Monetary Policy and

Inflation Expectations
High Frequency Evidence from Brazil

Carlos Goncalves, Mauro Rodrigues and Fernando Genta

WP/25/48

IMF Working Papers describe research in

progress by the author(s) and are published to
elicit comments and to encourage debate.
The views expressed in IMF Working Papers are
those of the author(s) and do not necessarily
represent the views of the IMF, its Executive Board,
or IMF management.

2025
FEB
 © 2025 International Monetary Fund WP/25/48

IMF Working Paper

Fiscal Affairs Department

Monetary Policy and Inflation Expectations: High-Frequency Evidence from Brazil

Prepared by Carlos Goncalves, Mauro Rodrigues and Fernando Genta

Authorized for distribution by Davide Furceri

February 2025

IMF Working Papers describe research in progress by the author(s) and are published to elicit
comments and to encourage debate. The views expressed in IMF Working Papers are those of the
author(s) and do not necessarily represent the views of the IMF, its Executive Board, or IMF management.

ABSTRACT: We investigate the impact of high frequency monetary policy shocks in Brazil using daily data
and Rigobon’ s identification via heteroskedasticity. We show that positive changes in interest rates cause
inflation expectations to decline and the exchange rate to appreciate. To the best of our knowledge, this
is the first paper to study how monetary policy affects inflation expectations in an emerging economy
using high frequency identification techniques.

JEL Classification Numbers: E52, E31

Keywords: Monetary policy; inflation expectations; Brazil

cgoncalves@[Link], mauror@[Link],
Author’s E-Mail Address:
fernandogentasantos@[Link]
WORKING PAPERS

Monetary Policy and Inflation

Expectations
High-Frequency Evidence from Brazil

Prepared by Carlos Goncalves, Mauro Rodrigues and Fernando Genta1

1
We thank Francesco Bianchi, Carlos Carvalho, Daniel Leigh, Rodrigo Valdez and participants in the IMF WHD
seminar for helpful suggestions. All remaining errors are our own.
IMF WORKING PAPERS Title of WP

Glossary
CB Central Bank
Copom Monetary Policy Committee (acronym in Portuguese)
FOMC Federal Open Markets Committee
HFI High-Frequency Identification
UIP Uncovered Interest Parity

INTERNATIONAL MONETARY FUND 2

IMF WORKING PAPERS Title of WP

Executive Summary
The last decade witnessed a substantial increase in the number of studies using high
frequency data to identify the effects of monetary policy shocks on the economy. The
overall lesson is that the old SVAR literature was unable to accurately represent the
impact of monetary policy shocks on the economy. But High Frequency Identification
(HFI) has rarely been deployed to study monetary transmission in Emerging Economies,
a gap we help bridge in this paper.

There are three distinctive features to our work. First, our main focus is on the impact of
monetary policy on daily inflation expectations. In much of this literature, daily shocks are
used to investigate monthly responses of prices and output. Here, the 𝑦 variables under
scrutiny – inflation expectations, the exchange rate and a measure of sovereign risk
premium – are themselves high frequency. Second, we use data from a high-debt
emerging economy, while most of the literature focuses on the US. Third, changes in
interest rates around monetary policy meetings are not automatically assumed as a
measure of exogenous monetary policy shocks. In most papers that use HFI, changes in
rates within a tight time window are automatically labeled as shocks. But "case study"
measures of monetary surprises are not immune to the endogeneity problem. Our paper
bears testament to that: here, the case study OLS regressions deliver different results
from our IV estimations exploiting heteroskedasticity. The paper is also related to the
“tight money paradox" literature kickstarted by Sargent and Wallace (1981) and Leeper
(1991). In essence, the idea in those articles is that the behavior of fiscal policy is key to
understand how changes in interest rates end up affecting inflation. If higher real interest
rates trigger an increase in debt levels and no response from primary balances (passive
fiscal policy in the words of Leeper), the public might expect the snowballing of debt to
end in monetization. Paradoxically then, monetary policy can be counterproductive:
higher interest rates would cause inflation expectations to go up. We do not find evidence
of this “unpleasant arithmetic” in the case of Brazil. At least not on average or for the
subperiods reported in the “Robustness” section.

INTERNATIONAL MONETARY FUND 3

Contents

1 Introduction 3

2 Data and empirical strategy 5

2.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 The identification strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Test of the identification assumptions . . . . . . . . . . . . . . . . . . . . . 9

3 Results 10
3.1 OLS: whole sample and "event study" . . . . . . . . . . . . . . . . . . . . 10
3.2 Effect of monetary surprises on inflation expectations using IV . . . . . . 12
3.3 Effect of monetary surprises on CDS and the exchange rate . . . . . . . . 13
3.4 Robustness tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4 Final remarks 15

List of Figures
1 Gross Debt/GDP in Brazil and other Emerging Economies . . . . . . . . 4
2 Full Sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3 Only Copom meetings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

List of Tables
1 Variance tests using TNC > TC . . . . . . . . . . . . . . . . . . . . . . . . 10
2 From naive OLS to case study coefficient . . . . . . . . . . . . . . . . . . . 11
3 OLS estimates: full sample and case study . . . . . . . . . . . . . . . . . . 11
4 IV through heteroskedasticity estimates . . . . . . . . . . . . . . . . . . . 12
5 Exchange rate and Risk: IV estimates . . . . . . . . . . . . . . . . . . . . . 13
6 Using survey measure of 𝜋 𝑒 . . . . . . . . . . . . . . . . . . . . . . . . . . 14
7 Excluding changes in inflation above P(95) and below P(05) . . . . . . . . 14
8 Moving window IV regressions: 3m interest rates . . . . . . . . . . . . . 14
9 Dropping FOMC dates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

