with
the widespread belief in a uniformly negative effect, as well as the absence of a well-documented long-run
positive effect in most econometric estimates. Central banks seldom leave interest rates alone long enough,
with a background of unchanged fiscal policy and few other shocks, to see the positive long-run effects. The
long-run neutrality of money was similarly not part of common experience-based doctrine until Friedman
(1968).
We are, in short, finally where Lucas started. Lucas built on a theory of neutral money, to describe
a short-run non-neutrality of money. In his model, higher money growth temporarily raises output, the
concern and common belief of his day. We finally have a complete theory of neutral nominal interest rates.
We want a theory of short-run non-neutrality in which higher interest rates temporarily lower inflation, the
concern and common belief today, as well as output.
No matter how one feels about fiscal theory, in its wake it is clear that we have to specify the question,
“what are the effects of monetary policy?” in a way that jointly specifies fiscal policy and the nature of
monetary-fiscal coordination. Even if one completely adopts the standard new-Keynesian point of view,
that central bank equilibrium-selection policies complete the monetary model and fiscal authorities will via
lump-sum taxes “passively” adjust fiscal surpluses to pay any fiscal consequences of monetary policy, the
size and nature of that fiscal policy adjustment needs to be examined in theory and in data.
I focus on the effects of monetary policy—a change in the nominal interest rate—with no change in
primary surpluses. Central banks cannot directly control fiscal surpluses. They cannot change tax rates or
spending. Monetary policy may well induce fiscal surpluses or deficits, but our central theoretical question
is, what can monetary policy do all by itself? If monetary policy only reduces inflation by inducing a fiscal
contraction, that’s news for economic theory and practitioners.
In data, history, and policy, monetary and fiscal policy are intertwined. Fiscal policy reacts to the
same events as monetary policy reacts to. An interest rate rise leads to higher interest costs on the debt,
which may induce a fiscal tightening. If successful at lowering inflation, the monetary tightening induces
a fiscal transfer to nominal bondholders who are repaid in more valuable money. The interest rate rise
might lead to a recession, which leads to stimulus, bank and business bailouts, and automatic stabilizers.
Inflation induced by monetary policy might lead to austerity and fiscal retrenchment. For evaluating history
or for predicting the likely course of the economy following a monetary policy shock, we want to include
such contemporaneous fiscal shocks and induced fiscal responses. For example, we can include a fiscal rule
describing surplus reactions to economic events, and calculate reactions to monetary policy shocks with such
a rule in place. That method is common in quantitative fiscal-theory model evaluation. But it does not
answer our theoretical question, what are the effects of monetary policy? I focus on monetary policy changes
without any change in fiscal surpluses to isolate the answer to this theoretical question.
One can also regard this focus as an orthogonalization: if we calculate the effects of “monetary policy”
as interest rate changes without changes in surpluses, and the effects of “fiscal policy” as changes in surpluses
and deficits without interest rate changes, we can add the two to understand history and to evaluate joint
fiscal and monetary policies.
There might be no model with all these desiderata in which higher interest rates lower inflation, even
temporarily. If so, that fact is important. A search that establishes that standard beliefs require induced
fiscal policy responses, irrational people, or an inherently unstable or non-neutral foundation, is just as
important as a search that finds a model to confirm standard beliefs.
You might think it’s easy. Just add some sticky prices. That turns out not to be the case. You might
think that standard models in use for decades satisfy this obvious desire, since the belief is so widespread.
That also turns out not to be the case. Both standard new-Keynesian and adaptive expectations models

18
produce a negative effect by positing a fiscal contraction coincident with the interest rate rise. Without that
contraction, even those models do not produce a negative effect of interest rates on inflation.

5.1. Higher Inflation in Simple Sticky Price Models

Figure 2 plots the response of my simplified rational expectations model to an unexpected permanent
interest rate rise, with no change in fiscal surpluses. The “sticky price” line gives the main message of this
section: Inflation rises, even in the short run. Sticky prices alone do not give us a negative effect. The real
interest rate rises (interest rate higher than following inflation), so output declines (not shown). But sticky
prices only draw out the positive response of inflation to interest rates. We maintain long-run neutrality:
The inflation rate eventually rises to match the interest rate.
1.2
Interest rate
1

0.8
Sticky price
Percent

0.6

0.4
Flexible or Lucas
0.2

-0.2
0 1 2 3 4 5
Time

Figure 2: Inflation response to a 1% permanent rise in the interest rate with no change in fiscal policy. Rational expectations
model. Parameters σκ = 1.

Start with the “flexible or Lucas” line. The flexible price model is (20)–(21). The responses solve

Et πt+1 = it (44)
∆Et+1 πt+1 = 0. (45)

At time 1, the interest rate rises unexpectedly from 0 to i1 = 1. There is no expected E0 π1 = i0 = 0 or

unexpected ∆E1 π1 inflation, so π1 = 0. After that, however, inflation rises to meet the higher interest rate
path, πt = it−1 = 11 = 1. The Lucas Phillips curve in this model (25) produces the same inflation result,
and a one-period movement in output.5
The “sticky price” line uses the forward-looking new-Keynesian Phillips curve. The response of inflation
to the interest rate solves (19)-(22):
1 σκ
Et πt+1 = πt + it (48)
1 + σκ 1 + σκ

5 The impulse-response solves

1
Et πt+1 = it + ∆Et πt (46)
σκ
∞
X
∆Et+1 πt+1 = ∆Et+1 ρj (it+j − πt+j+1 ). (47)
j=1

With only one unexpected movement at time 1 (t = 0 in the equation), so π2 = E1 π2 and so forth, (46) leaves π1 arbitrary
but then π2 = i1 + π1 /(σκ), π3 = i1 , π4 = i1 , etc. Now, use (47) to find π1 and the unique path. (47) reduces to
π1 = i1 − π2 = −π1 /(σκ). The unique solution is π1 = 0, and thus π2 = i1 , π3 = i1 , and so forth.

19
 ∞
X
∆Et+1 πt+1 = ∆Et+1 ρj (it+j − πt+j+1 ). (49)
j=1

In response to the unexpected permanent interest rate shock, it = i1 , t = 1, 2, . . ., there is a family of

solutions to (48) which we can index by π1 ,
1
πt+1 = i1 + t (π1 − i1 ) . (50)
(1 + σκ)
We can use (49) at time t = 0 to determine
∞
X
π1 = ρj (i1 − πj+1 ). (51)
j=1

Substituting from (50) and simplifying,

ρ
π1 = i1 (52)
1 + σκ
and the full unique solution is " #
1 + σκ − ρ
πt+1 = 1 − t+1 i1 . (53)
(1 + σκ)
Inflation rises to meet the nominal interest rate.
The real interest rate rises and output declines. If your objective is only to produce a model in which
the central bank can temporarily cool economic activity with interest rate rises, you have it. You might stop
here and say that we have indeed redone Lucas (1972a) with interest rate targets, since his purpose was to
understand how money growth affects the real economy.
But inflation still rises uniformly after the interest rate rise. Indeed, in period 1, inflation is greater
than the (zero) value of the frictionless model. Why? With sticky prices, higher nominal interest rates lead
to higher real interest rates. Higher real interest rates mean greater unfunded (by assumption of no change
in fiscal policy) interest costs on the debt. With no change in surpluses, higher real interest costs must come
from a higher unexpected period 1 inflation, which devalues outstanding debt. Equation (51) expresses this
result. Higher real interest rates, induced by sticky prices, are an inflationary force, since they raise debt
service costs. For this reason, as prices become stickier, κ → 0, first period inflation π1 rises further, π1 → ρ
(see (52)) which is just barely less than one. Stickier prices make inflation rise sooner.
Fiscal underpinnings matter crucially to these calculations. If we said that fiscal surpluses would always
rise to pay higher interest costs on the debt, then (49) has an extra surplus term on the right hand side, as
in the original (19), and the two terms on the right hand side are always equal. Then π1 = 0, no immediate
inflation rise, though inflation would still rise uniformly after that. If we could pair the interest rate rise with
even higher surpluses, we could predict lower inflation, π1 < 0. But the question we want to ask is, what
can higher interest rates do to lower inflation without fiscal support? The answer is, so far, higher interest
rates can produce a recession, but they raise inflation.
Even the slow rise of inflation is fragile, really the result of one-period rather than instantaneous
(floating-rate) debt. With instantaneous debt, as with very sticky prices, inflation rises immediately. To see
this effect, we can compute the same response in continuous time. From (40), for a generic interest rate path
{it }, the solution
Z t
πt = σκ e−σκτ it−τ dτ + e−σκt π0 (54)
τ =0
gives us a family of inflation paths indexed by π0 . Inflation is, due to sticky prices, a backward-looking sum
of nominal interest rates, plus a decaying transient. Only one of these paths satisfies the valuation equation
(42), which is in this case with no change in surpluses s̃t = 0,
Z ∞
0= e−rt (it − πt )dt. (55)
t=0

20
Average interest costs on the debt must be zero. This condition selects which path {πt } is the equilibrium.
Any period of positive real rates must be balanced by a period of negative real rates.
In the same case that the interest rate rises from 0 to a new constant value i, (54) reduces to
πt = (1 − e−σκt )i + e−σκt π0 . (56)
Plugging this into the valuation equation (55) with s̃t = 0 to determine π0 ,
Z ∞
i
0= − e−rt [(1 − e−σκt )i + e−σκt π0 ]dt (57)
r t=0
π0 = i, (58)
and then πt = 1, t ≥ 0 as well. Despite sticky prices, we pick the equilibrium in which inflation moves
instantly to match the interest rate.
Sticky prices do not imply sticky inflation. The few firms that can change price at any instant raise
their prices a lot. Much verbal intuition describes sticky inflation or inflation momentum. That needs a
different set of costs than those specified by current forward-looking sticky-price models.
Appealing to the new-Keynesian model will not help to produce a negative sign. This is the new-
Keynesian model. Solving the discrete-time model in new-Keynesian style, the central bank can produce
any value of first-period inflation π1 = π1∗ it wishes, by following an interest rate policy i1 = i∗1 + φ(π1 − π1∗ ),
where i∗1 = 1, the first point on the desired interest rate path, and π1∗ is the desired, possibly negative,
first-period inflation. Write it up as it = φπt + ut with ut = i∗t − φπt∗ . But any inflation path other than the
one we have already plotted requires a change in surpluses. Lower initial inflation requires a positive shock
to surpluses. If we phrase the question of a new-Keynesian (φ > 1, passive fiscal policy) model, “What is
the response of inflation to a monetary policy disturbance {ut } that produces an unexpected permanent rise
in the interest rate, and the associated passive fiscal policy requires no change in surpluses?” we have just
calculated the unique answer.6
For algebraic simplicity, I have so far used the model with no forward-looking term in the IS curve. I
asserted that simplification is not essential to the results, but it’s worth verifying that assertion. I add a
second important generalization: I allow lagged inflation in the Phillips curve. That is a common modification
in models such as Christiano, Eichenbaum, and Evans (2005), and it brings the model back toward the
adaptive expectations model which may lead one to hope for something like the latter’s short-run dynamics.
The model is
xt = Et xt+1 − σ(it − Et πt+1 ) (59)
πt = (1 − α)Et πt+1 + απt−1 + κxt (60)
ρvt+1 = vt + it − πt+1 (61)
it+1 = ηit + εi,t+1 . (62)
Equation (59) includes the forward-looking term in the IS equation. Equation (60) allows for lagged inflation
in the Phillips curve. Equation (61) specifies short-term debt and no change in surplus.
Figure 3 presents the response of this model to a permanent interest rate rise. I solve the model
numerically. The shaded area gives inflation paths for a grid of parameter values. I restrict parameters to
those that produce real eigenvalues. Sawtooth or sine-wave responses induced by complex eigenvalues can
produce some temporarily negative inflation responses, but are clearly unrealistic. As the figure shows, for
all such parameter values, this generalized model produces a steady rise in inflation. The static IS curve did
capture the results of this more complex model.

