Download The Output Composition Puzzle - compositional response of GDP to

Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
The Output Composition Puzzle - compositional response of GDP to
Australian monetary policy.
By Muheed Jamaldeen*
Abstract
The ‘output composition puzzle’ refers to economy-specific responses of components of GDP to domestic
monetary policy. This paper analyses the compositional response of Australian output to domestic monetary
policy using Structural Vector Auto Regression (SVAR) models estimated over the period 1986 to 2013 and
makes a distinction between the sensitivity of each GDP component, and its contribution to GDP deviation. This
study finds that investment is most sensitive (followed by net exports) to unanticipated increases in the cash
rate, and also contributes the most to output deviation. However, this study also finds that net exports are a
key driver of output reduction in the first year, and then play a role in dampening the investment contraction in
subsequent years. This suggests that the exchange rate channel is particularly important for the transmission of
Monetary Policy in Australia. In addition, this study finds that within investment, new dwelling investment is the
most sensitive to cash rate shocks. However, the rest of investment contributes more to contractions in output.
In contrast, durable consumption appears to be slightly more sensitive than its non-durable counterpart in the
first year. Interestingly, this study does not show evidence of a statistically significant ‘price puzzle’ or other
puzzles encountered in other SVAR studies.
Keywords: monetary policy, output composition, Structural Vector Auto Regression.
JEL classification: E52, E27, E20
*The majority of this research was conducted as part my completed Honours degree thesis at Macquarie University, NSW,
under the supervision of Prof. Jeffery Sheen, and Dr. Chris Heaton. I would like to thank my supervisors, as well as Dr. Natalia
Ponomoreva, and Assoc. Prof. Tony Bryant for their helpful comments and advice. Any errors or omissions remain my own.
Email: [email protected] or [email protected]
P. 0
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
1
Introductio n
Australian monetary policy has evolved considerably over the last few decades. The introduction of inflation
targeting in 1993, accompanied by central banking independence has brought monetary policy to the forefront
of economic management. Early studies such as Friedman and Schwartz (1963) postulated that monetary
policy changes drive changes in output. These views were reaffirmed by more recent studies such as Romer
and Romer (1999) and Christiano, Eichenbaum and Evans (1994). As a result of these research efforts and
many others, there is general consensus amongst economists that in the short-run monetary policy can
significantly influence the real economy (Bernanke and Gertler (1995)). With a few notable exceptions 1, there
has been limited research into the monetary policy impact on disaggregated GDP. Focusing only on aggregate
GDP would leave out important information regarding the differential dynamic responses of GDP components to
changes in monetary policy.
Given the ability of monetary policy to influence output in the short run, it is vital to understand how and why
output behaves in this way. Studies that explicitly explore the monetary transmission mechanism answer the
question ‘why’. This paper attempts to answer the question ‘how’ output responds the way it does by exploring
the differences in the responses of its key components. Answering the question ‘how’ would provide the
monetary authority with information about the sectors of the economy and the key agent behaviour that
deserves the central bank’s particular attention.
Angeloni et al (2003) examine differences between the US and the Euro Zone in terms of their compositional
GDP responses to monetary policy. They refer to the economy-specific response of the components of GDP as
‘the output composition puzzle’. The response is deemed a ‘puzzle’ because even though the conventional
understanding is that investment responds to interest rate changes (thereby contracting output); the authors
find that it is in fact consumption in the US, and investment in the Euro Zone that contributes the most to GDP
deviation. This paper adds to the literature on the output composition puzzle, by examining the response of
disaggregated GDP to Australian monetary policy using a Structural Vector Auto Regression (SVAR) model of
the economy. The nature of the disaggregated GDP response to monetary policy may provide policy makers
greater insight into structural properties of the economy (such as institutional arrangements) that influence the
GDP response.
1
International studies such as Bernanke and Gertler (1995), Erceg and Levin (2002), Angeloni et al (2003), Fujiwara
(2003), and Raddatz and Rigobon (2004) use disaggregated GDP.
P. 1
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
2
Literature re view
Since Sims (1980) introduced Vector Auto Regression, it has become a popular econometric method used to
model interactions in the macro-economy. However, it is not without its critics. Initially, identification of VAR
models used a recursive causal ordering of the variables. Cooley and Le Roy (1985) criticised this inherently
arbitrary atheoretical approach because it depended on the ordering of the variables. This could lead to findings
of reverse causation between some of the variables according to the ordering employed. In response to this
criticism, Bernanke (1986) and Sims (1986) introduced Structural VARs (SVARs) which utilised non-recursive
identification of the structural matrices instead.
Studies such as Bernanke and Gertler (1995), Radatz and Rigobon (2003), and Erceg and Levin (2002) utilise
disaggregated GDP in SVAR models of the US economy. However, none of these studies analyse the ‘output
composition puzzle’ per se. These papers focus on specific elements of GDP, in relation to specific research
questions.
The term ‘output composition puzzle’ was first coined by Angeloni et al (2003). The puzzle refers to the
difference in the compositional response of GDP to domestic monetary policy in different countries. They
conduct a comparative study of the US and the Euro Zone area, in which they compare the disaggregated GDP
response (to domestic monetary policy shocks) of the two areas, to illuminate differences in consumer and
investor
behaviour
between
the
two
regions.
They
use
three
methodologies:
SVAR
models,
large
macroeconomic structural equation models, and Dynamic Stochastic General Equilibrium (DSGE) models. In
addition to utilising existing large scale structural equation models for the US and, Euro Zone; the authors
construct country-level SVARs for the Euro Zone, and area SVARs for the US, and Euro Zone.
They find that large scale structural equation models for the US and Euro zone produce similar results to the
SVAR models. Consumption contributes most to output loss in the US, while investment plays this role in the
Euro Zone. From an empirical point of view, they find these two methodologies are sufficient to observe which
component is dominant. However, both methodologies are insufficient to ascertain the reason for this difference
between the US and the Euro Zone. For this they employ DSGE models to replicate the empirical results with an
attempt to trace back the structural features of each economy that cause this difference. A DSGE model is not
employed in this paper because it is not a comparative study with another economy and does not seek to
explain the ‘output composition puzzle’. Instead, this paper attempts to empirically characterise the
compositional response of Australian GDP to domestic monetary policy. The 2003 study concludes that the
DSGE model has difficulty accounting for the puzzle. The authors conclude by making a tentative assessment
P. 2
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
that consumers are responsible for the differences. Unfortunately they do not have a compelling explanation for
why this is the case. They posit that disposable income may be less responsive to monetary changes in the
Euro Zone than in the US, and argue that is likely to be due to the fact that the social safety net in Europe is
stronger. Though not reported, Angeloni et al (2003)
2
also construct a model for the UK, and find that
consumption contributes the most to GDP contractions in that economy.
In line with the Angeloni et al (2003) paper, Fujiwara (2003) examines the output compositional response to
monetary policy. The author constructs four VAR models based on Gordon and Leeper (1994), Christiano,
Eichenbaum and Evans (2001), Erceg and Levin (2002) and Leeper, Sims and Zha (1996). The study finds that
unlike the US and UK, investment makes the biggest contribution to change in Japanese GDP.
3
General specifications
Since VAR and SVAR were introduced in the 1980s, the technique has been widely applied in the literature. For
that reason, a technical discussion on the methodology is avoided and readers unfamiliar with the technique are
encouraged to refer to Hamilton (1994) for a rigorous exposition.
The variables3 considered in this study can be classified as follows:

External sector variables: US real GDP, Commodity prices

Domestic variables: trimmed mean inflation, cash rate, M1 monetary aggregate, real trade weighted index

Disaggregated components of domestic GDP.
Exte rna l se ctor va ri ab les
US Gross Domestic Product (US GDP)
Studies such as Gruen and Shuetrim (1994), de Roos and Russel (1996) and Beechey et al (2000) find that US
GDP has a strong relationship with Australian economic activity. Given its size and close relationship to the
Australian economy, shocks to US GDP can also be interpreted as shocks to demand for Australian exports over
the sample period being estimated.
2
3
See p 1286 of Angeloni et al (2003).
All variables, except inflation, and cash rate are in logs. The models are estimated in levels.
P. 3
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
Commodity Prices (Comp)
The inclusion of commodity prices in a stylised model is particularly important for the Australian economy
because commodities make up the majority of Australian export income, and have a significant impact on the
economy. Furthermore, commodity prices are also thought to account for the forward-looking nature of the
monetary authority. Sims (1992) posits that commodity prices act as an information signal to the central bank
regarding future inflation. This property is thought to resolve the ‘price puzzle’ often encountered in the SVAR
literature.
Dom esti c se cto r varia bl es
Trimmed mean inflation rate (Infl)
Including a measure of price is standard in studies that investigate economic activity. Following Dungey and
Pagan (2000), this paper includes the inflation rate instead of the price level. However, instead of CPI inflation,
this study includes – trimmed mean inflation; a key measure of underlying inflation. This is because the RBA in
its 1996 statement for the conduct of monetary policy (and numerous bulletins since) highlights the importance
of underlying measures of inflation. In addition, this series has the advantage of being less volatile compared to
other alternatives.
Monetary aggregate (M1)
This paper follows in the footsteps of studies such as Huh (1999), Cushman and Zha (1997) and Kim and
Roubini (2000) by including a narrowly defined monetary aggregate – M1. This variable is included in the model
for two reasons. Firstly, even though a monetary aggregate does not reflect the stance of monetary policy since
the advent of inflation targeting, it did have a weak role prior to that. Secondly, and crucially, it is included with
the intention of capturing the monetary interactions within the Australian economy.
Interest rate (Int)
The quarterly average of the interbank cash rate is used in this study to represent the monetary policy
instrument. Grenville (1997) and Dotsey (1987) contend that the cash rate is the chief instrument of monetary
policy since the floating of the AUD in December 1983.
Real trade weighted index (Rtwi)
Given that Australia is a small open economy and given its export profile, the real exchange rate is an
important indicator of Australian economic conditions. Studies such as Brischetto and Voss (1999) and Suzuki
(2004) use the US dollar bilateral nominal exchange rate for this purpose. The real trade weighted index is
more appropriate because it takes into account Australia’s broad relationship with its majo r trading partners
P. 4
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
and not just the US. Hence it is more reflective of Australia’s integration in the global economy. Furthermore,
since the real exchange rate incorporates the foreign price level as well as the nominal exchange rate, it is
more reflective of Australia’s international competitiveness 4. The RBA is likely to make some assessment of
Australia’s international competitiveness in its monetary policy decisions. This is best captured by the real
exchange rate5. For these reasons, this study utilises the real trade weighted index in line with Dungey and
Pagan (2000).
Mod el 1: Disa gg re ga ted G D P com ponents
The disaggregation of real GDP in Model 1 follows the standard accounting identity decomposition of five
components. All real GDP components enter the model in logs.

total household consumption (C)

private and public investment (Invst)

government consumption expenditure (G)