INTERNATIONAL MONETARY FUND 2

1 Introduction
The last decade witnessed a substantial increase in the number of studies using
high-frequency data to identify the effects of monetary policy shocks on the economy.
These include Agripino and Ricco (2021), Gertler and Karadi (2015), Bauer and Swanson
(2022, 2023), Nakamura and Steinsson (2018), amongst others. The overall lesson is that
the old SVAR literature, relying mostly on timing restrictions for identification, was
unable to accurately represent the impact of monetary policy shocks on the economy.
But High Frequency Identification (HFI) has rarely been deployed to study monetary
transmission in Emerging Economies1, a gap we aim to bridge here by focusing on a
large developing economy, Brazil.
This paper has three new features. First, our main focus is on the impact of monetary
policy on inflation expectations. The main variable of interest is the daily change in
market-based (not survey data) inflation expectations. In much of this literature, daily
monetary policy shocks are used to investigate monthly responses of prices and output.
Here, the variables under scrutiny – inflation expectations, the exchange rate and a
measure of sovereign risk premium (the so-called CDS) – are themselves high frequency
in nature. Second, we use data from a high-debt emerging economy, while most of the
literature focuses on the US2. Third, changes in interest rates around monetary policy
meetings are not automatically assumed to be a measure of exogenous monetary policy
shocks. In most papers using HFI, the change in rates within a tight time window
is automatically labeled as a shock. But as Rigobon and Sack (2004) convincingly
argue, "case study" measures of monetary surprises are not immune to the endogeneity
problem. Our paper bears testament to that: here the case study OLS regressions
deliver different results from our IV estimations exploiting heteroskedasticity.
Our work is also related to the “tight money paradox" literature kickstarted by
Sargent and Wallace (1981), and to its multiple offsprings, such as Leeper (1991) and
more recently Bianchi and Melosi (2019). In essence, the idea in those articles is that
monetary policy does not take place in a vacuum, and amongst other factors, the
behavior of fiscal policy is key to understanding how changes in interest rates end up
affecting inflation. If higher real interest rates trigger an increase in debt levels and
no response from primary balances (passive fiscal policy in the words of Leeper), the
public might expect the snowballing of debt to end in monetization. Paradoxically then,
monetary policy can be counterproductive: higher interest rates would cause inflation
expectations to go up.3
1A very recent exception is Checo et al (2024), which employs however a different empirical strategy.
2Cesa-Bianchi et al (2020) applies HFI to analyze monetary policy in the UK.
3Analyses of the so-called “tight money paradox" can be found in Loyo (1999), Bhattacharya and

INTERNATIONAL MONETARY FUND 3

 Can traces of the tight monetary paradox be found in Brazil’s post-2009 data? On
average, this does not appear to be the case.
Historically, Brazil has faced very high inflation from the 1980s until the mid-1990s.
In 1994, the country implemented a successful stabilization plan (the Real Plan), bringing
inflation down through an exchange rate peg. Then in 1999, following a sequence of
international crises (Asia, Russia), the government was forced to abandon the peg,
replacing it with an inflation-targeting cum floating exchange rate regime. In terms of
institutional progress, the formal independence of the central bank was granted only
very recently, in 2019, although de facto independence was arguably achieved much
earlier. Overall, our reading is that credibility building is an ongoing process and the
fact that long-term inflation expectations as of January 2025 were still 50 bps above the
target is a sign of imperfect credibility.
Moreover, government finances have been on shaky ground for many years: the
country has one of the largest public debts among emerging economies, deficits are
elevated, and real interest rates have been very high since the 1990s.4 This is illustrated
in Figure 1, which compares Brazil’s debt-GDP ratio with that of a set of emerging
countries during the last 15 years (which roughly coincides with our sample).5

Brazil Emerging Markets

100

80
Debt over GDP

60

40

2010 2013 2016 2019 2022 2010 2013 2016 2019 2022
Source: WEO, July 2024

Figure 1: Gross Debt/GDP in Brazil and other Emerging Economies

Kudoh (2003), Sims (2011), Uribe (2016), Andolfatto (2021), Werning (2021), amongst others.
4See Ayres et al (2022) for a discussion on Brazil’s recent monetary and fiscal history.
5General government gross debt, percent of GDP. Data from the IMF’s World Economic Outlook
Database, October 2024.

INTERNATIONAL MONETARY FUND 4

 Finally, simple estimates of the so-called Bohn rule6 for the country suggest passive
fiscal policy as the norm rather than the exception.
Given this background, one might conclude that inflation expectations cannot
possibly be tamed by higher interest rates. However, our estimated coefficients
indicate that a positive/negative interest rate surprise causes inflation expectations to
decline/increase by a nontrivial amount7.
We also test whether monetary policy influences the exchange rate. It has been
argued that by increasing the probability of default, a tighter monetary stance could
cause the exchange rate to depreciate. Blanchard (2005) and Favero and Giavazzi (2005)
make this argument explicitly for the case of Brazil circa 2002. Our analysis does not
support this conjecture for the 2009:2024 years as a whole.8 We also find no evidence
of the mechanism suggested for the "inverted-UIP": monetary policy tightenings do
not lead to higher risk premia in our sample (though, as we show later, this positive
association is what a naive OLS regression yields).
To be clear, this paper estimates average effects and it might very well be possible
that monetary policy was inefficient at specific moments in time within our sample.
Formally testing this is challenging, though. Robustness tests using different sample
periods corroborate our main finding: monetary policy works.