6 Bianchi and Melosi (2019) study a related question. They also study interest rate increases with no fiscal policy reaction,

in a more elaborate sticky-price model. They imagine a central bank that goes one step further, following an “active” Taylor
rule that reacts to current inflation with coefficient greater than one. Now, higher interest rates raise interest costs on the
debt, which leads as here to greater inflation. Their central bank raises interest rates even further, which if left alone leads to
explosive inflation. The possibility of a Markov switch back to a coordinated regime saves this asymptotic possibility. Since on
the path, future inflation is higher than current inflation, output falls.

21
Figure 3: Inflation response to a permanent interest rate rise, with no change in fiscal policy. Full model with forward-looking
IS curve and a lag in the Phillips curve. The shaded area shows impulse response functions for ρ = 0.99, η = 0.999, and other
parameters in a grid with α ∈ [0, 1], κ ∈ [0.01, 2], σ ∈ [0.01, 5], that produce real eigenvalues.

5.2. Transitory Interest Rates

The essential failure of the rational expectations sticky price model to produce a negative inflation
effect is not tied to the permanent interest rate increase shown in Figure 2 and 3. I only graph permanent
interest rate changes so that we can see long-run properties. However, transitory interest rate paths can give
a misleading appearance of a negative effect.
1.2
Interest rate
1 Flexible price
0.8

Sticky price
Percent

0.6

0.4

0.2

-0.2
0 1 2 3 4 5 6 7 8
Time

Figure 4: Response of the simple rational expectations sticky price model to a transitory interest rate path, with no change in
fiscal policy. Parameters σκ = 1, ρ = 0.99, and it = 0.7it−1 + εt .

Figure 4 plots the response of the simplified rational expectations model to a transitory interest rate
movement, with it = 0.7it−1 + εt . Inflation still rises uniformly. The only difference is that one doesn’t
notice long-run neutrality with a transitory shock.
At the cost of some algebra, relegated to the Appendix, we can find the response of the rational-
expectations sticky-price model (48)-(49) to an arbitrary interest rate path {it }, announced at period 1,
with no change in fiscal surplus:
∞ t
1−ρ X σκ X 1
πt+1 = t+1 ρj ij + ij (63)
(1 + σκ) j=1
1 + σκ j=1 (1 + σκ)t−j

22
and in particular
∞
1−ρ X j
π1 = ρ ij . (64)
1 + σκ j=1

In (63), all the coefficients are positive. Hence, any sequence of positive interest rates {it } generates
uniformly positive inflation response {πt }. In this sense, the positive response of the rational expectations
sticky price model is general and does not depend on the time-series process of the interest rates.
As usual, the analogous continuous time version is more elegant. The general solution to (54)-(55), for
an arbitrary interest rate path is given by:
Z ∞ Z t
πt = e−σκt r e−rτ iτ dτ + σκ e−σκτ it−τ dτ. (65)
τ =0 τ =0

See the Appendix for algebra. In (65) the coefficients of inflation on the interest rate are also all positive.
We can generate the appearance of a negative effect of interest rates on inflation, however. Inflation is
a long moving average of interest rates, with forward and backward terms. If interest rates were to rise in
the short run, and then turn towards a long-lasting negative value, then the future lower interest rates could
pull down even current inflation.
0.3

0.2

0.1
Interest rate
Percent

-0.1
Inflation
-0.2
-1 0 1 2 3 4 5 6 7 8 9 10
Time

Figure 5: Inflation response to a transitory rise in the interest rate. Parameters σκ = 0.1, r = 0.01.

Figure 5 presents an example. As in all figures, this is the response of inflation to the indicated interest
rate path, with no change in fiscal surpluses. I create an interest rate path with a hump shape rather
than an AR(1) shape to make the graph prettier, and to resemble paths often seen in VARs. I specify
it = e−0.35t − 0.6e−0.9t − 0.2. I then calculate the inflation path from (65). For an AR(1) interest rate
it = i0 e−ηt , (65) reduces to  
1 r + σκ −σκt
πt = σκe−ηt − ηe i0 . (66)
σκ − η r+η
(See the Appendix.) I add three such solutions to calculate Figure 5.
Figure 5 looks appealing. A higher nominal interest rate seems to produce lower inflation. As inflation
comes down, so does the nominal interest rate, both ending up lower. This looks a lot like the standard story
of a disinflation such as the 1980s. One could even read the plot as an account of 1980 through 2020—high
real interest rates (1980-1990), followed by negative real interest rates (2000-2020), which slowly fade away
(past the right end of the graph) when the economy settles back permanently lower inflation and nominal
interest rates. Until the next shock, that is.

23
 But a reading that higher interest rates cause the disinflation would be profoundly misleading. Inflation
declines initially because expected interest rates decline in the far future, despite, not because of, the short-
term interest-rate rise. Indeed, the positive interest rates drag inflation up from even more negative values.
If you want less inflation in this model, lowering interest rates immediately—the negative of Figure 2—is an
even more powerful tool. There is nothing in the mechanics of this model that resembles standard intuition,
that high real interest rates drive inflation down. Beware causal readings of impulse-response functions!
If we want a negative effect of interest rates on inflation, without a contemporaneous fiscal shock that’s
really doing the work, then, we will need to add ingredients beyond sticky prices.

5.3. Fiscal Requirements for Adaptive Expectations Models

The adaptive expectations model (5),

πt = (1 + σκ)πt−1 − σκit . (67)

produces standard intuition. Higher nominal interest rates lower inflation. Inflation or deflation left
unchecked will spiral away.
Debt still accumulates by (16),

ρvt+1 = vt + it − πt+1 − s̃t+1 . (68)

This is an identity, unaffected by expectations. Higher ex-post real interest rates it − πt+1 with no change
in surpluses would lead debt to spiral upward, violating the transversality condition or other bounds that
define an equilibrium.
If we ask the adaptive expectations model, “What is the response of inflation to a permanent unexpected
rise in the interest rate, with no change in fiscal surpluses?,” the downward spiral does not answer that
question. There is no answer to that question. The central bank cannot permanently raise interest rates
without higher fiscal surpluses. There are no multiple equilibria, which we can select based on fiscal backing.
(67) alone determines the inflation path, and if that path does not satisfy the present value debt equation,
or (68) and the transversality condition, the path can’t be an equilibrium
In reality, once a debt spiral emerges, central banks would eventually give in and abandon an interest
rate policy. Friedman (1968), describing a reduction in interest rates, wrote that the central bank would
eventually give in out of distaste for exploding inflation. Here, it is forced to give in by exploding debt and
deficits.
Pegs and spirals are good conceptual experiments, but not reasonable policies. The most standard
model that produces a disinflation from a higher interest rate combines adaptive expectations with a Taylor
rule. A positive shock to the Taylor rule starts an inflation decline, and then the Taylor rule automatically
lowers the interest rate to steady the incipient spiral. However, this model also requires tighter fiscal policy
to pay higher interest costs on the debt. Without such support, the disinflation does not occur.
It is easiest to exhibit this behavior in the continuous time version of the model. Uniting the continuous
time versions of (67) and (68) with a Taylor rule,

dπt
= σκ(it − πt ) (69)
dt
dvt
= rvt + it − πt (70)
dt
it = φπt + ut . (71)

(I retain for now short-term debt.) At time 0, the monetary policy shock ut rises suddenly and unexpectedly
from 0 and stays at the constant value ut = u0 . Substituting (71) in (69), inflation dynamics become

dπt
= −σκ(φ − 1)πt − σκu0 .
dt

24
The solutions for inflation, interest rate, real rate, and debt are
1 h i
πt = − 1 − e−σκ(φ−1)t u0
φ−1
1 h i
it = − 1 − φe−σκ(φ−1)t u0
φ−1
rt = e−σκ(φ−1)t u0
e−[r+σκ(φ−1)]t − 1
e−rt vt = .
r + σκ(φ − 1)

Figure 6 presents this result. The nominal and hence real interest rate rises, and inflation starts on its
downward spiral. Following the Taylor rule, the interest rate swiftly follows inflation down, and the economy
trends to the new lower inflation rate. This is a standard story of the 1980s, for example.
2

Disturbance u t Debt e -r t v t
1
Percent

Interest rate
-1 Inflation

-2
0 1 2 3 4 5
Time

Figure 6: Response to a 1% permanent monetary policy shock in the adaptive expectations model with a Taylor rule. Parameters
σκ = 1, φ = 1.5, ρ = 0.01.

The real interest rate is positive throughout the episode. Therefore, this path does not satisfy our fiscal
condition that the average real interest rate must be zero. With no change in surpluses, the greater interest
costs on the debt are simply rolled over, and debt increases without bound. The transversality condition
limT →∞ e−rT vT = 0 is violated, as shown.
Therefore, this simulation does not answer the question, what can the central bank do by itself, without
fiscal support? To achieve this outcome, fiscal policy must increase surpluses to pay the interest costs. If
this is the story of the 1980s, the story is a joint monetary-fiscal stabilization, with surpluses rising (as they
did), not a story of monetary policy acting alone.

