ISSN 1518-3548
Working Paper Series
Medium-Size Macroeconomic Model for the Brazilian Economy
Marcelo Kfoury Muinhos and Sergio Afonso Lago Alves
February, 2003
ISSN 1518-3548
CGC 00.038.166/0001-05
Working Paper Series
Brasília
n. 64
Feb
2003
p. 1 – 48
Working Paper Series
Edited by:
Research Department (Depep)
(E-mail: workingpaper@bcb.gov.br)
Reproduction permitted only if source is stated as follows: Working Paper Series n. 64
Authorized by Ilan Goldfajn (Deputy Governor for Economic Policy).
General Control of Subscription:
Banco Central do Brasil
Demap/Disud/Subip
SBS – Quadra 3 – Bloco B – Edifício -Sede – 2º subsolo
70074-900 Brasília – DF – Brazil
Phone: (5561) 414-1392
Fax: (5561) 414-3165
The views expressed in this work are those of the authors and do not reflect those of the Banco Central or its members.
Although these Working Papers often represent preliminary work, citation of source is required when used or reproduced.
As opiniões expressas neste trabalho são exclusivamente do(s) autor(es) e não refletem a visão do Banco Central do Brasil.
Ainda que este artigo represente trabalho preliminar, citação da fonte é requerida mesmo quando reproduzido parcialmente.
Banco Central do Brasil Information Bureau
Address:
Secre/Surel/Dinfo
SBS – Quadra 3 – Bloco B
Edifício -Sede, 2º subsolo
70074-900 Brasília – DF
Phones:
(5561) 414 (....) 2401, 2402, 2403, 2404, 2405, 2406
DDG:
0800 992345
FAX:
(5561) 321-9453
Internet:
http://www.bcb.gov.br
E-mails:
cap.secre@bcb.gov.br
dinfo.secre@bcb.gov.br
0HGLXP6L]H0DFURHFRQRPLF0RGHO
IRUWKH%UD]LOLDQ(FRQRP\
0DUFHOR.IRXU\0XLQKRV
6HUJLR$IRQVR/DJR$OYHV
$EVWUDFW
This paper presents a medium-scale macroeconomic model of the Brazilian
economy with more than 30 equations. Potential output is derived from a
Cobb-Douglas production function, while the demand side is divided into
estimated equations for: household consumption, investment in machinery
and construction, government spending and net exports. The estimated
Phillips curve has an interesting feature: a step dummy variable captures the
macroeconomic break in pass-through that occurred after the change of the
exchange rate regime in 1999. There are long-run equilibrium conditions for
the external and fiscal debt and also for the real interest rate. External and
supply shocks were simulated in order to generate impulse responses for the
medium size model.
Key words: macroeconomic model, interest rate equilibrium, and potential
output.
JEL Classification: E12, E27, F43, F47
*
We would like to thank Gil Riella for his outstand help in running the model in Matlab. Flávia Mourão
Graminho, Eduardo Loyo, Shad Turney and Andrew Levin also helped us in the estimations and with
suggestions. The views expressed in this work are those of the authors and do not reflect those of the
Banco Central do Brasil or its members.
**
Research Department, Central Bank of Brazil. E-mail addresses: marcelo.kfoury@bcb.gov.br and
sergio.lago@bcb.gov.br
3
,QWURGXFWLRQ
Small-scale macroeconomic models are very useful for forecasting the short run,
but they are not very useful for anchoring the key variables in the long run. They are not
able to answer questions about the macro equilibrium of the economy, nor to establish
fiscal or external constraints. Larger macroeconomic models work better in providing
information about the interaction of stabilization and growth in the medium run.
Questions concerning technology, investment, labor markets and the current account
balance are better addressed by a more comprehensive model. Micro-founded models
are also able to present long run properties consistent with economic agents’ optimal
behavior. On the other hand Keynesian models are important because they can be used
to simultaneously determine the equilibrium levels of output, employment, inflation,
current account, rate of investment and fiscal balance. However, in Keynesian models
the long-run equilibrium of some key variables such as the interest rate and exchange
rate are not endogenously determined.
Many Central Banks have built micro founded structural models. Examples
include the Bank of Canada’s QPM, the Bank of England’s MM (0DFURHFRQRPLF
0RGHO) and also the IMF’s Multimod. These models are in general divided into two
parts. A steady-state part assures long-run equilibrium, which is based on the optimal
behavior of economic agents, while a dynamic section describes the equilibrium path of
the economy using an error correction framework.
The steady-state model of the QPM is an overlapping-generation model with only
one good1. The Multimod is very similar and for the first five years uses the outcome
from the World Economic Outlook as a baseline. The dynamic section of the Multimod
uses a non-linear Phillips curve and also ensures long-run growth consistent with
sustainable external debt service.
The special features of the FRB-US are the non-arbitrage conditions in the
financial markets. In the goods sector, the expectational variables are model consistent.
The dynamic model is also based on an error-correction approach. VAR expectations
are also taken into account to describe transitory shocks. The steady-state section of the
model is not a dynamic general equilibrium model (DGE) but an ad-hoc baseline case.
4
Among the models in the Keynesian paradigm, one example is the Financial
Programming model of the International Monetary Fund, which uses the monetarist
approach to the balance of payments. This model was used in the creation of an entire
generation of IMF programs and is still being applied. The bottom-line is to set a goal
for the central bank’s net domestic assets as a way to avoid growth of the money supply
well above the floor for international reserves. The World Bank has a line of two-gap
growth models (domestic saving and external saving) called RMSM-X. In Brazil, IPEA
has set up a Keynesian macroeconomic model, based on the national accounts,
especially the balance of payments and the fiscal budget. A quarterly version of this
model has been released recently.
The Central Bank of Chile has built a Keynesian model very similar to the one
presented in this paper. The major difference between the models is in the derivation of
the steady-state equilibrium. In the Chilean model, consumption is divided into durable
and non-durable goods, which is a future goal for our model.
The main contributions of our model, compared to other macroeconomic models
developed in the Central Bank of Brazil, are:
-
Aggregate demand is calculated by estimating: (1) household consumption,
investment in (2) machinery and (3) construction, (4) net exports, (5)
government spending, (6) government taxes, (7) changes in inventories;
-
The model uses a Phillips curve that includes dummies for the structural
break in the pass-through coefficient in 1999 and a proxy for labor
productivity (unit labor cost);
-
Potential output is estimated by a Cobb-Douglas production function;
-
The model includes an estimated exchange rate error correction mechanism
converging to the Uncovered-Interest-Parity (UIP) equation on the long run,
measured in real terms, together with an equation for the risk premium, to
which responses for changes in fiscal and external conditions are added;2
-
The model includes ad-hoc steady-state conditions for the current account
deficit and the primary fiscal surplus.
1
It relies on indirect tax and share of imported consumption to include inflation and the exchange rate in
the model.
2
Muinhos, Alves and Riella (2002) have similar equations for UIP and Risk premium.
5
Some simulations for different Taylor rules and impulse responses for a
temporary cost-push shock are presented.
The paper is organized as follows. Section 2 presents the diagrams of the small
macroeconomic and the medium size models, showing the monetary policy
transmission mechanisms and presenting some discussion about the long run
equilibrium conditions for the external sector. Section 3 presents estimated and
calibrated equations for the demand, supply, external and monetary-fiscal blocks of the
model. Section 5 shows some simulation exercises and the last section concludes the
paper.
'LDJUDPVRI7KH7UDQVPLVVLRQ0HFKDQLVPVDQG(TXLOLEULXP&RQGLWLRQV
In order to compare the monetary transmission mechanisms of the medium-scale
and the small-scale models, it is necessary to explain the mechanisms in the latter
model, as shown in Figure 1. The model includes the traditional channel, via output gap,
and a second channel, via exchange rate. The IS curve shows that an increase in the real
interest rate will negatively affect the output gap, directly and indirectly via the term
structure of interest rates. A more negative output gap will decrease inflation via the
Phillips curve. By the UIP non-arbitrage condition, an increase in the interest rate
causes an appreciation of the exchange rate in the spot market, and, via the Phillips
curve, a decrease in imported prices will generate lower inflation.
)LJXUH
6PDOO0RGHO7UDQVPLWLRQ0HFKDQLVP
6
The two monetary transmission mechanisms described for the small model also
occur in the medium model, shown in Figure 23. But now it is possible to distinguish
between supply and demand effects. An increase in the interest rates will affect
household consumption and investment in construction and machinery through the term
structure, generating a decrease in aggregate demand. A higher interest rate will cause
an exchange rate appreciation and a decrease in net exports, decreasing aggregate
demand. On the supply side, the effects of a higher interest rate will take more time to
occur, because a lower level of investment will cause a decrease in the growth rate of
the capital stock, affecting potential output growth. The decrease in aggregate demand
leads to a drop in inflation through a more negative output gap. But this drop would be
partially offset by the decrease in potential output growth.
The exchange rate mechanism is still available in the medium size model. But
now the fiscal and external variables also affect the exchange rate via the risk premium.
An increase in the interest rate that worsens the fiscal accounts will generate an increase
in the risk premium and a depreciation of the exchange rate that might offset the
aggregate demand channel. The current account deficit also affects the risk premium
and consequently, the exchange rate and inflation. Rapid GDP growth may cause an
increase in inflation via the output gap and also via a worsening of the trade balance.
3
Although the main blocks of the medium model are represented in Figure 2, there are some interactions
between variables not shown in the figure in order to obtain a clean representation of the model.
Nevertheless, the model equations are commented in the text. As this is still a work in progress, our
blocks are subject to future improvements.
7
)LJXUH
0HGLXP0RGHO7UDQVPLWLRQ0HFKDQLVP
We also include the labor market as a monetary transmission channel. An increase
in the labor force may increase potential output via the Cobb-Douglas production
function. In addition, an increase in productivity measured by Total Factor Productivity
will increase GDP and decrease inflation, allowing for a loosening of monetary policy.
The absence of micro-founded behavior equations does not allow us to find
endogenous steady state values for variables such as the interest rate or exchange rate.
The exchange rate, for instance, is modeled with an error correction mechanism with
UIP as the level relation, driving the system to the long run equilibrium steady-state
exchange rate, defined as the exchange rate that leads to an ad-hoc long-run current
account/GDP ratio. This ratio, in turn, is consistent with a steady-state ratio of external
liabilities/GDP. For the interest rate, the use of the Taylor rule assures a long-run
equilibrium compatible with the inflation target and a neutral output gap.