2 Data and empirical strategy

2.1 Data
In the same vein as the Fed’s FOMC, in Brazil the monetary policy committee
(Copom is the acronym in Portuguese) meets at pre-scheduled dates. The committee
is composed of the Central Bank’s president and its board of directors, who discuss
the state of the economy for two days, typically a Tuesday and a Wednesday, and then
announce their decision on how much the policy rate should be adjusted after markets
close on Wednesday. Our 1 day window then will be comprised of variations in closing
prices from Wednesdays to Thursdays. The sample comprises all the weeks between
September 2009 and December 2024.
Since there is no futures market for the prime rate, we measure the change in interest
rates, Δ𝑖, using inter-bank deposit rate changes. For robustness, we look at different

6Available upon request; see Bohn (1998).

7All coefficients estimated using ID through heteroskedasticity are also highly statistically significant.
8Though this result is reversed when the data sample is restricted to the 2003-2008 period, as shown
in Goncalves and Guimaraes (2011).

INTERNATIONAL MONETARY FUND 5

maturities: 30, 90, 180 and 360 days.9
Brazil features a private market for inflation-indexed financial instruments that
allows us to tease out markets’ inflation expectations at different horizons: 1 year ahead
(our benchmark), but also 2, 3 and 5 years ahead.10
By subtracting from the expected return on nominal debt contracts the coupon of
inflation-indexed instruments of same maturity, we arrive at a measure of expected
inflation plus a risk premium. This is admittedly an imperfect measure of expected
inflation, given it also incorporates an inflation risk premium11.
The alternative is to use Central Bank’s weekly survey data, the so-called FOCUS
database. The problem with this strategy, in addition to the well-known issues with
surveys’ data on expectations (see Coibion et al (2020), for example) is that participants
update their forecasts in the Central Bank system mostly on Fridays. This means the
window around the COPOM meeting becomes too large: an entire week, instead of one
day. Nevertheless, when we employ this measure of inflation expectations, our results
barely change.
The other two dependent variables under scrutiny are (i) the variation in the closing
price of nominal exchange rate, and (ii) the Wednesday-to-Thursday change in CDS risk
premium.
In the case of the exchange rate, we simply replace Δ𝜋𝑡𝑒 by 𝐸Δ𝐸 𝑡
𝑡−1
: the percentage
change in the BRL/USD exchange rate between Wednesdays and Thursdays. Defined
this way, a positive (negative) Δ𝐸𝑡 implies a depreciation (appreciation) of the Brazilian
Real against the U.S. Dollar.12

2.2 The identification strategy

Our main goal is to estimate the effect of interest rates surprises (Δ𝑖) on inflation
expectations (Δ𝜋 𝑒 ). This is captured by the parameter 𝛽 in equation (1) below. However,
consistently estimating this effect is tricky. First, omitted variables might affect both
interest rates and inflation expectations. Second, reverse causality may impart a positive
bias on the OLS coefficient, as the Central Bank increases interest rates and bondholders

9Data can be downloaded from the Sao Paulo Stock Exchange webpage (BM&FBOVESPA, 2009-
2024a,b).
10Daily data on government bond yields were downloaded from a Bloomberg terminal. See Bloomberg
(2009-2024a,b,c) for non-inflation linked bond yields, and Bloomberg (2009-2024d,e,f) for inflation linked
bond yields. Alternatively, the Brazilian Financial and Capital Markets Association (2009-2024a,b,c)
provides free access to these series.
11Though Central Bank of Brazil (2014) argues it is a reliable measure of average inflation expectations.
In addition, if the risk-premium does not change much over time, its presence would not meaningfully
affect our estimations, since these are based on 1-day differences
12Data were obtained at the Central Bank of Brazil webpage (SGS Banco Central do Brasil, 2009-2024).

INTERNATIONAL MONETARY FUND 6

demand higher nominal yields when they expect higher inflation.

Δ𝜋𝑡𝑒 = 𝛼 + 𝛽Δ𝑖 𝑡 + 𝑢𝑡 (1)

Δ𝑖 𝑡 = 𝛾 + 𝛿Δ𝜋𝑡𝑒 + 𝑣 𝑡 (2)

Where 𝑢𝑡 and 𝑣 𝑡 are error terms with variances 𝜎𝑢𝑡 and 𝜎𝑣𝑡 , respectively. If 𝛿 ≠ 0, 𝛽
cannot be consistently estimated via OLS.
Rearranging the equations:

Δ𝜋𝑡𝑒 = (1 − 𝛿𝛽)−1 (𝛼 + 𝛽𝛾 + (𝛿𝑢𝑡 + 𝑣 𝑡 )) (3)

Δ𝑖 𝑡 = (1 − 𝛿𝛽)−1 (𝛼𝛿 + 𝛾 + (𝑢𝑡 + 𝛽𝑣 𝑡 )) (4)

Using (3) and (4) it can be easily shown that if one naively estimates (1), the estimated
parameter will differ from the true one as follows:

𝜎𝑢
𝛽d
𝑜𝑙𝑠 = 𝛽 + 𝛿(1 − 𝛿𝛽)
𝛽2 𝜎 𝑢 + 𝜎𝑣

Hence, when 𝛿 ≠ 0, 𝛽d 𝑜𝑙𝑠 → 𝛽 only as 𝜎𝑣 → ∞

To be clear, this is a potential problem for all empirical studies relying on variations
in interest rates around a narrow window of time.
Therefore, to estimate the effect of interest rate surprises on inflation expectations,
we follow Rigobon (2003)’s approach of identification through heteroskedasticity. As
shown in Rigobon and Sack (2004), identification is achieved if two conditions are
satisfied: (i) the variance of interest rate shocks (𝑣 𝑡 ) is higher in the subsample of daily
changes featuring Central Bank’s monetary policy committee meetings and (ii) no such
difference in variances is present for changes inflation expectations (𝑢𝑡 ).
That these are the identifying assumptions becomes clear from manipulating the
variance-covariance matrices for two partitions of the sample: C for Copom dates and
NC for the non-Copom dates. Define the variances of the shocks in equations (1) and
(2) in these two subsamples as:

