econstor
A Service of
zbw
Make Your Publications Visible.
Leibniz-Informationszentrum
Wirtschaft
Leibniz Information Centre
for Economics
Aguirre, Horacio; Blanco, Emilio
Working Paper
Financial stability and macroprudential policy: A
structural model evaluation of an emerging economy
Working Paper, No. 2016/71
Provided in Cooperation with:
Economic Research Department (ie), Central Bank of Argentina
Suggested Citation: Aguirre, Horacio; Blanco, Emilio (2016) : Financial stability and
macroprudential policy: A structural model evaluation of an emerging economy, Working Paper,
No. 2016/71, Banco Central de la República Argentina (BCRA), Investigaciones Económicas
(ie), Buenos Aires
This Version is available at:
http://hdl.handle.net/10419/168390
Standard-Nutzungsbedingungen:
Terms of use:
Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen
Zwecken und zum Privatgebrauch gespeichert und kopiert werden.
Documents in EconStor may be saved and copied for your
personal and scholarly purposes.
Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle
Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich
machen, vertreiben oder anderweitig nutzen.
You are not to copy documents for public or commercial
purposes, to exhibit the documents publicly, to make them
publicly available on the internet, or to distribute or otherwise
use the documents in public.
Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen
(insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten,
gelten abweichend von diesen Nutzungsbedingungen die in der dort
genannten Lizenz gewährten Nutzungsrechte.
www.econstor.eu
If the documents have been made available under an Open
Content Licence (especially Creative Commons Licences), you
may exercise further usage rights as specified in the indicated
licence.
ESTUDIOS BCRA
Working Paper 2016 | 71
Financial Stability and Macroprudential Policy:
A Structural Model Evaluation of an
Emerging Economy
Horacio Aguirre / Emilio Blanco
Central Bank of Argentina
January, 2016
Banco Central de la República Argentina
ie | Investigaciones Económicas
January, 2016
ISSN 1850-3977
Electronic Edition
Reconquista 266, C1003ABF
C. A. de Buenos Aires, Argentina
Phone: (+5411) 4348-3582
Fax: (+5411) 4348-3794
Email: investig@bcra.gob.ar
Web Page: www.bcra.gob.ar
The opinions in this work are an exclusive responsibility of his authors and do not necessarily reflect the position of
the Central Bank of Argentina. The Working Papers Series from BCRA is composed by papers published with the
intention of stimulating the academic debate and of receiving comments. Papers cannot be referenced without the
authorization of their authors.
Financial Stability and Macroprudential Policy: a Structural
Model Evaluation of an Emerging Economy
Horacio A. Aguirrey
Emilio F. Blancoz
January 2016
Abstract
We build a small structural open economy model, augmented to depict the credit market
and interest rate spreads (distinguishing by credit to …rms and families); monetary policy
with sterilized intervention in the foreign exchange market; and macroprudential policy as
capital requirements. We estimate the model using Bayesian techniques with quarterly data
for Argentina in 2003-2011; it can be extended to other emerging economies, allowing for
comparative empirical analysis. Results indicate that shocks to lending rates and spread
weigh on macroeconomic variables; likewise, the credit market is a¤ected by macroeconomic
shocks. Capital requirements, beyond their strictly prudential role, appear to have contributed to lower volatility of key variables such as output, prices, credit and interest rates. The
interaction of monetary policy, foreign exchange intervention and prudential tools appears
to be synergic: counting on a larger set of tools helps dampen volatility of both macroeconomic and …nancial system variables, taking into account the type of shocks faced during the
estimation period.
JEL classi…cation codes: E17, E51, E52, E58
Keywords: Macroprudential Policy, Semi-structural Model, Bayesian Estimation.
Previous versions of this paper appeared as BIS Working paper No. 504, and were presented at the XX
Meeting of the Network of Central Bank Researchers, CEMLA and Central Bank of the Dominican Republic, 2627 November, 2015. For useful comments and suggestions, we wish to thank J. Dorich, G. Escudé, E. Mendoza, C.
Montoro, F. Zanetti, anonymous referees, participants of the BIS Consultative Council of the Americas’ research
network on "Incorporating Financial Stability Considerations in Central Bank Policy Models" and in the following
meetings: XLVIII Annual Meeting of Asociación Argentina de Economía Política, XXIX Jornadas Anuales de
Economía del Banco Central del Uruguay, I Jornadas Nacionales de Econometría, Universidad de Buenos Aires,
and the economics seminar at Universidad de San Andrés. All views expressed are the authors’ own and do not
necessarily represent those of the Central Bank of Argentina (BCRA).
y
Economic research, BCRA. E-mail: haguirre@bcra.gob.ar
z
Economic research, BCRA, and Universidad de Buenos Aires. E-mail: emilio.blanco@bcra.gob.ar
1
1
Introduction
While …nancial stability has become increasingly important for monetary policy, standard models
fail to re‡ect the integration of both dimensions. So called “…nancial frictions” are gradually being
incorporated in Dynamic Stochastic General Equilibrium (DSGE) models, but the lack remains of
a workable model of mid- to small scale that includes a representation of …nancial intermediation.
Moreover, a wider set of tools, such as what has come to be known as "macroprudential" ones,
is increasingly being used by central banks–while standard monetary models tend to focus on a
single instrument, the short term nominal interest rate. We aim to incorporate …nancial stability
aspects and macroprudential tools into a small open economy model of the Argentine economy,
completely estimated and suitable for short-term forecasting and simulation exercises.
The …nancial side of the macroeconomy is built into central bank models in diverse ways,
without a single uni…ed and widespread framework. “Macro-modelling” options basically comprise …nancial accelerator e¤ects, collateral constraints, and the explicit representation of banking
intermediaries (see Roger and Vlcek, 2011, for a survey of central bank literature). Most current
models have generally been developed in North America and Europe; in Latin America, modeling
e¤orts have only recently been made for the depiction of …nancial issues in the macroeconomy.
Perhaps the …rst fully ‡edged DSGE model with the explicit interaction of banks and monetary policy, designed and calibrated for a Latin American country before the international crisis
brought these aspects to the foreground was that of Escudé (2008). He integrates both …nancial
and real features of the Argentine economy, including intermediation through banks, that lend to
families and whose deposits are subject to liquidity requirements. More recent modelling e¤orts,
in line with the literature ‡ourishing after the international …nancial crisis include Carvalho et
al. (2013), Garcia-Cicco and Kawamura (2014), González et al. (2013), among others.
The very same lack of an agreed framework to deal with …nancial stability in macroeconomic
models also justi…es the use of small structural ones, specially for applied work in central banks
and as a …rst approximation to the problem. As pointed out by Galati and Moessner (2011),
models that link the …nancial sector to the macroeconomy are far from having reached a stage
where they can be operationalized for analysis and simulation—but such tasks do call for workable
solutions even before a new "consensus model" is reached. For instance, Sámano Peñaloza (2012)
enlarges a small macroeconomic model for Mexico with a …nancial block in order to determine
the interplay of macroprudential and monetary policy; the former is introduced through capital
requirements. Szilagy et al. (2013) also add …nancial variables to a standard small model in
order to enrich the depiction of the Hungarian macroeconomy. Both of these models, while not
explicitly derived from …rst order conditions of an optimization problem, show the basic New
Keynesian structure.
We present an extension of Aguirre and Blanco (2013), who in turn build on previous works
done for Argentina (Elosegui et al, 2007; Aguirre and Grosman, 2010), while dealing with the
…nancial dimension largely after Sámano Peñaloza (2012). We augment an open economy version
of a semi structural New Keynesian model, to include explicit depiction of the credit market, active rates and interest rate spread; and an enriched description of monetary policy, with sterilized
intervention in the foreign exchange market. We estimate it using Bayesian techniques, allowing
us to incorporate our prior knowledge of the workings of this economy during the estimation
period (2003-2011). We also enhance the baseline model, introducing capital requirements under
di¤erent possible de…nitions, corresponding to alternative macroprudential rules, cyclical and
not, based on quantities and on prices. We aim to assess whether the interaction between monetary, foreign exchange and macroprudential policy helps dampen macroeconomic ‡uctuations
in any meaningful way during the estimation period.
2
Our modelling choices are closely related to our practical goals: if we had a theoretical
interest, we would pursue another modelling strategy. In the …rst place, we take an empirical
approach, in that a condition for model building is that parameters should all be estimated.
This contrasts with actual design and implementation of large scale DSGE models which, for all
the detail they provide, often rely to a substantial degree on calibration, and are naturally less
appropriate for estimation. Likewise, such models tend to be less workable in terms of forecasting:
typically, smaller models forecast better than larger ones, with di¤erent models being used for
di¤erent purposes (Canova, 2009; see Aguirre and Blanco, 2013, for forecasting with our model).
There is a place for representations of di¤erent sizes in a well-conceived modeling architecture,
and enlarging semi-structural models already in use may be more useful than starting DSGE
models from scratch (Roger and Vlcek, 2011). This is certainly relevant for central banks, where
a pragmatic approach may be favoured for the sake of incorporating …nancial stability in formal
models.
Thus, we have both descriptive and policy-oriented goals. As for the former, we wish to
improve the depiction of an economy where real aspects may not be dissociated from …nancial
ones, i.e. where the …nancial sector may play a role in either originating or transmitting shocks
(Borio, 2012). In this sense, our model involves an improvement from conventional comparable
ones in two ways: a richer description of monetary policy, with the central bank using both
interest rates and sterilized foreign exchange intervention, the monetary repercussions of which
are explicitly acknowledged; and credit market dynamics, capturing the interplay of credit and
interest rate spreads with the rest of the economy.
This framework can also be taken as a …rst approximation to enquire whether macroprudential policy may lead to better performance of certain key variables. In particular, we include a
macroprudential instrument (capital requirements) in addition to interest rates and foreign exchange intervention, so as to determine how it interacts with the other policy tools and whether
it may help smooth short run macroeconomic and …nancial market ‡uctuations. It is in regard
to the latter that our model contemplates …nancial stability considerations: reducing volatility
of both macroeconomic and …nancial variables is incorporated in policy rules as well as in the
criterion to evaluate their outcomes (see also section 2). As many emerging market economies
implement monetary, foreign exchange and macroprudential policy, the model also provides a
convenient framework for comparative empirical analysis.
There are, as is well known, limitations to what structural models can provide in terms of
policy and simulation exercises. However, we consider our proposal to be a reasonable trade-o¤
between tractability and ability to take the model to the data. This is all the more important
when we build and estimate a model that allows us to consider not only monetary policy and
macroprudential instruments, but also foreign exchange policy. Within the class of emerging
economies’ central bank policy models, ours is one of the few to consider those three dimensions
taken together. Indeed, emerging economies frequently use the exchange rate as a tool, but this
is seldom re‡ected in policy models. Finally, to the best of our knowledge, this paper and Aguirre
and Blanco (2013) are the …rst empirical assessments of the macroeconomic impact of prudential
regulations in Argentina, carried out in a completely estimated macroeconomic model.
The rest of the paper is organized as follows. Section 2 introduces macroprudential policy
and its features as applied in emerging market economies; and discusses alternative models and
how they relate to ours. Section 3 describes the baseline model; section 4 presents estimation and
impulse-response functions that illustrate the basic workings of the estimated model. Section
5 extends the model to include macroprudential policy in the form of capital requirements,
considering alternative formulations of the latter, with emphasis on macroeconomic and …nancial
performance associated to them. Section 6 concludes.
3
2
Macroprudential policy: a primer
Following the international …nancial crisis, there has been a change of perspective in monetary
policy frameworks, with the conventional focus being gradually rede…ned. Financial cycles are
becoming accepted as part of the functioning of market economies (Borio, 2012), whose consequences on stability have to be dealt with by central banks. Charging the central bank with
responsibility for …nancial stability is not su¢cient—appropriate tools, authorities and safeguards
are also needed (CGFS, 2011). Consequently, a double mandate is surging, with monetary and
…nancial stability as acceptable central bank targets. Roughly speaking, the introduction of a
…nancial stability mandate for central banks involves a move from a single focus for monetary
policy and a concern for the individual performance of …nancial institutions, to multiple targets
together with the oversight of …nancial institutions based on their potential systemic impact,
and even on the economy at large. This shift is schematically represented in Figure 1, where
two dimensions are sketched: the monetary policy framework, with either a single or a multiple
focus; and …nancial supervision and regulation, aimed at the individual risks of institutions or
at their systemic impact. Such shift has brought on the need to incorporate in formal models a
wider set of tools used by central banks, such as macroprudential measures.