0RGHOLQJWKH%UD]LOLDQ(FRQRP\
The IBGE4 began releasing a quarterly series of the income components of GDP in
the third quarter of 2001. The sample starts in 1991:01. This has made it possible to
construct and run more detailed macro models for the Brazilian economy. However,
4
Brazilian Institute of Geography and Statistics.
8
studies using these new quarterly data are very recent and the constructed models are
still based on a new Keynesian paradigm, using ad hoc relations between the variables
rather than micro-founded structural relationships. However, even considering that our
results are subject to the Lucas critique in some sense, it is still worth working on the
model, due to the fact that it can be used to simultaneously determine the levels of
output, employment, inflation, current account, rate of investment and fiscal balance.
And it is in line, in some parts, with other models developed using this new Brazilian
quarterly data, as in Cavalcanti, Kai and Carvalho (2002). The use of micro-founded
models is on our agenda for the next generation of the Central Bank of Brazil’s
structural models.
The estimation samples depend upon the availability and behavior of the data. We
choose not to homogenize the starting point of the estimations. We are estimating each
equation separately and if we do not consider the full series we would be throwing
information away. Nevertheless when there are severe structural breaks that cannot be
fixed with dummy variables, we decide to exclude the series before the break. Inflation
before 1994 is a good example of this problem. As we did not estimate the equations in
a system, simultaneity bias was avoided using two stage estimations or considering only
lagged variables on the right side of the equations.
7KH$JJUHJDWH'HPDQG6LGH
Aggregate demand is determined by its definitional identity, shown in Equation 1.
In this section, we will model each of its components and related variables such as
taxes, government expenditures and the fiscal deficit (primary concept).
<W = &W + *W + , W + ; WQHW + ∆6 W
:KHUH
<
RXWSXW
&
KRXVHKROGFRQVXPSWLRQ
*
JRYHUQPHQWFRQVXPSWLRQ
,
LQYHVWPHQW
W
W
W
W
; WQHW QHWH[SRUWV
∆6
W
LQYHQWRU\LQYHVWPHQW
9
(1)
Although the IBGE releases both nominal and real data, the real income
components do not sum to meet Equation 15. As a solution, we estimated real income
components using their nominal income share applied to real income. This method
guarantees Equation 1, in real terms, for the whole sample.
+RXVHKROG&RQVXPSWLRQ
Household consumption is by far the most important income component, since it
accounts for more than 60% of output. Although this initial model’s formulation is new
Keynesian, the traditional literature strongly indicates that consumption should be a
function of permanent rather than contemporaneous income. On the other hand,
empirical results suggest that some factors may also cause consumption decisions to be
based on contemporaneous income variables. For example, difficulties in obtaining
loans and weak forward-looking behavior by some agents may cause this behavior.
In this context, we used a very simple specification: a level equation, in
logarithms, with the ratio of consumption-to-disposable income on the left side, as
shown in Equation 2. In order to capture a permanent income effect, we used potential
output6, which we considered a reasonable measure. Contemporaneous income was also
tried, but it failed as a regressor7, so we decided to use the real interest rate to capture
the same behavior. An increase in the real interest rate should decrease the income
growth rate and, in response, the consumption growth rate. Theory indicates that we
should consider medium or long-term interest rates rather than short-term interest rates.
We could obtain the former considering the 6-month swap market, but this would force
us to use a smaller series since we only have 6-month interest rates from 1994Q4 on.
Therefore we decided to use the short-term interest rate. Additionally, we used a step
dummy variable to capture the increase in the consumption-to-disposable income ratio
after 1996, which may have resulted from an improved outlook related to the recent
5
Even when changing base period values in order to guarantee the income identity in some quarters, there
are always some periods in which the income identity does not fit. This is probably due to the fact that
the individual series are not deflated with the same deflator as the income series.
6
See Section 3.2.
7
We tried a weighted average of permanent and contemporaneous income, with weights to be estimated
in the regression, but contemporaneous income was not significant. This is probably due to colinearity
between GDP and potential GDP.
10
stabilization of the economy. The outcome of the estimation, which used an outlier
pulse dummy for 1994Q18, is shown in Table 1, in.
3
&
&
ln GW = α 0 + α1 ⋅ ln WG−1 + α 2 ⋅ UW −1 + ∑ βL ⋅ 6HDVL + α 3 ⋅ '96
L =1
<W −1
<W −2
(2)
:KHUH
&
KRXVHKROGFRQVXPSWLRQ
<W G
GLVSRVDEOHSRWHQWLDOLQFRPH <W
7W G
GLUHFWWD[HV
<
SRWHQWLDORXWSXW
U
VKRUW
W
W
W
G
= <W − 7W G
WHUP
π
6HOLF
UDWH U =
− 3 π −
4
∑ 4
=0
UHDO
, IURP 199141 WR 199541
W
W
W
W
LQWHUHVW
, IURP 19954 2 RQ
L
L
6HOLFW
DQQXDOL]HG%UD]LOLDQRYHUQLJKWQRPLQDOLQWHUHVWUDWH
'
VWHSGXPP\IURP\HDURQ
7DEOH
(TXDWLRQ
0HWKRG 2/6
6DPSOH WR
(VWLP9DOXH
6W'HYLDWLRQ
W
39DOXH
-0.308
0.045
-6.873
0.000
0.254
0.123
2.068
0.046
-0.737
0.262
-2.817
0.008
0.060
0.020
2.933
0.006
0.080
0.018
4.559
0.000
0.103
0.017
6.070
0.000
0.055
0.018
3.060
0.004
0.168
0.040
4.172
0.000
R2 = 0.782
R2Ajust. = 0.740
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV ) S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW) S
&RHIILFLHQW
α0
α1
α2
β1
β2
β3
α3
D94:01
8
Without the dummy variable, all fitting and residual (serial correlation and heteroskedasticity test)
statistics seemed to be acceptable, but we nevertheless observed a huge residual value in 1994Q1 of
about 3 times the regression standard error. This could indicate an outlier. Running the regression with a
pulse dummy for that period, we find that the new coefficients do not significantly change, but fitting
and residual statistics are much better.
9
We considered two measures for the ex-ante real interest rate to correct distortions caused during the
hyperinflation period. In post Real Plan period, the 4-quarter inflation average would represent an
adaptative ex-ante inflation expectation with 75% (calibrated) backward looking. If this procedure were
to be used for the previous period, it would lead to false negative real interest rate, with huge absolute
values, from 1994Q3 to 1995Q1. It is due to the fact that the former procedure is incorrect to generate
ex-ante interest rate in hyperinflationary periods, during which past values averages systematically
underestimate current inflation.
11
,QYHVWPHQW
Like household consumption, investment has an important role in determining
economic activity since it affects both output directly, and potential output indirectly via
capital accumulation. We break the investment series into its construction and
machinery components and model each separately. These two series are best explained
by the 6-month real interest rate. Even though the sample size for these series is smaller,
as explained in the last section, the regressions seemed to have good explanatory power
- good fit and good residual statistics – and we considered them as acceptable
representations of reality. Construction and machinery investment are represented in
Equations 3 and 4, respectively, and estimation results are shown in Tables 2 and 3,
respectively. Total investment is determined by the identity in Equation 5.
The behavior of investment seems to have been passed through changes related to
the way its components (machinery and construction) were affected by the real interest
rate and seasonal movements from 1999 on. Up to 2000Q2, construction investment is
affected by 2-lags of the real interest rate; from 2000Q2 forward, it is also affected by 1lag of the real interest rate. With respect to machinery investment, it does not seem to be
affected by the real interest rate in 1995.
3
,&
,&
6
6ZDS 6
U
'
α
βL ⋅ 6HDVL + β 4 ⋅ 6HDV1 ⋅ '1999
ln G W = α 0 + α1 ⋅ ln GW −1 + α 2 ⋅ UW 6ZDS
+
⋅
⋅
+
∑
3 W −1
2000
−2
L =1
<W −1
<W −2
(3)
3
,0
,0
6
ln G W = $0 + $1 ⋅ ln GW −1 + $2 ⋅ UW 6ZDS
⋅ '1996 + ∑ %L ⋅ 6HDVL + %4 ⋅ 6HDV2 ⋅ '1999
−1
L =1
<W −1
<W − 2
(4)
, = ,& + ,0
(5)
W
W
W
:KHUH
,
WRWDOLQYHVWPHQW
W
,&
FRQVWUXFWLRQLQYHVWPHQW
,0
PDFKLQHU\LQYHVWPHQW
<
UHDOLQFRPH
W
W
W
UW 6ZDS PHGLXP
UXQ
UHDO
, XQWLO 199541
π
LW6ZDS 6 3
=
− π W −L
IURP
4
RQ
,
1995
2
4
∑
L=0 4
W
UW
6ZDS 6
LQWHUHVW
12
UDWH
LW6ZDS DQQXDOL]HGPRQWKVZDSLQWHUHVWUDWH
'\\\\
VWHSGXPP\IRU\HDU³\\\\´RQ
7DEOH
(TXDWLRQ
0HWKRG 2/6
6DPSOH WR
(VWLP9DOXH
6W'HYLDWLRQ
W
39DOXH
-1.393
0.312
-4.466
0.000
0.336
0.159
2.112
0.047
-1.010
0.502
-2.011
0.057
-2.300
0.629
-3.658
0.001
0.097
0.034
2.806
0.011
0.161
0.019
8.554
0.000
0.124
0.020
6.351
0.000
0.073
0.029
2.524
0.020
R2 = 0.831
R2Ajust. = 0.775
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV ) S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW) S
&RHIILFLHQW
α0
α1
α2
α3
β1
β2
β3
β4
7DEOH
(TXDWLRQ
0HWKRG 2/6
6DPSOH WR
(VWLP9DOXH
6W'HYLDWLRQ
W
39DOXH
-1.458
0.309
-4.716
0.000
0.497
0.120
4.142
0.000
-1.625
0.717
-2.265
0.033
0.171
0.023
7.568
0.000
0.212
0.025
8.562
0.000
0.119
0.021
5.600
0.000
-0.065
0.028
-2.303
0.031
R2 = 0.902
R2Ajust. = 0.877
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV ) S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW) S
&RHIILFLHQW
Α0
Α1
Α2
Β1
Β2
Β3
Β4
1HW([SRUWV
Since nominal exports and imports, in US dollars, are modeled in Sections 3.3.3
and 3.3.4, modeling real exports and imports is quite simple. We only have to transform
the currency into Brazilian real and deflate the series, as described in Equations 6 and 7.
Net exports are determined by the definition in Equation 8.