 𝜎𝑢𝐶 , if 𝑡 ∈ 𝐶  𝜎𝑣𝐶 , if 𝑡 ∈ 𝐶

 

 
𝜎𝑢𝑡 = ; 𝜎𝑣𝑡 =
 𝜎𝑢𝑁
 , if 𝑡 ∈ 𝑁  𝜎𝑣𝑁
 , if 𝑡 ∈ 𝑁
 

INTERNATIONAL MONETARY FUND 7

Calculating the difference in the variance-covariance matrices, one gets:
" #
1 𝜎𝑣𝐶 − 𝜎𝑣𝑁 𝐶 + 𝛿2 (𝜎𝑢𝐶 − 𝜎𝑢𝑁 𝐶 ) 𝛽(𝜎𝑣𝐶 − 𝜎𝑣𝑁 𝐶 ) + 𝛿(𝜎𝑢𝐶 − 𝜎𝑢𝑁 𝐶 )
Ω𝐶 − Ω 𝑁 𝐶 =
(1 − 𝛽𝛿)2 · 𝛽2 (𝜎𝑣𝐶 − 𝜎𝑣𝑁 𝐶 ) + 𝜎𝑢𝐶 − 𝜎𝑢𝑁 𝐶

Imposing the identifying assumptions:

𝜎𝑣𝐶 > 𝜎𝑣𝑁 (5)

𝜎𝑢𝐶 = 𝜎𝑢𝑁 (6)

One gets:

𝜎 𝐶 − 𝜎𝑣𝑁 𝐶 1 𝛽 
 
Ω𝐶 − Ω 𝑁 𝐶 = 𝑣  
(1 − 𝛽𝛿)2  · 𝛽 
 2 
 
Thus allowing for two ways of identifying 𝛽:
" #
𝜔11 𝜔12 𝜔12 𝜔22
c𝐶 − Ω
Ω š 𝑁𝐶 = 𝛽=
⇒b 𝛽=
or, alternatively b
· 𝜔22 𝜔11 𝜔12

The same can be achieved in an instrumental variable setting.13 Let:

√
 Δ𝑖 𝑡 / 𝑇𝐶 , if 𝑡 ∈ 𝐶



Δ𝐼 = √
 Δ𝑖 𝑡 / 𝑇𝑁
 , if 𝑡 ∈ 𝑁

And,
√
 Δ𝑖 𝑡 / 𝑇𝐶 , if 𝑡 ∈ 𝐶


𝑧 𝑡𝑖

= √
 −Δ𝑖 𝑡 / 𝑇𝑁
 , if 𝑡 ∈ 𝑁

where 𝑇𝐶 is the size of the subsample in which Copom meetings occur, and 𝑇𝑁 represents
the number of observations in its complement. If the sizes of the subsamples are different,
the explained variable also needs to be normalized as follows:

√
 Δ𝜋𝑡𝑒 / 𝑇𝐶 , if 𝑡 ∈ 𝐶


e𝑡𝑒 =

Δ𝜋 √
 Δ𝜋𝑡𝑒 / 𝑇𝑁
 , if 𝑡 ∈ 𝑁


13For further details, see Rigobon and Sack (2004).

INTERNATIONAL MONETARY FUND 8

 Then the first stage, using the system of equations, takes the following form:

1 1 1 𝛿+𝛽 2 𝐶
𝑝𝑙𝑖𝑚 (𝑧 ′Δ𝐼) = (Δ𝑖 𝐶 )′(Δ𝑖 𝐶 ) − (Δ𝑖 𝑁 𝐶 )′(Δ𝑖 𝑁 𝐶 ) = ( ) .(𝜎𝑣 − 𝜎𝑣𝑁 𝐶 ) > 0 (7)
𝑇 𝑇𝐶 𝑇𝑁 𝐶 1 − 𝛿𝛽

Where the last equality comes from the identifying assumption discussed.
Similarly:

1 1 1 𝛽
𝑝𝑙𝑖𝑚 (𝑧 ′Δ𝜋 𝑒 ) = (Δ𝑖 𝐶 )′(𝑢 𝐶 ) − (Δ𝑖 𝑁 𝐶 )′(𝑢 𝑁 𝐶 ) = ( ).(𝜎𝑢𝐶 − 𝜎𝑢𝑁 𝐶 ) = 0 (8)
𝑇 𝑇𝐶 𝑇𝑁 𝐶 1 − 𝛿𝛽

2.3 Test of the identification assumptions

Variance tests allow us to check if the identifying assumptions just discussed are
valid. According to expression (5), the variance of Δ𝑖 𝑡 should be higher in subset 𝐶 than
in subset 𝑁 𝐶 and, following (6), there should be no such difference in variances for the
explained variables Δ𝜋𝑡𝑒 , Δ𝑡 and Δ𝐸𝑡 .
Table 1 shows the ratio between variances in set 𝐶 and set 𝑁 for all variables used,
along with a 99% confidence interval.
The variances for Δ𝑖 𝑡 differ markedly across the subsamples. For Δ𝜋𝑡𝑒 , Δ𝑡 and
Δ𝐸𝑡 , conversely, we cannot reject the null of equal variances. The exception is the
2-year inflation expectations. Hence the identifying conditions hold in most cases and,
accordingly, the instrumental variables strategy employed should be able to identify
true causal effects.