exports (Ex); and

imports (Im).
Angeloni et al (2003) model only consumption and investment. They include government expenditure and net
exports into a residual term defined ‘rest of GDP’. They argue that this is a parsimonious shortcut to implicitly
impose the constraint of the accounting identity. Since Australia is a small open economy and because net
exports play an important role, this paper splits ‘rest of GDP’ into its components, in order to capture the
potential impact that monetary policy may have on net exports. This also has the added advantage of
familiarity with the standard textbook decomposition of GDP.
Mod el 2: Disa gg re ga ted G D P com ponents
While Models 1 will be based on the accounting identity, Model 2 will delve deeper into the compositional
response of household consumption and investment. Following Erceg and Levin (2002), and Bernanke and
Gertler (1995), Household consumption is disaggregated into durable consumption (Durc) and non-durable
consumption (Non-Durc), and all sector investment is disaggregated into new dwelling investment (House) and
rest of investment (Rest-Invest). Non-durable consumption (Non-Durc) is total household final consumption
4
The nominal trade weighted index was considered, but not chosen because it does not have information on relative
prices, which are key to understanding the response of net exports to changes in monetary policy via the exchange rate
channel.
5
Note that throughout this study the real exchange rate and real trade weighted index will be used interchangeably.
P. 5
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
expenditure minus clothing, footwear, furnishings and purchase of vehicles 6. Rest of investment (Rest-Invest) is
all sector investment minus new dwelling investment. Hence, it includes the remainder of private and public
investment. In order to retain the accounting identity, government consumption expenditure, exports, and
imports are also included in the model. Again, all real GDP components enter the model in logs. The
consumption decomposition intends to explore whether goods that have a longer life (durables) respond
differently to monetary policy. The investment disaggregation attempts to explore the extent and timing of the
new dwelling investment response.
4
Model 1
Model 1 is an 11-variable VAR model estimated in levels over the period 1986:03 to 2012:03. This sample
period was chosen due to availability of trimmed mean inflation data, and to exclude data volatility subsequent
to the floating of the Australian dollar. In addition to consumption, investment, government expenditure,
exports, and imports; all external and, domestic sector variables are included. The model is large given the
sample size. However, this enables the modeller to minimise Omitted Variable Bias (OVB) 7 to which small
SVARs are often subject. Thus, the large dimensions of the model intend to capture the important interactions
within the Australian economy in an attempt to avoid OVB.
4. 1
M o d e l s p ec i f i c s
A constant term is included to account for the mean of the series in the VAR 8. An impulse dummies for 2002:04
is included in the M1 equation to account for an outlier in the series.
Lag length tests suggest an order 1 or 5 lags 9. Since the data is quarterly, to ensure that the model is not overparameterised (while maintaining reasonable degrees of freedom), a lag order of four is selected after
concurrently
considering
lag
length
criterion,
autocorrelation
tests and
stability
of
the
model.
The
autocorrelation tests were conducted on the reduced form VAR of four lags10. Most equations do not show
evidence of first or fourth order autocorrelation.
6
This follows the definition in studies such as Fisher and Voss (2004) and Tan and Voss (2000)
7
See Christiano et al (1998) and Sims (1992) for a discussion on the susceptibility of small VARs to OVB.
8
See Hamilton (1994)
9
Akaike’s Information Criteria (AIC) suggests five lags, while Schwarz (or Bayesian) Information Criterion (SC) and the
Hannan-Quin Criterion (HQ) suggest one when estimated with a maximum of five lags.
10
M1 (weak first order), and Import (fourth order) equations exhibit some autocorrelation. However, increasing the lag
order of the VAR is not feasible given the number of observations and parameters. Moreover, the five lag VAR does no t
reduce autocorrelation, and the impulse responses are noisy and unusable. Autocorrelation test results are available on
request and are not reported for the sake of brevity.
P. 6
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
St ab il it y and st at iona ri ty
Unit Root tests suggest that most variables are non-stationary I(1) processes
11
. This paper follows the work of
the majority of SVAR studies and estimates the reduced form VAR in levels. Even though the VECM approach
would improve the efficiency of the estimates, it was avoided for four reasons. Firstly, in line with Sims (1980)
and Sims, Stock and Watson (1990), estimation in levels still provides consistent estimates and avoids
discarding relevant information about possible long-run relationships between the variables in the model.
Secondly, it is the most common approach in the SVAR literature and hence has the advantage of comparability
of identifying restrictions and results (Brischetto and Voss, 1999). Thirdly, estimation in levels is adopted to not
enforce possibly incorrect restrictions on the model (Hamilton, 1994). Fourthly, even if some of the variables
are integrated, stability of the VAR12 would imply system stationarity thus making the impulse responses valid.
The stability of the VAR process is assessed to ascertain the reliability of the estimates from the model 13. A VAR
process is considered stable if its reverse characteristic polynomial has no roots in or on the complex unit circle;
and importantly, stability implies a stationary VAR (Lutkephol, 2006).
4. 2
I d e n t i f i c at i o n
To recover the structural parameters (A and B matrices) from the reduced form VAR (and estimate an SVAR),
restrictions need to be imposed on the matrices describing the contemporaneous relationships, thereby
overcoming the identification problem. This paper adopts a non-recursive identification structure as there are
clearly identifiable relationships between the system variables.
Where K is the number of variables; K (K-1) /2 restrictions need to be imposed (on the A matrix) for justidentification. In this case, that amounts to 55 restrictions in addition to those imposed on the B matrix. On
theoretical grounds, 77 restrictions are imposed on the contemporaneous relationships between the variables.
The A matrix describes the contemporaneous relationships between the 11 variables in the Y vector of
dependent variables. Some domestic sector equations are discussed in detail below. The 0s restrict the
corresponding system variable from having a contemporaneous impact on that equation. The B matrix, which
describes the contemporaneous relationships between the structural shocks (Bεt), is set to a diagonal of freely
estimated parameters with no contemporaneous relationships.
11
Unit root tests are available on request. Not reported for the sake of brevity.
12
A VAR (p) process is stable if its reverse characteristic polynomial has no roots in or on the complex unit circle. This
implies stationarity. See Lutkepohl (2006).
13
See Appendix for modulus of Eigen values of the reverse characteristic polynomial.
P. 7
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
External sector
Both equations from the external sector
0
0
0
0
0
0
0
0
0
0  USgdpt 
1
0
0
0
0
0
0
0
0
0
31
a
32
1
0
0
0
0
0
0
0
0
41
a
42
0
1
0
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
61
a
62
0
0
0
1
0
0
0
0
0
71
a
72
0
0
0
0
1
0
0
0
0
a
82
a
83
a
84
a
85
a
86
a
87
1
0
0
0
0
a
93
a
94
a
95
a
96
a
97
a
1
a
0
a
0
0
0
0
0
0
a
1
a
a
a
a
a
a
a
 1
a

a

a
 0

AYt   a
a

 0
 0

 0
a

have no domestic variables affecting
21
them. Such a lag structure has become
a standard assumption in the smallopen-economy SVAR literature and is
referred to as “block exogeneity”. This
reflects Australia’s small open economy
status, which implies that it cannot
affect external sector variables. US GDP
contemporaneously impacts Commodity
111
102
a
112
113
114
115
116
117
98
118
109
119
a
910
1110
0
1
 Compt 