∆ OQ(; ) = ∆ OQ(([SRUWV ) − ∆H − π
W
OQ(0
W
W
)=
W
OQ(,P SRUWV ) − H −
W
; WQHW = ; W− 0 W
13
W
(6)
W
W
(7)
(8)
:KHUH
([SRUWV QRPLQDO86GROODUH[SRUWV
W
,PSRUWV QRPLQDO86GROODULPSRUWV
W
H
H[FKDQJHUDWH ORJDULWKP
W
π
W
,3&$LQIODWLRQUDWH
*RYHUQPHQW&RQVXPSWLRQ
Brazil’s fiscal surplus (primary concept) is defined as total tax revenues minus
non-interest government expenditures. The former are divided, as usual, into direct and
indirect taxes and the latter are divided into government investment and government
consumption.
We model total taxes as a function of lagged total taxes and lagged income in a
simple specification described in Equation 9. In order to capture the most recent
behavior of tax policy, the estimation sample was very short; the result is shown in
Table 4. A step dummy was used to model a level change that occurred from 1999 on
and a pulse dummy was used to capture a 1997:04 outlier. Direct taxes, modeled in
order to have a measure of disposable income (<W − 7W G ), are modeled by a similar
specification, described in Equation 10. Its output is shown in Table 5. Government
investment, as a ratio of total taxes, is modeled as a AR(3) process described in
Equation 11, with outlier dummies. The result is shown in Table 6. Government
consumption is calculated as a residual, in Equation 12, since we determined an
exogenous path for the fiscal deficit (primary concept) as a GDP ratio (annual average).
3
7W7 = α 0 + α 1 ⋅ 7W7−1 + ∑ β L ⋅ <W −L
(9)
L =1
7W G = α + α ⋅ 7W −G + ∑ β L ⋅ <W −L
(10)
, W*
, W*−1
, W*−2
, W*−3
α
α
α
α
=
+
⋅
+
⋅
+
⋅
0
1
2
3
7W7
7W7−1
7W7−2
7W7−3
(11)
L =
3
( )
3
3
M =0
M =1
*W = ∑ 7W7− M − )635W ⋅ ∑ <W − M − ∑ *W − M
M =0
:KHUH
7W7
WRWDOWD[
7W G
GLUHFWWD[
, W*
JRYHUQPHQWLQYHVWPHQW
14
(12)
)635
W
ILVFDOGHILFLW SULPDU\FRQFHSW
7DEOH
(TXDWLRQ
0HWKRG 2/6
6DPSOH WR
(VWLP9DOXH
6W'HYLDWLRQ
W
39DOXH
-25.297
12.177
-2.077
0.053
-0.391
0.097
-4.050
0.001
0.464
0.079
5.889
0.000
-0.290
0.083
-3.475
0.003
0.371
0.079
4.697
0.000
-4.782
1.004
-4.762
0.000
-10.449
1.684
-6.205
0.000
R2 = 0.926
R2Ajust. = 0.899
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV ) S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW) S
&RHIILFLHQW
α0
α1
β1
β2
β3
D99
D97:04
7DEOH
(TXDWLRQ
0HWKRG 2/6
6DPSOH WR
&RHIILFLHQW
(VWLP9DOXH
6W'HYLDWLRQ
W
39DOXH
α0
-29.538
11.458
-2.578
0.020
α1
-0.449
0.104
-4.333
0.000
β1
0.342
0.075
4.546
0.000
β2
-0.231
0.083
-2.779
0.013
β3
0.288
0.078
3.675
0.002
D99
4.224
1.000
4.226
0.001
D97:04
-9.587
1.674
-5.726
0.000
R2 = 0.899
R2Ajust. = 0.863
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV ) S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW) S
7DEOH
(TXDWLRQ
0HWKRG 2/6 :KLWH+HWHURVNHGDVWLFLW\&RQVLVWHQW6WDQGDUG(UURUV &RYDULDQFH
6DPSOH WR
&RHIILFLHQW
(VWLP9DOXH
6W'HYLDWLRQ
W
39DOXH
α0
0.031
0.012
2.497
0.022
α1
0.509
0.142
3.584
0.002
α2
0.447
0.173
2.581
0.019
α3
-0.320
0.171
-1.871
0.078
D96:02
0.032
0.003
9.171
0.000
D97:04
0.072
0.002
31.423
0.000
R2 = 0.859
R2Ajust. = 0.820
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV ) S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW) S
,QYHQWRU\,QYHVWPHQW
We define the inventory dynamics as follows:
6 = (1 − δ )⋅ 6 −1 + ∆6 −1
W
W
15
W
(13)
:KHUH
6
LQYHQWRU\OHYHO
W
∆6
LQYHQWRU\LQYHVWPHQW
δ
GHSUHFLDWLRQUDWHLQYDULDQWRYHUWLPH
W
The basic hypothesis is that firms produce in order to maintain a minimum
inventory as a long run time invariant demand ratio
6
, where Zt = Yt - ∆St. Keeping
=
this assumption in mind and dividing both sides of Equation 13 by Zt , we obtain the
result described in Equation 14, considering that =W +1 = =W ⋅ (1 + JW=+1 ) .
In the steady state, the ratio
equilibrium ratio
6
should converge. Hence, in Equation 15,
=
∆6
depends on the depreciation rate plus the quarterly Z growth.
6
6W +1 6W
6
6 ∆6
− = − JW=+1 ⋅ W +1 − δ ⋅ W + W
=W +1 =W
=W +1
=W = W
(δ + J W= )⋅ =6W
+1
W
=
∆6 W
=W
∴
∆6 W
= δ + J W=+1
6W
(
(14)
)
:KHUH
=W +1 = =W ⋅ (1 + JW=+1 )
J W=
TXDUWHUO\= JURZWKUDWH
W
Assuming that the inventory dynamics over the last decade (1991/2001) have
behaved on average as indicated by Equation 15, we can estimate two latent variables
related to inventory formation: the initial inventory (1991:01) and the depreciation rate
(δ). Note that if the ratio
∆6
is supposed to be constant over the sample, the sum of the
6
quadratic deviations from the sample average must be minimized. In this context, we
just have to find the values of those two non-observed variables that minimize the
previous sum, since the values of J W=+1 are known. Doing this we find 6 and
δ DQQXDOEDVLV
16
(15)
(VWLPDWLQJD'\QDPLF,QYHQWRU\,QYHVWPHQW0RGHO
A deviation of inventory investment from the long run relationship means that the
economy is growing faster or slower than expected, should inventory investment be
below or above the long run relationship, respectively10. Consider the case, for example,
when the growth of demand is slower than expected. Since firms, on the other side, have
already decided to produce in order to achieve the expected demand plus the long run
inventory investment, there must be a positive gap between actual and long run
∆6
6
ratio. What happens next? Firms decide to produce less, running in the opposite
direction, correcting the gap until it is closed again. This may produce an over shooting
∆6
gap, as has been observed in the Brazilian inventory
6
dynamic behavior for the
investment series. On the other hand, production decisions may also be negatively
affected by the real interest rate.
Hence, we model the
∆6
gap as a function of it own lags and of the real interest
6
W
W
∆6
is modeled with the potential Zt growth rate, considering that it
6
must equal the potential output growth rate in the long run, as shown in Equation 16.
Table 7 shows the estimations, which include a dummy to reflect a reduction in the level
that occurred from 1994Q1 to 1997Q4.
W
rate. Potential
W
JW = ∑ α L ⋅ JW −L + α 4 ⋅ (UW −3 − U HT )
3
(16)
L =1
:KHUH
J
W
LQYHQWRU\LQYHVWPHQWUDWLRJDSGHILQHGDV J
W
=
<
∆6
− δ +
− 1
6
< −1
W
W
<
SRWHQWLDORXWSXW
U HT
VKRUWUXQHTXLOLEULXPUHDOLQWHUHVWUDWH
W
10
W
W
Indeed, it can be used as a leading indicator for the output gap, due to the cited reasons.
17
7DEOH
(TXDWLRQ
0HWKRG 2/6
6DPSOH WR
(VWLP9DOXH
6W'HYLDWLRQ
W
39DOXH
-0.930
0.126
-7.377
0.000
-0.793
0.139
-5.685
0.000
-0.622
0.117
-5.320
0.000
-0.431
0.221
-1.947
0.059
-0.052
0.014
-3.691
0.001
R2 = 0.646
R2Ajust. = 0.597
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV ) S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW) S
&RHIILFLHQW
α1
α2
α3
α4
D9497
7KH6XSSO\6LGH
The supply side was modeled using a traditional Cobb-Douglas production
function approach and a Phillips type curve.
&REE'RXJODV3URGXFWLRQ)XQFWLRQ
A Cobb-Douglas production function, with capital and labor, was modeled as
described in Equation 17.
< = $ ⋅ (. ⋅ XFL
W
W
W
W
)α ⋅ (/ )1−α
W
W
W
:KHUH
/ = 3($ ⋅ (1 − X
W
W
. = (1 − δ )⋅ .
W
W
)
4
W
−1 + ∑ β ⋅ ,
L
L
4
∑β
L
=2
L
=2
W
−L
=1
3($
ODERUIRUFH $JH≥\HDUV
X
XQHPSOR\PHQWUDWH
XFL
LQVWDOOHGLQGXVWULDOFDSDFLW\XVHG
δ
GHSUHFLDWLRQUDWH
11
Released on a monthly basis by the %UD]LOLDQ ,QVWLWXWH RI *HRJUDSK\ DQG 6WDWLVWLFV (IBGE), the
3RSXODomR(FRQRPLFDPHQWH$WLYD (PEA) is the potential labor force in the economy, accounting for
employed people and unemployed who looked for a job in the last 30 days, both older than 15 years.
12
Released on a monthly basis by the %UD]LOLDQ,QVWLWXWHRI*HRJUDSK\DQG6WDWLVWLFV (IBGE).
13
Released on a quarterly basis by the %UD]LOLDQ)XQGDomR*HW~OLR9DUJDV (FGV).
14
We considered, for simplification sake, that capital inventory depreciates by the same rate estimated in
the inventory investment section.
18
(17)
The total factor productivity (TPF) series is extracted as a residual of Equation 17,
since the GDP, PEA, u, uci and αW series are known. It is important to emphasize here
that the TPF series (At) depend only on the latent values of βL and on the initial capital
inventory (.). Knowing the TPF series, we can define potential output as in Equation
<
18. The output gap15, written in a logarithmic form K = ln
<
W
W
W
, can be easily derived
as described in Equation 1916.