5.4. Adaptive Expectations With a Fiscal Constraint

In response to our question, the effect of interest rates on inflation holding fiscal policy constant, the
adaptive expectations model cannot realistically produce a permanent inflation change, by any path of
interest rates .
This point is again easier to see in the continuous time version of the model. Denoting rt ≡ it − πt ,
(69)-(70) and the transversality condition are
dπt
= −σκrt (72)
dt
dvt
= rvt + rt (73)
dt

25
 lim Et e−rT vT = 0. (74)
T →∞

The symbol r without subscript represents the steady state real interest rate, and point of linearization. The
symbol rt represents variation of the real interest rate around this steady state value. Normally the surplus
s̃t appears on the right hand side of (73) but I omit it as the question holds fiscal policy constant. We are
calculating a response function, with all variables zero before time 0, and a single unexpected shock to a
new interest rate path at that date.
The adaptive expectations model is determinate: Given the interest rate path, there is only one equi-
librium inflation path. Thus, rather than select among equilibria, fiscal policy creates a constraint on the
set of interest rate paths that the central bank can follow, in order not to create a debt spiral. A nominal
rate path gives rise to one inflation path, so is equivalent to a chosen real rate path. It is simpler to look at
the real rate choice and then find the implied inflation and nominal rates.
Equations (73) and (74) imply the familiar restriction that the average interest cost is zero
Z ∞
0= e−rτ rτ dτ. (75)
τ =0

This is now a restriction on the real interest rates that the central bank may choose.
Given a real interest rate path, the inflation solution (72) is backward-looking,
Z t
πt = −σκ rτ dτ. (76)
τ =0

Inflation is proportional to the sum of all past real interest rates.

Define the long-run inflation rate π∞ ≡ limT →∞ πt , so
Z ∞
π∞ = −σκ rτ dτ. (77)
τ =0

To lower long-run inflation in (77), then, the central bank picks a real rate path {rt }, subject to the fiscal
constraint (75). Given the real rate path, we find inflation from (76), and the nominal rate from it = rt + πt .
We see right away that in the limit r → 0, the adaptive expectations model cannot produce any permanent
inflation or disinflation—a value of π∞ other than zero—in the absence of a change in fiscal policy. The
right hand sides of (77) and (75) are the same. Positive real interest rates that push down inflation also
build up the debt. They must be balanced by negative real rates to bring down the debt, but those real
rates raise inflation right back to where it was.
A positive steady state real rate r > 0 offers an apparent avenue for permanent disinflation. But, since
r is small, this result is fragile, and the resulting policies are unrealistic. Equation (75) downweights real
rates in the future. So, to lower long-run inflation, the central bank must first lower real interest rates,
building up inflation, to produce a period of low interest costs, which lowers the debt. Then the central
bank turns around and raises real interest rates, driving inflation down, using the accumulated savings on
the debt to pay the higher interest costs. The weighted integral in (75) allows a longer period of future high
interest rates than the initial period of low interest rates, which then overall drives inflation down relative
to its initial value.
But with small r, the opportunity requires large swings in rates and inflation to produce a small
permanent reduction in inflation. Moreover, the period of low rates and high inflation must come first,
before the final period of high rates and lower inflation. Imagine a central banker announcing that the plan
for combatting inflation is first to make inflation worse, to drive down the debt, and only then turn around
to cure the initial inflation plus the deliberate addition to inflation. This is highly unrealistic, and not at all
consistent with the classic intuition we are trying to rescue.
When we pose the question as, “Can higher interest rates lower inflation, without a change in fiscal
policy,” not even the most classic adaptive expectations model, which fits the narratives of central bankers
and the policy world, can do it.

26
5.5. A Model with Long-Term Debt that Produces a Negative Effect
Figure 7 offers a simulation of the only current model I know of in which higher interest rates with no
change in fiscal policy produce a negative short-run inflation effect.
1.5

Interest rate i
1

0.5 , short debt

0

-0.5 Inflation
Percent

-1

-1.5 Output x
-2

-2.5

-3
0 1 2 3 4 5 6 7 8 9 10
Time

Figure 7: Response of inflation to an interest rate shock, and no change in fiscal policy, with long term debt. In the base case,
debt has a geometric maturity structure, decaying at rate 0.9t with t = maturity. The short-term debt case is one-period debt
only.

The model is

xt = Et xt+1 − 0.5(it − Et πt+1 ) (78)

πt = Et πt+1 + 0.5xt (79)
n
ρvt+1 = vt + rt+1 − πt+1 − s̃t+1 (80)
n
Et rt+1 = it (81)
n
rt+1 = 0.9qt+1 − qt . (82)

Here I use the full rational-expectations model, i.e. including the Et xt+1 term in (78), as analytic solutions
are not insightful. I include long-term debt with a geometric maturity structure. The face value of zero
(j)
coupon bonds of maturity j, Bt = (1 − ω)ω j Bt declines at rate ω = 0.9. The symbol vt is the real market
n
value of nominal government debt. The symbol rt+1 represents the ex-post nominal return on the portfolio
of government debt. Equation (81) prices long-term bonds with the expectations hypothesis. Equation (82)
links the log price qt of the government bond portfolio to its rate of return. This is a simplified version of
the model in Cochrane (2021), which in turn simplifies Sims (2011).
Again, I raise interest rates, I leave fiscal surpluses unchanged, and I calculate the inflation and output
responses. Inflation declines temporarily! Inflation then rises in the long run, fulfilling long-run neutrality.
This is the pattern we have been seeking.
Sims (2011) calls the pattern of Figure 7 “stepping on a rake” and offers it as a parable of the 1970s
inflation cycles, in which higher interest rates temporarily lowered inflation, but inflation came back. We can
call the pattern “unpleasant interest-rate arithmetic,” a successor to Sargent and Wallace (1981) unpleasant
monetarist arithmetic. (Sargent and Wallace focus on seignorage in a model with real debt, in which the
central bank controls money supply. In this model, there is no seignorage or money, government debt is
nominal, and the central bank follows an interest rate target.) Unpleasant interest-rate arithmetic is here

27
a negative sum, or inequality proposition: The central bank gets more long-run inflation than it saves in
short-run inflation. (Woodford (2001) has the first statement of this effect.)
Long-term debt is the crucial innovation relative to the simple models of previous figures, and its
inclusion produces the negative sign. With long-term debt, but no ability to change surpluses, the central
bank can lower inflation now, but by raising inflation later. Raising long-run interest rates, and thus long-run
inflation, devalues long-run debt. Since surpluses haven’t changed, lowering the value of long-term debt must
raise the value of short-term debt. But short-term debt can only become more valuable via a lower price
level.
You can see the mechanism directly with flexible prices. The real interest rate is a constant r, and there
is a geometric maturity structure in which debt of maturity j has face value ϕBe−ϕj . There is a constant
nominal interest rate i. The government debt valuation equation, stating that the market value of debt
equals the present value of surpluses, reads
R∞
ϕB j=0 e−ij e−ϕj dj ϕ B
Z ∞
= = Et e−rj st+j dj.
P i+ϕP j=0

Imagine an unexpected permanent rise of the nominal interest rate from i to i0 with no change in surpluses.
The price level then jumps by
P0 i+ϕ
= 0 .
P i +ϕ
With perpetuities, ϕ = 0, this is a powerful effect. A rise from i = 1% (0.01) to i = 2% (0.02) cuts the price
level in half. A 1% larger inflation then starts right away. Shorter maturity debt gives a smaller response.
For example, if debt has a 4 year half life (the face value of debt declines to e−1 by a 4 year maturity), then
ϕ = 0.25, so a rise from 1% to 2% nominal rate implies 0.26/0.27 ≈ 10% price level drop. That is still a
large cumulative disinflation inflation from a 1% interest rate rise.
As with the short-term debt case, there is nothing essential about a permanent interest rate shock to
generating this example. Any persistent shock will also lower long-term bond prices and raise long-term
inflation, raising the value of short-term bonds. The initial disinflation is smaller, but still present.
A Taylor-type rule, in which the interest rate reacts to inflation, automatically adds this sort of response
to the inflationary effect of fiscal or other shocks. In this way, a Taylor-type rule spreads the inflation of such
shocks over time. With the forward-looking Phillips curve, smoothing inflation over time reduces output
volatility, so it is a good policy.
Taylor emphasizes that his rule works well in a variety of models, and that robustness rather than strict
optimality in a particular model is its virtue. Raising interest rates in response to inflation can eliminate the
instability of the adaptive expectations model; it can eliminate the indeterminacy of the rational expectations
model; and it can reduce volatility in the rational-expectations fiscal-theory model.
The graph is attractive, but this mechanism of this model is more limited and quite different from
conventional intuition about monetary policy. 1) The negative effect of interest rates on inflation in this
model only holds for unexpected interest rate rises. Lower inflation breaks out when a higher interest rate is
announced, not when it happens. 2) The effect vanishes when governments borrow short term, as illustrated
by Figure 7 in the line marked “π, short debt.” More generally, the size of the negative effect depends on
the maturity structure of debt. 3) The negative effect requires long-lasting interest rate increases, that raise
long-term nominal bond yields. 4) Stickier prices reduce the negative effect of interest rates on inflation, and
the effect is strongest with flexible prices. With sticky prices, the higher real interest rates add interest costs
of the debt, an inflationary force. 5) In this model, inflation declines immediately and then recovers, where
in the common intuition higher interest rates lower future inflation gradually with “long and variable lags.”
These features are not necessarily counterfactual. They are unknown. Nobody has looked to see if
the negative effect of interest rates on inflation, holding constant fiscal policy, is quantitatively linked to
announcement, maturity, persistence, bond price movements, price stickiness, and fiscal events as the model
predicts. Looking would be a valuable empirical project.

28
 But as much as the result looks standard, the model’s mechanism is completely different from standard
intuition, that higher nominal rates raise real rates due to sticky prices, real interest rates depress de-
mand, and through a Phillips curve lower future inflation. It does not produce something like the adaptive-
expectations dynamics in the short run, which then turn around and become stable when some suitable
friction or information problem is resolved. And it will be a long time before we write opeds, Fed chairs ex-
plain, and we teach to undergraduates that the central mechanism by which the central bank can temporarily
lower inflation is to rearrange the real payoffs to different maturities of nominal government debt!