Figure 1
Financial regulation &
supervision
Individual risk
Single focus
Systemic &
macroeconomic risk
Inflation targeting
Microprudential
policy
Monetary
policy
framework
Monetary and
financial stability
Macroprudential
policy
Multiple focus
Macroprudential policy is far from being a well-de…ned concept, but a generic term for measures whose goal extends beyond safeguarding the solvency or liquidity of …nancial institutions,
to cover their link with macroeconomic performance—recognizing possible spillovers from the
…nancial system to the economy at large, and vice versa. Many di¤erent measures can be considered as macroprudential, ranging from capital and liquidity requirements as a function of
certain "cyclical" variables, to loan-to-value ratios, dynamic provisions and other tools that may
incorporate to a certain extent the state of the …nancial system or the economy as an input to
determine whether to soften or tighten regulations on banks. However broad in scope, measures
taken under a macroprudential approach share a number of features: they are aimed at limiting
systemic risk and spillovers from the …nancial system to the macroeconomy (and vice versa);
they take into account externalities of individual …nancial …rms, such as interconnection, procyc4
licality, and common exposures; as a consequence, the …nancial system is considered as a whole,
and systemic risk is treated as endogenous.
A common theme running through macroprudential analysis is that prevention is key: central
banks and supervisors should act before the turn of the cycle, as critical pressures build up but
before a crisis breaks out. In particular, countercyclical macroprudential policy aims at: i)
strengthening the …nancial system so that it is better prepared to face the downturn of the
(…nancial and business) cycle; ii) smoothing the cycle, preventing imbalances from accumulating
during the “boom” phase.
Emerging market economies have used macroprudential instruments more extensively, and
earlier, than industrial economies (Lim et al., 2011), out of a long experience with crises. Indeed,
one can trace elements of a macroprudential approach since at least the 1990s, used to address
systemic risk following several episodes, such as the "Tequila" crisis in 1995, the Southeast Asia
crisis in 1997, Russia in 1998, Brazil in 1999, Ecuador in 2000, Argentina in 2001. For these
countries, such elements are part of a broader “macro-…nancial” stability framework that also
comprises management of the exchange rate and the capital account, and which are part of the
move from the Northeast to the Southwest quadrant of Figure 1. Moreover, the international
…nancial crisis has increased the number of advanced countries that put in place macroprudential
policies within a more formal framework; the European Systemic Risk board is a case in point.
Thus, what we examine here may bear parallels with developing economies at large, and even
be useful for industrial economies that are implementing broader sets of measures aimed at
containing system-wide risk in …nancial markets.
Likewise, the role of the exchange rate for monetary and …nancial stability is substantial in
emerging market economies, well beyond what is articulated in conventional frameworks. Actually, measures like systematic foreign exchange intervention or liquidity supply through multiple
instruments have long been part of the policy "toolbox" in developing countries, even in those
which implement in‡ation targeting regimes. Based on the experience of Brazil, Chile, Colombia
and Perú, Chang (2008) shows that in‡ation targeting in Latin America di¤er systematically
from the "Taylor rule cum pure ‡oating" formula supposedly associated to the scheme. Far
from being a deviation from "best practice" in monetary policy by the countries in the region,
it obeys to the need to shield their economies from abrupt changes in international …nancial
conditions through measures such as international reserve accumulation. This type of policies
has been called "unconventional", but the label applies largely to industrial economies, whereas
in the developing world such measures are not necessarily associated to exceptional responses in
the face of the international …nancial crisis (Kawamura and García-Cicco, 2014, present a formal
analysis of such responses). This is why we …nd it important to analyze the interplay of both
macroprudential policy, monetary policy and foreign exchange intervention.
In Argentina, several crisis episodes have evidenced the close connection between …nancial
system soundness and macroeconomic performance. The most traumatic one was perhaps the
demise of the currency board, in place from 1991 to 2001. The main features of such experience
exceed the scope of this paper; we note here that the so called "convertibility" regime showed
bluntly how the implementation of microprudential policy, even by state-of-the-art standards,
may not be enough to safeguard the …nancial system from both adverse shocks and the presence
of "hidden" mismatches in a …nancially dollarized economy. In keeping with the maintenance
of a peg to the US dollar, the private sector became progressively indebted in foreign currency,
even if, on aggregate, its revenues were denominated in pesos. The government also became
progressively indebted in foreign currency. Notably, both private and public agents displayed
behaviour that seemed to consider that favourable external conditions, as seen in the …rst half of
the 1990s, would last inde…nitely. Successive emerging economies’ crises hit the country’s ability
5
to access foreign …nancing, and deteriorated its competitiveness, ultimately leading to recession
and a multiple crisis, on the …scal, foreign exchange and banking fronts.
Since 2003, a major feature of the Argentine macroprudential policy toolkit have been direct
and indirect measures limiting foreign currency exposure of …nancial institutions. In so far as
foreign exchange intervention limits the variability of a certain class of assets that weigh on
…nancial system dynamics, it can also be considered as part of the macroprudential "toolkit" in
a broader sense. A similar reasoning applies to capital ‡ows regulation. Macroprudential policy
also includes building up a capital bu¤er through pro…t reinvestment mechanism; loan-to-value
ratios for certain types of credit; valuation of public sector securities in …nancial institutions’
balance sheets; liquidity requirements and deposit insurance. We …nd it particularly important
to analyze how foreign exchange intervention interacts with monetary policy and more standard
macroprudential tools such as capital requirements. In this paper, we look at capital requirements
implemented in di¤erent ways: as a function of the credit-to-GDP gap, the output gap or interest
rate spreads, or set exogenously from the point of view of the macroeconomy; this allows to gauge
the potential e¢cacy of policy implemented with or without concern for cyclical variables, as
well as based alternatively on price or quantity indicators.
Models that integrate the most widely used monetary policy analysis framework—the New
Keynesian one—with macroprudential tools have only recently been developed, and a uni…ed
approach is lacking. Angeloni and Faia (2013) look at instruments and policy rules of a central
bank that aims not only at price stability but also at …nancial stability in a Diamond-Dybvig
setting: they examine how interest rate and capital ratio rules interplay. Covas and Fujita
(2009) analyze how a productivity shock is transmitted when …nancial intermediaries are subject
to alternative capital requirements. Angelini et al. (2010), Denis et al. (2010) are recent examples
that inquire about the interaction between monetary policy and macroprudential tools, and …nd
that introducing a new policy rule in coordination with monetary policy helps to reduce the
variance of output and in‡ation. Indeed, a frequent concern is to what extent both types of
policy may be considered complements or substitutes.
Cecchetti and Kohler (2014) propose an enlarged aggregate demand-aggregate supply system with both interest rates and capital requirements; they use a game-theoretic approach to
investigate the optimal degree of coordination between both policy tools, in a static, theoretical
framework. They show that both type of instruments are full substitutes, in the sense that if the
ability to use one is limited, the other can “…nish” the job. When a …nancial stability objective
is contemplated, that characteristic depends on the coordination between them—under full coordination, substitutability remains. In turn, and in the context of a comprehensive discussion of
…nancial stability and monetary policy, Agenor and Pereira da Silva (2013) analyze whether monetary and macroprudential policy are complementary in an small macroeconomic model: they
…nd them to be so, and have to be calibrated jointly, accounting for the type of credit market
imperfections observed in middle income countries and for the fact that macroprudential regimes
may a¤ect in substantial ways the monetary transmission mechanism. Végh (2014) argues that
both foreign exchange intervention and reserve requirements act in the sense of allowing interest
rate policy to achieve other goals: thus, for emerging market countries facing a sudden stop,
exchange rate intervention may be used to “defend” the local currency, so that interest rates do
not necessarily have to be raised with that aim, while reserve requirements are changed in order
to in‡uence credit market conditions—this gives monetary policy higher degrees of freedom to
act countercyclically. Once again, none of these models are based on the same structure; and in
the case of Cecchetti and Kohler (2014) and Agenor and Pereira da Silva (2013) nor are they
6
derived from the explicit solution of microeconomic problems of households and …rms1 .
Finally, a word is also in order regarding the isomorphism between …nancial stability issues, at
which macroprudential measures aim, and DSGE models (or models like ours, which are based
on them). Financial stability ultimately re‡ects the sustainability of …nancial intermediaries’
operations and its interaction with the macroeconomy. For example, the subprime crisis put in
the foreground the relationship between asset prices, credit growth and macroeconomic performance, and whether it may lead to unstable behavior of the variables involved. Such dynamics,
however, are extremely di¢cult to represent in models based on linear approximations around
steady states, and which are solved to yield stable solutions. Thus, "…nancial frictions" turn
out to be a device that allows for explicit representation of credit market variables in DSGE
models, but that does little by the way of modelling the potential transition from the normal
functioning of the system to a …nancial crisis. Such transition calls for non-linear techniques
applied to "macro…nancial" models, something that recent works are developing (Bianchi and
Mendoza, 2013). Therefore, there certainly is a gap between …nancial stability analysis and what
can be described by models that depict "well behaved" cyclical deviations around a steady state.
With this caveat in mind, the following sections present a model inspired by the New Keynesian
tradition that incorporates macroprudential policy.
3
The baseline model
Following work by Elosegui et al. (2007) and Aguirre and Grosman (2010), our baseline model
is a small structural open economy model with a Taylor-type rule and foreign exchange market
intervention, with the monetary e¤ects that these imply. It already incorporates a money market
equation, providing a natural starting point for the introduction of a simpli…ed …nancial block,
where we describe credit market conditions in the manner of Sámano Peñaloza (2012).
The standard macroeconomic block of the model comprises an IS-type equation (1), a Phillips curve (5) and a Taylor-type rule (6)—the …rst two of which can be obtained as log-linear
approximations of …rst order conditions of consumers’ and …rms’ optimization problems in a
monopolistic competition setting where price adjustments are sluggish. The IS equation contains
output growth, and not the output gap, as endogenous variable, due exclusively to empirical considerations; and it is augmented to re‡ect the impact of open economy variables, namely the real
exchange rate, on consumption decisions and hence on output; it also includes a lagged growth
term, that can be related to the assumption that preferences over consumption exhibit habit
formation (Fuhrer, 2000). The IS (1) also contains the spread between the active rate of interest
(charged for taking credit) and the short term interest rate; as in Sámano Peñaloza (2012) and
Szylagy et al (2013), this term aims at capturing the impact of credit market conditions on
aggregate demand, as it represents the extra cost above the short term interest rate that the
non …nancial private sector has to pay to banks in order to obtain resources2 ; alternatively, the
sum of the short term rate and the spread may be interpreted as the active rate that the private
sector pays to obtain funds. The average spread is made up of those corresponding to …rms and
households’ credit. An additional term in the IS corresponds to the e¤ect of …scal impulse on
aggregate demand,which is just a convenient way of depicting …scal shocks, but which serves no
direct purpose to the exercises in this paper. We leave for further work more disaggregation of
demand in the model (consumption, investment, exports and imports), which could help have
1
Recent contributions to the study of macroprudential policy in macroeconomic models in the Latin American
case include Carvalho et al (2013) and González et al. (2013)
2
An alternative speci…cation is to include credit directly in the IS curve; this is work in progress and will be
incorporated in further versions.
7
a better characterization of credit to di¤erent agents and consider shocks to foreign demand in
the analysis.
In turn, the Phillips curve (5) evidences the e¤ect of foreign prices in the domestic economy,
through an "imported in‡ation" component via the real exchange rate; the inclusion of the latter
in both the IS and Phillips curves is derived analytically by Galí and Monacelli (2005). Lagged
in‡ation in the Phillips curve has been found empirically signi…cant by many studies and can be
thought of as a consequence of the ability of …rms to adjust prices according to lagged in‡ation
(Galí and Gertler, 1999). Lagged output gap in the Phillips curve is basically due to empirical …t,
something that turns up in estimates of other economies (Galí et al., 2001), and may be justi…ed
in relation to GDP data being released with lags (Pincheira and Rubio, 2010). The Taylor rule
(6) also includes a coe¢cient on nominal exchange rate depreciation, so that the central bank’s
behavior not only depends on the output gap and in‡ation. Two terms account for the central
bank’s involvement with …nancial stability: the short term rate also depends on its own lagged
values, showing a desire to smooth interest rate movements; and on the "credit gap", i.e. the
di¤erence between current credit to the private sector and its steady state value (more on this
below).