 Ct 


 Invstt 
 Gt 


 Ext 
 Im 
t


 Inflytmt 
 M 1t 


 Intt 
 Rtwit 


Prices to stylise the demand side shocks to commodity prices. Commodity prices are assumed to have an
instantaneous impact on all domestic equations (except the monetary aggregate), due to their expeditious
impact on the Australian economy.
GDP components
There are two key assumptions for these equations. Firstly, with the exception of government consumption
expenditure14, both external variables are allowed to contemporaneously affect Australian GDP. It is therefore
assumed that global economic conditions contemporaneously impact the Australian economy. This approach is
common in most Australian SVAR studies, due to the small and open nature of Australia’s economy. Secondly,
following Raddatz and Rigobon (2003), it is assumed that shocks to individual components of GDP take at least
one quarter to affect the other components.
Domestic sector
4
Inflytm t = a 82 Comp +a 83 C +a 84 Invst +a 85 G +a 86 Ex +a 87 Im +
t
t
t
t
t
t
 Γ Y +b
8
t-j
εt
Inflytm
88
(1.1)
j=1
It is assumed that inflation is contemporaneously impacted by commodity prices and Australian GDP only.
Commodity prices transmit into the economy rapidly given their dominance in Australia’s export profile. For
instance, an increase in commodity prices is likely to impact on Australian firms’ marginal cost (just as they
affect foreign firms’). Of course, they also do impact income, demand, and thereby output, but we cannot be
sure which effect dominates.
14
Since the government faces information lags and because spending decisions are made ahead of time it is assumed
that external sector variables do not contemporaneously impact Australian government’s spending decisions.
P. 8
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
It is unlikely that an increase in U.S. GDP will contemporaneously impact domestic inflation. However, it is
reasonable to assume that Australian GDP would contemporaneously impact on inflation. This assumption is
common in both international (see Bernanke and Blinder 1992, Kim and Roubini 2000) and domestic SVAR
studies (see Brischetto and Voss 1999; Dungey and Pagan 2000; Dungey and Fry 2003; Berkelmans 2005).
4
M1t =a 93 C +a 94 Invst +a 95 G +a 96 Ex +a 97 Im +a 98 Inflytm +a 910 Int
t
t
t
t
t
t
t
 Γ Y +b
9
t-j
εt
M1
99
(1.2)
j=1
Although excluded from some studies, a monetary aggregate is included in this study to capture the monetary
interactions within the economy. This is in line with SVAR studies such as Brischetto and Voss (1999), Kim and
Roubini (2000) and Cushman and Zha (1997), which allow GDP, inflation and the interest rate to have a
contemporaneous impact on the monetary aggregate, thus specifying the equation as a standard money
demand equation.
4
Int t =a 102 Comp +a 109 M1 +
t
t
Γ
10
Yt-j +b1010 ε t
Int
(1.3)
j=1
The interest rate equation is interpreted as the policy reaction function of the RBA and it is assumed that the
cash rate is representative of the monetary policy stance of the RBA. However, it is important to note that this
specification is limited, in that the monetary authority is assumed to not respond to any other variables other
than those specified. This is one of the shortcomings of the SVAR framework as it is likely that the RBA has a
whole range of variables in its information set that it uses to make monetary policy decisions.
Modern monetary policy is complex and has numerous components to it. Hoover and Jorda (2001) succinctly
summarise these components and their interactions as shown below.
T abl e 1.
C o m p o n e n t s of m o n e t a r y p o l i c y
Policy maker
Public
Systematic
Anticipated
Known policy reaction function
Unanticipated
Surprise shift to a new known policy
reaction function
Unsystematic
Credible announcement of a transitory,
atypical setting of a policy instrument
Random shock to policy reaction function
Source: Hoover and Jorda (2001). Emphasis added.
In the context of the SVAR framework, the monetary policy shock only captures the unanticipated and
unsystematic (exogenous) component of monetary policy. It is therefore important to note that the conclusions
derived from SVAR studies largely relate to unanticipated unsystematic monetary policy shocks.
P. 9
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
This paper adopts information timing restrictions for the cash rate equation, and assumes that the monetary
authority is allowed to contemporaneously react to as much information available at the time (M1 and
commodity prices). Hence, any information not available at the time of the decision will be excluded from the
contemporaneous relationships and only be available in lags. Although exchange rate information would be
contemporaneously available, inflation information is not, and therefore real exchange data would not be
available. As Dungey and Pagan (2000) argue, even though the RBA would pay attention to Australia’s
international competitiveness (as captured by the real trade weighted index), it is unlikely that the RBA would
react to contemporaneous movements in such a variable and the relevant norm may well be some average of
the past values. For this reason, the real exchange rate enters the equation in lag only.
Moreover, in line with Sims (1992) it is assumed that commodity prices represent the inflationary expectations
and forward-looking behaviour of the monetary authority.
 G Y +b
4
Ex
Rtwi =a USgdp +a Comp +a C +a Invst +a G +a
t
111
t
112
t
113
t
114
t
115
t
116
Im
+a
t
117
+a Inflytm +a M1 +a Int +
t
118
t
119
t
1110
Rtwi
e
t-j
11
t
j=1
1111
t
(1.4)
Finally, the real exchange rate is assumed to respond contemporaneously (as well as with lags) to all the
domestic and foreign variables, reflecting the precision and speed with which information feeds into the
exchange rate market. This is a standard assumption in most SVAR studies, both domestic and international.
R es u l ts
Given the identification of the structural matrices, the estimated coefficients and standard errors of the A matrix
of the SVAR are reported below. Broadly speaking, the signs of most coefficients appear to be consistent with
economic theory, and the standard errors imply that most estimates are significant. The non-recursive
identification adopted for Model 1 is 22 more restrictions than required for just-identification. As is often the
case with large SVAR models with restrictions based on economic theory, the LR test 15 for over-identification
suggests that, in comparison to a just-identified model, the restrictions are not vastly supported by the data.
However, as shown in Section 6, this does not affect the overall conclusion of this paper as the impulse
responses are largely robust to alternative identification schemes (such as recursive identification). The nonrecursive identification is adopted since it is not arbitrary, not affected by the ordering of the variables, and
produces more sensible impulse responses.
15
SVAR is over-identified with 21 degrees of freedom. LR Test: Chi 2( 21): 71.19, Prob:0.00
P. 10
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
T abl e 2.
US GDP
1.00
1.69
-0.21
-2.16
0.00
-1.28
-1.85
0.00
0.00
0.00
-0.36
A m a t r i x es t i m a t e s a n d s t a n d a r d e r r o r s
Comp
0.00
1.00
0.01
0.07
0.00
-0.03
0.02
-1.74
0.00
-0.47
0.21
C
0.00
0.00
1.00
0.00
0.00
0.00
0.00
5.01
-1.56
0.00
-0.65
I
0.00
0.00
0.00
1.00
0.00
0.00
0.00
-1.40
0.01
0.00
-0.04
A matrix parameter estimates
G
Ex
Im
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
1.00
0.00
0.00
0.00
1.00
0.00
0.00
0.00
1.00
3.66
1.81
-3.16
1.10
0.31
-1.73
0.00
0.00
0.00
0.51
0.24
-0.76
A matrix standard errors
US GDP
Comp
C
I
G
Ex
Im
0.00
0.00
0.00
0.00
0.00
0.00
0.00
1.11
0.00
0.00
0.00
0.00
0.00
0.00
0.07
0.01
0.00
0.00
0.00
0.00
0.00
0.33
0.03
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.35
0.03
0.00
0.00
0.00
0.00
0.00
0.25
0.02
0.00
0.00
0.00
0.00
0.00
0.00
0.68
9.52
1.99
4.48
2.03
2.49
0.00
0.00
1.17
0.23
0.59
0.25
0.51
0.00
2.14
0.00
0.00
0.00
0.00
0.00
0.58
0.04
0.53
0.12
0.25
0.11
0.17
Computed using MLE Scoring Algorithm (see Amisano & Giannini (1992)).
Inflytm
0.00
0.00
0.00
0.00
0.00
0.00
0.00
1.00
-0.02
0.00
0.00
M1
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
1.00
-40.37