6. Fiscal Constraints

The calculations so far add inflation dynamics produced by particular models together with fiscal
constraints on what any model can do. A look directly at those fiscal constraints clarifies the challenge.
With long-term debt, in response to a shock that happens at time 0, and with sticky prices so that
there are no jumps in the price level, the decompositions (19) and (43) generalize to
Z ∞ Z ∞ Z ∞
e−(r+ϕ)t πt dt = − e−rt s̃t dt + e−rt (1 − e−ϕt )(it − πt )dt. (83)
t=0 t=0 t=0

(t+j)
(Cochrane (2023) section 3.5.) Here ϕ models debt with a geometric maturity structure, Bt = ϕBt e−ϕj .
I impose the expectations hypothesis. To allow time-varying risk premiums one places the the expected
nominal return on the government bond portfolio in place of the interest rate.
This equation unites the three important fiscal mechanisms. On the left, a rise in expected future
inflation devalues the claims of long-term bondholders, and deflation gives them a windfall. This mechanism
takes the place of a price level jump, but acts via more plausible drawn-out inflation which devalues long-term
bonds as they come due. On the right, we have surpluses. Those are zero for my definition of “monetary
policy.” Finally, we have the effect of interest costs on the debt, which flow on to the government budget as
debt rolls over.
For a given path of interest rates, this equation restricts the path of inflation. For a rational expectations
model with multiple equilibria, it determines the equilibrium inflation path. For an adaptive expectations
model, it restricts what interest rate paths the central bank may follow without inducing an unsustainable
debt spiral.
With no change in surpluses and for short term debt ϕ = ∞, expression (83) simplifies to
Z ∞
0= e−rt (it − πt )dt. (84)
t=0

With short-term debt and no price-level jumps, inflation cannot devalue outstanding debt directly. Then
the long-run average of real interest rates and interest costs on the debt must be unchanged. For any model
of inflation dynamics, the average weighted difference between inflation and interest rates must be zero.
For perpetual debt ϕ = 0, (83) simplifies to
Z ∞
e−rt πt dt = 0. (85)
t=0

In this case, both government and bondholders are fully insured against changes in real interest rates, as
neither has to roll over existing debt. With long-term debt, however, inflation devalues long-term nominal
promises, and deflation delivers a windfall to bondholders. Thus, with no change in surpluses, the long-term
average of inflation itself must be unchanged.
In these expressions, and the intermediate cases in between, unpleasant interest-rate arithmetic is
generic. Any model in which higher interest rates lower short-run inflation must produce higher subse-
quent inflation. Monetary policy is limited to moving inflation around through time, lower at some time in
return for higher at another time.

29
 With short-term debt in (84), inflation averages zero around the nominal interest rate path. With
long-term debt in (85), inflation itself averages zero around zero response, which makes it a little easier to
get lower inflation in the short run in return for more inflation later, as I did in Figure 5. But no model can
produce uniformly negative inflation from higher nominal interest rates.

7. Evidence and Experience

Stability and determinacy challenge conventional doctrines, but they are not obviously contrary to his-
torical experience. Indeed, recent decades provide a nearly ideal set of experiments to distinguish monetarist,
adaptive-expectation Keynesian, new-Keynesian, and fiscal theory views. And they do so via the robust and
basic properties of the models, not via ephemeral correlations.
While belief in a powerful negative and direct effect of higher interest rates on inflation is strong,
particularly in central banks and policy circles, the actual evidence for such an effect is weak.

7.1. Great Experiments

We have just seen something close to an interest rate peg. From 2008 to 2016 in the US, from 2008 to
2022 in Europe, and from 1995 to 2022 in Japan, interest rates were effectively stuck at zero. They could
not move much in the downward direction, they did not move in the upward direction. Central bankers gave
“forward guidance” that interest rates would not move, at least not promptly and more than one for one
with observed inflation.
When these countries hit the zero (or effective lower) bound, many economists, commenters, central
banks, and international monetary institutions, armed with thinking that I represent by the adaptive expec-
tations model, predicted a “deflation spiral” as depicted in the left-hand panel of Figure 1. They continued
to worry that a spiral would break out during the decades of zero interest rates. New-Keynesians had a
different worry. They had warned for 20 years that a zero bound would lead to volatile multiple-equilibrium
inflation as depicted in the right-hand panel of Figure 1. By contrast, the model with rational expectations
and fiscal theory says inflation is stable and determinate at a peg.
Neither unstable spirals nor additional multiple-equilibrium inflation volatility broke out. Inflation just
batted around 1-2% the whole time, if anything less volatile than it had been when central banks could
move interest rates.7 It is a pretty decisive experiment in favor of the view that inflation can be stable and
determinate at a peg, or with interest rates that move less than one for one with inflation.
True, long-run fiscal policy was not in great shape in the 2010s. But there wasn’t much news. Moreover,
real interest rates and interest costs on the debt fell, unexpectedly relative to long-term bond prices in 2007,
were negative for most of the period, and were forecast or feared to remain negative pretty much permanently.
Low interest costs on the debt act like surpluses (see (19)) to hold down inflation. At 100% debt to GDP,
1-2% negative real interest rates in place of 1-2% positive real interest rates have the same deflationary effect
as 2-4% of GDP additional primary surpluses. The era also featured r∗ and r < g debates, and the question
whether government debt needs to be repaid at all.
From 2021 through 2022, analysts similarly worried that central banks’ year of inaction, and then
another half year of interest rates below inflation, would lead inflation to spiral upward. They worried that
only a sustained dose of interest rates higher than current and past inflation, with a steep recession, could
bring inflation down, repeating the early 1980s. (See, for example, most of Bordo, Cochrane, and Taylor
(2022).) Yet, from 2022 to summer 2023, inflation declined with interest rates still below current inflation,
as it did previously in 1948, 1951, and 1975. On the other hand, the outbreak of inflation following a huge
$5 trillion fiscal stimulus, and its gentle decline afterwards despite interest rates below inflation, mirrors the
simplest fiscal-theory response to a fiscal shock.8

7I summarize here more detailed analysis in Cochrane (2017), Cochrane (2018), and Cochrane (2023) Ch. 20.
8 See Cochrane (2022a), Cochrane (2024) for simulations.

30
 What about historical pegs that preceded inflation? These episodes are central to Friedman’s (1968)
argument that pegs are unstable. Well, quiet (the opposite of volatile) inflation under a peg also requires
sustainable fiscal policy. Most governments with interest rate pegs and spiraling inflation used the peg to
hold down interest costs of the debt in a time of high and rising fiscal problems. One can read such pegs as
part of a financial repression regime designed to boost demand for government debt and hold down interest
costs, not a market interest rate in free and open financial markets with solvent fiscal policy. The US
interest rate peg of the 1940s and early 1950s was explicit in that aim, and came with restrictions on private
interest rates. And even so, that peg lasted remarkably long, relative to models that predict instability or
indeterminacy. The peg lasted through nearly 20% inflation in 1947, which also faded away with no spiral.
Those interest rate pegs also typically included exchange rate controls and trade restrictions.
Fiscal theory and monetarism agree that if the government prints a trillion dollars to finance a large
deficit, and people do not trust the debt to be repaid, inflation will result. But if the Fed gives a trillion
dollars in money, while taking back a trillion dollars of treasury securities, monetarism predicts exactly the
same inflation. Fiscal theory predicts no inflation to first order. We just saw as clean a version of that
experiment as one could ask for. In the covid expansion, the Fed monetized about $3 trillion of a $5 trillion
expansion. Inflation quickly followed, peaking at 8%. In the Quantitative Easing era, the Fed monetized
about $4 trillion of government debt, with no additional deficits, and produced no visible inflation.

7.2. Interest Rates and Inflation

My theoretical uncertainty about the effects of interest rates on inflation seems awkward given the
firmness of many analysts’ and central bankers’ beliefs. But econometric analysis actually gives weak evidence
that higher interest rates lower inflation.
When monetary policy shocks reduce inflation at all, such reductions are delayed more than a year and
then are small and marginally statistically significant or insignificant. Ramey (2016) provides a comprehen-
sive review and update. See especially Ramey’s Figures 1, 2, and 3, with small, sometimes positive, delayed,
and insignificant inflation effects, and note the disappearance of any effect after 1983. Romer and Romer
(2023) Figure 5 and Figure 7 summarize the results of the narrative approach. After a higher interest rate
or monetary tightening, inflation rises slightly for a year, then gently declines, but just about two standard
errors below zero. Moreover, these estimates include credit and quantitative shocks in the 1950s and 60s,
and the last shocks are in 1981 and 1988 (Table 2). The “price puzzle” that higher interest rates want to
raise inflation dogged VAR analysis for decades, requiring careful orthogonalization to produce even this
mild inflation reduction. It may have been trying to tell us something. The empirical evidence that higher
interest rates lower output and employment is much stronger. The question here is whether they lower
inflation.
Several other considerations weaken even this much evidence. Most of all, none of these estimates tries
to orthogonalize monetary to fiscal policy, and so do not answer our basic question. That omission is natural,
as without a fiscal theory in mind, nobody thought that it was important to do so. It is plausible that fiscal
contractions accompany monetary contractions. If nothing else, fiscal policy may tighten to pay higher
interest on the debt. Fiscal policy is also likely to tighten to control inflation at the same time as monetary
policy does so. Fiscal stimulus is a common response to deflation, and austerity is a common response to
inflation. The early 1980s included a major Social Security reform, large tax reforms, spending controls,
and growth-enhancing deregulation. Sharply rising primary surpluses in the late 1980s and 1990 paid higher
interest costs on the debt and paid for the real interest costs and bondholder windfall of the 1980s. Most
successful disinflations combine monetary, fiscal, and often microeconomic reform. Failed stabilizations often
include an attempted monetary tightening that is not accompanied by fiscal and microeconomic reform.
Nakamura and Steinsson (2018) eloquently summarize the hard identification and computation troubles
of contemporary empirical work. To identify an interest-rate shock, as opposed to an endogenous response,
we must assume that regressions control for all variables that the central bank sees and responds to. But
central banks watch more variables than we can ever include. Nakamura and Steinsson use 9/11 as a clear
example: The Fed lowered interest rates after the terrorist attack, likely reacting to its consequences for
output and inflation. But VARs, even those that include interest rates to capture market expectations,
register the event as an exogenous shock.