Macroeconomic Block
gty =
y
1 Et gt+1
y
2 gt 1
+
bt
3r
+
spreadH
t +
F
c
ebtri
t
4
5 sf t
6 (spreadt 1 )
+ "yt
(1)
gty : output growth rate, r : real interest rate, etri : trilateral real exchange rate (RER), sf :
…scal surplus to GDP ratio, and interest rate spread is de…ned as3
H
spreadt =
spreadFt
where
bit
= biH;act
t
bit
= biF;act
t
spreadH
t
spreadFt
(2)
(3)
(4)
H;act
F
spreadH
: nominal active rate
t : spread - household credit, spreadt : spread - …rms credit, it
- households, iF;act
:
nominal
active
rate
…rms,i
:
nominal
(passive)
interest
rate
t
t
1 Eb t+1
bt =
2
t
+
where
= 1
2 bt 1
+ a3 yt
1
+ a4 ebtri
t + "t
(5)
1
: in‡ation, Ebt+1 : expected in‡ation, yt : output gap
bit =
a
b
1 it 1
+
2 yt
: annual in‡ation,
GDP ratio de…ned as
+
a
3 Et b t+1
+
b +
4 t
d + "i
t
5 CRt
: $/USD depreciation rate, CR : Non …nancial private sector credit to
d t = CR
dH
dF
CR
t + CRt
(7)
CRH : Household’s credit to GDP ratio, CRF : Firm’s credit to GDP ratio
3
Where
H
and
F
(6)
are calibrated
1
:
2
8
Foreign exchange conditions and policy, as well as the money market, are described in equations (8)-(12). A modi…ed uncovered interest rate parity (UIP) condition (8) considers the e¤ects
of central bank operations in the foreign exchange market: the nominal exchange rate depends
on expected depreciation, the di¤erence between the local and the international interest rate,
and a country risk premium that is made up of an endogenous component and an exogenous
shock. The former is determined by interventions in the currency market: the central bank
intervenes by buying or selling international reserves, and issuing or withdrawing bonds from
circulation in order to sterilize the e¤ects of intervention on the money supply. Monetary e¤ects
naturally require an LM curve: equation (12) describes equilibrium in the money market, which
may be estimated for narrower or broader de…nition of monetary aggregates. How exchange rate
intervention is instrumented is described by equation (11), whereby the central bank buys or
sells international reserves in reaction to nominal exchange rate variability; equation (9) shows
to what extent such intervention is sterilized.
FX Policy Block
bit = bit + ! 1 Etbt+1 + (1
! 1 )bt + ! 2bbt + ! 3 res
c t + bt
i : international interest rate, b : CB bonds to GDP ratio,
: international reserves to GDP ratio
bbt =
=
1
1
m
m+b
res
c t + ebdt
1
m
bt
(8)
: exogenous risk-premium, res
(9)
(10)
m: money to GDP ratio4
res
ct =
ct 1
1 res
m
bt =
1 it
b +
2 bt
b + "res
t
(11)
2 t
+
b +
3 bt
b + "m
t
(12)
4 t
This speci…cation merits some further explanation. Introducing a policy of sterilized intervention can be thought of as "augmenting" or modifying the uncovered interest rate parity (8);
actually, what we have is a new equation for the determination of the nominal exchange rate—
after all, the purpose of sterilized intervention is precisely to "block" in a way the conditions
imposed by UIP in its normal form. This modi…ed UIP can be rationalized as follows: domestic
agents may invest in both local and foreign currency-denominated bonds, which are not perfect substitutes; returns of bonds in pesos have to compensate for expected depreciation; in turn,
bonds in foreign currency pay the international rate but re‡ect a liquidity risk. It may further be
assumed that not all actors that participate in the foreign currency market optimize on the base
of fundamentals; some of them decide on the past performance of the currency (and are called
“chartists”); this is behind the expected depreciation term in (8), which corresponds to agents
that act on fundamentals, and the current depreciation term, which corresponds to “chartists”.
4
The parameter
is calibrated equal to 0:5833
9
In turn, the endogenous component of risk premium in (8) is determined by interventions
in the currency market: the central bank buys or sells international reserves, and issues or
withdraws bonds from circulation in order to sterilize the e¤ects of intervention on the money
supply. The consequent change in the endogenous risk premium may be rationalized as re‡ecting
both counterparty (bbt ) and exchange rate risk ( [
rest ): to hold a higher stock of bonds, local
investors demand a higher rate (this would not be the same as holding bonds issued abroad,
re‡ecting a di¤erent counterparty); changes in international reserves are associated to changes
in exchange rate risk, as when it intervenes, the central bank modi…es the foreign currency
volatility5 . Other rationalizations could read as follows: regarding the presence of bbt , if central
bank bond issuance is interpreted as postponed liquidity supply, higher bonds today may mean
higher liquidity tomorrow and, therefore, a higher interest rate rate today; international portfolio
adjustment could be considered costly, depending on the relative holdings of bonds in pesos and in
foreign currency, and so central bank intervention using reserves actually changes the endogenous
risk premium and, with it, the exchange rate (Sierra, 2008).
Central bank interventions are ruled by a "propensity" to avoid exchange rate movements to
a certain extent as measured by the 2 coe¢cient in (11), in keeping with the aim of a managed
‡oating regime of smoothing short term "excessive" ‡uctuations of the nominal exchange rate.
Thus, any external …nancial shocks are smoothed by the central bank in line with its aim of
minimizing short run disruption in the foreign exchange market. A desire to act gradually is
re‡ected by the autoregressive 1 coe¢cient, which can be rationalized on the grounds of …nancial
stability.
Having characterized the basic macroeconomic dynamics, together with central bank policy
in the money and foreign exchange markets, the following step is to consider lending rates and
credit. In the model, credit—strictly, the credit-to-GDP gap—is basically a function of output
growth and the lending interest rate, as shown in both credit market equilibrium equations,
one referred to household (consumption) credit and the other to corporate (commercial) credit
(13). In turn, equation (15) describes active (lending) rates as a function of the output gap,
non performing loans and the short term rate; the spread emerges naturally as the di¤erence
between the lending and money market rate. This speci…cation is consistent with empirical
results for the Argentine economy that spread depends negatively on growth and positively on
non-performing loans (Aguirre et al, 2014). As before, lending rates are considered for both
commercial and consumption loans. Non performing loans are a function of economic activity, in
line with their observed cyclical behavior. Credit as previously de…ned also feeds back into the
"macroeconomic block" of the model through its inclusion in the interest rate rule (6); this, of
course, is not the only way in which the quantity of credit may directly a¤ect the macroeconomy
(it could, for instance, directly impact on output in (1)), but in this speci…cation we consider
only one channel that, albeit indirect, is related to …nancial stability considerations on the part
of the central bank—a feature which, in our view, is relevant for the estimation period. As noted,
lending rates as de…ned in (15) do a¤ect economic activity through the inclusion of the interest
rate spread in (1). Finally, exogenous variables follow autoregressive processes: the international
interest rate, the exogenous component of risk premium in (8), foreign in‡ation, two measures of
the bilateral exchange rate, the …scal balance and potential output. Unless otherwise indicated,
all variables are expressed as deviations from steady state values, denoted by a circum‡ex.
5
Including exports and imports in the model, as pointed out earlier, could also help better characterize the
evolution of international reserves.
10
Financial Block
H
dH
CR
t
bty
= AH
1 g
1
Hd
bH;act
AH
2 it 1 + A3 CRt
dF
CR
t
= AF1 gbty
1
dF
A2biF;act
t 1 + A3 CRt
H
1
1
+ "HCR
t
+ "Ft CR
(13)
(14)
\t
biH;act
= B1 Delinq
t
B2 gbty
1
+ B3bit + "Hact
t
(15)
F
B2 gbty
1
+ B3bit + "Ft act
(16)
\t
biF;act
= B1 Delinq
t
Delinq H : ratio of non performing loans to household credit, Delinq F : ratio of non performing loans to …rms credit
H
\t
Delinq
F
\t
Delinq
=
DH \ H
1 Delinq t 1
+
DH y
bt 1
2 g
+ "HDelinq
t
(17)
=
DF \ F
1 Delinq t 1
+
DF y
bt 1
2 g
+ "Ft Delinq
(18)
Identities
\
U S;R + c e
U S;E
ebd t + c1 e\
t
2
t
bit Et bt+1
tri
ec
t
rbt
d
ed
bt + ct
t
y
gbt
yt + gby t
bt
(19)
(20)
bt
m
b t + bt +
(21)
(22)
gbty
(23)
eU S;R : USD/REAL RER, eU S;E : USD/EURO RER, : international in‡ation, g y : potential
output growth rate, g y : GDP growth rate, : money growth rate
Exogenous variables6
ib t =
bt =
ct =
U S;R
e\
t =
U S;E
e\
t =
c =
sf
t
6
gby t =
Parameter
7
b
bt
1i t 1
+ "it
(24)
1
+ "t
(25)
2
3
ct
1
+ "t
\
U S;R
t 1
4e
\
U S;E
t 1
5e
+
(26)
U S;R
"et
(27)
eU S;E
+ "t
(28)
sf
c
6 sf t 1 + "t
gy
by
7 g t 1 + "t
(29)
(30)
is calibrated
11
4
Estimation
We estimate this baseline version of the model (equations 1-30) completely through Bayesian
techniques7 , based on quarterly data and for the 2003Q3-2011Q3 period; this is the longest period
spanning an homogeneous macroeconomic policy regime—the currency board regime adopted in
1991 was abandoned during the 2001-2002 crisis, after which a managed ‡oating regime was
adopted. The estimation period also includes the breakout of the international …nancial crisis
and the response to it, such as the successive rounds of quantitative easing in the United States:
this is relevant for any policy assessment in emerging markets, since the latter felt the impact of
changing international …nancial conditions. In the case of Argentina, repercussions were felt since
mid-2007, and managed ‡oating exchange rate policy proved important in stabilizing the local
money market following the external shock; also, several measures were put in place to provide
liquidity in local currency to …nancial institutions. As noted, both foreign exchange operations
and their e¤ect on the money market are represented in our model. Also, shock t in equation
(8) can be thought of as representing the kind of exogenous increase in risk premium associated
to the events from 2007 onwards.
Bayesian techniques prove particularly useful for the kind of situation described in the above
paragraph: if one knows that structural change has taken place, this information can be included
in a way not allowed by classical estimation methods. Bayesian statistics allows researchers to
incorporate a priori information on the problem under study, thus potentially improving the
e¢ciency of estimates—and re‡ecting a frequent concern of both analysts and policy makers
regarding how to include what they know from experience about the economy in a formal framework. Under this approach, parameters are interpreted as random and data as …xed. Both
features are particularly relevant when the sample size is small due to structural breaks, as it
is the case of Argentine economy in the period we focus on. De…ne 2
as the vector of
parameters. Given the prior information g( ), the observed data YT = [Y1; Y2 ; :::; YT ] and the
sample information f (YT = ), the posterior density—transition from prior to posterior—of the
parameters is given by Bayes’ rule:
g ( =YT ) =
g ( =YT ) =
f (YT = ) g ( )
f (YT )
R
f (YT = ) g ( )
f (YT = ) g ( ) d
Notice that f (YT ) (the marginal likelihood) is constant, hence the posterior density is proportional to the product of the likelihood function f (YT = ) and the prior density. The inclusion of
prior information allows then to generate a more "concave" density, which is crucial for parameter identi…cation when the information contained in the data is considered insu¢cient; in
other words, if we want to know which alternative model parameters are more likely to have
been obtained from the sample used, providing a priori information improves the ability to
identify them correctly.
The modes of the posterior distributions can be easily computed using standard optimization
routines—in our case we choose a Monte-Carlo based approach. However, obtaining the whole
posterior distributions is considerably more di¢cult, requiring the calculation of complex multivariate integrals. For this reason, many algorithms have been developed to compute samples
7
Model solution, estimation and stochastic simulations were performed using the Dynare 4.3.3 software platform
in Matlab.
12
of the posterior distributions by e¢ciently using available information. The most popular is the
Random Walk Metropolis-Hastings algorithm, which we use in our estimation. The algorithm
applies a random walk as a jumping process to explore the posterior distribution of the parameters. We used two chains of 50,000 replications each. The variance of the jumps is calibrated
to achieve an acceptation rate between 0:2 and 0:4, which is considered an acceptable target to
ensure that the search is global.