0.05
Int
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.05
1.00
-0.01
Rtwi
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
1.00
Inflytm
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.01
0.00
0.01
M1
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
11.83
0.08
Int
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.02
0.00
0.00
Rtwi
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
Impu lse responses
Having estimated the SVAR model, the impulse responses16 to an unanticipated unsystematic cash rate shock of
17 basis points is reported below.
The impulse response for private consumption is sluggish and takes around two years to reach its full effect.
One possible explanation for the sluggish response is the interplay of substitution and wealth effects of
consumption. An increase in the cash rate would increase the cost of consumer borrowing and debt servicing,
causing consumers to substitute away from consumption (substitution effect). It is also plausible that monetary
contraction would increase the return to a given stock of savings (wealth effect). Via the wealth effect it is
probable that consumers increase current and future consumption. The ultimate impact on consumption
therefore depends on the relative magnitude of the substitution and wealth effects. In this case it appears that
the wealth effect has some impact in the first year after the shock.
The investment response is also quite sluggish. The maximum effect of the investment response is reached in
the fifth quarter, and is much larger than the contraction in consumption. The sluggish response is likely to be
because firms are unable to change their investment plans in response to an unanticipated cash rate innovation
16
Since the GDP components are in logs, their impulse response can be converted percentages.
P. 11
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
(which is precisely what an SVAR innovation represents). Moreover, if firms have priced in an unexpected
increase in their cost of capital (in the form of capital risk), most firms are likely to proceed with the project
anyway. Similar behaviour is observed by Bernanke and Gertler (1995) for the US economy 17.
Given that the government consumption expenditure equation is not fully specified (since this is not a study
examining Fiscal Policy such as Dungey and Fry (2003)), one cannot draw too many conclusions from its
impulse response. Overall, as one would expect, the impulse response suggests that there is only marginal
deviation in government consumption expenditure.
The impulse responses of exports and imports are easier to understand when viewed alongside the response of
the real exchange rate. The real exchange rate appreciates instantaneously by 0.19% due to the interest rate
shock. This overshooting of the exchange rate is consistent with the Uncovered Interest Parity (UIP) rule 18 with
expectations. Moreover, the response is indicative of price rigidities in the short run. Interestingly, the response
does not reveal any evidence of the “exchange rate puzzle 19” or the “forward premium puzzle 20” that appear in
some studies.
Due to the AUD appreciation, imports reach their peak increase in the third quarter. Exports, decline from
quarters 2 to 6; subsequently returning to the baseline. Overall, this implies that net exports would fall within
the first year of the cash rate innovation. Subsequently, the real AUD begins to return to its baseline, driving
exports up and imports down. This result is consistent with the ‘J-curve’ representation of the net exports
response to exchange rate shocks. Importantly, the increase in net exports offsets part of the contractionary
impact of investment.
The impulse response of inflation is sluggish, and the maximum effect of the monetary policy contraction is only
felt in the 10th quarter. Interestingly, the inflation response does not reveal statistically significant evidence of
what is known as the ‘price puzzle’ - a consistent increase in inflation subsequent to an interest rate increase.
This suggests that the monetary policy shock has been sufficiently indentified. Similar to what Sims (1992)
observes it is likely that commodity prices play a role in eradicating the price puzzle prevalent in most SVAR
studies.
17
They use inventory investment in the model and hence interpret it as inventory build-up.
18
See Blanchard and Sheen (2007) pp. 423 for a comprehensive discussion of the UIP with expectation which accounts
for the overshooting.
19
Immediate depreciation of the exchange rate following an increase in the interest rate.
20
Persistent appreciation of the exchange rate following a monetary policy contraction.
P. 12
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
The response of the monetary aggregate (M1) is consistent with economic theory. The disturbance to the cash
rate results in an instantaneous fall in money supply. The response closely follows the cash rate and returns to
the baseline once the cash rate innovation has died out. Interestingly, the model does not exhibit the ‘Liquidity
Puzzle’ which is apparent in some SVAR studies 21.
I m p u l s e r es p o n s e s t o c o n t r a c t i o n a r y m o n e t a r y p o l i c y – M o d e l 1
Fig ur e 1.
Investment
Consumption
Government expenditure
0.40%
0.40%
0.40%
0.20%
0.20%
0.20%
0.00%
0.00%
1
5
9
13
17
-0.20%
0.00%
1
21
5
9
13
17
21
1
-0.20%
-0.20%
-0.40%
-0.40%
-0.60%
-0.60%
-0.80%
-0.80%
-0.40%
5
9
13
17
21
13
17
21
17
21
-0.60%
-0.80%
Imports
Exports
Inf lation
0.40%
0.40%
15.0%
0.20%
0.20%
10.0%
0.00%
0.00%
1
5
9
13
17
5.0%
1
21
5
9
13
17
21
-0.20%
-0.20%
0.0%
-0.40%
-0.40%
-5.0%
-0.60%
-0.60%
-10.0%
-0.80%
-0.80%
-15.0%
1
Monetary aggregate M1
0.20%
40.0%
0.60%
30.0%
0.40%
0.00%
-0.20%
1
5
-0.40%
9
13
17
21
9
Real trade weighted index
Cash rate
0.40%
5
20.0%
0.20%
10.0%
-0.60%
0.00%
0.0%
-0.80%
1
1
-1.00%
5
9
13
17
21
5
9
13
-0.20%
-10.0%
-1.20%
-1.40%
-1.60%
-20.0%
-0.40%
-30.0%
-0.60%
The vertical axis shows deviations from the baseline level prior to the cash rate innovation, while the horizontal axis shows the number of quarters after the disturbance.
The solid lines represent point estimates of the impulse responses, while the dotted lines are 95% confidence intervals based on Hall’s (1992) bootstrap method utilising
250 bootstrap replications. Note that US GDP and Commodity Prices do not have impulse responses as the cash rate does not enter either equation.
21
The presence of the “liquidity puzzle” (or lack thereof) does not however negate the validity of the monetary policy
function, because the monetary aggregate, however narrowly defined, does not represent the monetary policy stance in
Australia during the majority of the sample period.
P. 13
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
4. 3
S ensi ti vi t y vs. C on tr ib u ti on
The impulse responses from Model 1 reflect the extent to which each GDP component diverges from its baseline
due to a monetary policy shock. In other words, the impulse responses show the sensitivity to an increase in
the cash rate. The sensitivity to a monetary policy shock does not necessarily mean that the output response is
driven by the most sensitive component. For instance, a component might be highly sensitive, but only make
up a small share of GDP, and therefore have a limited impact on economic activity. It is therefore important to
distinguish between ‘sensitivity to shocks’ and ‘contribution to GDP contraction’.
In the standard SVAR framework, the reduced form residuals are converted to structural shocks by imposing
the identifying restriction matrices A and B. This implies that the Wold moving average form of the VAR can be
written as a function of the structural shocks.
yt  0 A1 B t  1 A1 B t 1  2 A1B t 2  .....
(1.5)
Based on the above exposition, the impulse response function (IRF) can be derived as follows.
IRFy 
yt  s
  s a j b jj
 jt
(1.6)
Where aj is the jth column of the A-1 matrix that represents the contemporaneous relationships between the
system variables and b jj is the elements of the matrix describing the contemporaneous relationships of the
structural shocks. Since all GDP components are in logs we can rewrite this as follows:
IRFk 
 log yk ,t  s
 j ,t