31
 Nakamura and Steinsson also emphasize the oft-forgotten lesson of 1980s time-series econometrics, that
estimating 10-year responses from monthly VARs raised to the 120th power is dangerous. Such responses rely
crucially on one having in hand the persistent state variables, and accurately measuring their near-unit-root
dynamics. Direct regressions (“local projections”) of variables of interest on shocks are potentially more
reliable, but less parsimonious and therefore less well measured. They force one to recognize just how few
non-overlapping samples we have at the horizons of interest.
Is there even such a thing as an exogenous monetary policy shock? Central banks always say they are
reacting to something. At best, central banks might react to something that has no effect on inflation or
(separately) output. But nobody has taken this old suggestion seriously, preferring to try to find ephemeral
shocks that are orthogonal to everything. Romer and Romer (2023) define a shock as a time that Federal
Reserve officials start to take inflation seriously. But this is more a non-linear or state-dependent response
to inflation than a conventional flip-of-the-coin shock.
Moreover, the vast majority of monetary policy shocks in empirical work consist of events in which the
Federal Funds rate did not move at all, but forecasting variables predict a few basis points movement, or the
funds rate moves, say 25 bp, but the forecasting variables predict a few basis points more or less movement.
The measured effects of monetary policy on inflation then consist of the average inflation rate, say, three
years after such small regression errors. This is naturally a tenuous historical basis on which to measure the
question, what if the Fed were to surprise everyone with a large and sustained interest rate rise?
The vast majority of current estimates also find that identified monetary policy shocks lead to short-lived
interest rate responses. They therefore do not measure the effect of the persistent interest rate movements
that we need to evaluate the long-term debt mechanism, or to examine the long-run Fisher prediction that
inflation should eventually rise. They do not measure the effect of the persistent high real interest rates of
the one great episode, the early 1980s.
No matter how measured, responses to shocks completely miss the central ingredient of an intervention
such as the early 1980s. By design, “shocks” are deviations from a rule, not changes in rule or “regime”
itself that may durably change expectations. If the art of reducing inflation is to convince people that a
new regime has arrived, then the response to any monetary policy “shock” orthogonal to a stable “rule”
completely misses that policy. Clarida, Galı́, and Gertler (2000) famously attribute the end of 1970s inflation
to a change in the rule, a change in φπ in it = φπ πt + φx xt + ut , not to a big shock ut .
Uribe (2022) evaluates via an identified VAR the “neo-Fisherian” hypothesis that persistently higher
interest rates raise inflation. He identifies a permanent shock as one that increases both the nominal interest
rate and inflation in the long run. He finds that shock raises inflation and nominal interest rates in the
short run. Similarly, Schmitt-Grohé and Uribe (2022) find that permanent interest rate shocks depreciate
the currency. In both papers, transitory interest rate movements lead to the standard disinflation and
appreciation.
Their results are good news for Figure 2, in which a persistent interest rate rise raises inflation even in
the short run. But they are bad news for Figure 4 in which a transitory rate rise also raises inflation, and
Figure 7 in which a permanent interest rate rise lowers short-run inflation with long-term debt. They are
also bad news for the view that the 1980s succeeded via a highly persistent monetary policy shock. However,
Schmitt-Grohé and Uribe also did not attempt to find monetary policy shocks orthogonal to fiscal policy
changes, so their results do not directly bear on our question.
In sum, the empirical literature evaluating whether higher interest rates lower inflation is tenuous despite
enormous effort, and is not addressed at our central questions. Thus, current empirical work leaves open a
possibility: The widespread faith that higher interest rates reliably lower inflation, without fiscal help, may
simply not be true.

8. Paths to Follow

In sum, despite 50 years of intertemporal general equilibrium macroeconomics since Lucas (1972a), we
still don’t have well-agreed-on economic model of the conventional belief that higher nominal interest rates,

32
without a change in fiscal policy, can lower inflation even temporarily. The best I have been able to show
here—the rational-expectations fiscal-theory model with long-term debt—is far from the power of and works
by utterly different mechanisms than standard intuition, and is evidently far from widely accepted.
Naturally, one wants to add additional frictions and model ingredients in the hope of producing a
stronger effect, and one that perhaps matches standard intuition. But it’s not as easy as it looks.
One line of new-Keynesian literature explores a return to adaptive expectations, or complex learning
and expectation formation schemes which have a similar effect.9 This line of work has not been extended to
ask our question—whether higher interest rates can lower inflation without fiscal help—but it is a natural
avenue to explore. Others, especially in the policy and commentary world, simply assert that we should
simply revert to adaptive expectations ISLM and regard everything since as basically a failed effort. (For
example, Blinder (2024).)
In my little tour, adaptive or backward-looking expectations change the sign of the response of inflation
to interest rates by changing the basic stability and determinacy properties of the model. That’s major
surgery, not an epicycle. It requires an eigenvalue to cross one. Small parameter changes around the
frictionless rational benchmark won’t have the required effect. Conversely, the basic sign and stability of the
backward-looking expectations model must change along the way back to the frictionless rational benchmark.
Adaptive expectations is bounded away from a rational frictionless view.
Less than rational expectations are surely a reasonable ingredient for episodes, regime changes, or
transient dynamics. But if they are a minimum necessary ingredient for the basic sign and stability properties
of monetary policy, that undermines the robustness of our analysis. If this is our simplest model, we admit
that there is no simple economic model, no analogue to MV=PY, that delivers the basic sign and operation
of monetary policy. Maybe this is where we end up, but it’s at least worth considering if another path is
possible.
Whether people are truly “rational” and what that term even means does not really matter here. The
issue is expectations in the model, not expectations in the world. The issue is whether the expectations
in our simplest basic theory should be “model-consistent,” whether the expectations of that model are the
same as the expectations in that model (Lucas (1991)). Are economists really that much better forecasters
than everyone else? What happens if people catch on? Related, the key point of the “anchored” expectations
view is fundamentally whether expectations in the model are reactive. Rational expectations react instantly
to model shocks, as the model forecast changes. “Anchored” expectations in the policy view do not move.
You can’t tell just by looking out the window. Rational expectations may seem adaptive. Expectations
must be, in equilibrium, functions of variables that people observe, and past inflation is a likely to be
prominent among those variables. The point of “rational expectations” is only that those forecasting rules
are likely to change when the policy rule changes, as Lucas (1976) famously pointed out10 in his “critique.”
Adaptive expectations may even be model-consistent, until you change the model.
For example, consider the debate over 2021-2023 inflation. Survey, bond market, forecaster, and central
bank expectations were slow to move as observed inflation picked up. One mightPproclaim them adaptive, a
∞
slow-moving average of lagged inflation, as Friedman speculated in 1968, πte = j=1 αj πt−j . But the same

9 Prominent examples include Gabaix (2020) (see comments at Cochrane (2016)), Garcı́a-Schmidt and Woodford (2019), and

Bianchi-Vimercati, Eichenbaum, and Guerreiro (2022). Examples with learning: Marcet and Nicolini (2003) study recurrent
hyperinflations with learning. They explain why bouts of hyperinflation break out in the context of large and stable inflation,
with no large changes in seigniorage. In Orphanides and Williams (2005), learning contributes to inflation dynamics, and
allows private sector inflation expectations to provide useful information to the central bank. In Williams (2006), learning can
undermine the desirable properties of a price-level target, that people expect a period of inflation to follow deflation. In Bullard
and Mitra (2002), learning changes the optimal policy rule. Sargent (2001) “Conquest of American Inflation” is a classic example
in which learning by the Fed takes center stage. Evans and Honkapohja (2001) consider learning models broadly. McCallum
(2009a), McCallum (2009b), argue that the new-Keynesian explosive threat equilibrium selection is learnable, and fiscal theory
is not; Cochrane (2009) argues the opposite.
10 Sargent (1971) and Lucas (1972b) made the point earlier in the Phillips curve context. They showed that in a π =
t
απt−1 + κxt + ut regression, α < 1 coefficients can easily occur even when the true Phillips curve has rational expectations.
The coefficient α shifts if the monetary policy process shifts.

33
expectations could well have been rational: Output was temporarily higher than the reduced post-pandemic
potential, so current inflation was higher than rationally expected future inflation (πt = Et πt+1 + κxt ). In
both interpretations, expected inflation is lower than current inflation.
“Rational,” “model-consistent,” or “reactive” expectations do not mean “clairvoyant,” or even partic-
ularly good forecasts. Rational people may be quite bad at forecasting inflation, in life and in models. With
noisy signals, rational forecast errors can even be serially correlated (Coibon and Gorodnichenko (2012)).
Inflation, like stocks, may simply be hard to forecast. The Fed and sophisticated market participants are
pretty bad at forecasting inflation too. One can regard rational, model-consistent, expectations as simply a
humility principle reflecting the fact that economists armed with structural models do not outperform the
supposedly irrational agents we survey.
Of course, empirical performance rather than theoretical beauty is the most important test. To move
back towards adaptive expectations, we would also need to face the failures of the adaptive expectations /
unstable approach during the zero bound era, in 1970s stagflation, in the relatively rapid end of inflation
in 1982, in the success of 1990s inflation targets, in the ends of hyperinflations (Sargent (1982)), and in the
failure of inflation to spiral upwards in 2022-2023, among other episodes.
One is drawn to add other model ingredients. Surely in the DSGE smorgasbord there are enough
ingredients to come up with a temporary negative sign. That is, I think, the right answer, in the sense that
we need to know if it works or that it does not. My point is, it has not yet been done. Many model-implied
monetary-policy response functions have been computed of course, but they do not yet hold fiscal policy
constant in an interesting way, or look into the footnote about lump-sum taxes to see just what those are
and to what extent inflation reduction comes from an implicit fiscal contraction. The literature that puts
fiscal theory in explicit DSGE models is an exception; see for example most recently Bianchi and Melosi
(2022), Chen, Leeper, and Leith (2022), and Leeper (2021). This literature is explicit about fiscal-monetary
coordination. However, it has focused on switching between a highly restricted active-fiscal and active-money
regimes, and so far has not addressed the question whether higher interest rates can lower inflation with no
change in fiscal policy.
One easily jumps to consider capital with adjustment costs, financial frictions, credit constraints, portfo-
lio adjustment costs, money, liquidity effects in government bonds, additional price and wage-setting frictions,
strategic complementarities, individual heterogeneity, and other model complications. But the negative re-
sponse of inflation to interest rates should be a robust and deeply rooted phenomenon, one that will not
vanish if, for example, the US changes working capital financing traditions or the downpayment rules on
mortgages. In his Nobel Lecture, Lucas (1996) cites David Hume for understanding the neutrality and
non-neutrality of money in 1752. Velde (2009) documents a beautiful non-neutrality episode in 1724 France,
with a monetary and financial system utterly unlike our own.
It may to be possible to find in these generalizations sufficient conditions to deliver the negative sign,
with enough model complications. Our goal though is the minimum necessary conditions, that apply most
broadly and robustly, that we can put at the basis of any good theory of inflation and monetary policy
under interest rate targets. That is a harder goal. Again, Lucas (1972a) is a great example. In his economy,
the flexible price version leads to super-neutrality: An increase in money just raises the price level. Lucas
put in one “friction,” imperfect information about aggregates, leading to a confusion between relative and
aggregate price movements.
Adding to the difficulty, standard new-Keynesian models have struggled to produce the kind of response
one sees in VAR estimates, and that represent traditional beliefs, even allowing a contemporaneous “passive”
fiscal response. In the estimates and common beliefs, after an interest rate rise inflation either rises or stays
constant for a year, and then falls over time. Interest rates, like money, act with long and variable lags. In
the theory, inflation jumps down, and rises over time (Ball (1994) first drew attention to this contrast. The
Phillips curve is πt = βEt πt+1 + κxt . Higher interest rates lower xt , which lowers πt relative to Et πt+1 .
Equilibrium selection chooses a πt lower than it otherwise would have been, but inflation rises over time.)
Current medium-scale new-Keynesian models can reproduce something like standard beliefs and VARs,
if not by the standard mechanisms. Christiano, Eichenbaum, and Evans (2005), Christiano, Eichenbaum,