The priors chosen are based on the posterior distributions from an estimation performed for
the pre-crisis, currency board period. The set of observed variables Y is:
H
F
c ; ebU S;R ; ebU S;E ; CR
d H ; CR
d F ; biH;act ; biF;act ; Delinq
\ ; Delinq
\ ]
Y = [b; bi; bi ; b ; gby ; b; m;
b res;
c sf
See Annex I for a description of variables’ de…nitions and data sources.
Table 1 presents parameter estimates8 ; Table 2 contains the standard deviation of shocks.
parameters
1
3
4
1
2
3
4
5
6
1
2
3
4
5
6
1
2
3
4
5
!1
!2
!3
Table 1: Baseline model
parameter estimates
prior mean post. mean conf. interval
0:300
0:264
0:233
0:305
0:050
0:078
0:062
0:094
0:100
0:065
0:051
0:078
0:300
0:526
0:455
0:599
0:500
0:397
0:340
0:456
0:170
0:136
0:125
0:149
0:200
0:109
0:084
0:133
0:300
0:113
0:071
0:159
0:300
0:123
0:075
0:169
0:500
0:937
0:882
0:989
0:500
0:741
0:617
0:873
0:500
0:320
0:283
0:362
0:700
0:972
0:945
0:999
0:700
0:711
0:651
0:773
0:500
0:658
0:544
0:764
0:700
0:573
0:519
0:623
0:000
0:021
0:016 0:057
0:000
0:025
0:012
0:038
0:200
0:083
0:064
0:101
0:000
0:007
0:005
0:010
4:000
5:911
5:598
6:262
0:100
0:008
0:002
0:014
1:000
0:178
0:000
0:380
prior
beta
norm
beta
beta
beta
norm
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
norm
norm
beta
norm
norm
beta
norm
pstdev
0:100
0:035
0:050
0:100
0:200
0:050
0:100
0:100
0:100
0:200
0:200
0:200
0:200
0:200
0:200
0:200
0:200
0:200
0:100
0:200
1:500
0:050
1:000
8
It is worth mentioning that we estimated alternative speci…cations of equations (10) and (11) in terms of lagged
variables and signs of parameters of interest, and selected the one with the best goodness-of-…t, as measured by
the posterior odds ratio.
13
parameters
1
2
3
4
1
2
AH
1
AH
2
AH
3
B1H
B2H
B3H
DH
1
DH
2
AF1
AF2
AF3
B1F
B2F
B3F
DF
1
DF
2
"i
y
"g
"y
"i
"
"RP
U S;R
"e
U S;E
"e
"
"m
"res
"sf
"CR;H
"act;H
"Delinq;H
"CR;F
"act;F
"Delinq;F
Table 1 (cont.): Baseline model
parameter estimates
prior mean post. mean conf. interval
1:200
1:203
1:137
1:270
0:500
0:553
0:477
0:623
0:500
0:031
0:023
0:038
0:500
0:665
0:635
0:695
0:700
0:982
0:964
0:998
0:100
0:138
0:116
0:159
0:300
0:401
0:385
0:417
0:100
0:066
0:056
0:078
0:300
0:379
0:365
0:397
0:300
0:069
0:048
0:092
0:300
0:169
0:145
0:194
0:300
0:228
0:179
0:279
0:500
0:810
0:761
0:856
0:300
0:472
0:419
0:518
0:300
0:333
0:319
0:343
0:100
0:110
0:091
0:129
0:300
0:410
0:392
0:427
0:300
0:018
0:010
0:025
0:300
0:230
0:212
0:249
0:300
0:215
0:153
0:275
0:500
0:912
0:894
0:929
0:300
0:455
0:424
0:485
Table 2: Baseline model
standard deviation of shocks
prior mean post. mean conf. interval
0:050
0:003
0:002
0:004
0:050
0:024
0:014
0:040
0:050
0:015
0:011
0:019
0:050
0:002
0:001
0:002
0:050
0:009
0:008
0:011
0:050
0:022
0:013
0:032
0:050
0:073
0:061
0:082
0:050
0:046
0:035
0:057
0:050
0:011
0:008
0:013
0:060
0:038
0:033
0:044
0:050
0:105
0:096
0:115
0:050
0:005
0:003
0:005
0:100
0:114
0:101
0:127
0:050
0:006
0:005
0:008
0:050
0:009
0:007
0:011
0:100
0:202
0:187
0:215
0:050
0:007
0:005
0:009
0:050
0:011
0:008
0:013
14
prior
norm
beta
norm
norm
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
prior
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
pstdev
0:300
0:200
0:300
0:100
0:200
0:050
0:050
0:050
0:050
0:100
0:100
0:100
0:200
0:100
0:050
0:050
0:050
0:100
0:100
0:100
0:200
0:100
pstdev
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
With this fully estimated model, we look at impulse-response functions in order to understand
its basic dynamics, with emphasis on how the credit market block interacts with the rest of
the economy. Following positive shocks to lending rates—both for commercial and consumption
credit—(Figure 1), credit decreases and the interest rate spread increases—the short term interest
rate increases, but to a lesser degree than the active rate. As expected, each line of credit reacts
more strongly to an increase of its own rate. This a¤ects the real side of the economy, with a
negative e¤ect on output growth. As the short term interest rate increases, the nominal exchange
rate depreciates—the impact on UIP means that a higher local rate, with no change in the
international interest rate, translates into a higher expected depreciation of the local currency.
Pass-through from the exchange rate to domestic prices entails a fall on the real interest rate.
The central bank acts by gradually increasing the short term rate and intervening in the foreign
exchange market to reduce foreign exchange volatility.
A shock to the passive rate (Figure 2), translates immediately into a higher real (short term)
interest rate, which goes together with (initial) nominal and real exchange rate appreciation;
output is also a¤ected. The central bank reacts by (initially) buying reserves and sterilizing the
monetary e¤ect of its operations by issuing bonds. In the credit market, lending rates go up
while credit diminishes—spreads are reduced as the active rate is raised less than one-to-one with
respect to the passive rate. We are aware that both exercises are just a crude approximation
at describing the interplay between the credit market and the macroeconomy, and that certain
aspects that are very relevant for …nancial stability analysis are omitted here—for example, the
e¤ect of passive rates on deposit growth9 .
9
In this model, a higher passive rate means only a higher opportunity cost of holding transactional money, but,
by construction, no e¤ect on savings deposits (which are not included); however, this can be very signi…cant.
15
Figure 2
Accumulated responses to 1 s.d. shock to the short term interest rate
Baseline Model
Short term IR Shock
0.004
0.003
0.003
0.002
0.002
0.001
0.001
1
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
35
37
39
41
43
45
47
-0.001
-0.001
-0.002
Inflation
Output growth
Real short term interest rate
Nominal exchange rate change
Baseline Model
Short term IR Shock
0.025
0.020
0.015
0.010
0.005
1
6
11
16
21
26
31
-0.005
-0.010
-0.015
-0.020
Money
CB bonds
16
Reserves
36
41
46
49
Figure 2 (cont.)
Accumulated responses to 1 s.d. shock to the short term interest rate
Baseline Model
Short term IR Shock
0.001
0.001
1
6
11
16
21
26
31
36
41
46
-0.001
-0.001
-0.002
-0.002
-0.003
Household Credit
Household lending rate
Household interest rate spread
Firm Credit
Firm lending rate
Firm interest rate spread
Figure 3
Accumulated responses to 1 s.d. shock to the Household lending rate
Baseline Model
Household lending rate Shock
0.001
0.001
0.000
0.000
1
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
35
37
39
41
43
45
47
-0.000
-0.000
-0.001
-0.001
-0.001
-0.001
Inflation
Output growth
Real short term interest rate
17
Nominal exchange rate change
49
Figure 3 (cont.)
Accumulated responses to 1 s.d. shock to the Household lending rate
Baseline Model
Household lending rate
Shock
0.002
0.002
0.001
0.001
1
6
11
16
21
26
31
36
41
36
41
46
-0.001
-0.001
-0.002
-0.002
-0.003
Money
CB bonds
Reserves
Baseline Model
Household lending rate Shock
0.007
0.006
0.005
0.004
0.003
0.002
0.001
1
6
11
16
21
26
31
46
-0.001
Household Credit
Household lending rate
Household interest rate spread
Firm Credit
Firm lending rate
Firm interest rate spread
18
Figure 4
Accumulated responses to 1 s.d. shock to the Firm lending rate
Baseline Model
Firm lending rate Shock
0.001
0.001
0.000
0.000
1
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
35
37
39
41
43
45
47
-0.000
-0.000
-0.001
-0.001
-0.001
-0.001
Inflation
Output growth
Real short term interest rate
Nominal exchange rate change
Baseline Model
Firm lending rate Shock
0.002
0.002
0.001
0.001
1
6
11
16
21
26
31
-0.001
-0.001
-0.002
-0.002
-0.003
Money
CB bonds
19
Reserves
36
41
46
49
Figure 4 (cont.)
Accumulated responses to 1 s.d. shock to the Firm lending rate
Baseline Model
Firm lending rate Shock
0.008
0.007
0.006
0.005
0.004
0.003
0.002
0.001
1
6
11
16
21
26
31
36
41
46
-0.001
-0.002
Household Credit
Household lending rate
Household interest rate spread
Firm Credit
Firm lending rate
Firm interest rate spread
This exercise can also be done to analyze how a real shock is transmitted throughout the
rest of the economy and the credit market (Annex 2). A positive shock to the IS curve increases
output and in‡ation; the short term interest rate increases, in nominal terms—basically due to
the reaction required by the Taylor rule—but decreases in real terms. This leads to real exchange
rate appreciation so the central bank buys reserves to "resist" it and issues bonds to sterilize the
monetary e¤ects of its operations. In turn, both types of credit increase, the lending rates fall,
and so do both spreads.
It is worth noting that, in the cases of shock to the lending rate and to output, the spread is
countercyclical in the sense that higher (lower) spread entails lower (higher) credit and output10 .
In contrast, when the short term interest rate is shocked, the spread appears to be procyclical—
while credit also goes down, since the active rate is going up, the spread is reduced. Our
interpretation is that in the latter case the e¤ect of decreased credit demand, together with
lower output associated to a higher real rate, more than o¤sets the direct expansionary impact
of a lower spread. In all of the three cases, credit is procyclical.
Thus, even a relatively simple speci…cation as this appears at least to be partly indicative
of how the credit market interacts with the rest of the economy and with monetary policy. As
shown by the exercises above, it is not only the traditional "transmission mechanism" of shocks
that should be looked at, but the addition of both foreign exchange operations and the credit
market reveal new channels that are relevant to the explanation of cyclical impulses.
Regarding how cycles are transmitted throughout the economy, we can also look at suggestive
results from the relationship between the macroeconomic and the …nancial blocks of the model:
10
This agrees with the empirical …nding of Aguirre et al (2013) for the Argentine economy in 1996-2012, that
output growth has a negative e¤ect on interest rate spread, also indicating countercyclicality.
20
we compare the variability of credit following a shock to output growth, and the variability of
the latter in the face of a credit shock. Table 3 shows the standard deviation of growth following
shocks to consumption and commercial credit: growth is more variable in the face of a shock to
corporate credit than one to household credit. But both types of credit are ten to twenty times
more volatile following a shock to output growth than in the opposite case. This suggests that
impulses coming from the real side of the economy weigh more heavily on the …nancial system
than the other way around.
Table 3
Standard deviations of responses to shocks of selected variables after
10 quarters
20 quarters
30 quarters
Consumption credit
Output
0:00015
0:00022
0:00022
Corporate credit
Output
0:00029
0:00042
0:00042
Output growth
Consumption credit
0:00338
0:00251
0:00204
Corporate credit
0:00297
0:00223
0:00181
5
The extended model: macroprudential policy
Of the many di¤erent measures that can be considered as "macroprudential", we will focus on
one of the most basic …nancial system regulations—a capital adequacy ratio—, and will consider
several variants. These range from a purely exogenous ratio from the macroeconomic point
of view (thus akin to conventional prudential regulation) to rules according to which adequate
capital depends on macroeconomic or …nancial system variables. This allows us to examine
cyclical measures and others that do not directly depend on a cyclical variable, as well as pricebased vis-a-vis quantity-based rules. Such measures are macroprudential in addition to the
managed ‡oating foreign exchange regime: in so far as such policy limits variability of a certain
class of assets that weigh on …nancial system dynamics, foreign exchange intervention can also
be considered part of the macroprudential "toolkit".