( )
<W = $W ⋅ (. W ⋅ XFL IH ) ⋅ /WIH
α
[
W
1−α W
(18)
]
KW = α W ⋅ ln (XFLW ) − ln (XFL SH ) + (1 − α W )⋅ [ln (1 − X W ) − ln (1 − X )]
:KHUH
/WIH = 3($W ⋅ (1 − X )
X
QDWXUDOXQHPSOR\PHQWUDWH
IH
IXOOHPSOR\PHQWLQGH[
Note that each quarter’s investment contributes to capital formation up to 4
quarters ahead. The intuition behind this modeling is that certain investments may be
converted into capital faster than others. However, we restricted the investment
percentage lag to be between 2 and 4 quarters.
In this model, we use a dynamic behavior for α, the capital share yield. Actually
this series had a structural break in 1994 as shown in Graph 1, which plots annual
values17. In order to obtain α quarterly values, we considered an alternative procedure
that assured a smooth pattern and annual quarterly values average restriction, as
described and justified in Appendix 1. A visual comparison between annual and
quarterly estimated values is presented in Graph 2. Due to the fact that the IBGE has
only released data up to 2000, we used α forecasts simulated by Equation 26 in order to
estimate the output gap and potential output through 2001.
15
16
“Hiato” in Portuguese.
Note that output gap can be view as a weighted average of the utility capacity gap and the employment
rate gap. However, in the simulations, we did not model X and XFL . Hence, the simulations of the
W
W
output gap are obtained from aggregate demand, output and the potential output described here.
17
Released on a annual basis by the %UD]LOLDQ,QVWLWXWHRI*HRJUDSK\DQG6WDWLVWLFV (IBGE) on “Table 4 Composição do Produto Interno Bruto sob as três óticas - 1996-2000”, Contas Nacionais do Brasil.
19
(19)
In the following, we will present some results based on Solow’s growth model
considering the effect of a volatile path for α. First, we rewrite the potential output
equation considering labor efficiency (Et), as in Equation 21. Equation 22 shows the last
equation transformed into a logarithm first difference. It is a well-known result that, in
steady state, N W IH and ( should grow at the same rate. Accepting that the first
W
difference in the natural logarithm equals the growth rate, we may expect that in the
steady state or in a sufficiently large sample, Equation 22 converts to Equation 2318.
*UDSK6WUXFWXUDOEUHDNRIFDSLWDOVKDUH\LHOG
( ) ⋅ (( )
\W = N W IH
α
1−α W
W
( ) ( )
∆ ln(\W ) = α W ⋅ ∆ ln NW IH + ln NW IH−1 ⋅ ∆(α W ) + (1 − α W )⋅ ∆ ln((W ) − ln((W −1 )⋅ ∆(α W )
]
∆ ln(\W ) = ∆ ln N W IH + ln N W IH−1 − ln((W −1 ) ⋅ ∆(α W )
:KHUH
1
(W = ( $W )1−α
\W =
W
<W
/WIH
NW IH =
.W ⋅ XFL IH
/WIH
Regarding Equation 23, where both sides are functions of non-observed variables
(., βL, XFLIH and X )19, its empirical validation depends heavily upon the estimates – or
calibrations – of those latent variables. Hence, we decided that the estimation of these
variables should include an optimization process aimed at reducing the deviations
between both sides, over a sample in which we could validate the main hypothesis of
this result, namely that N W IH and ( grow by approximately the same rate. There was
W
18
A particular case is when αt is time invariant, the second term on the right side vanishes and we obtain
the known result that
19
N W IH and \ grow by the same rate as well as ( .
W
W
Remember that δ is already estimated.
20
2000.I
(21)
W
( ) [( )
1999.I
1998.I
1997.I
1996.I
1995.I
1994.I
2000
1993.I
0.45
1992.I
0.45
W
1991.I
0.48
*UDSK4XDUWHUO\D VHULHV
1990.I
0.48
1999
0.50
1998
0.50
1997
0.53
1996
0.53
1995
0.55
1994
0.55
1993
0.58
1992
0.58
1991
0.60
1990
0.60
(22)
(23)
another restriction that should be considered here. However, it is a restriction present in
the Phillips curve modeled in the next section. After modeling this curve, we will
present the latent variables estimation process.
3KLOOLSV&XUYH0RGHOLQJ
A Phillips curve, as usual, should consider an expected inflation rate, a measure of
the level of activity such as the output gap, and also a pass-through component, to
capture changes in import prices. Concerning this last component, we model a structural
break in the pass-through coefficient after the move to a floating exchange rate regime
in January 1999. We assume that, under the new regime, movements in exchange rates
will not be perceived to be as permanent as they were in the crawling peg regime. Thus,
expecting this coefficient to be smaller under the floating exchange rate regime, we
introduced a step dummy in a non-linear pass-through coefficient in order to capture the
structural break.
With respect to inflation expectations, we should make some remarks concerning
administrative prices. In Brazil, administrative prices have a high weight in the IPCA20
basket, averaging around 30% of the total index. Forecasting these prices is more
accurate one year ahead, mainly because of the readjustment clauses contained in the
contracts governing these prices. In this context, our Phillips curve models just the "free
prices"21, which should respond to monetary policy22. But free price inflation
expectations must be a function of full inflation, with backward and forward
components. A final feature of the specification is a verticality long run restriction: there
must be no intercept coefficient, and backward and forward coefficients and the passthrough coefficient must sum to 1. The absence of an intercept coefficient is the final
restriction to be used on the latent variables estimation mentioned earlier. This
restriction should be considered in the estimation described in the next section.
Attempting to capture all those features, the Phillips curve specification is given in
Equation 2423.
20
Measured and released on a monthly basis by the Brazilian Institute of Geography and Statistics
(IBGE), IPCA is the consumer price index chosen for the purpose of gauging yearly inflation targets in
the Inflation Target system.
21
Purging government prices.
22
In our simulations, we considered the government inflation forecasts up to one year ahead based on the
contracts. But, for longer forecasting horizons, we assume that government prices should move together
with free prices.
23
The Phillips curve estimation output will be shown in Table 8 Section 3.2.3.
21
W
IUHH
=
1
⋅
W −1
+ (1 −
1
−
3
−
4
⋅ ' IO )⋅ (W
W +1
+
2
⋅ KW −1 + (
3
+
4
⋅ ' IO )⋅
(
HW +
W
)
:KHUHDOOYDULDEOHVDUHVSHFLILHGLQORJDULWKPV
π W IUHH
IUHHLQIODWLRQUDWHFRQVLGHULQJWKH,3&$
π
IXOOLQIODWLRQUDWHFRQVLGHULQJWKH,3&$
W
π *
IRUHLJQLQIODWLRQUDWHFRQVLGHULQJWKH8633,
K
RXWSXWJDS KLDWRLQ3RUWXJXHVH
H
H[FKDQJHUDWH
'
VWHSGXPP\EHIRUHH[FKDQJHUDWHUHJLPHFKDQJLQJDQGDIWHU
W
W
W
IO
(VWLPDWLRQRI1RQ2EVHUYHG9DULDEOHV
As commented in Boone, Juillard, Laxton and N’Diaye (2002), NAIRU estimation
processes that do not exploit information about inflation may result in inefficient
historical measures of the NAIRU, biased parameter estimates, as well as inefficient
forecasts of the NAIRU. In this work, rather than taking into account the path of the
NAIRU, we assumed a constant natural rate of unemployment. Nevertheless, as this last
critique suggests, we must use inflation information to estimate the latent variables. In
this sense, we considered a Phillips curve restriction described as follows:
As previously mentioned, the latent variables estimation process must
approximate both sides of Equation 23, over a sample in which we could validate the
main hypothesis of this result, meaning that N W IH should converge. On the other hand,
Equation 24 should have no intercept. We assume here that the only reason why the
Phillips curve should have a significant intercept is from a misestimated output gap.
Hence, in each interaction of the estimation process, there are two phases: the first
consists of a optimization process in which we estimate the latent variables in order to
minimize the quadratic sum of the difference between both sides of Equation 23, as
described in System 25. The sample used in the optimization process is from 1995:1 to
2001:4, because in this period N W IH seemed to be very stable.
22
(24)
min ∑
W
{ ln(\ )− [ ln(N )+ [ln(N )− ln(( )]⋅ ( )]}
IH
W
IH
W −1
W
2
W −1
W
(25)
6XEMHFW WR
0≤
L
XFL IH DQG X ≤ 1
The second phase of each interaction consists of estimating Equation 24 with an
intercept. Assuming that the only misestimated variable is the output gap generated in
the first phase, and assuming that this error should be related to the XFLHI , the intercept
value should be equal to α 2 ⋅ ε K ⋅ α , where ε K ⋅ α is the XFLHI error multiplied by average
αW. Therefore, we could estimate the XFLHI error, and a new measure of XFLHI . With the
new XFLHI estimated, we could run phase one again but with the restriction that XFLHI
should be equal to theestimated value.
The process generates another value for the natural rate of unemployment that
should be consistent with the imposed XFLHI. With new estimates, we could do phase 2
again restarting the cycle of interactions. It is important to note that, in each second
phase, the intercept got less and less significant. With the estimated parameters, we
could estimate24 Equation 24 without an intercept term, but including two outlier
dummies, as shown in Table 8. In convergence, the process estimated:
. 0 = 904.25 XFL IH = 84.93% X = 5.29% β 4 = 0 β 3 = 0 β 2 = 1
7DEOH
(TXDWLRQ
0HWKRG 76/6
6DPSOH WR
(VWLP9DOXH
6W'HYLDWLRQ
W
39DOXH
0.212
0.122
1.743
0.096
0.311
0.062
5.009
0.000
0.510
0.148
3.452
0.002
-0.453
0.149
-3.046
0.006
0.026
0.009
2.927
0.008
0.018
0.008
2.079
0.050
R2 = 0.809
R2Ajust. = 0.763
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV 1⋅5 S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW) S
&RHIILFLHQW
α1
α2
α3
α4
α99:04
α00:03
24
This equation was estimated using Two-Stage Least Squares with lagged inflation and inflation
forecasts made by a univariated model, as instrumental variables for the forward component. The
results are robust and we could confirm that pass-through coefficient reduced from about 51% to 6%
after changing the exchange rate regime.