34
and Trabandt (2016), Christiano, Eichenbaum, and Trabandt (2018), and Smets and Wouters (2007), are
classic examples. But they include a long list of additional ingredients. Many of those ingredients are also
quite different from standard microeconomic specifications, troubling both for the project that microfoun-
dations should be micro founded, and for whether the specifications really are policy invariant. Some of the
ingredients which seem crucial to producing the inflation response include: 1) A strong one-period habit in
utility, which essentially adds a time derivative on the relation between consumption growth and interest
rates. 2) Sticky wages, driven by labor monopoly power, in contrast to the usual microeconomic worry about
labor monopsony. 3) Firms must borrow money to pay the next period’s wage bill in advance. Higher real
interest rates then raise marginal costs in the Phillips curve. This raises inflation now πt relative to expected
future inflation βEt πt+1 , but with equilibrium selection and timing rules holding πt from moving, that means
βEt πt+1 falls. 4) Capital with adjustment costs that depend on investment growth, not the investment to
capital ratio as in most microeconomic and asset pricing literature. 5) Wages and prices are indexed, so
that those who cannot adjust nonetheless react to inflation. This modification puts lagged inflation in the
Phillips curve. But few prices and wages in the US are actually indexed. More recent literature that matches
impulse response functions, for example, Auclert, Rognlie, and Straub (2020), adds still more ingredients, in
this case heterogeneous agents.
Are all these ingredients necessary for the basic sign and pattern of monetary policy in standard New-
Keynesian models? And what will we need when we separate monetary and fiscal policy? Those are good and
important questions, which have not yet been answered. Authors in this literature (and perhaps editors and
referees) have been good at adding ingredients, not so much in seeing which ingredients can be subtracted
while keeping the important story intact.
The Phillips curve seems an obvious weak point, and locus of improvement in this quest. In my
little models, the Phillips curve is the central source of price dynamics. That point is particularly clear in
continuous time, where the IS curve is the same for rational and adaptive expectations. Yet the Phillips
curve has not achieved great theoretical and empirical clarity, despite decades of dedicated work.
Its status is ephemeral in policy analysis. The Phillips curve is said to be “flat,” κ ≈ 0 and unreliable.
Then inflation depends only expected inflation. But expected inflation can’t just come from speeches, an-
nouncements, or interest rates interpreted as signals of central bank intentions. Signals of what? Eventually
some central bank action must lower inflation.
Mankiw and Reis (2002), (2006), (2007) pursue a “sticky information” Phillips curve. It is particularly
attractive in this context since they motivate it to reverse the Ball (1994) puzzle and to produce inflation
that rises over time when output is high. It is essentially a generalization of the Lucas (1972a) model, that
allows price-setters take more than a period to learn the aggregate price level. They relate output to current
inflation relative to a distributed lag of expected inflation rather than the previous period’s inflation. It
solves the problem that Lucas (1972a) only produces a one-period output movement. However, Mankiw and
Reis (2002) merge their Phillips curve with money supply control; Mankiw and Reis (2006) doesn’t present
any results; and Mankiw and Reis (2007), which merges their Phillips curve with an interest rate rule, doesn’t
actually produce the desired result. Inflation and output move essentially together, and inflation is high and
declining when the level of output is high. Perhaps this attractive ingredient will in the future produce the
desired sign and pattern of the inflation response, but it does not yet do so.
In addition to wondering what ingredients to put in, then, perhaps the Phillips curve is one we can
take out or avoid. Our goal is to understand the dynamic relationship between interest rates and inflation,
πt = a(L)it . The Phillips curve came from thinking about output and employment effects of a given inflation.
Lucas (1972a) had a theory of inflation, M V = P Y . He wanted a theory of how inflation affects output.
Interest rate based models reverse the causal logic. They use the IS equation to describe how interest rates
move output, and then the Phillips curve to describe how output affects inflation. (With the usual caveats
for causal readings of equilibrium conditions.) But the Phillips curve wasn’t designed to be the central
mechanism for inflation dynamics.
Moreover, Lucas wrote following Friedman, in the view that monetary policy shocks accounted for the
majority of business cycle fluctuations. The VAR literature and history since the 1980s emphatically overturn
that view. Monetary policy shocks account for tiny forecast errors of output. The 2008 and 2020 recessions

35
were obviously not caused by monetary policy shocks. Understanding how monetary policy contributes to
or controls inflation without tying it to a specific and large output effect is more interesting today than it
would have been in 1972.
Consider, then, what kind of inflation dynamics would we observe in an endowment economy? The
Phillips curve approach says that with constant output and a constant real interest rate, nominal interest
rates translate immediately to expected inflation just as if prices are completely flexible. Are there really no
inflation dynamics in an endowment economy, or one with a constant real interest rate? Perhaps somewhere
in the new interest in supply chain dynamics, (for example Lane (2022) ), reallocation shocks (for example,
Guerrieri et al. (2021) ) or production networks (for example, Minton and Wheaton (2022), Rubbo (2022),
and La’O and Tahbaz-Salehi (2023)) a qualitatively different model of inflation dynamics will emerge.
Constraining all this speculation, however, two large obstacles stand in the way of the goal that higher
nominal interest rates reduce inflation, without fiscal help. First, i = r + π e , so a 1 percentage point higher
nominal interest rate i must raise the real interest rate r by more than percentage point to lower expected
inflation. Unlike MV=PY, it is not enough for the higher nominal rate to be split between a real rate and
expected inflation.
Second, the basic fiscal constraints on the path of inflation presented in Section 6, and (83) in particular
remain. There we found that monetary policy in every model of inflation under an interest rate target is
limited to unpleasant arithmetic. It can only shift inflation over time, producing lower inflation at one date
in return for higher inflation at another date.
There is again another possibility: It might not be true. The long-term debt model, with unpleasant
interest-rate arithmetic or “stepping on a rake” may be as close as we come to standard intuition. Otherwise,
nominal interest rate rises, with no change in fiscal policy, may not lower inflation even in the short run.
Of course actual monetary policy has fiscal effects, via seigniorage, via liquidity frictions on government
debt generally, via induced fiscal policies to pay interest on debt and windfalls to bondholders, by inducing
stimulus in recessions and austerity following inflations. Monetary policy with such fiscal effects may produce
interesting inflation responses. Perhaps experience of monetary policy’s actual effectiveness comes through
these fiscal channels. But understanding “monetary policy” via its contemporaneous or induced fiscal effects
obviously leads us to a quite different understanding of its economic mechanism, and of macroeconomic
policy more generally.

9. Conclusion

What is the dynamic effect of interest rates on inflation, πt = a(L)it , in our world, in which central
banks set nominal interest rates, provide abundant reserves at market interest, do not control money supplies,
do not make equilibrium-selection threats, and cannot directly change fiscal policy? And, of course, after
that, how do interest rates then affect output, employment, and other variables?
I have followed one line of thought on these questions to its logically inevitable conclusion: Rational
expectations and fiscal underpinnings of monetary policy imply that inflation is stable and determinate in
the long run. Those features imply that pegs are possible, and that higher interest rates, without a change
in fiscal policy, eventually raise inflation. There may well be a short-run negative effect of interest rates on
inflation. I show one such model, but we need to know if there are better yet simple models of that effect.
Or, we need to accept the limited power of that model, and that otherwise higher nominal interest rates
without changes in fiscal policy raise inflation.
Thus, as I see it, we have made a lot of progress. We’re finally at the launch pad. We have, at last,
a fully economic theory of inflation under interest rate targets. It is stable, and neutral in the long run. A
simple sticky price elaboration gives output effects, and we have one model of a temporary negative inflation
effect. We have a firm foundation for elaboration and exploration. However, the theory does not behave in
the slightest the way verbal analysis of policy analysts and central banks does, which is largely an elaboration
of adaptive expectations ideas. The basic sign, stability, and economic mechanisms are completely opposite.

36
 But all this is novel, sort of known but not widely recognized, or known but controversial. Thus,
even these basic questions are, in fact, still up for grabs. These are the good old days. Low-hanging fruit
abounds. We’re really at the beginning stages where simple models need exploration, not, as it appears, in a
mature stage where essentials are settled and all there is to do is to add to the immense stock of complicated
epicycles. However, given the state of actual agreed-on knowledge, central banks’ proclamations of detailed
technocratic ability to manipulate delicate frictions is really not justified.11

11 For a lovely example, see the chart at https://www.ecb.europa.eu/mopo/intro/transmission/html/index.en.html. Lucas

(1979) is a good antidote to hubris.

37
References
Auclert, Adrien, Rodolfo Rigato, Matthew Rognlie, and Ludwig Straub. 2022. “New Pricing Models, same
Old Phillips Curves?” Manuscript .
Auclert, Adrien, Matthew Rognlie, and Ludwig Straub. 2020. “Micro Jumps, Macro Humps: Monetary
Policy and Business Cycles in an Estimated HANK Model.” Manuscript.
Ball, Laurence. 1994. “Credible Disinflation with Staggered Price-Setting.” The American Economic Review
84 (1):282–289.
Bianchi, Francesco and Leonardo Melosi. 2019. “The Dire Effects of the Lack of Monetary and Fiscal
Coordination.” Journal of Monetary Economics 104:1–22.
———. 2022. “Inflation as a Fiscal Limit.” Manuscript .
Bianchi-Vimercati, Riccardo, Martin Eichenbaum, and Joao Guerreiro. 2022. “Fiscal Policy at the Zero
Lower Bound without Rational Expectations.” Manuscript .
Blinder, Alan S. 2024. A Monetary and Fiscal History of the United States, 1961–2021. Princeton, NJ:
Princeton University Press.
Bordo, Michael, John H. Cochrane, and John Taylor. 2022. How Monetary Policy Got Behind the Curve—and
How to Get Back. Stanford, CA: Hoover Institution Press.
Bullard, James and Kaushik Mitra. 2002. “Learning about monetary policy rules.” Journal of Monetary
Economics 49 (6):1105–1129.
Calvo, Guillermo A. 1983. “Staggered prices in a utility-maximizing framework.” Journal of Monetary
Economics 12 (3):383–398.
Chen, Xiaoshan, Eric M. Leeper, and Campbell Leith. 2022. “Strategic interactions in U.S. monetary and
fiscal policies.” Quantitative Economics 13 (2):593–628.
Christiano, Lawrence J., Martin Eichenbaum, and Charles Evans. 2005. “Nominal Rigidities and the Dynamic
Effects of a Shock to Monetary Policy.” Journal of Political Economy 113:1–45.
Christiano, Lawrence J., Martin S. Eichenbaum, and Mathias Trabandt. 2016. “Unemployment and Business
Cycles.” Econometrica 84 (4):1523–1569.
———. 2018. “On DSGE Models.” Journal of Economic Perspectives 32 (3):113–40.
Clarida, Richard, Jordi Galı́, and Mark Gertler. 2000. “Monetary Policy Rules and Macroeconomic Stability:
Evidence and Some Theory.” Quarterly Journal of Economics 115:147–180.
Cochrane, John H. 1998. “A Frictionless View of U.S. Inflation.” NBER Macroeconomics Annual 13:323–384.
———. 2009. “Can Learnability Save new-Keynesian Models?” Journal of Monetary Economics 56 (8):1109–
1113.
———. 2011. “Determinacy and Identification with Taylor Rules.” Journal of Political Economy 119:565–
615.
———. 2016. “Comments on ‘A Behavioral New-Keynesian Model’ by Xavier Gabaix.” Manuscript.
———. 2017. “The New-Keynesian Liquidity Trap.” Journal of Monetary Economics 92:47–63.
———. 2018. “Michelson-Morley, Fisher, and Occam: The Radical Implications of Stable Quiet Inflation at
the Zero Bound.” NBER Macroeconomics Annual 32:113–226.
———. 2021. “A Fiscal Theory of Monetary Policy with Partially Repaid Long-Term Debt.” Review of
Economic Dynamics 45:22–40.