We enlarge the model’s …nancial block by adding a capital adequacy ratio, which we …rst
de…ne as an exogenous rule (31)- its level does not depend on variables explicitly modelled11 .
We take this to be, a priori, the most representative way of depicting capital requirements in
Argentina during the estimation period, as they were not de…ned as a function of macroeconomic
variables, but of risk-weighted assets of …nancial institutions (BCRA, 2014) The capital adequacy
ratio (CAR) is then included in the equation describing the actives rates (32) and (33); we
hypothesize that higher capital requirements will be associated with higher lending rates, since
each additional loan has to be "backed" by more equity. The new equations are as follows.
Capital Adequacy Ratio
First Option: Exogenous
[t =
CAR
0
+
[
1 CARt 1
+ "CAR
t
(31)
where CAR : capital adequacy ratio
11
Strictly speaking, of course, capital requirements are always endogenous from the point of view of …nancial
institutions, as they depend on their risk-weighted assets.
21
H
\t
biact;H
= B1H Delinq
t
F
\t
biact;F
= B1F Delinq
t
B2H gbty
B2F gbty
1
1
[ t + "Hact
+ B3Hbit + B4 CAR
t
(32)
[ t + "Ft act
+ B3F bit + B4 CAR
(33)
We then estimate model 2 with macroprudential policy using Bayesian techniques; as with
the baseline model, we estimate using quarterly data of the Argentine economy for the 2003Q32011Q3 period. Estimates of parameters and standard deviations are shown in Tables 4 and
5.
parameters
1
3
4
1
2
3
4
5
6
1
2
3
4
5
6
1
2
3
4
5
!1
!2
!3
1
2
3
4
1
2
Table 4: Model 2, exogenous CAR
parameter estimates
prior mean post. mean conf. interval
0:300
0:215
0:180
0:246
0:050
0:032
0:006
0:062
0:100
0:141
0:118
0:170
0:300
0:323
0:290
0:360
0:500
0:459
0:401
0:518
0:170
0:217
0:185
0:249
0:200
0:158
0:108
0:211
0:300
0:166
0:124
0:206
0:300
0:260
0:161
0:354
0:500
0:962
0:931
0:992
0:500
0:709
0:609
0:832
0:500
0:364
0:295
0:447
0:700
0:962
0:928
0:998
0:700
0:905
0:827
0:961
0:500
0:220
0:113
0:317
0:700
0:626
0:533
0:743
0:000
0:013
0:009 0:036
0:000
0:024
0:005
0:043
0:200
0:077
0:045
0:106
0:000
0:005
0:001
0:010
4:000
5:595
4:733
6:500
0:100
0:010
0:003
0:016
1:000
0:240
0:002
0:458
1:200
0:952
0:828
1:061
0:500
0:692
0:589
0:820
0:500
0:027
0:020
0:035
0:500
0:738
0:694
0:779
0:700
0:976
0:954
0:998
0:100
0:128
0:102
0:156
22
prior
beta
norm
beta
beta
beta
norm
beta
beta
beta
beta
beta
beta
beta
beta
beta
beta
norm
norm
beta
norm
norm
beta
norm
norm
beta
norm
norm
beta
beta
pstdev
0:100
0:035
0:050
0:100
0:200
0:050
0:100
0:100
0:100
0:200
0:200
0:200
0:200
0:200
0:200
0:200
0:200
0:200
0:100
0:200
1:500
0:050
1:000
0:300
0:200
0:300
0:100
0:200
0:050
Table 4 (cont.): Model 2, exogenous CAR
parameter estimates
parameters prior mean post. mean conf. interval prior
AH
0:300
0:377
0:360
0:390
beta
1
H
A2
0:100
0:098
0:076
0:122
beta
AH
0:300
0:414
0:396
0:436
beta
3
H
B1
0:300
0:099
0:075
0:123
beta
B2H
0:300
0:254
0:230
0:281
beta
B3H
0:300
0:239
0:159
0:318
beta
H
B4
0:300
0:145
0:120
0:170
beta
DH
0:500
0:819
0:787
0:850
beta
1
DH
0:300
0:374
0:328
0:419
beta
2
AF1
0:300
0:385
0:353
0:416
beta
AF2
0:100
0:099
0:070
0:132
beta
AF3
0:300
0:459
0:433
0:489
beta
B1F
0:300
0:023
0:011
0:033
beta
B2F
0:300
0:244
0:184
0:302
beta
B3F
0:300
0:261
0:186
0:303
beta
F
B4
0:300
0:134
0:098
0:171
beta
DF
0:500
0:907
0:885
0:932
beta
1
DF
0:300
0:473
0:436
0:510
beta
2
0:500
0:011
0:010
0:012
beta
0
0:700
0:378
0:288
0:478
beta
1
"i
y
"g
"y
"i
"
"RP
U S;R
"e
U S;E
"e
"
"m
"res
"sf
"CR;H
"act;H
"Delinq;H
"CR;F
"act;F
"Delinq;F
"CAR
Table 5: Model 4, exogenous CAR
standard deviation of shocks
prior mean post. mean conf. interval
0:050
0:003
0:002
0:004
0:050
0:019
0:011
0:028
0:050
0:018
0:014
0:022
0:050
0:001
0:001
0:002
0:050
0:010
0:008
0:012
0:050
0:035
0:024
0:046
0:050
0:062
0:054
0:069
0:050
0:042
0:035
0:049
0:050
0:013
0:010
0:016
0:060
0:031
0:023
0:038
0:050
0:109
0:095
0:122
0:050
0:004
0:003
0:005
0:100
0:122
0:111
0:131
0:050
0:007
0:005
0:008
0:050
0:008
0:006
0:010
0:100
0:167
0:154
0:179
0:050
0:007
0:005
0:009
0:050
0:012
0:009
0:014
0:050
0:014
0:011
0:017
23
prior
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
gamma
pstdev
0:050
0:050
0:050
0:100
0:100
0:100
0:100
0:200
0:100
0:050
0:050
0:050
0:100
0:100
0:100
0:100
0:200
0:100
0:200
0:200
pstdev
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
0:035
We use the estimated model to try to gain some understanding of the potentially stabilizing
properties of prudential policy. Are capital adequacy ratios associated to less volatility in the
macroeconomy and the …nancial system? In order to answer this question, we compute the
estimated variability of selected variables under the "exogenous" CARs, and compare them with
the baseline model (model 1). At this point, it is worth remembering that, by construction,
estimated models re‡ect the type and magnitude of shocks that the economy underwent during
the estimation period; so by showing variability under the estimated policy , we approximate
the economy’s performance associated to it in the face of the particular shocks occurred.
A number of objections to the exercise may be raised. One could argue to what extent we can
use a small structural model, not explicitly derived from optimizing behavior of agents, to assess
alternative policies. As policies change, so do responses of agents, something not necessarily
captured by our behavioral equations. It should be noted, however, that the model is built with
rational expectations so, at least at the level of aggregation we are working with, responses do
incorporate expectations consistent with the model’s structure.
In connection to the above, it could be pointed out that results in a structural model such as
this one are subject to the "Lucas critique"—with estimated parameters being biased as there
is no guarantee of invariance to policy changes. This requires some methodological clari…cation:
using a "micro founded" model would not, in and of itself, assure such invariance and, with it,
unbiased results—even if this is usually taken for granted in the use of DSGE models. This is a
purely empirical question12 —and as practitioners know, parameters in macroeconomic models are
usually re-estimated or re-calibrated periodically, implicitly violating the same condition they are
assumed to satisfy. Macroeconomic models, whether large or small, are in practice subject to this
bias—the question is how large it is, and how it compares to that of alternative models. Largescale DSGE frameworks, for instance, are ridden with problems of identi…cation and estimation—
with certain key parameters or relationships being neither "micro founded" nor estimated. So
that while we cannot rule out that the model presented here is indeed subject to the Lucas
critique, in our view it represents an acceptable trade-o¤ between empirical tractability (with all
parameters being estimated) and full analytical development that can (only theoretically) bring
the model closer to invariance to selected policy interventions.
With the previous points in mind, we compute standard deviations of macroeconomic and
…nancial variables under models 1 and 2. We do the exercise for: in‡ation, output growth,
local short term interest rates, the real trilateral (trade-weighted) exchange rate, money growth,
international reserves, credit (total and by line), lending interest rates (average and by credit
line), non performing loans (by credit line) and capital requirements. The comparison in Table 6
suggests a lower volatility during the estimation period under an exogenous capital requirement
for most of the variables considered.
12
As found by Ericsson and Irons (1995), macroeconomic models are typically subject to the Lucas critique in
practice; the econometric condition to be satis…ed is that of superexogeneity, something that is independent of
whether the model was derived from …rst order conditions of an optimization problem or not.
24
Table 6 : Observed and estimated standard deviations of selected variables
Variable
Model 1
Model 2
Baseline
Exogenous CAR
0:058
0:031
i
0:013
0:012
gy
0:057
0:047
tri
e
0:096
0:057
m
0:220
0:184
res
0:550
0:507
CR
0:262
0:239
CRH
0:128
0:137
F
CR
0:224
0:191
iact
0:017
0:016
iact;H
0:018
0:019
act;F
i
0:018
0:015
Delinq H
0:115
0:076
Delinq F
0:197
0:157
CAR
0:015
These results suggest that capital requirements do ful…ll their expected role of decreasing risktaking: non performing loans’ volatility is between 20% and 35% lower when capital requirements
are implemented than when they are not. Volatility of credit is reduced on average 9% under
capital adequacy ratios, a result driven by lower volatility of …rms’ credit; in the same breath,
lending rates to …rms are 14% lower when prudential policy is implemented (but the same does
not apply to consumption lending rates). In addition, there appears to be a macroeconomic
impact of capital requirements: growth volatility is 17% lower under CARs than without them;
and in‡ation volatility appears to be reduced by half when CAR is put in place. While we
are fully aware of the suggestive nature of our results, we believe they point in a direction
of macroeconomic signi…cance of macroprudential measures; indeed, our …gures are of the same
order of magnitude that Covas and Fujita (2009) …nd for growth volatility reduction with countercyclical capital requirements.This leads to discuss alternative rules to "exogenous" CARs, that
are a function of macroeconomic or …nancial variables, with cyclical behaviour or not.
5.1
Examining alternative capital adequacy rules
Using as a starting point the results obtained in model 2, we propose three alternative CAR rules:
(i) a function of the output gap (34); (ii) a function of the credit-to-GDP gap (35), which is the
standard way in which countercyclical capital regulation is currently being designed under Basel
III (Drehmann and Tsatsaronis, 2014); or (iii) the interest rate spread (36). These alternatives,
labeled respectively as models 3, 4 and 5, correspond to di¤erent policy concerns: risk taken
by banks is moderated by higher requirements, which may be more related to macroeconomic
(model 3) or …nancial system performance (models 4 and 5). The main di¤erence in motivation
between models 4 and 5 is whether quantity-based or price-based indicators perform better in
terms of early warning of crises (Shin, 2013).
25
Second Option: Endogenous
[t =
CAR
0
+
1 CARt 1
[
+
^t
2y
[t =
CAR
0
+
1 CARt 1
[
+
2 CRt
[t =
CAR
0
+
1 CARt 1
[
+
2 spreadt
+ "CAR
t
(34)
d + "CAR
t
(35)
+ "CAR
t
(36)
For each model, we: use the estimated parameters of model 213 ; and change the capital adequacy ratio in order to implement the three alternative rules outlined in the previous paragraph,
taking estimated coe¢cients for each of them (such coe¢cientes are obtained from estimations
of models 3-5). We once again compute the standard deviation of selected variables for all the
for models we compare. Certainly, there are well-known limits to this strategy (see previous section), but we believe it is a useful …rst approximation to the assess potential gains associated to
implemeting explicit macroprudential rules. Results generally indicate that capital requirements
as a function of macroeconomic (output gap) or …nancial system variables (credit gap, interest
rate spread) deliver lower volatility than purely "exogenous" ones. For most macroeconomic and
monetary variables (in‡ation, growth, real exchange rate, money and the short term interest
rate), a capital adequacy rule that changes with the output gap delivers the lowest volatility.