yk ,t  s 1
; where yk, = C, Invst, G, Ex, Im
 j ,t yk
(1.7)
By the national accounting identity, it is therefore possible to recover total output by aggregating each GDP
component’s IRF after weighting the response by its share of GDP 22.
y
 log GDPt  s GDPt  s 1
1 yk

  k ,t  s ,
 j ,t
 j ,t GDPt
 j ,t yk GDPt
k
(1.8)
In other words, the Weighted Impulse Response Function (WIRF) can be used to recover total output, and
thereby examine which component contributes most to output contractions. The figure below shows IRFs and
WIRFs. Net-exports are computed as the difference in the exports, and imports responses. Implied GDP is
calculated as the sum of the WIRFs of the GDP components.
22
2012:03 shares of GDP are used to compute the WIRFs. C = 54%, I = 29%, G = 18%, Ex = 22%, and Im = 21%
P. 14
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
Fig ur e 2.
U n - w e i g h t e d I R F v s . W ei g h t e d I R F : M o d e l 1
Weighted Impulses
Un-weighted Impulses
0.20%
0.20%
0.10%
0.10%
0.00%
0.00%
1
5
9
13
17
1
21
5
9
13
17
21
-0.10%
-0.10%
-0.20%
-0.20%
-0.30%
-0.30%
-0.40%
-0.40%
C
C
I
G
I
G
NX
Implied GDP
NX
Figure 2 shows the importance of distinguishing between sensitivity, and contribution. It is clear that the
combination of the fall in net exports and investment drives down output till the fifth quar ter. The subsequent
increase in net exports (driven by the falling real exchange rate) offsets the contractionary effect of investment
till the 12th quarter. It is also clear that investment, and net exports are highly sensitive to cash rate shocks.
The implied GDP response reaches a sustained trough in the first and second years. This is broadly consistent
with Australian SVAR studies such as Dungey and Pagan (2000), and Brischetto and Voss (1999).
The WIRFs can also be used to calculate the Cumulative Contribution Ratio (CCR) which shows the composition
of GDP variation due to the monetary policy shock. Angeloni et al (2003), Fujiwara (2003), and Linde (2003)
use a similar measure 23 because they argue that it smooths out some of the possible noise in the responses,
particularly in the initial periods. Moreover, it can show the net impact after accounting for variables such as
net exports that have a positive impact on GDP. Let Ck,h be defined as:
yk 1
yk
 j yk GDP