INTERNATIONAL MONETARY FUND 9

 Table 1: Variance tests using TNC > TC

Ratio of variances 99% CI

Δ𝑖 Maturity 12m 2.33 [1.65; 3.42]
Maturity 6m 3.40 [2.42; 4.99]
Maturity 3m 4.53 [3.21; 6.65]
Maturity 1m 5.25 [3.73; 7.69]
Δ𝜋 𝑒 1 year 1.23 [0.88; 1.81]
2 years 1.59 [1.13; 2.34]
3 years 1.33 [0.95; 1.96]
5 years 1.11 [0.79; 1.63]

Δ𝐸 0.92 [0.65; 1.35]

Δ𝑟𝑖𝑠 𝑘 0.85 [0.60; 1.24]
Notes: The table displays tests of the identification assumptions based on Rigobon’s (2003) method-
ology. Specifically, we check for differences in variances across set C (weeks with Copom meeting)
and set N (weeks without Copom meeting), using daily data between September 2009 and August
2024. The middle column of the table shows the ratio between variances in set C and in set N. The
last column presents the 95% confidence interval for this ratio. We show statistics for the change in
nominal interest rates (Δ𝑖 ), the change in inflation expectations (Δ𝜋 𝑒 ), and the percentage change in
the BRL/USD exchange rate (Δ𝐸 ). Δ𝑖 is measured by changes in interbank deposit rate. We use three
maturities: 1 year, 6 months and 3 months. Inflation expectations are measured taking the difference
between inflation linked and non-inflation linked government bonds. In this case, we calculate the
variance ratio for 1, 2 and 5-year maturities. For all variables, differences were computed between
Wednesday’s and Thursday’s values in every week within our sample.

3 Results

3.1 OLS: whole sample and "event study"

Figures 2 and 3 below show, respectively, the correlation pattern between Δ𝑖 and
Δ𝜋 𝑒 for the whole sample and for the dates of Copom meetings only (as in the case
study literature).
A naive interpretation of the correlation pattern displayed on the left chart in
table 2 might lead to the conclusion that increases in interest rates backfire: inflation
expectations increase with positive interest rate surprises. And even when looking
at Copom meetings only (chart on the right), one would not be able to conclude that
tightenings lead to lower inflation expectations. The slope, though negative, is not
statistically significant. Thus an event-study strategy would suggest no impact of
monetary policy on inflation expectations.

INTERNATIONAL MONETARY FUND 10

 Whole Sample Copom days
t Student(745) = 2.94, p = 3.40e−03, r Pearson = 0.11, CI95% [0.04, 0.18], n pairs = 747
t Student(117) = −1.05, p = 0.30, r Pearson = −0.10, CI95% [−0.27, 0.08], n pairs = 119
300
200 30
100 20
0 10
0

Variation in 1y expected inflation

Variation in 1y expected inflation

0.2 0.2

0.0 0.0

−0.2
−0.2

−0.4 −0.2 0.0 0.2 0.4

0 5 1015
−0.50 −0.25 0.00 0.25 0306090120 Variation in 3m interest rates
Variation in 3m interest rates
loge(BF01) = 1.43, ρPearson = −0.09, CIHDI
posterior JZS
loge(BF01) = −1.40, ρPearson = 0.11, CI95%
posterior HDI JZS
[0.04, 0.17], r beta = 1.41 95% [−0.27, 0.08], r beta = 1.41

Figure 2: Full Sample Figure 3: Only Copom meetings

Table 2: From naive OLS to case study coefficient

We next report OLS estimates of Δ𝜋 𝑒 on Δ𝑖 for the whole sample. Table 3 displays
the results for four different measures of changes in inflation expectations and four
different measures of changes in interest rates: 1 month, 3 months, 6 months and 1 year.
All coefficients are positive.

Table 3: OLS estimates: full sample and case study

Interest rate measure = Δß

1 month 3 months 6 months 12 months
Full Sample 𝛽 0.16* 0.25*** 0.24*** 0.21***
(se) (0.07) (0.05) (0.03) (0.02)
Copom Sample 𝛽 -0.13 -0.09 -0.02 +0.01
(se) (0.11) (0.09) (0.06) (0.05)
Notes: This table displays OLS regression results of changes in inflation
expectations (Δ𝜋 𝑒 ) against changes in nominal interest rates (Δ𝑖). Con-
stants are included but not reported. We use daily data from Brazilian
assets between September 2009 and December 2024. Δ𝜋 𝑒 is computed as
the difference between Wednesday’s and Thursday’s observations. We
run regressions using different measures of monetary shocks but always
𝑒
the same y-variable, Δ𝜋1𝑦𝑒 𝑎𝑟
. Numbers in parentheses are Newey-West
standard errors.

These results were already hinted at in the correlation charts above. The positive
and statistically significant coefficients in the first row could be hastily (and mistakenly)
interpreted as evidence of the tight money paradox. The second row suggests this to be
likely due to endogeneity – market future interest rates increase when expected inflation

INTERNATIONAL MONETARY FUND 11

increases – but if we were to stop at the case study regression, the verdict would still be
very unfavorable to monetary policy: no impact on inflation expectations.
In the next subsection, we present the IV estimations exploiting the data’s het-
eroskedasticity. Using this technique, we uncover a strong and consistent textbook-like
impact of monetary policy on inflation expectations and the exchange rate.