Instead, for most …nancial system variables (credit and lending rates), it is capital requirements
as a function of interest rate spread that are associated to lower variance; the exception here is
non-peforming loans, with lower variability under model 3 (CAR as function of output gap). Perhaps not surprisingly, a CAR that depends on output performs better (under this criterion) for
macroeconomic variables, while a rule that depends on interest rates works better for …nancial
system variables.
Table 7 : Standard deviations of selected variables - Calibrated Models
Variable Model 2
Model 3
Model 4
Model 5
Exogenous Endogenous Endogenous
Endogenous
CAR
CAR ( y)
CAR (cred)
CAR (spread)
0:0307
0:0301
0:0312
0:0312
i
0:0116
0:0111
0:0117
0:0122
gy
0:0473
0:0471
0:0475
0:0476
etri
0:0572
0:0571
0:0574
0:0572
m
0:1836
0:1762
0:1822
0:1937
res
0:5065
0:5067
0:5043
0:5072
CR
0:2392
0:2392
0:2393
0:238
H
CR
0:1372
0:1372
0:1373
0:1369
CRF
0:1907
0:1907
0:1908
0:1902
iact
0:0164
0:0179
0:0172
0:016
act;H
i
0:0191
0:0206
0:0199
0:0186
iact;F
0:0152
0:0165
0:0159
0:015
H
Delinq
0:0757
0:0755
0:0759
0:0757
Delinq F
0:1571
0:157
0:1574
0:157
CAR
0:0153
0:0233
0:020
0:0465
13
See Table A.3 and Table A.4 in Annex 3 for the same exercise with estimated (instead of calibrated) models.
26
In order to gain a comprehensive assessment of these results and the policies associated to
them, we aggregate the di¤erent variability measures by summing up variances of the variables
considered—thereby using ad hoc "loss" functions. As those functions are not derived from the
utility of a representative consumer, they do not indicate anything in terms of social welfare,
but we interpret them instead as embodying alternative evaluation criteria of an analyst or
policymaker whose concern is for the volatility of selected macroeconomic and …nancial variables.
Generally, the loss function we use here is de…ned as follows.
We considered several loss functions, combining di¤erent macro g y ; ; etri and …nancial
i; iact ; iact;H ; iact;F ; CRH ; CRF ; CAR variables. Thus, an example of loss function could be:
L
!g
y
2
gy
+!
2
+ !i
2
i
+ ! CAR
2
CAR
where 2gy is the variance of output gap, 2 is the variance of in‡ation, 2i is the variance of
short term interest rate and 2CAR is the variance of the capital adequacy ratio, and ! are the
y
corresponding weights such as ! g + ! + ! i + ! CAR = 1:
Initially, we assign equal weights to all components of the function, considering in all cases in‡ation, output growth, the short term interest rate and real exchange rate depreciation, together
with: consumption credit, commercial credit, and commercial credit and capital requirements.
To consider lending rates, we also look at the sum of in‡ation, output growth, real exchange rate
depreciation and: consumption lending rate and credit; commercial lending rate and credit. To
focus on macroeconomic variables and central bank’s instruments, we consider output growth,
in‡ation, the short term interest rate and capital adequacy ratios. When only macroeconomic
variables and interest rates are involved, capital requirements that vary with the output gap
tend to yield lower volatility, whereas when …nancial system variables are included, either CAR
as a function of spread or of the credit gap obtain lower aggregate variability. However, when
the variabiliy of capital requirements themselves is considered, "exogenous" requirements yield
lower "losses" (Table 8, panel a).
A related exercise has to do with changing weights in the terms of the loss function. First, we
compute aggregate volatility with higher weights on macroeconomic variables (output growth,
in‡ation, real exchange rate depreciation), grouping: both macroeconomic and …nancial system
variables (Table 8, panel b); and macroeconomic variables and interest rates only (Table 8, panel
c). In both cases, capital requirements that vary with the output gap show the lowest volatility.
This suggests, perhaps naturally, that a CAR that changes with an indicator of the business
cycle is more apt to do better when the main concern is macroeconomic performance.
Next, we compute aggregate volatility with higher weights on …nancial system variables
(interest rates, credit), once again grouping macroeconomic and …nancial system variables (Table
8, panel d), and macroeconomic and interest rates only (Table 8, panel e). In panel d, CAR as
a function of interest rates spread delivers lower volatility in two cases, whereas in two others a
better perfomance is found for capita requirements that change with the output gap, or even the
"exogenous" CAR, In panel e, "exogenous" capital requirements are associated to lower volatility.
Thus, when more weight is put on …nancial system indicators’ performance, there appears to be
a role for macroprudential policy based on interest rate spreads.
27
28
Table 8: Loss Functions of alternative models
Variables
Model 1
Model 2
Model 3
Model 4
Model 5
Considered
Baseline Exogenous CAR Endogenous CAR Endogenous CAR Endogenous CAR
in Loss Function
(y)
(cred)
(spread)
1
a) Equal weights ! = n
gy ;
0:00661
0:00318
0:00312
0:00323
0:00324
g y ; iact
0:00355
0:00251
0:00254
0:00255
0:00252
y
act
g ; ;i
0:00666
0:00319
0:00314
0:00324
0:00325
g y ; ; iact ; etri
0:00688
0:00323
0:00317
0:00328
0:00329
y
act
g ; ; i; i
0:00710
0:00355
0:00352
0:00362
0:00361
g y ; ; i; CAR
0:00679
0:00355
0:00379
0:00377
0:00555
g y ; ; i; etri ; CRH
0:03250
0:02564
0:02563
0:02577
0:02560
y
tri
F
g ; ; i; e ; CR
0:06607
0:04305
0:04302
0:04318
0:04291
g y ; ; i; etri ; CRF ; CAR
0:04319
0:04342
0:04347
0:04500
y
act
tri
H
g ; ; i ; e ; CR
0:03250
0:02564
0:02563
0:02577
0:02560
g y ; ; iact ; etri ; CRF
0:06607
0:04305
0:04302
0:04318
0:04291
act;H
act;F
y
tri
1
4
i
i
i
g
e
=!
= 15
b) Weights: Macro variables ! = ! = !
= 15 ; Financial variables ! = !
y
tri
H
g ; ; i; e ; CR
0:00532
0:00298
0:00297
0:00301
0:00300
y
tri
F
g ; ; i; e ; CR
0:00756
0:00415
0:00414
0:00418
0:00416
g y ; ; i; etri ; iact;H
0:00425
0:00175
0:00173
0:00177
0:00176
g y ; ; i; etri ; iact;F
0:00425
0:00175
0:00174
0:00178
0:00177
act;H
act;F
y
1
5
; Financial variables ! i = ! i
= !i
= 12
c) Weights: Macro variables ! g = ! = 12
g y ; ; ; i; iact;H
0:00280
0:00136
0:00133
0:00138
0:00138
g y ; ; i; iact;F
0:00280
0:00136
0:00133
0:00138
0:00138
tri
act;H
act;F
y
1
5
2
e
i
i
i
g
= 15 ; Financial variables ! = !
=!
= 15
d) Weights: Macro variables ! = ! = 15 and !
y
tri
H
g ; ; i; e ; CR
0:00701
0:00696
0:00695
0:00698
0:00695
y
tri
F
g ; ; i; e ; CR
0:01821
0:01281
0:01280
0:01283
0:01276
g y ; ; i; etri ; iact;H
0:00167
0:00081
0:00082
0:00083
0:00081
y
tri
act;F
g ; ; i; e ; i
0:00166
0:00076
0:00077
0:00078
0:00077
act;H
act;F
y
5
1
; Financial variables ! i = ! i
= !i
= 12
e) Weights: Macro variables ! g = ! = 12
g y ; ; i; iact;H
0:00075
0:00042
0:00043
0:00043
0:00043
g y ; ; i; iact;F
0:00075
0:00042
0:00043
0:00043
0:00043
In general, results suggest that for the 2003-2011 period, the interaction of monetary and foreign exchange policy (interest rate rules plus foreign exchange intervention) and macroprudential
policy (capital requirements) generated lower volatility of key macroeconomic and …nancial variables than if no macroprudential policy would have been implemented. As shown above, for a
considerable set of macroeconomic and …nancial system variables, we …nd lower volatility associated to the implementation of capital adequacy ratio rules under di¤erent de…nitions. When
measures of aggregate volatility are computed, capital requirements that are modelled as functions of macroeconomic or …nancial system variables (the credit gap, output growth or spread)
generally outperform an "exogenous" formulation of the capital adequacy rations.
What do we make of these …ndings? First and foremost, measures that contain risk in
the …nancial system also have an in‡uence on macroeconomic performance—evidence for the
relevance of macroprudential policy design. Just as the managed ‡oating regime has been found
to be optimal for the Argentine economy in a large scale DSGE model (Escudé, 2009) and
to deliver lower observed variability of macroeconomic variables than alternative regimes in a
fully estimated model (Aguirre and Grosman, 2010), an enhanced policy package that includes
regulation of the …nancial system further contributes to lowering the volatility of certain variables.
Furthermore, there appears to be some "division of labour" among CAR formulations: when
macroprudential regulation is a function of cyclical indicators (the output gap), it appears to be
more helpful to dampen macroeconomic ‡uctuations; but when it is based on …nancial system
indicators or even "exogenous", it may more directly help reduce …nancial system volatility.
Finally, it is worth pointing out that capital requirements that do not depend directly on
macroeconomic or …nancial system variables yield lower volatility than when no such policy is in
place. Rationalizing lower aggregate variability of the exogenous CAR rule is at least twofold. On
the one hand, in an economy with a relatively small …nancial system, where credit barely reaches
15% of GDP by the end of the sample period, there does not appear to be a clear advantage of
putting in place a rule that links capital requirements to some indicator of the state of the real
economy or of the …nancial system at large. We hypothesize that this may have to do with a more
signi…cant in‡uence from the real economy to the …nancial system than otherwise—something
that calls for further work to be properly established. This is also consistent with the model
of Angeloni and Faia (2013), who …nd that banking sector risk is more stable under a "…xed"
capital regime. On the other hand, we cannot rule out that, since the CAR rule actually in place
during the estimation period14 is more similar to that of model 2 (exogenous) than to a function
of macroeconomic or aggregate …nancial system variables, this may imply a generally better …t
to data (in this case, through lower variance) when compared to rules that were actually not in
place. However, a measure of comparative …t like logarithmic data densities suggests that the
model with CAR as a function of credit would be the one of choice (Table 9). Of course, we
may advance further by computing optimal policy and comparing it with what is reported; even
within the limits of a small structural model, this could shed some more light on the interplay
of monetary, foreign exchange and macroprudential policy.
14
Capital ratios in the Argentine …nancial system are a functions of the risk of the di¤erent type of assets held
by …nancial institutions. See BCRA (2014) for details.
29
Table 9
Log data densities of alternative models
Model
Log data density
Baseline
1207:69
Exogenous CAR
1316:30
Endogenous CAR (y)
1318:77
Endogenous CAR (cred)
1324:89
Endogenous CAR (spread)
1301:44
6
Concluding Remarks
Based on our previous work (Aguirre and Blanco, 2013), we estimated a small macroeconomic
model of the Argentine economy, augmented—in its baseline version—to include explicit depiction of the credit market, active rates and interest rate spread; and an enriched description of
monetary policy, with sterilized intervention in the foreign exchange market. In this paper, we
present a somewhat more detailed speci…cation of the …nancial sector, distinguishing credit by
type (commercial or consumption) and making non performing loans endogenous. Compared
to current analyses of the interaction of monetary and macroprudential policy, we provide a
framework that explicitly allows for the interaction of foreign exchange intervention, interest
rate policy and macroprudential policy—something that, to our knowledge, is only dealt with by
Escudé (2014) for the case of capital controls. This feature is particularly relevant in emerging
market economies, where foreign exchange intervention is frequently implemented, but rarely
included in macroeconomic models, and even less in those that extend the framework to include
…nancial stability issues.