h
yk 1
yk
 s 1 
 j yk GDP
k

Ck , h
h
s 1
yk 1
yk
 j yk GDP

h
GDP
1
 s 1  GDP
j

h
s 1
 WIRF
 IRF
h
s 1
k
h
s 1
(1.9)
Implied GDP
This gives the fraction of the variation in GDP attributable to GDP component k cumulated over h quarters.
Thus, CCRk,h24 can be computed as follows;
CCRk ,h 
Ck ,h
C
k ,h
(1.10)
k
The table below shows the CCRs based on the point estimates, and upper and lower error bands (given in
parentheses). Within the first year, investment, and net exports contribute the most to deviations in output.
This indicates that the exchange rate channel of monetary policy transmission is important in the short run, and
23
They refer to it as the “contribution ratio”.
24
Note that CCR is bounded between 0 and 1.
P. 15
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
(based on the CCRs in parentheses) may be important beyond the first year. However, over time, consumption
grows in importance.
T abl e 3.
C u m u l a t i v e C o n t r i b u t i o n R a t i os ( C C R )
Horizon (h)
Household consumption
Investment
9%
73%
(62%
60%)
69%
(62%
77%)
61%
(57%
59%)
52%
(51%
30%)
44%
(48%
39%)
4
(12%
8
9%)
13%
(14%
12
2%)
17%
(17%
16
8%)
21%
(19%
20
(20%
5
17%)
23%
12%)
Government
consumption
4%
(6%
7%)
6%
(7%
0%)
9%
(9%
4%)
10%
(9%
10%)
12%
(10%
10%)
Net exports
14%
(19%
24%)
12%
(16%
21%)
13%
(18%
28%)
17%
(20%
42%)
21%
(23%
39%)
Model 2
In order to better understand the driving forces of the consumption, and investment responses, Model 2
disaggregates consumption into durable and non-durable consumption, and investment into new dwelling and
rest of investment. To preserve the accounting identity, government consumption expenditure, exports, and
imports are retained in this model. All external and domestic sector variables are included as before. Given that
this model has 13 variables (as opposed to 11 in the previous model) the lag length was reduced to three, in
order to preserve degrees of freedom. The usual diagnostic tests of autocorrelation, and stability were
conducted 25 on the reduced form VAR of three lags. The results were largely similar to Model 1, suggesting
reasonable specification. Other aspects, such as the sample period, deterministic terms, lag structure, and nonrecursive identification scheme are unchanged from Model 1.
The following figure shows the impulse responses due to an unanticipated monetary policy shock of 34 basis
points. The impulses show that new dwelling investment is the most sensitive component to the rate increase in
this model. This is likely to be due to the cash rate being transmitted through to lending in the housing market.
It is therefore unsurprising that markets and commentators pay close attention to developments in the housing
market. In contrast, durable consumption appears to be slightly more sensitive than its non-durable
counterpart in the first year. However, the responses are broadly similar from the second year onwards. This is
likely to be because consumers do not rely heavily on credit for durable consumption, and as such the
transmission of monetary policy to consumer credit is likely to be limited. The wealth effect (through cash
deposits) is likely to have a greater impact on consumption. This is evident in the sluggish durable and nondurables response, which was also observed in aggregate consumption in Model 1. Driven by real exchange rate
deviations, net exports exhibit the ‘J curve’ representation, which is also similar to Model 1.
25
Autocorrelation tests are not reported for the sake of brevity. Stability tests are in the Appendix.
P. 16
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
The responses of the domestic variables appear to be consistent with those obtained from the previous model.
However, the point estimates of the inflation response reveal a small, albeit statistically insignificant ‘price
puzzle’. The response of the monetary aggregate and real exchange rate respectively reveals no evidence of
the ‘liquidity puzzle’ or the ‘exchange rate puzzle’. This suggests that Model 2 has been reasonably well
specified, and the cash rate innovation sufficiently identified.
Fig ur e 3.
I m p u l s e r es p o n s e s t o c o n t r a c t i o n a r y m o n e t a r y p o l i c y : M o d e l 2
Durable consumption
Non-durable consumption
New dwelling investment
0.40%
0.40%
0.40%
0.20%
0.20%
0.20%
0.00%
1
5
9
13
17
21
-0.20%
0.00%
0.00%
1
5
9
13
17
21
1
-0.20%
-0.20%
-0.40%
-0.40%
-0.40%
-0.60%
-0.60%
-0.60%
-0.80%
-0.80%
-0.80%
Government expenditure
Remainder of investment
0.40%
0.40%
0.20%
0.20%
0.00%
0.00%
1
5
9
13
5
17
17
21
13
17
21
Exports
0.20%
0.00%
5
9
13
17
21
1
-0.20%
-0.20%
-0.20%
-0.40%
-0.40%
-0.40%
-0.60%
-0.60%
-0.60%
-0.80%
-0.80%
-0.80%
Imports
5
Inf lation
0.40%
13
0.40%
1
21
9
9
Monetary aggregate M1
10.0%
0.40%
0.20%
0.20%
5.0%
0.00%
-0.20%
0.00%
1
5
9
13
17
0.0%
21
1
-0.20%
1
5
9
17
21
13
17
21
-0.40%
5
9
13
17
21
-0.60%
-5.0%
-0.80%
-0.40%
-1.00%
-10.0%
-0.60%
-1.20%
-1.40%
-0.80%
-15.0%
-1.60%
Cash rate
Real trade weighted index
100.00%
1.20%
80.00%
1.00%
0.80%
60.00%
0.60%
40.00%
0.40%
20.00%
0.20%
0.00%
1
5
9
13
17
21
0.00%
-20.00%
-0.20%
-40.00%
-0.40%
1
5
9
13
The vertical axis shows deviations from the baseline level prior to the cash rate innovation, while the horizontal axis shows the number of quarters after the disturbance.
The solid lines represent point estimates of the impulse responses, while the dashed lines are 95% confidence intervals based on Hall’s (1992) bootstrap method utilising
250 bootstrap replications. Note that US GDP and Commodity Prices do not have impulse responses as the cash rate does not enter either equation.
P. 17
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
5. 1
S ensi ti vi t y vs. C on t r ib u ti on
Similar to the approach adopted for Model 1, the weighted impulse responses of consumption and investment
components are computed to distinguish sensitivity from contribution to contractions in their respective
26
aggregates. As such, the weights are shares of the respective aggregate in 2012 .