3.2 Effect of monetary surprises on inflation expectations using IV

The main IV results are shown in Table 4. Now all estimates are negative and display
small standard-errors (majority of p-values are smaller than 1%). Positive interest rate
surprises at different maturities are associated with lower inflation expectations at all
horizons.

Table 4: IV through heteroskedasticity estimates

Dependent variable = Δ𝜋 𝑒

1 year 2 years 3 years 5 years

Δ𝑖 -0.27*** -0.64*** -0.69*** -0.67***

(1 month) (0.07) (0.07) (0.08) (0.07)

Δ𝑖 -0.26*** -0.60*** -0.69*** -0.69***

(3 months) (0.07) (0.06) (0.07) (0.07)

Δ𝑖 -0.21*** -0.45*** -0.53*** -0.67***

(6 months) (0.05) (0.05) (0.06) (0.08)

Δ𝑖 -0.20*** -0.43*** -0.50*** -0.53***

(12 months) (0.06) (0.07) (0.07) (0.06)

N 768 768 768 768

Notes: This table displays IV regression results of changes in inflation
expectations (Δ𝜋 𝑒 ) against changes in nominal interest rates (Δ𝑖) using four
measures of expectations and also changes in interest rates at four different
maturities.

In terms of magnitudes, using Table 4 a 100 basis point surprise in the interest rate
leads to a reduction of between 0.2 to 0.3 percentage points in the 1-year ahead measure
of inflation expectations.

INTERNATIONAL MONETARY FUND 12

3.3 Effect of monetary surprises on CDS and the exchange rate
We now evaluate the impact of monetary shocks on the exchange rate and a risk
measure. In theory, an increase in interest rates could lead to a currency depreciation
via higher probability of default. The results of the previous subsection indicate that
even if that were to be true, monetary tightenings would still bring down inflation
expectations.
IV Regressions featuring both Δ𝐸 and Δ𝐶𝐷𝑆 as the dependent variable do not lend
credence to the "higher rates leading to depreciation via increased risk" story. Positive
interest rate surprises cause an appreciation of the Brazilian currency (the Real) against
the U.S. dollar (and do not seem to affect the CDS either).

Table 5: Exchange rate and Risk: IV estimates

Interest rate measure = Δß

1 month 3 months 6 months 12 months
Δ𝐸 -5.64*** -5.10*** -3.93*** -3.42***
(se) (0.91) (0.75) (0.63) (0.69)
Δ𝐶𝐷𝑆 0.01 -0.03 -0.05 -0.11*
(se) (0.06) (0.05) (0.04) (0.05)
Notes: This table displays IV results using the same technique as before.
We report the coefficients using the 5-year CDS. The results with changes
in 1-year CDS are available upon request.

3.4 Robustness tests

We test the robustness of our results through three additional checks14. First, in
Table 6 we present the results when, instead of using market measures of inflation
expectations, we resort to expected inflation from the Central Bank’s survey with market
participants. Specifically, models are run using the average of 12 months ahead inflation
expectation reported by all survey participants (top row) and only the from the top 5
forecasters (bottom row). The benefit of using this measure is that it is not contaminated
by the risk premia. The disadvantage, as mentioned before, is its weekly frequency.
Reassuringly, the results obtained using Brazilian securities are also borne out using
survey data instead. Out of 8 entries in Table 6, only one is not statistically significant.
Interestingly, coefficients are larger in the upper row, which uses the median of all
survey respondents.
The second robustness exercise consists of excluding episodes with very large swings
in expected inflation. Specifically, we exclude all Δ𝜋 above the percentile 0.95 and below
14Here, we report results for two measures of interest rate surprises only: 12m and 3m

INTERNATIONAL MONETARY FUND 13

 Table 6: Using survey measure of 𝜋 𝑒

Interest rate measure = Δß

1 month 3 months 6 months 12 months
Δ𝜋 𝑒𝑠𝑢𝑟𝑣𝑒 𝑦 -0.60*** -0.54*** -0.43*** -0.36***
(se) (0.17) (0.13) (0.11) (0.12)
Δ𝜋 𝑒𝑠𝑢𝑟𝑣𝑒 𝑦𝑡𝑜𝑝5 -0.25*** -0.20*** -0.13** -0.09
(se) (0.10) (0.08) (0.06) (0.07)
Notes: This table displays the results of monetary policy shocks on inflation
expectations reported by market participants to the CB. The first row
includes all participants; the second row includes forecasts from the top-5
forecasters only.

percentile 0.05. This reduces the sample to 689 observations and makes it harder to
identify the impact of monetary policy since variability in the variable of interest is
curtailed.
Table 7: Excluding changes in inflation above P(95) and below P(05)

Δ𝑖 used Δ𝜋 𝑒 1𝑦 Δ𝜋 𝑒 2𝑦 Δ𝜋 𝑒 3𝑦 Δ𝜋 𝑒 5𝑦 Δ𝐸
𝐸
3 months -0.07* -0.47*** -0.59*** -0.61*** -2.00**
(se) (0.04) (0.05) (0.04) (0.06) (0.69)
12 months -0.12*** -0.39*** -0.46*** -0.46*** -0.94*
(se) (0.03) (0.05) (0.04) (0.05) (0.48)
Number of observations is now 689.

The third test focuses on sub-samples. We report results for six different 10-year
moving windows with start dates in 2009, 2010, 2011, 2012, 2013 and 2014. We use the
3-month interest rate in all specifications. Out of the 18 entries in Table 8, only one is
borderline significant. Moreover, over time monetary policy effectiveness seems to have
increased judging by the second and third rows.