Bayesian estimation techniques allow us to incorporate our prior knowledge of the workings of
this economy during the estimation period (2003-2011). Looking at impulse-response functions of
the estimated model, we gain an intuitive understanding of the model’s dynamics—whether they
conform to hypotheses regarding the response of macroeconomic (activity, prices, exchange rates)
and …nancial (money, credit) variables to di¤erent shocks. Higher lending rates are associated
to higher spread, lower credit and output growth; in turn, higher output implies lower interest
rate spread and higher credit. Impacts from the credit market to the rest of the economy
should be further investigated to see whether a hypothesis of “…nancial cycles” (Borio,2012)
may apply during the estimation period. Likewise, the …nancial system (in this highly aggregate
representation) is a¤ected by macroeconomic shocks: in particular, credit behaves in a procyclical
way (in line, for instance, with evidence by Bebczuk et al, 2011). Assessing the impact of changes
in international …nancial conditions is also part of further work to be done.
In Aguirre and Blanco (2013) we looked at forecast performance, showing our estimated
model predicts quarterly output growth, annual interest rates and quarterly foreign exchange rate
depreciation with signi…cantly higher accuracy than: a conventional "three equation plus UIP"
macroeconomic model; and a model with sterilized intervention (but no "…nancial block)—this
was evaluated for 1-, 2- and 4-step out-of-sample forecasts, and using RMSE and MAE forecast
evaluation criteria. We also looked at whether macroprudential policy helped macroeconomic
performance in any meaningful way during the estimation period. Here we advance further in this
kind of evaluation, considering the aggregate volatility of macroeconomic and …nancial system
variables.
Just as previous results show that macroeconomic volatility is reduced when foreign exchange
intervention is implemented in addition to interest rate rules (Escudé, 2009; Aguirre and Gros30
man, 2010), we …nd that capital requirements may a¤ect not only solvency or liquidity conditions,
but also macroeconomic variables at large; over and above their strictly prudential role, they contribute to desirable cyclical macroeconomic property—smoothing output, price, interest rate and
credit volatility over the business cycle. This is found when comparing fully estimated models
with alternative capital adequacy rules during the 2003-2011 period. These results suggest that
the interaction of monetary policy, foreign exchange intervention and prudential tools is, an a
way, synergic; they enhance the …ndings of Agenor and Pereira da Silva (2013), who point out
that for the sake of macroeconomic and …nancial stability, monetary and macroprudential policy
are largely complementary; and illustrate the conclusion of Cecchetti and Kohler (2014), for
whom the linkages between monetary policy and macroprudential tools open the way for the improvement of both macroeconomic and …nancial system performance. Our …ndings extend such
notions in a possible sequence of availability of tools: from interest rates to foreign exchange
intervention and capital requirements, more tools at the disposal of a central bank may help
reduce volatility.
Thus, the discussion may not be so much between interest rate and macroprudential measures
as complements or substitutes; instead, the question is whether counting on a larger set of tools
helps the central bank achieve more desirable outcomes in terms of policymakers’ preferences
or objectives. Here, our …ndings are in line with the literature developed so far, which appears
to point toward a positive answer—quali…ed, of course, by the di¤erent analytical settings and
actual experiences on which each study has been developed. Even within the limitations of
small structural models for simulation exercises, in our assessment results suggest a likely role
for regulation of the …nancial system in dampening macroeconomic ‡uctuations in a developing
economy like Argentina.
31
References
[1] Agenor, P. R. and L. Pereira da Silva (2013): “In‡ation Targeting and Financial Stability:
A Perspective from the Developing World,” Banco Central do Brasil Working Paper 324,
September.
[2] Aguirre, H., T. and E. Blanco (2013): “Credit and Interest Rate Spreads in a Small Macroeconomic Model of the Argentine Economy,” BIS Consultative Council of the Americas’research network on “Incorporating Financial Stability Considerations in Central Bank
Policy Models”.
[3] Aguirre, H., T. Burdisso, F. Grillo and E. Giupponi (2014): “Interest rate spreads in an
emerging economy under di¤erent macroeconomic regimes,” paper presented at the 5th BIS
CCA research conference "Challenges from changing international …nancial conditions",
Bogotá, May.
[4] Aguirre, H. and N. Grosman (2010): “A Note on Managed Floating in a Small Economic
Model,” paper presented at the XLV annual meeting of the Argentine Economic Association,
Buenos Aires.
[5] Angelini, P., S. Neri and F. Panetta (2010): “Monetary and Macroprudential Policies,”
Discussion Paper, Bank of Italy.
[6] Angeloni, I. and E. Faia (2013): “Capital regulation and monetary policy with fragile banks,”
Journal of Monetary Economics, vol. 60(3), pages 311-324.
[7] Banco Central de la República Argentina (2014): Marco normativo. Diciembre de 2014,
December.
[8] Bebczuk R., Burdisso T., Carrera, J. and M. Sangiácomo (2011): “A New Look into Credit
Procyclicality: International Panel Evidence,” BCRA Working Paper 2011j 55, Central
Bank of Argentina.
[9] Bianchi, J., and E. Mendoza (2013): “Optimal Time-consistent Macroprudential Policy,”
NBER Working Paper 19704, December.
[10] Borio, C. (2012): “The …nancial cycle and macroeconomics: What have we learnt?,” BIS
Working Papers No 395, Bank for International Settlements.
[11] Canova, F. (2009): “How Much Structure in Empirical Models?,” Palgrave Handbook of
Applied Econometrics, T. Mills y K. Patterson (eds.), vol. 2, pp. 68-97.
[12] Carvalho, F., M. Castro and S. Costa (2013): “Traditional and Matter-of-fact Financial
Frictions in a DSGE Model for Brazil: the role of macroprudential instruments and monetary
policy,” Banco Central do Brasil Working Paper 336, November.
[13] Cecchetti, S. and M. Kohler (2014): “When Capital Adequacy and Interest Rate Policy Are
Substitutes (And When They Are Not),” International Journal of Central Banking, Vol. 10
No. 3, September, pp. 205-231.
[14] Chang, R (2008): “In‡ation Targeting, Reserves Accumulation, and Exchange Rate Management in Latin America,” Banco de la República de Colombia, Borradores de Economía
no 487.
32
[15] Committe on the Global Financial System (2011): “Practical approaches to designing and
conducting macroprudential policy,” Bank for International Settlements, January.
[16] Covas, F. and S. Fujita (2009): “Time-varying capital requirements in a general equilibrium model of liquidity dependence,” Working Papers 09-23, Federal Reserve Bank of Philadelphia.
[17] Dennis, B., L. Clerc and B. Mojon (2011): “Macroprudential Policy and the Conduct of
Monetary Policy,” Banque de France Occasional Paper.
[18] Drehmann, M. and K. Tsatsaronis (2014): “The credit-to-GDP gap and countercyclical
capital bu¤ers: questions and answers,” BIS Quarterly Review, Bank for International Settlements, March.
[19] Elosegui, P., G. Escudé, L. Garegnani, J. Sotes (2007): “Un Modelo Económico Pequeño
para Argentina,” Serie Estudios BCRA No. 3, Central Bank of Argentina
[20] Ericsson, N. y J. Irons (1995): “The Lucas critique in practice: theory without measurement,” en K. D. Hoover (ed.), Macroeconometrics: Developments, Tensions and Prospects,
Dordrecht: Kluwer Academic Press.
[21] Escudé, G. (2008): “ARGEM: un modelo de equilibrio general dinámico y estocástico para
la Argentina,” Serie Estudios BCRA No. 5, Central Bank of Argentina
[22] Escudé, G. (2009): “ARGEMmy: An Intermediate DSGE Model Calibrated/Estimated for
Argentina: Two Policy Rules are Often Better than One,” BCRA Working Paper Series
2009 42, Central Bank of Argentina.
[23] Escudé, G. (2014): “The possible trinity: Optimal interest rate, exchange rate, and taxes on
capital ‡ows in a DSGE model for a small open economy,” Economics - The Open-Access,
Open-Assessment E-Journal, vol. 8, pages 1-58.
[24] Galati, G. and R. Moessner (2011): “Macroprudential Policy - A Literature Review,” BIS
Working Paper no. 337.
[25] Galí, J., and M. Gertler (1999): “In‡ation dynamics: a structural econometric analysis,”
Journal of Monetary Economics, 44, pp. 195-222.
[26] Galí, J. and T. Monacelli (2005): “Monetary policy and exchange rate volatility in a small
open economy,” Review of Economic Studies, vol 72(3) pp. 707-734
[27] Gerali, A., S. Neri, L. Sessa, and F. Signoretti (2010): “Credit and Banking in a DSGE
Model of the Euro Area,” Temi di Discussione 740, Bank of Italy, Economic Research
Department.
[28] González, A., F. Hamman and D. Rodríguez (2013): “Macroprudential policies in a commodity exporting economy,” paper presented at the …rst plenary meeting of the of the BIS
Consultative Council of the Americas’ research network on "Incorporating Financial Stability Considerations in Central Bank Policy Models”, Mexico, November.
[29] Lim, C., F. Columba, A. Costa, P. Kongsamut, A. Otani, M. Saiyid, T. Wezel, and X. Wu
(2011): “Macroprudential Policy: What Instruments and How to Use Them? Lessons from
Country Experiences,” IMF Working Paper WP/11/38, October.
33
[30] Mandelman, F. (2009): “Business Cycles and Monetary Regimes in Emerging Economies:
A Role for a Monopolistic Banking Sector,” Working Paper 2006-17b, Federal Reserve Bank
of Atlanta.
[31] Markovic, B.(2006): “Bank Capital Channels in the Monetary Transmission Mechanism,”
Bank of England Working Papers 313.
[32] Pincheira, P., and H. Rubio (2010): “The Low Predictive Power of Simple Phillips Curves
in Chile: A Real-Time Evaluation,” Working Paper 559, Central Bank of Chile.
[33] Roger, S. and Vlcek, J. (2011): “Macro…nancial Modeling at Central Banks: Recent Developments and Future Directions,” IMF Working Papers 12/21, International Monetary
Fund.
[34] Sámano Peñaloza, D. (2012): “In the Quest of Macroprudential Policy Tools,” Central Bank
Award Rodrigo Gomez 2011, Center for Latin American Monetary Studies.
[35] Shin, H.S. (2013): “Procyclicality and the Search for Early Warning Indicators,” IMF Working Papers 13/258, International Monetary Fund.
[36] Sierra, J. (2008):“Imperfect …nancial integration, uncovered interest parity and central bank
foreign exchange reserves,” University of Southern California, mimeo.
[37] Szilágyi,K., Baksa, D., Benes, J., Horváth, A., Köber, C. and Soós G. (2013): “The Hungarian Monetary Policy Model,” MNB Working Papers 2013/1, Magyar Nemzeti Bank.
[38] Végh, C. (2014): “Política cambiaria, monetaria y macroprudencial a través del ciclo económico,” Revista de Economía, Vol. 21, No 2, Banco Central del Uruguay, November.