38
———. 2022a. “Fiscal Histories.” Journal of Economic Perspectives 36 (4):125–146.
———. 2022b. “Inflation Past, Present and Future: Fiscal Shocks, Fed Response and Fiscal Limits.” In How
Monetary Policy Got Behind the Curve—and How To Get Back, edited by Michael D. Bordo, John H.
Cochrane, and John Taylor. Stanford, CA: Hoover Institution Press.

———. 2023. The Fiscal Theory of the Price Level. Princeton NJ: Princeton University Press.
———. 2024. “Fiscal Narratives for US Inflation.” Manuscript URL https://www.johnhcochrane.com/
research-all/sims-comment.
Coibon, Olivier and Yuriy Gorodnichenko. 2012. “What Can Survey Forecasts Tell Us about Information
Rigidities?” Journal of Political Economy 120 (1):116–159.

Evans, George W. and Seppo Honkapohja. 2001. Learning and Expectations in Macroeconomics. Princeton:
Princeton University Press.
Friedman, Milton. 1968. “The Role of Monetary Policy.” The American Economic Review 58:1–17.
———. 1969. “The Optimum Quantity of Money.” In The Optimum Quantity of Money and Other Essays.
Chicago: Aldine, 1–50.
Gabaix, Xavier. 2020. “A Behavioral New Keynesian Model.” American Economic Review 110:2271–2327.
Galı́, Jordi. 2009. Monetary Policy, inflation, and the Business Cycle: An introduction to the new Keynesian
Framework. Princeton University Press.

Garcı́a-Schmidt, Mariana and Michael Woodford. 2019. “Are Low Interest Rates Deflationary? A Paradox
of Perfect-Foresight Analysis.” American Economic Review 109:86–120.
Gordon, Robert J. 1970. “The Recent Acceleration of Inflation and Its Lessons for the Future.” Brookings
Papers on Economic Activity 1970 (1):8–47.

———. 1976. “Recent developments in the theory of inflation and unemployment.” Journal of Monetary
Economics 2 (2):185–219.
Guerrieri, Veronica, Guido Lorenzoni, Ludwig Straub, and Iván Werning. 2021. “Monetary Policy in Times
of Structural Reallocation.” Manuscript .
King, Robert G. 2000. “The New IS-LM Model: Language, Logic, and Limits.” Federal Reserve Bank of
Richmond Economic Quarterly 86 (3):45–104.
Lane, Philip R. 2022. “Inflation Diagnostics.” URL https://www.ecb.europa.eu/press/blog/date/2022/
html/ecb.blog221125~d34babdf3e.en.html.
La’O, Jennifer and Alireza Tahbaz-Salehi. 2023. “Optimal Monetary Policy in Production Networks.” Econo-
metrica 90:1295–1336.
Leeper, Eric M. 1991. “Equilibria Under ‘Active’ and ‘Passive’ Monetary and Fiscal Policies.” Journal of
Monetary Economics 27:129–147.
———. 2021. “Shifting Policy Norms and Policy Interactions.” Manuscript .

Lipsey, Richard G. 1960. “The Relation between Unemployment and the Rate of Change of Money Wage
Rates in the United Kingdom, 1862-1957: A Further Analysis.” Economica 27 (105):1–31.
Lucas, Robert E. 1972a. “Expectations and the Neutrality of Money.” Journal of Economic Theory 4:103–
124.
———. 1972b. “Testing the Natural Rate Hypothesis.” In The Econometrics of Price Determination, edited
by Otto Eckstein. Washington, DC: Federal Reserve Board, 50–59.

39
———. 1976. “Econometric Policy Evaluation: A Critique.” Carnegie-Rochester Conference Series on Public
Policy 1:19–46.
———. 1979. “A Report to the OECD by a Group of Independent Experts OECD, June 1977: A Review.”
Carnegie-Rochester Conference Series on Public Policy 11:161–168.

———. 1991. Models of Business Cycles. Wiley.

———. 1996. “Nobel Lecture: Monetary Neutrality.” Journal of Political Economy 104 (4):661–682.
Mankiw, N. Gregory and Ricardo Reis. 2002. “Sticky Information versus Sticky Prices: A Proposal to
Replace the New Keynesian Phillips Curve.” Quarterly Journal of Economics 117:1295–1328.

———. 2006. “Pervasive Stickiness.” American Economic Review 96 (2):164–169.

———. 2007. “Sticky Information in General Equilibrium.” Journal of the European Economic Association
5 (2-3):603–613.
Marcet, Albert and Juan P. Nicolini. 2003. “Recurrent Hyperinflations and Learning.” American Economic
Review 93 (5):1476–1498.

McCallum, Bennett T. 1981. “Price Level Determinacy with an Interest Rate Policy Rule and Rational
Expectations.” Journal of Monetary Economics 8:319–329.
———. 2009a. “Inflation Determination with Taylor Rules: Is New-Keynesian Analysis Critically Flawed?”
Journal of Monetary Economics 56 (8):1101–1108.

———. 2009b. “Rejoinder to Cochrane.” Journal of Monetary Economics 56 (8):1114–1115.

Minton, Robert and Brian Wheaton. 2022. “Hidden Inflation in Supply Chains: Theory and Evidence.”
Manuscript, Harvard University and UCLA .
Nakamura, Emi and Jón Steinsson. 2018. “Identification in Macroeconomics.” Journal of Economic Per-
spectives 32 (3):59–86.
Nelson, Edward. 2020. Milton Friedman and Economic Debate in the United States, 1932–1972, Volume 2.
Chicago: University of Chicago Press.
Orphanides, Athanasios and John C. Williams. 2005. “Inflation scares and forecast-based monetary policy.”
Review of Economic Dynamics 8 (2):498–527. Monetary Policy and Learning.

Phelps, Edmund S. 1967. “Phillips Curve, Expectations of Inflation and Optimal Unemployment Over
Time.” Economica 34:254–281.
Ramey, Valerie. 2016. “Macroeconomic Shocks and Their Propagation.” In Handbook of Macroeconomics
Vol. 2, edited by John B. Taylor and Harald Uhlig. Amsterdam: Elsevier, 71–162.

Romer, Christina D. and David H. Romer. 2023. “Does Monetary Policy Matter? The Narrative Approach
After 35 Years.” NBER working Paper 31170 .
Rotemberg, Julio J. 1982. “Sticky Prices in the United States.” Journal of Political Economy 90 (6):1187–
1211.

Rubbo, Elisa. 2022. “Networks, Phillips Curves and Monetary Policy.” Manuscript, University of Chicago .
Samuelson, Paul A. and Robert M. Solow. 1960. “Analytical Aspects of Anti-Inflation Policy.” The American
Economic Review 50 (2):177–194.
Sargent, Thomas J. 1971. “A Note on the ”Accelerationist” Controversy.” Journal of Money, Credit and
Banking 3 (3):721–725.

40
———. 1982. “The Ends of Four Big Inflations.” In Inflation: Causes and Effects, edited by Robert E. Hall.
University of Chicago Press, for the NBER, 41–97.
———. 2001. The Conquest of American Inflation. Princeton, NJ: Princeton University Press.
Sargent, Thomas J. and Neil Wallace. 1975. “‘Rational’ Expectations, the Optimal Monetary Instrument,
and the Optimal Money Supply Rule.” Journal of Political Economy 83 (2):241–254.
———. 1981. “Some Unpleasant Monetarist Arithmetic.” Federal Reserve Bank of Minneapolis Quarterly
Review 5:1–17.
Schmitt-Grohé, Stephanie and Martı́n Uribe. 2022. “The effects of permanent monetary shocks on exchange
rates and uncovered interest rate differentials.” Journal of International Economics 135:103560.

Sims, Christopher A. 2011. “Stepping on a Rake: The Role of Fiscal Policy in the Inflation of the 1970s.”
European Economic Review 55:48–56.
———. 2024. “Origins of US Inflation Since 1950.” Manuscript, presented at 2024 AEA Session “Under-
standing the Return of Inflation” .

Smets, Frank and Raf Wouters. 2007. “Shocks and Frictions in US Business Cycles: A Bayesian DSGE
Approach.” American Economic Review 97:586–606.
Steinsson, Jón and Emi Nakamura. 2013. “Price Rigidity: Microeconomic Evidence and Macroeconomic
Implications.” Annual Review of Economics :133–163.

Taylor, John B. 1993. “Discretion Versus Policy Rules in Practice.” Carnegie-Rochester Conference Series
on Public Policy 39:195–214.
———. 1999. “The Robustness and Efficiency of Monetary Policy Rules as Guidelines for Interest Rate
Setting by the European Central Bank.” Journal of Monetary Economics 43:655–679.
Uribe, Martı́n. 2022. “The Neo-Fisher Effect: Econometric Evidence from Empirical and Optimizing Mod-
els.” American Economic Journal: Macroeconomics 14 (3):133–62.
Velde, François R. 2009. “Chronicle of a Deflation Unforetold.” Journal of Political Economy 117 (4):591–
634.
Wicksell, Knut. 1898. Geldzins und Güterpreise. Jena, Germany: Gustav Fischer.