Table 8: Moving window IV regressions: 3m interest rates

2009:2019 2010:2020 2011:2021 2012:2022 2013:2023 2014:2024

Δ𝜋 𝑒 1𝑦 −0.25 ∗∗∗ −0.13 *** −0.17 ** −0.18 ** −0.21 ** −0.19 *
(se) 0.06 0.05 0.08 0.09 0.10 0.11
Δ𝜋 𝑒 5𝑦 −0.61 *** −0.56 *** −0.79 *** −0.81 *** −0.85 *** −0.91 ***
(se) 0.06 0.06 0.09 0.10 0.11 0.13
Δ𝐸
𝐸 −2.85 *** −3.67 *** −5.50 *** −5.90 *** −7.70 *** −8.79 ***
(se) 0.73 0.73 0.97 1.06 1.25 1.40

Finally, there is a significant number of instances in which the Copom meeting

coincides with FOMC Meetings in the US: 32 from 2009 to the end of 2024. If variables

INTERNATIONAL MONETARY FUND 14

that are potentially relevant to explain inflation expectations in Brazil, such as commodity
prices, display higher variance on the days of FOMC that are also Copom days, a bias
could be introduced to our estimates. Our fourth robustness exercise hence consists of
dropping those common dates from the sample.
The new C-subsample now has 90 data points, whereas the NC-subsample is kept
unchanged. Results presented in Table 9 show this paper’s main message remains in
this smaller sample.

Table 9: Dropping FOMC dates

Δ𝜋 𝑒 and Δ𝐸
𝐸

1 year 2 years 3 years 5 years FX

Δ𝑖 -0.35*** -0.82*** -0.89*** -0.82*** -5.90***

(3 month) (0.07) (0.08) (0.08) (0.08) (0.95)
Δ𝑖 -0.27.*** -0.65*** -0.72*** -0.71*** -4.98***
(12 months) (0.08) (0.11) (0.12) (0.11) (1.03)

N 736 736 736 736 736

Notes: This table displays IV regression results of changes in inflation
expectations and the exchange rate against changes in nominal interest
rates. The methodology is always Rigobon’s identification through het-
eroskedasticity, but the sample is smaller because we drop days in which
the Copom meeting coincides with the FOMC.

4 Final remarks
The large and thriving literature on HFI of monetary policy shocks has two important
missing parts. First, it is too US-centric (with few exceptions). Arguably, some emerging
economies’ characteristics can impair monetary transmission, such as shallow credit
markets, lack of fiscal credibility, risk premia, etc. Second, inflation expectations have
been largely neglected, though they are a crucial determinant of actual inflation in
both theoretical and empirical work. This paper bridges these two gaps by using data
from an emerging economy to assess the impact of monetary surprises on inflation
expectations.
The resurgence of models featuring characteristics of the FTPL also sparked a
renewed interest in the unpleasant arithmetic logic of Sargent and Wallace (1981). Can
monetary policy backfire if fiscal policy is active and debt is elevated? We do not find
systematic evidence of that using data from Brazil.

INTERNATIONAL MONETARY FUND 15

References
Andolfatto, D. (2021). “Is It Time for Some Unpleasant Monetarist Arithmetic?"
Federal Reserve Bank of St. Louis Review, vol. 103(3), pages 315-332, July.
Ayres, J.; Garcia, M.; Guillen, D.; Kehoe, P. J. (2022). “The case of Brazil." Kehoe, T. J.;
Nicolini, J. P. (eds.) A Monetary and Fiscal History of Latin America, 1960-2017. University
of Minnesota Press.
Bauer, M. Swanson, E. (2022). "A Reassessment of Monetary Policy Surprises
and High-Frequency Identification", NBER Working Papers 29939, National Bureau of
Economic Research.
Bauer, M. Swanson, E. (2023). "An Alternative Explanation for the Fed Information
Effect", American Economic Review, vol 113, pages 664-700.
Bhattacharya, J.; Noritaka K. (2002). “Tight money policies and inflation revisited."
Canadian Journal of Economics, vol. 35(2), pages 185-217, May.
Bianchi, Francesco Melosi, Leonardo, 2019. "The dire effects of the lack of monetary
and fiscal coordination," Journal of Monetary Economics, Elsevier, vol. 104(C), pages 1-22.
Blanchard, O. (2005). “Fiscal Dominance and Inflation Targeting: Lessons from
Brazil." Giavazzi, F.; Goldfajn, I.; Herrera, S. (eds.) Inflation Targeting, Debt, and the
Brazilian Experience, 1999 to 2003. Cambridge: MIT Press.
Bloomberg (2009-2024a). “Anbima Brazil Govt Bond Fixed Rate 1 Year - BZAD1Y
Index." Brazilian Financial and Capital Markets Association.
Bloomberg (2009-2024b). “Anbima Brazil Govt Bond Fixed Rate 2 Year - BZAD2Y
Index." Brazilian Financial and Capital Markets Association.
Bloomberg (2009-2024c). “Anbima Brazil Govt Bond Fixed Rate 3 Year - BZAD3Y
Index." Brazilian Financial and Capital Markets Association.
Bloomberg (2009-2024d). “Anbima Brazil Govt Bond IPCA Rate 1 Year - BZAA1Y
Index." Brazilian Financial and Capital Markets Association.
Bloomberg (2009-2024e). “Anbima Brazil Govt Bond IPCA Rate 2 Year - BZAA2Y
Index." Brazilian Financial and Capital Markets Association.
Bloomberg (2009-2024f). “Anbima Brazil Govt Bond IPCA Rate 3 Year - BZAA4Y
Index." Brazilian Financial and Capital Markets Association.
BM&FBOVESPA (2009-2024a). “Swap DI x pré 180 dias (nominal yield curve)." Brazil-
ian Financial and Capital Markets Association. The daily data is available for download at