23
0RGHOLQJWKH&RPSRQHQWVRIWKH&REE'RXJODV3URGXFWLRQ)XQFWLRQ
Regarding the fact that there was no prior information about the future dynamics
of αt, we simply estimated an ARIMA (3;1;0) model with no intercept in order to avoid
a non-justified trend, as shown in Equation 26. The estimation outcome is described in
Table 9. The total factor productivity (At) was modeled, in logarithm, by a seasonal
ARIMA (2;1;0), as shown in Equation 27 and its estimation outcome is described in
Table 10. PEA was modeled in logarithms with an autoregressive component, linear
trend and seasonality. In an effort to account for a level change that occurred after
1994:3, we introduced a step dummy. The specification is shown in Equation 28 and the
estimation outcome is described in Table 11. And, finally, the capital inventory,
obtained by the estimated parameters into its definition in Equation 5, is shown in
Equation 29.
∆α = β 1 ⋅ ∆α
W
W
+ β 2 ⋅ ∆α
−1
W
+ β 3 ⋅ ∆α
−2
W
(26)
−3
3
∆ ln ($ ) = β 0 + β 1 ⋅ ∆ ln ($ −1 ) + β 2 ⋅ ∆ ln ($ − 2 ) + ∑ α ⋅ 6HDV
W
W
W
L
L
L
(27)
=1
3
ln (3($ ) = β 0 + β 1 ⋅ ln (3($ −1 ) + β 2 ⋅ 7UHQG 91:01 + ∑ α ⋅ 6HDV + β 3 ⋅ '94:03
W
W
L
L
. = 0.98 ⋅ .
W
7DEOH
W
−1
+,
W
L
−2
(TXDWLRQ
0HWKRG 2/6
6DPSOH WR
(VWLP9DOXH
6W'HYLDWLRQ
W
39DOXH
2.200
0.117
18.784
0.000
-1.926
0.176
-10.930
0.000
0.619
0.090
6.851
0.000
R2 = 0.985
R2Ajust. = 0.983
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV ) S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW) S
&RHIILFLHQW
β1
β2
β3
24
(28)
=1
(29)
7DEOH
(TXDWLRQ
0HWKRG 2/6
6DPSOH WR
(VWLP9DOXH
6W'HYLDWLRQ
W
39DOXH
-0.052
0.012
-4.294
0.000
0.745
0.182
4.082
0.000
-0.380
0.148
-2.568
0.017
0.083
0.020
4.160
0.000
0.080
0.017
4.626
0.000
0.056
0.017
3.316
0.003
R2 = 0.673
R2Ajust. = 0.602
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV ) S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW) S
&RHIILFLHQW
β0
β1
β2
α1
α2
α3
7DEOH
(TXDWLRQ
0HWKRG 2/6
6DPSOH WR
(VWLP9DOXH
6W'HYLDWLRQ
W
39DOXH
6.2527
1.805
3.464
0.001
0.6225
0.109
5.710
0.000
0.0014
0.000
2.947
0.006
-0.0058
0.003
-2.017
0.051
0.0041
0.003
1.434
0.160
0.0027
0.003
0.979
0.334
0.0090
0.004
2.470
0.018
R2 = 0.987
R2Ajust. = 0.985
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV ) S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW) S
&RHIILFLHQW
β0
β1
β2
α1
α2
α3
β3
([WHUQDO%ORFN
The exchange rate and sovereign risk premium are modeled in the next two
subsections on a monthly basis in order to capture their movements more precisely. In
quarterly analysis, important information may be lost. In that case, it is important to reestimate quarterly coefficients in order to keep the same impulse response features,
which are not guaranteed when we consider the same autoregressive and error
correction coefficients in the monthly and quarterly specifications.
The procedure is very simple. Long run coefficients, excluding those that are
autoregressive, should be the same in both frequencies, and convergence velocities must
be the same as well. If in the monthly specification the autoregressive coefficient is 0.7,
for instance, it should be (0.7)^3 in the quarterly specification. Another example is the
error correction term. If in the monthly specification it takes 9 months to achieve halflife, for instance, it should take 3 quarters in the quarterly specification.
25
When it is not that easy to derive quarterly coefficients, we recommend a
optimization procedure, in which the coefficients are chosen in order to minimize a
fitting function25 between the quarterly average of monthly values of the original
impulse response and quarterly values of the quarterly impulse response, generated by
the coefficients to be determined.
([FKDQJH5DWH
We modeled the exchange rate with an equation based on a UIP non-arbitrage
condition. As described completely in Muinhos, Alves and Riella (2002)26, there is a
strong short run first difference relationship between the Brazilian exchange rate, CBond spread over treasury27 and interest rate differential, all in nominal vales28. But,
surprisingly, despite the fact that all coefficients are significant and have the expected
sign, they are all greater than unity, in absolute values, as predicted by UIP, even when
correcting for the sovereign risk premium. This may be a result of frictions and
assymetric information.
Nevertheless, the UIP condition should prevail in the long run. With this in mind,
we carry out an error correction model for the first difference of the real exchange rate ,
capturing the short and long run dynamics. The long run and the error correction first
difference specifications are described in Equation 30a and 30b, respectively. Their
outcome estimations are shown in Table 12a and 12b, respectively.
We observed that, in the long run level specification; permanent shocks to the risk
premium produce an over-shooting behavior, since there is a strong contemporaneous
response that then decreases after one period. It is interesting to note that the permanent
coefficient is very close to the 1, as predicted by UIP. As reported in the empirical
literature, the real interest rate differential is not significant, but has the correct sign. As
25
Absolute errors sum, squared errors sum, fourth power errors sum, and so on, depending on the
influence of smaller errors is intending to affect the fitting function.
26
In this paper, we comment about the UIP puzzle, the literature about exchange rate, and some results
cited here.
27
The authors found out that C-Bond spread over treasury should embody the information of the
Brazilian sovereign risk and should be free of exchange rate risk, as justified in the cited paper.
28
A first difference logarithm equation is the left-hand side variable, because the authors could not reject
the null hypothesis that exchange rate has a unit root in the used sample. The exchange rate
expectation was modeled as a lagged exchange rate plus the expected inflation differential, in order to
maintain the real exchange rate constant. The risk premium was modeled as a linear function of the CBond spread over treasury. And, instead of imposing a unitary interest rate differential coefficient,
with negative signal, they dropped this arbitrage condition and estimated the coefficients.
26
a solution, we imposed a UIP predicted coefficient equal to -1. Regarding the
expectation term, we considered an adaptative weighted average with a backwardlooking component, a forward-looking component and a long run equilibrium real
exchange rate. The latter is calibrated as the real exchange rate necessary to achieve an
ad hoc current account surplus in the long run in each simulation.
The short run specification, in first difference, is purely backward looking, but no
theory coefficient was imposed. It is interesting to note that real exchange rate changes
are affected by the change of one-lagged real interest rate differential instead of the
contemporaneous differential. All coefficients were significant and with the correct
sign, but as in Muinhos, Alves and Riella (2002), much greater than the129 as predicted
by UIP. And the error correction term is slightly greater than one, indicating an overshooting returning to the equilibrium, with a vanishing oscillatory behavior, which
confirms empirical evidence in Brazil. Note that those coefficients represent monthly
behavior, which is more volatile than the quarterly behavior. When quarterly
coefficients are calculated by the procedure described previously, this volatility is
smoothed. We also added two dummy variables in order to capture outliers.
ε W = − α − α ⋅ ε HT + α ⋅ ε W − + α ⋅ (W ε W + − (UW − UW I )+ α ⋅ 6&%RQGW + α ⋅ 6&%RQGW − + µ
∆ε W = $ ⋅ ∆ε W − + $ ⋅ ∆ (UW − − UW −I )+ $ ⋅ ∆6&%RQG W + $ ⋅ µ W −
:KHUHDOOYDULDEOHVDUHVSHFLILHGLQORJDULWKPV
ε
W
UHDOH[FKDQJHUDWH
ε HT
HTXLOLEULXPUHDOH[FKDQJHUDWH
U
VKRUWUXQUHDOLQWHUHVWUDWH UW
W
I
=
6HOLFW W,3&$
− ∑ −
L =
IRUHLJQVKRUWUXQUHDOLQWHUHVWUDWH UW
U
W
6&%RQG &%RQGVSUHDGRYHUWUHDVXU\
W
µ
29
W
HUURUWHUPVXSSRVHGWREHUDQGRP
In absolute values.
27
I
=
)HG)XQGVW W33,
− ∑ −L
L =
W
(30a)
(30b)
0HWKRG 76/6
W
−L
7DEOHD (TXDWLRQD
6DPSOH 0DUWR6HS
,QVWUXPHQWV
L = 1WR12 U −1 − U −1 6&%RQG
I
W
&RHIILFLHQW
α
α
α
α
(VWLP9DOXH
0.692
0.278
20.559
-19.620
W
6W'HYLDWLRQ
0.072
0.073
7.097
7.085
W
−L
L = 1WR12
7
9.578
3.811
2.897
-2.769
39DOXH
0.000
0.000
0.006
0.009
R2 = 0.978
R2Ajust. = 0.976
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV 2EV 5VTXDUHG S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW2EV 5VTXDUHG S
0HWKRG 76/6
W −
(
)
7DEOHE (TXDWLRQE
6DPSOH $SUWR2FW
,QVWUXPHQWV
UW − − UW −I 6&%RQG W −L L = WR
&RHIILFLHQW
$
$
$
$
'
'
(VWLP9DOXH
0.826
-3.900
16.384
-1.317
-0.058
-0.058
W −
&XU$FF*'3W −L L = WR
6W'HYLDWLRQ
0.134
2.015
7.016
0.281
0.020
0.021
7
6.171
-1.936
2.335
-4.683
-2.892
-2.812
39DOXH
0.000
0.061
0.025
0.000
0.006
0.008
R2 = 0.859
R2Ajust. = 0.840
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV ) S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW) S
5LVN3UHPLXP
We modeled C-Bond spread over treasury, used in Equations 30 and 31, in order
to capture sovereign risk perceptions generated by fiscal variables, external trade and
solvency/liquidity variables.30 The downward trend in the C-bond yield curve as it gets
closer to its maturity was not considered in the simulations. Using a parsimonious
criterion, we focused on relevant variables and avoided over fittingestimations. In the
best-fit estimation, foreign reserves (%GDP), public debt (%GDP) and current account
balance (%GDP) coefficients were significant and representative of fiscal variables,
external trade and solvency/liquidity indicators
31
. The specification is described in
Equation 31 and its output estimation, by TSLS, is shown in Table 13.
30
31
For a detailed description of the treatment on the risk premium see Muinhos, Alves and Riella (2002),
Defying intuition, exchange rate volatility did not have significant explanatory power for the risk
premium.