———. 1965. Interest and Prices. New York: Augustus M. Kelley.

Williams, John C. 2006. “Monetary Policy in a Low Inflation Economy with Learning.” In Monetary
Policy in an Environment of Low Inflation; Proceedings of the Bank of Korea International Conference
2006. Seoul, Korea: The Bank of Korea, 199–228. Reprinted in Federal Reserve Bank of San Francisco
Economic Review, 2010.

Woodford, Michael. 1994. “Monetary Policy and Price Level Determinacy in a Cash-in-Advance Economy.”
Economic Theory 4:345–380.
———. 1995. “Price-Level Determinacy Without Control of a Monetary Aggregate.” Carnegie-Rochester
Conference Series on Public Policy 43:1–46.

———. 2001. “Fiscal Requirements for Price Stability.” Journal of Money, Credit and Banking 33:669–728.
———. 2003. Interest and Prices. Princeton: Princeton University Press.

41
 “Expectations and the Neutrality of Interest Rates”
Online Appendix
John H. Cochrane
May 9, 2024

A1. Solving the Model for Arbitrary Interest Rates

The model is

xt = −σ(it − Et πt+1 ) (A1)

πt = Et πt+1 + κxt (A2)
ρvt+1 = vt + it − πt+1 − s̃t+1 (A3)
T
lim ρ vT = 0 (A4)
T →∞

We want to calculate the impulse-response function for a generic path {it }. All variables are zero until
time 1. At time 1 we set off a sequence {i1 , i2 , ...}. There is no change to surpluses, so s̃t = 0. Given π1 ,
the other πt follow since there is no more uncertainty. Equations (A1)-(A2) give us a set of possible paths
of inflation indexed by π1 . We use (A3) and (A4) to choose π1 .
This section establishes the following results for this impulse-response function. For an arbitrary se-
quence {i1 , i2 , ...},
∞ t
1 X
j σκ X 1
πt+1 = t+1 (1 − ρ) ρ ij + i .
t−j j
(1 + σκ) j=1
1 + σκ j=1 (1 + σκ)

For an AR(1) it = ηit−1 + εt ,

  
(1 − ρ) ρ σκ 1 σκ t
πt+1 = + − η i1 .
(1 + σκ) (1 − ρη) 1 − η (1 + σκ) (1 + σκ)t 1 − η (1 + σκ)

For η = 1, i.e. a one-time permanent increase in the interest rate,

" #
(1 + σκ − ρ)
πt+1 = 1 − t+1 i1
(1 + σκ)

Now, to derive these results. Eliminating output from (A1)-(A2),

1 σκ
Et πt+1 = πt + it . (A5)
1 + σκ 1 + σκ
(A6)

Iterating forward (A5), after the shock at time 1, (for t ≥ 1),

t
1 σκ X 1
πt+1 = t π1 + ij . (A7)
(1 + σκ) 1 + σκ j=1 (1 + σκ)t−j

In the case of AR(1), it = ηit−1 + εt , we have the not very elegant expression
t
1 σκ X η j−1
πt+1 = t π 1 + i1
(1 + σκ) 1 + σκ j=1 (1 + σκ)t−j

42
 1
1 (1+σκ)t
− ηt σκ
πt+1 = t π1 + 1 i1
(1 + σκ) (1+σκ) − η 1 + σκ
If η = 1, so πt = π1 , t > 1, this reduces to
1
πt+1 = i1 + t (π1 − i1 ) .
(1 + σκ)

Now, we need to find π1 . Iterating (A3) forward,

ρt vt = (0 − π1 ) + ρ (i1 − π2 ) + ρ2 (i2 − π3 ) + ρ3 (i3 − π4 ) + ...

Thus, the condition ρt vt → 0 is

∞
X
π1 = ρj (ij − πj+1 ) .
j=1

Debt is devauled to pay the higher interest costs that result from higher real interest rates. Now plug inflation
from (A7),
∞ j
!!
X
j 1 σκ X 1
π1 = ρ ij − π +
j 1
i
j−k k
(1 + σκ) 1 + σκ
j=1 k=1 (1 + σκ)

∞ ∞ ∞ j
X 1 X σκ X j X 1
π1 = − ρj j
π 1 + ρ j
ij − ρ i
j−k k
(1 + σκ) 1 + σκ j=1
j=1 j=1 k=1 (1 + σκ)
∞ ∞ ∞ ∞
X 1 X σκ X X j 1
π1 = − ρj π
j 1
+ ρ j
i j − ρ i
j−k k
j=1 (1 + σκ) j=1
1 + σκ (1 + σκ)
k=1 j=k
∞ ∞ ∞
!
X
j 1 X
j σκ X k 1
π1 = − ρ π +
j 1
ρ ij − ρ ρ ik
j=1 (1 + σκ) j=1
1 + σκ 1 − 1+σκ
k=1
ρ
!! ∞
(1+σκ) σκ 1 X
π1 = − ρ π1 + 1 − ρ ρj ij
1 − (1+σκ) 1 + σκ 1 − 1+σκ j=1
∞
ρ 1−ρ X j
π1 = − π1 + ρ ij
1 + σκ − ρ 1 + σκ − ρ j=1
∞
X
(1 + σκ − ρ) π1 = −ρπ1 + (1 − ρ) ρj ij
j=1
∞
1−ρ X
π1 = ρj ij . (A8)
1 + σκ j=1

For an AR(1)
∞
1 − ρ X j j−1 ρ 1−ρ
π1 = ρ η i1 = i1 .
1 + σκ j=1 1 + σκ 1 − ρη

For η = 1
ρ
π1 = i1 .
1 + σκ

With π1 , we now have the general solution. Using (A8) in (A7),

∞ t
1 X
j σκ X 1
πt+1 = t+1 (1 − ρ) ρ ij + i .
t−j j
(1 + σκ) j=1
1 + σκ j=1 (1 + σκ)

43
 For the AR(1)
t
η j−1
 
1 1−ρ ρ σκ X
πt+1 = t i1 + i1
(1 + σκ) 1 + σκ 1 − ρη 1 + σκ j=1 (1 + σκ)t−j

1 t
σκ (1+σκ)t − η
 
1−ρ 1 ρ
πt+1 = t +
i1 1 i1
1 + σκ 1 − ρη
(1 + σκ) 1 + σκ 1+σκ −η
  
(1 − ρ) ρ σκ 1 σκ t
πt+1 = + − η i1
(1 + σκ) (1 − ρη) 1 − η (1 + σκ) (1 + σκ)t 1 − η (1 + σκ)

For η = 1, " #
(1 + σκ − ρ)
πt+1 = 1 − t+1 i1 .
(1 + σκ)

A1.1. Continuous time solution

Here I derive the general solution for inflation given an arbitrary path of interest rates in continuous
time. The model is (38) - (41)
Et dπt = −σκπt dt + σκit dt, (A9)
dvt = (rvt + it − πt − s̃t )dt. (A10)
T
lim Et ρ vT = 0
T →∞

We study a single shock at time 0, followed by perfect foresight, with s˜t = 0. Solving the first and second
equations separately, we have (54)-(55)
Z t
πt = σκ e−σκτ it−τ dτ + e−σκt π0 (A11)
τ =0
Z ∞
0= e−rt (it − πt )dt. (A12)
t=0

Express (A11) as
Z t
πt = σκ e−σκ(t−τ ) iτ dτ + e−σκt π0 . (A13)
τ =0

Substitute in (A12) and solve for π0 :

Z ∞ Z ∞ Z t Z ∞
−rt −rt −σκ(t−τ )
0= e it dt − σκ e e iτ dτ dt − π0 e−rt e−σκt dt. (A14)
t=0 t=0 τ =0 t=0
Z ∞ Z ∞ Z t Z ∞
0= e−rt it dt − σκ e−rt e−σκ(t−τ ) iτ dτ dt − π0 e−rt e−σκt dt. (A15)
t=0 t=0 τ =0 t=0
Z ∞ Z ∞ Z ∞ Z ∞
0= e−rt it dt − σκ e−rt e−σκ(t−τ ) iτ dτ dt − π0 e−rt e−σκt dt. (A16)
t=0 τ =0 t=τ t=0
Z ∞ Z ∞ Z ∞ Z ∞
0= e−rt it dt − σκ eσκτ iτ e−rt e−σκt dτ dt − π0 e−rt e−σκt dt. (A17)
t=0 τ =0 t=τ t=0
∞ ∞
e−(r+σκ)τ
Z Z
1
0= e−rt it dt − σκ eσκτ iτ dτ − π0 . (A18)
t=0 τ =0 r + σκ r + σκ
∞ ∞
e−(r+σκ)τ
Z Z
1
0= e−rt it dt − σκ eσκτ iτ dτ − π0 . (A19)
t=0 τ =0 r + σκ r + σκ

44
 Z ∞ Z ∞
σκ 1
0= e−rt it dt − e−rτ iτ dτ − π0 . (A20)
t=0 r + σκ τ =0 r + σκ
Z ∞
r 1
0= e−rt it dt − π0 . (A21)
r + σκ t=0 r + σκ
Z ∞
π0 = r e−rt it dt. (A22)
t=0

Then, back to (A11),

Z t Z ∞
−σκτ −σκt
πt = σκ e it−τ dτ + e r e−rτ iτ dτ. (A23)
τ =0 τ =0
or Z t Z ∞
−σκ(t−τ ) −σκt
πt = σκ e iτ dτ + e r e−rτ iτ dτ. (A24)
τ =0 τ =0

For an AR(1)
it = i0 e−ηt ,
we have Z t Z ∞
−σκ(t−τ ) −ητ −σκt
πt = σκ e i0 e dτ + e r e−rτ i0 e−ητ dτ. (A25)
τ =0 τ =0
 Z t Z ∞ 
−σκt (σκ−η)τ −σκt −(r+η)τ
πt = σκe e dτ + e r e dτ i0 . (A26)
τ =0 τ =0

e(σκ−η)t − 1 −σκt
 
r −σκt
πt = σκ e + e i0 . (A27)
σκ − η r+η
 
σκ r −σκt
πt = [e−ηt − e−σκt ] + e i0 . (A28)
σκ − η r+η
   
σκ −ηt r σκ
πt = e + − e−σκt i0 . (A29)
σκ − η r + η σκ − η
   
σκ −ηt r(σκ − η) − σκ(r + η) −σκt
πt = e + e i0 . (A30)
σκ − η (r + η)(σκ − η)
 
σκ −ηt η(r + σκ) −σκt
πt = e − e i0 . (A31)
σκ − η (r + η)(σκ − η)
 
1 r + σκ −σκt
πt = σκe−ηt − ηe i0 . (A32)
σκ − η r+η

45