34
Annex 1. Description of variables and data sources
Table A.1
Variable
gy
i
i
m
res
sf
eU S;R
eU S;E
CR
CRH
CRF
iact
iact;H
iact;F
Delinq
CAR
Description
GDP growth seasonally adjusted, 1993 base year series
In‡ation, change in consumer price index and in composite index (wages and wholesale prices)
Domestic passive interest rate - …xed term deposits in AR pesos, 30-59 day maturity
Foreign in‡ation,changes in: average main commercial partners US, Brazil and
Euro-zone CPI
Foreign interest rate - USD Libor, 3 months
Bilateral exchange rate depreciation (US dollar, AR pesos)
Money: currency in circulation in AR peso million as a percentage of GDP
International reserves: in USD millions as a percentage of GDP
Fiscal surplus: revenues minus spending (primary)
nominal exchange rate US dollar, BR real
nominal exchange rate US dollar, euro
Credit: Ratio of non …nancial private sector credit (Households and Firms) to GDP
Household Credit: Pledge lending, Personal loans, Private securities and Accrued
resources on loans in domestic and foreign currency (as a ratio to GDP)
Firm Credit: Overdrafts and Discounts loans in domestic and foreign currency (as a ratio to GDP)
Interest rates on loans granted to the non-…nancial private sector - avg. iact;H and iact;F
Interest rates on Pledge lending, Personal loans, Private securities and Accrued resources on loans
granted to the non-…nancial private sector
Interest rates on Overdrafts and Discounts loans granted to the non-…nancial private sector
Non performing loans as a percentage of non-…nancial private sector credit
Tier 1 capital compliance / Risk weighted assets (…nancial system)
Source
National accounts (INDEC)
INDEC
Central Bank of Argentina
FRED and Bloomberg
Bloomberg
Bloomberg
Central Bank of Argentina
Central Bank of Argentina
Ministry of Economy
Bloomberg
Bloomberg
Central Bank of Argentina
Central Bank of Argentina
Central Bank of Argentina
Central Bank of Argentina
Central Bank of Argentina
Central Bank of Argentina
Central Bank of Argentina
Central Bank of Argentina
Annex 2. Accumulated responses to 1 s.d. shock to the IS curve
Inflation
.010
Output growth
.03
.008
Real short term interest rate
.004
.000
.000
.02
.006
-.002
.004
.01
-.004
-.004
.002
-.008
-.006
.00
.000
-.012
-.008
-.002
-.01
5
10 15 20 25 30 35 40 45 50
Nominal exchange rate change
.002
-.010
5
Money
10 15 20 25 30 35 40 45 50
-.016
5
CB bonds
10 15 20 25 30 35 40 45 50
5
Reserves
10 15 20 25 30 35 40 45 50
Households Credit
.01
.06
.008
.008
.00
.04
.004
.006
-.01
.02
.000
-.02
.00
-.004
-.03
-.02
-.008
.004
.002
36
-.04
-.04
5
10 15 20 25 30 35 40 45 50
.000
-.002
-.012
5
Firm Credit
10 15 20 25 30 35 40 45 50
-.004
5
Households Lending rate
.008
.002
.006
.001
10 15 20 25 30 35 40 45 50
Firms Lending rate
.002
-.001
-.001
-.002
-.002
-.002
-.004
.000
-.003
-.003
-.004
-.002
5
10 15 20 25 30 35 40 45 50
Firms Interest rate spread
.001
.000
-.001
-.002
-.003
-.004
5
10 15 20 25 30 35 40 45 50
Households Interest rate spread
.000
.000
.002
10 15 20 25 30 35 40 45 50
.001
.000
.004
5
-.006
5
10 15 20 25 30 35 40 45 50
-.004
5
10 15 20 25 30 35 40 45 50
5
10 15 20 25 30 35 40 45 50
Annex 3. Parameter estimates of alternative models
No CAR
1
3
4
1
2
3
4
5
6
1
2
3
4
5
6
1
2
3
4
5
!1
!2
!3
1
2
3
4
1
2
AH
1
AH
2
AH
3
B1H
B2H
B3H
B4H
DH
1
DH
2
AF1
AF2
AF3
0:264
0:078
0:065
0:526
0:397
0:136
0:109
0:113
0:123
0:937
0:741
0:320
0:972
0:711
0:658
0:573
0:021
0:025
0:083
0:007
5:911
0:008
0:178
1:203
0:553
0:031
0:665
0:982
0:138
0:401
0:066
0:379
0:069
0:169
0:228
0:810
0:472
0:333
0:110
0:410
Table A.2. Models 2-5: Posterior means
Exogenous Endogenous CAR Endogenous CAR
CAR
output gap
credit gap
0:215
0:184
0:206
0:032
0:035
0:009
0:141
0:071
0:073
0:323
0:445
0:356
0:459
0:383
0:493
0:217
0:156
0:172
0:158
0:145
0:265
0:166
0:128
0:110
0:260
0:311
0:268
0:962
0:962
0:968
0:709
0:813
0:516
0:364
0:307
0:315
0:962
0:968
0:984
0:905
0:912
0:928
0:220
0:363
0:253
0:626
0:616
0:898
0:013
0:009
0:039
0:024
0:030
0:019
0:077
0:088
0:130
0:005
0:008
0:003
5:595
5:948
6:196
0:010
0:010
0:007
0:240
0:100
0:148
0:952
0:988
1:002
0:692
0:621
0:765
0:027
0:028
0:029
0:738
0:744
0:478
0:976
0:978
0:987
0:128
0:065
0:106
0:377
0:437
0:335
0:098
0:058
0:114
0:414
0:464
0:446
0:099
0:093
0:103
0:254
0:201
0:187
0:239
0:127
0:243
0:145
0:230
0:153
0:819
0:783
0:797
0:374
0:304
0:339
0:385
0:366
0:388
0:099
0:059
0:211
0:459
0:553
0:377
37
Endogenous CAR
credit spread
0:243
0:063
0:075
0:353
0:346
0:150
0:164
0:206
0:374
0:960
0:370
0:223
0:971
0:851
0:166
0:556
0:033
0:028
0:111
0:008
5:698
0:010
0:098
0:960
0:618
0:024
0:662
0:973
0:071
0:399
0:117
0:455
0:079
0:212
0:168
0:144
0:812
0:396
0:325
0:029
0:440
No CAR
B1F
B2F
B3F
B4F
0:018
0:230
0:215
DF
1
DF
2
0:912
0:455
0
1
2
Table (Cont.) Models 2-5: Posterior means
Exogenous Endogenous CAR Endogenous CAR
CAR
output gap
credit gap
0:023
0:028
0:031
0:244
0:279
0:225
0:261
0:162
0:273
0:134
0:281
0:317
0:907
0:898
0:890
0:473
0:316
0:459
0:011
0:010
0:015
0:378
0:704
0:587
0:155
0:025
3
38
Endogenous CAR
credit spread
0:023
0:220
0:272
0:186
0:912
0:416
0:011
0:295
0:154
0:753
Annex 4. Evaluation of alternative macroprudential rules using estimated models
We present an alternative way of assessing macroprudential policy implemented as di¤erent
capital adequacy rules. We estimate each model, from 2 to 5, completely, instead of just using
the estimated coe¢cients of model 2 and changing the parameters of equation (31), as shown
in section 5.1. Estimated coe¢cients of models 2-5 are shown in Table A.2. We compute
standard deviations of macroeconomic and …nancial variables under each estimated model, plus
model 1. We do the exercise for: in‡ation, output growth, local short term interest rates,
the real trilateral (trade-weighted) exchange rate, money growth, international reserves, credit
(total and by line), lending interest rates (average and by credit line), non performing loans
(by credit line) and capital requirements. The comparison in Table A3 suggests the lowest
volatility during the estimation period under an endogenous capital requirement (output gap,
model 3) for the following variables: international reserves, average, consumption and commercial
lending interest rates, and consumption non-performing loans. In turn, capital requirements as
a function of interest rate spreads (model 5) deliver lower growth, deposit interest rate, money
growth and commercial non-performing loans than alternative policies. When capital adequacy
is implemented based on the credit-to-GDP gap (model 4), it shows the lowest variability for
in‡ation, real exchange rate depreciation and capital requirements. An "exogenous" CAR (model
2) delivers the lowest standard deviations of average and commercial credit. Finally, using no
capital requirements but monetary and foreign exchange policy (model 1) is associated to the
lowest variability of consumption credit.
In addition, as in section 5.1, we also compute ad-hoc loss functions, aggregating volatility
of selected variables associated to each type of macroprudential policy (table A.4). When all
variables are given equal weight (panel a), CAR as a function of interest rate spread is linked to
the lowest volatility when output growth, in‡ation and interest rates are included; if, however,
we consider a larger set of variables in the loss functions, (with the exchange rate and credit
market variables), then an "exogenous" CAR formulation yields lower volatility. The latter
rule also delivers lower volatility than the alternatives when both macroeconomic (including the
exchange rate) and …nancial system variables are considered; this applies to both higher weights
on macroeconomic variables (panel b) and on …nancial variables (panel c). However, when only
growth, in‡ation and interest rates are arguments of the loss function, we …nd that the lowest
volatility is obtained in three of four speci…cations under a CAR that depends on interes rate
spread.
Compared to the results of section 5.1, aggregate volatility here is more frequently associated
to an "exogenous" rule; this seem, to some extent, driven by the inclusion of the the exchange
rate and of credit. That a rule that is not a function of cyclical conditions yields a better result
may have to do with the fact that it was such type of rule that was in place during the estimation
period; or with the relatively low impact of shocks from the …nancial system to the macroeconomy
than otherwise (see the discussion in section 5.1). In any case, this alternative evaluation suggests
the robustness of the …nding that implementation of some type of macroprudential policy is
bene…cial in terms of macroeconomic volatility reduction.
39
40
i
gy
etri
m
res
CR
CRH
CRF
iact
iact;H
iact;F
Delinq H
Delinq F
CAR
Table A.3 : Observed and estimated standard deviations of selected variables
Model 1
Model 2
Model 3
Model 4
Model 5
Baseline Exogenous CAR Endogenous CAR (y) Endogenous CAR (cred) Endogenous CAR (spread)
0:058
0:031
0:033
0:029
0:037
0:013
0:012
0:013
0:020
0:011
0:057
0:047
0:059
0:061
0:042
0:096
0:057
0:073
0:045
0:073
0:220
0:184
0:193
0:151
0:134
0:550
0:507
0:503
0:661
0:608
0:262
0:239
0:252
0:277
0:252
0:128
0:137
0:144
0:141
0:137
0:224
0:191
0:196
0:232
0:208
0:017
0:016
0:013
0:020
0:017
0:018
0:019
0:014
0:021
0:018
0:018
0:015
0:015
0:021
0:018
0:115
0:076
0:069
0:082
0:069
0:197
0:157
0:129
0:176
0:128
0:015
0:032
0:013
0:035
41
Table A.4: Loss Functions of alternative models
Variables
Model 1
Model 2
Model 3
Model 4
Model 5
Considered
Baseline Exogenous CAR Endogenous CAR Endogenous CAR Endogenous CAR
in Loss Function
(y)
(cred)
(spread)
1
a) Equal weights ! = n
gy ;
0:0066
0:0032
0:0045
0:0046
0:0031
g y ; iact
0:0036
0:0025
0:0036
0:0042
0:0020
y
act
g ; ;i
0:0069
0:0035
0:0047
0:0050
0:0034
g y ; ; iact ; etri
0:0161
0:0068
0:0100
0:0071
0:0087
y
act
g ; ; i; i
0:0071
0:0036
0:0049
0:0055
0:0035
g y ; ; i; CAR
0:0068
0:0036
0:0057
0:0052
0:0044
g y ; ; i; etri ; CRH
0:0324
0:0254
0:0308
0:0270
0:0273
y
tri
F
g ; ; i; e ; CR
0:0659
0:0430
0:0483
0:0609
0:0518
g y ; ; i; etri ; CRF ; CAR
0:0432
0:0493
0:0611
0:0531
y
act
tri
H
g ; ; i ; e ; CR
0:0325
0:0256
0:0308
0:0270
0:0276
g y ; ; iact ; etri ; CRF
0:0661
0:0431
0:0483
0:0609
0:0520
act;H
act;F
y
tri
1
4
i
i
i
g
e
=!
= 15
b) Weights: Macro variables ! = ! = !
= 15 ; Financial variables ! = !
y
tri
H
g ; ; i; e ; CR
0:0053
0:0030
0:0040
0:0031
0:0035
y
tri
F
g ; ; i; e ; CR
0:0076
0:0042
0:0052
0:0054
0:0051
g y ; ; i; etri ; iact;H
0:0043
0:0018
0:0026
0:0018
0:0023
g y ; ; i; etri ; iact;F
0:0042
0:0017
0:0026
0:0018
0:0023
act;H
act;F
y
1
5
i
i
i
g
=!
= 12
c) Weights: Macro variables ! = ! = 12 ; Financial variables ! = !
y
act;H
g ; ; ; i; i
0:0028
0:0014
0:0019
0:0020
0:0013
g y ; ; i; iact;F
0:0028
0:0014
0:0019
0:0020
0:0013
tri
act;H
act;F
y
1
5
2
and ! e = 15
; Financial variables ! i = ! i
= !i
= 15
d) Weights: Macro variables ! g = ! = 15
y
tri
H
g ; ; i; e ; CR
0:0070
0:0070
0:0080
0:0075
0:0071
y
tri
F
g ; ; i; e ; CR
0:0182
0:0128
0:0138
0:0188
0:0153
g y ; ; i; etri ; iact;H
0:0017
0:0008
0:0011
0:0010
0:0009
g y ; ; i; etri ; iact;F
0:0017
0:0008
0:0011
0:0010
0:0009
act;H
act;F
y
5
1
; Financial variables ! i = ! i
= !i
= 12
e) Weights: Macro variables ! g = ! = 12
g y ; ; i; iact;H
0:0008
0:0005
0:0005
0:0007
0:0004
g y ; ; i; iact;F
0:0008
0:0004
0:0005
0:0007
0:0004