INTERNATIONAL MONETARY FUND 16

[Link]
data/consultas/mercado-de-derivativos/precos-referenciais/taxas-referenciais-
bm-fbovespa/.
BM&FBOVESPA (2009-2024b). “Swap DI x pré 360 dias (nominal yield curve)." Brazil-
ian Financial and Capital Markets Association. The daily data is available for download at
[Link]
data/consultas/mercado-de-derivativos/precos-referenciais/taxas-referenciais-
bm-fbovespa/.
Brazilian Financial and Capital Markets Association (2009-2024a). “Term Structure
and Break Even Inflation (IPCA): 1-Year Break Even Inflation (252 Business Days)."
Brazilian Financial and Capital Markets Association. The daily data is available
for download at [Link]
[Link] and the API to build the database is available at [Link]
[Link]/en/precos-indices/apis-de-precos/titulos-publicos/#face-value-
(vna).
Brazilian Financial and Capital Markets Association (2009-2024b). “Term Structure
and Break Even Inflation (IPCA): 2-Year Break Even Inflation (504 Business Days)."
Brazilian Financial and Capital Markets Association. The daily data is available
for download at [Link]
[Link] and the API to build the database is available at [Link]
[Link]/en/precos-indices/apis-de-precos/titulos-publicos/#face-value-
(vna).
Brazilian Financial and Capital Markets Association (2009-2024c). “Term Structure
and Break Even Inflation (IPCA): 3-Year Break Even Inflation (726 Business Days)."
Brazilian Financial and Capital Markets Association. The daily data is available
for download at [Link]
[Link] and the API to build the database is available at [Link]
[Link]/en/precos-indices/apis-de-precos/titulos-publicos/#face-value-
(vna).
Central Bank of Brazil (2014). “Breaking the Break-even Inflation Rate." Inflation
Report, December 2014.
Cesa-Bianchi, A.; Thwaites, G.; Vicondoa, A. (2020). “Monetary policy transmission
in the United Kingdom: A high frequency identification approach." textitEuropean
Economic Review, vol. 123, 103375.
Checo, A.; Grigoli,F. and Sandri, D (2024). "Monetary Policy Transmission in
Emerging Markets: Proverbial Concerns, New Evidence," BIS WP 1170.

INTERNATIONAL MONETARY FUND 17

 Coibion, Olivier, Yuriy Gorodnichenko, and Tiziano Ropele (2020). “Inflation
Expectations and Firm Decisions: New Causal Evidence” Quarterly Journal of Economics,
vol. 135(1), 165–219.
Favero, C. A.; Giavazzi, F. (2005). “Inflation Targeting and Debt: Lessons from
Brazil." Giavazzi, F.; Goldfajn, I.; Herrera, S. (eds.) Inflation Targeting, Debt, and the
Brazilian Experience, 1999 to 2003. Cambridge: MIT Press.
Getler, M; Karadi, P. (2015). "Monetary Policy Surprises, Credit Costs, and Economic
Activity", American Economic Journal: Macroeconomics, vol 7, pages 44-76.
Goncalves, C.; Guimaraes, B. (2011). "Monetary policy, default risk and the exchange
rate in Brazil." Revista Brasileira de Economia
Loyo, E. (1999). “Tight Money Paradox on the Loose: a Fiscalist Hyperinflation."
Harvard Kennedy School, unpublished manuscript, August.
Miranda-Agrippino, S.; Ricco, G. (2021). "The transmission of monetary policy
shocks." American Economic Journal: Macroeconomics, vol 13, pages 74-107.
Nakamura, E.; Steinsson, J. (2018). "High-Frequency Identification of Monetary
Non-Neutrality: The Information Effect." The Quarterly Journal of Economics, Volume 133,
Issue 3, Pages 1283–1330.
Rigobon, R. (2003). “Identification through Heteroskedasticity." Review of Economics
and Statistics, vol. 85(4), pages 777-792, November.
Rigobon, R.; Sack, B. (2004). “The impact of monetary policy on asset prices." Journal
of Monetary Economics, vol. 51(8), pages 1553-1575, November.
Sargent, T.; Silber, W. (2022). “Inflation, Deficits and Paul Volcker." Wall Street Journal,
March 3, 2022.
Sargent, T. J.; Wallace, N. (1981). “Some unpleasant monetarist arithmetic." Quarterly
Review, Federal Reserve Bank of Minneapolis, vol. 5(Fall).
Sims, C. A. (2011). “Stepping on a rake: The role of fiscal policy in the inflation of
the 1970s." European Economic Review, vol. 55(1), pages 48-56, January.
SGS Banco Central do Brasil (2009-2021). “Exchange rate - Free - United States Dollar
(sale) - 1." Banco Central do Brasil. The data is available for download at [Link]
[Link]/sgspub/localizarseries/[Link]?method=prepararTelaLocalizarSer
(accessed August 6, 2021).
Uribe, M. (2016). “Is the Monetarist Arithmetic Unpleasant?" NBER Working Paper
22866, National Bureau of Economic Research.

INTERNATIONAL MONETARY FUND 18

 Werning, I. (2021). “Recalculating Sargent and Wallace’s Unpleasant Arithmetic
using Interest Rates." MIT.
World Economic Outlook Database (2009-2019). “General government gross debt,
percent of GDP." International Monetary Fund. The data is available for download at
[Link] (accessed
Feb 1, 2022).

INTERNATIONAL MONETARY FUND 19