28
6&%RQG = α 0 + α 1 ⋅ 6&%RQG −1 + α 2 ⋅ ∆ Re V / *'3 +
W
W
W
(31)
+ α 3 ⋅ ∆3' / *'3 + α 4 ⋅ &XU$F / *'3
W
W
:KHUH
5HV*'3
IRUHLJQUHVHUYHV *'3
3'*'3
SXEOLFGHEW *'3
W
PRQWK
PRQWK
W
&XU$F*'3 PRQWKDFFXPXODWHGFXUUHQWDFFRXQWEDODQFH *'3
PRQWK
W
7DEOH (TXDWLRQ
6DPSOH -DQWR'HF
,QVWUXPHQWV
6&%RQG W − 5H V3,%W 5H V3,%W ∆'/73,%W 7&RU3,%W H HW −
0HWKRG 76/6
&RHIILFLHQW
α0
α1
α2
α3
α4
(VWLP9DOXH
6W'HYLDWLRQ
W
39DOXH
-0.0004
0.0044
-0.0821
0.9348
0.8597
0.0472
18.2134
0.0000
-0.8396
0.4274
-1.9646
0.0536
0.1360
0.0788
1.7259
0.0890
-0.2536
0.1074
-2.3608
0.0212
R2 = 0.874
R2Ajust. = 0.867
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV ) S
-DUTXH%HUD1RUPDOLW\7HVW S
White Heteroskedasticity Test: F = 0.910 (p =0.543)
([SRUWV
In this section and in the next, we present our nominal net export modeling in US
dollars. For simplification sake, we modeled export and import quantities. Prices are
modeled as ARMA processes, as described in Muinhos, Alves and Riella (2002).
Equation 32 presents the quarterly estimates for the export quantity index. The sample
starts in 1988 and the coefficients and the t statistics are in Table 14. In the literature
there are some papers that also estimate the price (real exchange rate) and income
(world GDP) elasticities for exports. Pastore e Pinotti (1999) e Gonzaga e Bevilacqua
(1997) found similar coefficients for the income elasticity. However the price elasticity
of 0.14 was smaller than found for those papers. Pastore e Pinnoti (1999) for example
estimated at 0.24 for the price elasticity and 0.81 for the world income elasticity.
3
exp = α 0 + α1 ⋅ exp −1 + α 2 ⋅ \ * + α 3 ⋅θ −1 + α 4 ⋅ OS[W + ∑ β ⋅ 6HDV +
W
W
W
W
W
M
+ α 5 ⋅ '91:03
29
=1
M
M
(32)
:KHUH
H[S
\
TXDQWLWDWLYHLQGH[RIH[SRUWVIRUSHULRGW
W
ZRUOG*'3IRUSHULRGWPHDVXUHGDVWKHORJRIWKHZRUOGLPSRUWV
W
θ
UHDOH[FKDQJHUDWHIRUSHULRGW
W
OS[W
SULFHLQGH[RIH[SRUWVIRUSHULRGW
6HDV
VHDVRQDOGXPPLHV
W
M
'
RXWOLHUGXPP\IRU
7DEOH (TXDWLRQ
0HWKRG 2/6
&RHIILFLHQW (VWLP9DOXH
6W'HYLDWLRQ
α0
-0.572
0.804
α1
0.559
0.081
α2
0.445
0.085
α3
0.139
0.054
α4
-0.257
0.137
β1
-0.090
0.024
β2
0.151
0.026
β3
0.098
0.022
α5
-0.239
0.056
6DPSOH 1988:01 to 2001:02
7
39DOXH
-0.711
0.481
6.865
0.000
5.249
0.000
2.561
0.014
-1.874
0.067
-3.729
0.001
5.822
0.000
4.554
0.000
-4.307
0.000
R2 = 0.954
R2Ajust. = 0.946
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV ) S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW) S
,PSRUWV
Equation 33 presents the estimations of the import quantity index, with
coefficients and t statistic value shown in Table 15. The quantity index for imports
presents a structural break in the first half of nineties, which makes it necessary to
introduce a level dummy in order to avoid a unit root process.
Our coefficient for the real exchange rate is smaller that the one usually seen in
the literature. However the income-elasticity is closer to other estimations. Pastore e
Pinotti (1999) found the price-elasticity of (-0,96) and their income elasticity is 1,02
(taking into account industrial production). Even considering a level dummy after 1993,
it seems that the income elasticity still presents a structural break after that year. When
we shrink the sample, this coefficient almost doubles.
3
LPS W = α 0 + α 1 ⋅ LPS W −1 + α 2 ⋅ \ W + α 3 ⋅ θ W −1 + ∑ β M ⋅ 6HDV M +
M =1
+ α 4 ⋅ 'LPS + ∑ β DD:WW ⋅ 'DD:WW
DD:WW
30
(33)
:KHUH
LPS
W
\
TXDQWLWDWLYHLQGH[IRULPSRUWVLQSHULRGW
GRPHVWLF*'3LQSHULRGW
W
θ
UHDOH[FKDQJHUDWHLQSHULRGW
W
6HDV
VHDVRQDOGXPPLHVIRUWKHSHULRGM
'LPS
VWHSGXPP\WKDWLVXQWLODQGDIWHUEHLQJLQEHWZHHQ
'DDWW
RXWOLHUGXPPLHVIRUDQG
M
7DEOH (TXDWLRQ
0HWKRG 2/6
&RHIILFLHQW
(VWLP9DOXH
6W'HYLDWLRQ
6DPSOH WR
7
39DOXH
α0
-3.077
1.610
-1.911
0.063
α1
0.568
0.081
7.039
0.000
α2
1.170
0.413
2.831
0.007
α3
-0.191
0.082
-2.337
0.024
β1
-0.102
0.038
-2.675
0.011
β2
-0.016
0.036
-0.442
0.661
β3
0.024
0.040
0.595
0.555
'imp
0.332
0.099
3.367
0.002
'95:03
-0.189
0.094
-2.004
0.051
'97:01
-0.305
0.090
-3.382
0.002
'99:01
-0.206
0.089
-2.302
0.026
R2 = 0.985
R2Ajust. = 0.982
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV ) S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW) S
)RUHLJQ'LUHFW,QYHVWPHQW
Equation 34 presents the estimated equation for Foreign Direct Investment, with
outcomes shown in Table 16. The presence of profit and the first difference of the risk
premium in the FDI equation are important, not only in terms of significance but also
with expected sign. An increase in the risk premium is a leading indicator of a decrease
in FDI, while an increase in profit remittances is an indicator of an increase in FDI.
)', = α 0 + α 1 ⋅ )',
W
W
−1
+ α 2 ⋅ ∆( 6&%RQG −1 ) + α 3 ⋅ \ −1 + α 4 ⋅ OXFUR −1
W
W
:KHUH:
FDIt
Foreign Direct Investment in period t , in 2000 US$;
∆(SCBondt)
first difference in the spread of C-Bond in period t;
31
W
(34)
yt
GDP in period t;
lucrot
net profit in the Balance of Payment in period t, is 2000 US$.
7DEOH (TXDWLRQ
0HWKRG: 2/6
6DPSOH: WR
&RHIILFLHQW (VWLP9DOXH
6W'HYLDWLRQ
7
39DOXH
α0
-18.840
8.663
-2.175
0.041
α1
0.616
0.095
6.478
0.000
α2
-14.936
4.015
-3.720
0.001
α3
3.942
1.906
2.068
0.051
α4
0.454
0.109
4.164
0.000
R2 = 0.904
R2Ajust. = 0.887
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV ) S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW) S
0RQHWDU\DQG)LVFDO%ORFN
For simulation purposes, the interest rate follows a standard Taylor rule
described in Equation 35, where γ1 is the weight on the persistence of the interest rate, γ2
is the weight on inflation and γ3 is the weight on the output gap. The variable LW(T is the
long run equilibrium of the interest rate, and it was set to be around 6%. In the baseline
scenario, the values were chosen in an ad-hoc manner and γ1 is 0.8, γ2 is 1.3 and γ3 is 0.8.
4
HW
(T .
LW = γ 1 ⋅ LW −1 + (1 − γ 1 )⋅ γ 2 ⋅ ∑ π W −L − π W7−arg
L
+ γ 3 ⋅ KW −1 + LW
L =1
(
)
(35)
Although the traditional method of forecasting 6-month rates is by extracting
information from the term structure, empirical results suggest that, due to the low
liquidity in the market for futures contracts, yield curve information is not a good
forecaster for future 6-month rates. Hence, we modeled the 6-month interest rate as a
function of the contemporaneous Selic rate and contemporaneous and lagged risk
premium values, as described in Equation 36. Outlier dummies were also used. The
estimation outcome is shown in Table 17. Fiscal debt can be broken into three
components: external fiscal debt, internal debt indexed to the change in exchange rate
plus a risk premium, and internal debt denominated in the Selic rate. Thus, we modeled
these fiscal debt components, subtracting the fiscal surplus, as in Equation 37.
6ZDS6 = α 0 + α 1 ⋅ 6ZDS6 −1 + α 2 ⋅ 6HOLF + α 3 ⋅ 6HOLF −1 + α 4 ⋅ ∆6&%RQG
W
W
W
32
W
W
(36)
'
([W
'
,QW
W
W
(
= ' −1 ⋅ 1 + L
([W
W
= ' −1
,QW 1R
W
exchange rate variation
W
' ='
W
W
W
W
(
,QW :LWK
([W
)⋅ (1 + 5LVN )
⋅ (1 + 6HOLF )+
⋅ (1 + 5LVN )⋅ (1 +
exchange rate variation
W
+ ' −1
I
W
+'
,QW
W
− )6
(
W
)
W
7DEOH (TXDWLRQ
0HWKRG 2/6
6DPSOH 1988:01 to 2001:02
&RHIILFLHQW (VWLP9DOXH 6W'HYLDWLRQ W
39DOXH
α0
0.029
0.008
3.599
0.001
α1
0.731
0.122
5.999
0.000
α2
0.416
0.151
2.764
0.008
α3
-0.293
0.078
-3.763
0.000
α4
2.054
0.487
4.215
0.000
D98:08
-0.038
0.019
-2.030
0.047
D99:01
0.043
0.013
3.230
0.002
D99:02
0.072
0.014
5.270
0.000
R2 = 0.955
R2Ajust. = 0.950
%UHXVFK*RGIUH\6HULDO&RUUHODWLRQ/07HVW ODJV ) S
-DUTXH%HUD1RUPDOLW\7HVW S
:KLWH+HWHURVNHGDVWLFLW\7HVW) S
6LPXODWLRQV
The model is simulated in a Matlab/Simulink environment until 2100:4, but we
will only show the first 30 years of results for simplicity. Our closure rule is an ad-hoc
end-point for the current account/GDP ratio, which brings us to a long-run equilibrium
value for the real exchange rate. The current account surplus was set at 0% for the last
period of the simulation. The primary fiscal surplus follows an exogenous vanishing
path to the long run. We also assumed that world and domestic growth converge in the
long run.
In the first simulation, whose graphics are in the Appendix (see Simulation 1),
we set different weights for the Taylor rule. The baseline simulation presents γ1, γ2 and
γ3 as 0.85, 1.30 and 0.30, respectively. A more aggressive rule against inflation sets γ1,
γ2 and γ3 equal to 0.85, 1.50 and 0.10, respectively, meaning more weight in the
inflation gap from the target and less weight in the output gap. An opposite rule with
higher weight on the output gap is γ1, γ2 and γ3 as 0.85, 1.10 and 0.50, respectively.
The results show a good convergence of the model. Inflation goes toward the target,
GDP grows close to the potential, and fiscal debt is decreasing in the medium run. The
33
(37)
comparisons between the three Taylor rules show that the more aggressive monetary
policy leads to lower GDP growth and a lower fiscal surplus.
The second simulation (see Simulation 2) is presented in terms of impulse
responses. Three types of impulse shocks are simulated: a temporary positive shock to
the nominal Selic interest rate, a temporary positive shock to the C-Bond spread and a
temporary positive shock to administered prices.
An increase in the Selic interest rate has the greatest impact on inflation with a
lag of 7 quarters. Risk premium shocks affect inflation through different channels. The
first is the exchange rate channel, which causes an increase in inflation via the pass
through channel. The second channel is the medium run interest rate; in this channel, an
increase in the risk premium causes an increase in the medium term interest rate and a
corresponding slight decrease in inflation (Swap06) via a decrease in the output gap and
the GDP growth rate. But the average impact of an increase in the risk premium on
inflation is positive until it vanishes in the long run.
Administered price shocks cause an increase in inflation, as expected.
Administered price shocks instantly decrease the real interest rate, increasing the GDP
growth rate and inflation. But this is followed by an increase in the nominal interest
rate, in order to bring inflation back to the target, which increases the real interest rate,
decreasing the GDP growth rate. This oscillatory path of the GDP growth rate
continues, depending on the weight of the output gap in the Taylor rule, but in a
vanishing path.
Those temporary shocks have temporary effects on the real exchange rate, but
permanent effects on the nominal exchange rate. Positive risk premium and
administered price shocks depreciate the nominal exchange rate while positive nominal
interest rate shocks cause nominal appreciation.
Positive nominal interest rate shocks increase the sovereign risk premium by
worsening the public debt. Indeed, the effects of positive nominal interest rate shocks on
public debt, although not permanent, take so long to vanish that they appear permanent.
Positive administered price shocks decrease the public debt, as expected, via the
inflationary tax effect.
34
&RQFOXVLRQDQG1H[W6WHSV
The objective of this paper was to present the main features of the Keynesian
macroeconomic model in development at the Central Bank of Brazil. As this paper is
still a work in progress, we have many more steps to accomplish and close conclusions.
The model with disaggregated demand and potential output with a production function
demonstrates good convergence. We still can detect problems with the import and
consumption equations. The simulations simultaneously brought about consistent paths
for output, employment, inflation, the current account, the rate of investment and the
fiscal balance. However the long-run equilibrium of some variables are dependent on
the end-points for the interest rate and exchange rate. Another problem that we have
with this kind of model is that it is not robust to the Lucas critique. Some of the
parameters may vary through the sample period due to policy changes. Aware of this
limitation we still consider Keynesian models useful tools for identifying the
transmission mechanism of the monetary policy. The simulations have to be considered
with caution especially for Brazilian economy, because there are many cases of
structural breaks and policy swings.
As future goals we can point out:
• Consumption disaggregated in durable and non-durable goods;
• A forward looking rational expectations term for inflation in the Phillips curve
and for the exchange rate in the UIP equation;
• More equations for the wage sector, using the Phillips curve that includes the
unit labor cost;
• A more structured fiscal block;
• A production function with more than one kind of capital.
35
%LEOLRJUDSK\
Agenor, Pierre-Richard and Montiel Peter (1996) “Development Macroeconomics”
Princeton University Press
Alves, Sergio A. Lago (2001) “Evaluation of the Central Bank of Brazil Structural
Model's Inflation Forecasts in an Inflation Targeting Framework” %DQFR&HQWUDOGR
%UDVLO:RUNLQJ3DSHU6HULHV nº 16.
Bansal, Ravi e Magnus Dahlquist (1999) “The Forward Premium Puzzle: Different
Tales from Developed and Emerging Economies” &(35'LVFXVVLRQ3DSHU2169.
Bogdanski, Joel, Tombini, Alexandre e Werlang Sergio (2000) “Implementing Inflation
Targeting in Brazil” %DQFR&HQWUDOGR%UDVLO:RUNLQJ3DSHU6HULHV nº1.
Boone, Laurence, Michel Juillard, Douglas Laxton and Papa N’Diaye (2002) “How
Well Do Alternative Time-Varying Parameter Models Os The NAIRU Help
Policymakers Forecasts Unemployment And Inflation In The OECD Countries?”
,0) :RUNLQJ 3DSHU, presented at the Eighth International Conference of The
Society for Computational Economics, CEF2002 (Aix en Provence, France,
June/2002).
García, Carlos, Pablo García, Igal Magendzo e Jorge Restrepo (2002) “A Medium-Sized
Macroeconometric Model of the Monetary Transmission Mechanism in Chile”
&RQIHUHQFLD 0RGHORV 'H (TXLOLEULR *HQHUDO 3DUD /D (FRQRPtD &KLOHQD,
organized by Central Bank of Chile (Santiago, Abril/2002)
Cavalcanti, Marco A. F. H., Hamilton Kai and Leonardo Carvalho (2002) “Principais
Características do Modelo Macroeconômico do IPEA” ,3($, seminary presented on
June 26, 2002
Gonzaga Gustavo e Bevilacqua Afonso (1997) Relatório Consultoria da Banco Central
do Brasil, mimeo
Reis, Eustaquio, Cavalcanti Marco Antônio, Castro, Alexandre Rossi Jr. Jose Araújo
Emerson e Hernandes Beatriz (1999) “Model for Projections and Simulations of the
Brazilian Economy” ,3($7H[WRSDUD'LVFXVVmRQ
McCallum, Ben. (1994) “ A Reconsideration of the Uncovered Interest Parity
Relationship” -RXUQDORI0RQHWDU\(FRQRPLFV vol. 33 pp 105-132
36
Meredith, G. & Chinn, M. (1998): Long-Horizon Uncovered Interest Rate Parity,
1%(5:RUNLQJ3DSHU 6797
Min, Hong G. (1998). “Determinants of Emerging Markets Bond Spread: Do Economic
Fundamentals Matter” World Bank mimeo.
Muinhos, Marcelo & Alves, Sergio A. L. e Riella, Gil (2002) “Modelo Estrutural Com
Setor Externo: Endogenização do Prêmio de Risco e do Câmbio” %DQFR&HQWUDOGR
%UDVLO:RUNLQJ3DSHU6HULHV no 42.
Muinhos, Marcelo & Freitas, Paulo e Araújo, Fabio (2001) “Uncovered Interest Parity
with Fundamentals: A Brazilian Exchange Rate Forecast Model” %DQFR&HQWUDOGR
%UDVLO:RUNLQJ3DSHU6HULHV nº19.
Pastore Afonso e Pinotti Maria Cristina (1999) Boletim Periódico - ACC Pastore
Consultoria
Wadhwani, Sushil B. (1999) - &XUUHQF\3X]]OHV LSE Lecture on 16 September 1999
37
$SSHQGL[ 2EWDLQLQJD4XDUWHUO\9DOXHV)RU7KH3URGXFWLRQ)XQFWLRQ
In order to obtain quarterly values for α, three alternatives were available. The
first is to maintain the annual values in each quarter. However, as the resulting quarterly
series present a step shaped pattern, with abrupt level changes on the first quarter of
every year, this alternative was discarded because we expect a smoother behavior. A
natural choice, as a second alternative, is to consider a filtered series, obtained by a HP
filter, for instance, instead of the original one. Again, however, this resulted in a
undesired behavior: although the average quarterly values of each year should be equal
to the original annual values, this was not the case when using the regular filtering
process. Hence, we considered a third alternative that assured the following two
features: the smoothness and the restriction on the average of the annual quarterly
values. This alternative was based on the quarterly data generating process, based on an
annual frequency data, presented in Alves (2001) and is described in System 38.
$
≡ Annual capital share yield series, with n observations
$
W
≡ Particular value for α $ in year t : t ∈ [1, n ]
One wishes to estimate the quarterly capital share yield series α 4 such as:
4
4
W4
4
,
W 4
≡ Quarterly capital share yield series, with 4n observations
≡ Particular value for
4
in quarter Q of year t : t ∈ [1, n ], Q ∈ [1, 4]
series should ensure that:
Minimize
Subject to
L = ∑∑ (∆ 2
Q
W
=1
4
∑
Q =1
4
W4
4
4
4
W4
=1
= 4⋅
38
4
W
,
)
2
∀ t ∈ [1, Q ]
(38)
$SSHQGL[ 6LPXODWLRQ*UDSKLFV
6LPXODWLRQ
,QIODWLRQ
WD \ORU
WD \ORU
WD \ORU
*'3
WD\ORU
WD\ORU
WD\ORU
5HDO,QWHUHVW5DWH
WD \ORU
WD\ORU
39
WD \ORU
5HDO([F5DWH
WD\ORU
WD\ORU
1RPLQDO([F5DWH
WD\ORU
WD \ORU
WD\ORU
WD \ORU
3XEOLF'HEW
WD\ORU
WD\ORU
40
WD\ORU
6LPXODWLRQ
(Quarters in horizontal axis)
,QIODWLRQ
6K6HOLF
6K5LVN
6K$GP
*'3
6K6HOLF
6K5LVN
6K$GP
1RPLQDO,QWHUHVW5DWH
6K6HOLF
6K5LVN
41
6K$GP
5,6. EE
6K6HOLF
6K5LVN
6K$GP
5HDO([F5DWH
6K6HOLF
6K5LVN
6K$GP
1RPLQDO([F5DWH
6K6HOLF
6K5LVN
42
6K$GP
3XEOLF'HEW
6K6HOLF
6K5LVN
6K$GP
5HDO,QWHUHVW5DWH
6K6HOLF
6K5LVN
43
6K$GP
Banco Central do Brasil
Trabalhos para Discussão
Os Trabalhos para Discussão podem ser acessados na internet, no formato PDF,
no endereço: http://www.bc.gov.br
Working Paper Series
Working Papers in PDF format can be downloaded from: http://www.bc.gov.br
1
Implementing Inflation Targeting in Brazil
Joel Bogdanski, Alexandre Antonio Tombini e Sérgio Ribeiro da Costa
Werlang
Jul/2000
2
Política Monetária e Supervisão do Sistema Financeiro Nacional no
Banco Central do Brasil
Eduardo Lundberg
Jul/2000
Monetary Policy and Banking Supervision Functions on the Central
Bank
Eduardo Lundberg
Jul/2000
3
Private Sector Participation: A Theoretical Justification of the Brazilian
Position
Sérgio Ribeiro da Costa Werlang
Jul/2000
4
An Information Theory Approach to the Aggregation of Log-Linear
Models
Pedro H. Albuquerque
Jul/2000
5
The Pass-through from Depreciation to Inflation: A Panel Study
Ilan Goldfajn e Sérgio Ribeiro da Costa Werlang
Jul/2000
6
Optimal Interest Rate Rules in Inflation Targeting Frameworks
José Alvaro Rodrigues Neto, Fabio Araújo e Marta Baltar J. Moreira
Jul/2000
7
Leading Indicators of Inflation for Brazil
Marcelle Chauvet
Set/2000
8
The Correlation Matrix of the Brazilian Central Bank’s Standard
Model for Interest Rate Market Risk
José Alvaro Rodrigues Neto
Set/2000
9
Estimating Exchange Market Pressure and Intervention Activity
Emanuel-Werner Kohlscheen
Nov/2000
10
Análise do Financiamento Externo a Uma Pequena Economia
Carlos Hamilton Vasconcelos Araújo e Renato Galvão Flôres Júnior
Mar/2001
11
A Note on the Efficient Estimation of Inflation in Brazil
Michael F. Bryan e Stephen G. Cecchetti
Mar/2001
12
A Test of Competition in Brazilian Banking
Márcio I. Nakane
Mar/2001
44
13
Modelos de Previsão de Insolvência Bancária no Brasil
Marcio Magalhães Janot
Mar/2001
14
Evaluating Core Inflation Measures for Brazil
Francisco Marcos Rodrigues Figueiredo
Mar/2001
15
Is It Worth Tracking Dollar/Real Implied Volatility?
Sandro Canesso de Andrade e Benjamin Miranda Tabak
Mar/2001
16
Avaliação das Projeções do Modelo Estrutural do Banco Central do
Brasil Para a Taxa de Variação do IPCA
Sergio Afonso Lago Alves
Mar/2001
Evaluation of the Central Bank of Brazil Structural Model’s Inflation
Forecasts in an Inflation Targeting Framework
Sergio Afonso Lago Alves
Jul/2001
Estimando o Produto Potencial Brasileiro: Uma Abordagem de Função
de Produção
Tito Nícias Teixeira da Silva Filho
Abr/2001
Estimating Brazilian Potential Output: A Production Function
Approach
Tito Nícias Teixeira da Silva Filho
Ago/2002
18
A Simple Model for Inflation Targeting in Brazil
Paulo Springer de Freitas e Marcelo Kfoury Muinhos
Abr/2001
19
Uncovered Interest Parity with Fundamentals: A Brazilian Exchange
Rate Forecast Model
Marcelo Kfoury Muinhos, Paulo Springer de Freitas e Fabio Araújo
Maio/2001
20
Credit Channel without the LM Curve
Victorio Y. T. Chu e Márcio I. Nakane
Maio/2001
21
Os Impactos Econômicos da CPMF: Teoria e Evidência
Pedro H. Albuquerque
Jun/2001
22
Decentralized Portfolio Management
Paulo Coutinho e Benjamin Miranda Tabak
Jun/2001
23
Os Efeitos da CPMF sobre a Intermediação Financeira
Sérgio Mikio Koyama e Márcio I. Nakane
Jul/2001
24
Inflation Targeting in Brazil: Shocks, Backward-Looking Prices, and
IMF Conditionality
Joel Bogdanski, Paulo Springer de Freitas, Ilan Goldfajn e
Alexandre Antonio Tombini
Ago/2001
25
Inflation Targeting in Brazil: Reviewing Two Years of Monetary Policy
1999/00
Pedro Fachada
Ago/2001
26
Inflation Targeting in an Open Financially Integrated Emerging
Economy: the case of Brazil
Marcelo Kfoury Muinhos
Ago/2001
17
45
27
Complementaridade e Fungibilidade dos Fluxos de Capitais
Internacionais
Carlos Hamilton Vasconcelos Araújo e Renato Galvão Flôres Júnior
Set/2001
28
Regras Monetárias e Dinâmica Macroeconômica no Brasil: Uma
Abordagem de Expectativas Racionais
Marco Antonio Bonomo e Ricardo D. Brito
Nov/2001
29
Using a Money Demand Model to Evaluate Monetary Policies in Brazil
Pedro H. Albuquerque e Solange Gouvêa
Nov/2001
30
Testing the Expectations Hypothesis in the Brazilian Term Structure of
Interest Rates
Benjamin Miranda Tabak e Sandro Canesso de Andrade
Nov/2001
31
Algumas Considerações Sobre a Sazonalidade no IPCA
Francisco Marcos R. Figueiredo e Roberta Blass Staub
Nov/2001
32
Crises Cambiais e Ataques Especulativos no Brasil
Mauro Costa Miranda
Nov/2001
33
Monetary Policy and Inflation in Brazil (1975-2000): a VAR Estimation
André Minella
Nov/2001
34
Constrained Discretion and Collective Action Problems: Reflections on
the Resolution of International Financial Crises
Arminio Fraga e Daniel Luiz Gleizer
Nov/2001
35
Uma Definição Operacional de Estabilidade de Preços
Tito Nícias Teixeira da Silva Filho
Dez/2001
36
Can Emerging Markets Float? Should They Inflation Target?
Barry Eichengreen
Fev/2002
37
Monetary Policy in Brazil: Remarks on the Inflation Targeting Regime,
Public Debt Management and Open Market Operations
Luiz Fernando Figueiredo, Pedro Fachada e Sérgio Goldenstein
Mar/2002
38
Volatilidade Implícita e Antecipação de Eventos de Stress: um Teste
para o Mercado Brasileiro
Frederico Pechir Gomes
Mar/2002
39
Opções sobre Dólar Comercial e Expectativas a Respeito do
Comportamento da Taxa de Câmbio
Paulo Castor de Castro
Mar/2002
40
Speculative Attacks on Debts, Dollarization and Optimum Currency
Areas
Aloisio Araujo e Márcia Leon
Abr/2002
41
Mudanças de Regime no Câmbio Brasileiro
Carlos Hamilton V. Araújo e Getúlio B. da Silveira Filho
Jun/2002
42
Modelo Estrutural com Setor Externo: Endogenização do Prêmio de
Risco e do Câmbio
Marcelo Kfoury Muinhos, Sérgio Afonso Lago Alves e Gil Riella
Jun/2002
46
43
The Effects of the Brazilian ADRs Program on Domestic Market
Efficiency
Benjamin Miranda Tabak e Eduardo José Araújo Lima
Jun/2002
44
Estrutura Competitiva, Produtividade Industrial e Liberação
Comercial no Brasil
Pedro Cavalcanti Ferreira e Osmani Teixeira de Carvalho Guillén
Jun/2002
45
Optimal Monetary Policy, Gains from Commitment, and Inflation
Persistence
André Minella
Ago/2002
46
The Determinants of Bank Interest Spread in Brazil
Tarsila Segalla Afanasieff, Priscilla Maria Villa Lhacer e Márcio I. Nakane
Ago/2002
47
Indicadores Derivados de Agregados Monetários
Fernando de Aquino Fonseca Neto e José Albuquerque Júnior
Set/2002
48
Should Government Smooth Exchange Rate Risk?
Ilan Goldfajn e Marcos Antonio Silveira
Set/2002
49
Desenvolvimento do Sistema Financeiro e Crescimento Econômico no
Brasil: Evidências de Causalidade
Orlando Carneiro de Matos
Set/2002
50
Macroeconomic Coordination and Inflation Targeting in a TwoCountry Model
Eui Jung Chang, Marcelo Kfoury Muinhos e Joanílio Rodolpho Teixeira
Set/2002
51
Credit Channel with Sovereign Credit Risk: an Empirical Test
Victorio Yi Tson Chu
Set/2002
52
Generalized Hyperbolic Distributions and Brazilian Data
José Fajardo e Aquiles Farias
Set/2002
53
Inflation Targeting in Brazil: Lessons and Challenges
André Minella, Paulo Springer de Freitas, Ilan Goldfajn e Marcelo Kfoury
Muinhos
Nov/2002
54
Stock Returns and Volatility
Benjamin Miranda Tabak e Solange Maria Guerra
Nov/2002
55
Componentes de Curto e Longo Prazo das Taxas de Juros no Brasil
Carlos Hamilton Vasconcelos Araújo e Osmani Teixeira de Carvalho de
Guillén
Nov/2002
56
Causality and Cointegration in Stock Markets: The Case of Latin
America
Benjamin Miranda Tabak e Eduardo José Araújo Lima
Dez/2002
57
As Leis de Falência: uma Abordagem Econômica
Aloisio Araujo
Dez/2002
58
The Random Walk Hypothesis and the Behavior of Foreign Capital
Portfolio Flows The Brazilian Stock Market Case
Benjamin Miranda Tabak
Dez/2002
59
Os Preços Administrados e a Inflação no Brasil
Francisco Marcos R. Figueiredo e Thaís Porto Ferreira
Dez/2002
47
60
Delegated Portfolio Management
Paulo Coutinho e Benjamin Miranda Tabak
Dez/2002
61
O Uso de Dados de Alta Freqüência na Estimação da Volatilidade e
do Valor em Risco para o Ibovespa
João Maurício de Souza Moreira e Eduardo Facó Lemgruber
Dez/2002
62
Taxa de Juros e Concentração Bancária no Brasil
Eduardo Kiyoshi Tonooka e Sérgio Mikio Koyama
Jan/2003
63
Optimal Monetary Rules: The Case of Brazil
Charles Lima de Almeida, Marco Aurélio Peres, Geraldo da Silva e Souza e
Benjamin Miranda Tabak
Jan/2003
48