The figure below demonstrates that even though durables and non-durables are relatively similar in their
sensitivity to monetary policy shocks, non-durables contribute the most to the contraction in aggregate
consumption. New dwelling investment is significantly more responsive to monetary policy shocks than the rest
of investment. However, non-new dwelling investment makes a much greater contribution to contraction in
aggregate investment. This suggests that even though the housing market is highly sensitive to cash rate
shocks, the RBA should not rely heavily on the housing investment to transmit monetary policy to the economy.
Fig ur e 4.
U n - w e i g h t e d I R F v s . W ei g h t e d I R F : M o d e l 2
Un-weighted - Consumption
0.15%
Weighted - Consumption
0.15%
0.10%
0.10%
0.05%
0.05%
0.00%
-0.05%
1
5
9
13
17
0.00%
21
-0.05%
-0.10%
-0.10%
-0.15%
-0.15%
Durc
13
17
21
Non-Durc
Weighted - Investment
0.30%
0.10%
1
5
9
13
17
-0.10%
21
-0.30%
-0.30%
-0.50%
-0.50%
-0.70%
1
5
9
13
17
21
-0.70%
House
6
9
Durc
0.10%
-0.10%
5
Non-Durc
Un-weighted - Investment
0.30%
1
Non-House
House
Non-House
Robustness
SVAR models are popular frameworks used for Macroeconomic analysis. However, just like any econometric
model, the results are only as good as the model itself. It is therefore imperative to test the sensitivity of the
baseline results to alternative econometric specifications. Lag length of the VAR, identification, deterministic
terms, and sample period are each altered to test the robustness of Model 1 impulse responses
27
26
Durables are 11% of aggregate consumption. New dwellings are 10% of aggregate investment.
27
Model 2 was tested, and yielded similar robustness to Model 1. Not reported for the sake of brevity.
P. 18
.
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
The figure below shows the impulse responses to a range of sensitivity checks: 2 and 3 lag VARs, an alternative
identification scheme - Choleski decomposition (recursive identification
29
(since some of the series appear to be trending), and a 3 lag VAR
28
), a trend term in the baseline VAR
restricted to the sample period 1994:01 to
2012:03 (post inflation targeting period).
Fig ur e 5.
T es t i n g r o b u s t n e s s of i m p u l s e r e s p o n s es
0.20%
0.20%
0.20%
Investment
Consumption
0.10%
Government expenditure
0.10%
0.00%
0.10%
0.00%
0
4
8
12
16
20
0.00%
0
4
8
12
16
20
0
-0.10%
-0.10%
-0.10%
-0.20%
-0.20%
-0.20%
-0.30%
-0.30%
-0.30%
-0.40%
-0.40%
-0.40%
0.20%
0.20%
4
8
12
16
20
-0.10%
4
8
12
16
20
-0.20%
-0.20%
-0.30%
-0.30%
-6.00%
-0.40%
-0.40%
-8.00%
40.00%
4
8
12
16
20
10.00%
-0.60%
0.00%
8
16
20
Real trade weighted index
0.40%
0.30%
20.00%
-0.40%
4
0.50%
Cash rate
30.00%
0
0
-4.00%
M1
-0.20%
12
0.00%
-2.00%
0.00%
20
2.00%
0
-0.10%
0.20%
16
Inflation
0.00%
0
12
4.00%
0.10%
0.00%
8
6.00%
Imports
Exports
0.10%
4
0.20%
0.10%
4
8
12
16
20
0.00%
-0.80%
-10.00%
-1.00%
-20.00%
-0.20%
-1.20%
-30.00%
-0.30%
3 lags
10 12 14 16 18 20
0
2 lags
-0.10%
0
4
8
12
16
20
Notes: The vertical axis shows deviations from the baseline level prior to the cash rate innovation of one standard deviation, while the
horizontal axis shows the number of quarters after the disturbance. The solid lines represent point estimates of the impulse responses. Note
that US GDP and Commodity Prices do not have impulse responses as the cash rate does not enter either equation.
Sub-sample
Recursive VAR
Trend in VAR
Broadly speaking, the responses of the GDP components are relatively consistent across all five estimated
VARs. In particular, investment and imports appear to be the most sensitive. In terms of the direction of the
investment and imports responses, the sub-sample model (with 3 lags) appears to be the only anomaly.
However, the extent of the deviation appears consistent with other models. Interestingly, nearly all the models
(except the recursive and 2 lag VARs) do not reveal substantial evidence of the ‘price puzzle’. It is likely that
this is due to the inclusion of commodity prices, identification scheme of the cash rate innovation, and the
28
The ordering of the variables is identical to that in Model 1.
29
Given the reduced sample size, the number of lags had to be reduced for the SVAR to be estimated.
P. 19
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
30
sample period . The responses of the monetary aggregate do not reveal strong evidence of the ‘liquidity
puzzle’ either. There is some inconsistency in the responses of the real exchange rate (mainly the sub -sample
model). However there is no evidence of the ‘exchange rate puzzle’ as all responses are positive following the
cash rate innovation. Broadly speaking, these sensitivity simulations suggest that the impulse responses
obtained from the baseline Model 1 are generally robust, and therefore one can be reasonably conf ident of the
results obtained.
7
Conclusio n
Angeloni et al (2003) examine differences between the US and the Euro Zone in terms of their compositional
responses of GDP to domestic monetary policy. They refer to the economy-specific response of the components
of GDP as ‘the output composition puzzle’. The response is deemed a ‘puzzle’ because, even though the
conventional understanding is that investment responds to interest rate changes (thereby changing output);
the authors find that it is in fact consumption in the US (and UK), and investment in the Euro Zone that
contributes the most to GDP deviation. This paper examines the output composition puzzle for Australia using
two Structural Vector Auto Regression (SVAR) models.
Model 1 is an eleven variable SVAR containing consumption, investment, government consumption expenditure,
exports, and imports. It is estimated over the period 1986:03 to 2012:03 due to constraints imposed by the
availability of underlying inflation data, and parameter instability in the periods subsequent to floating of the
currency. This paper makes a distinction between the sensitivity of each GDP component, and its contribution to
GDP deviation. Sensitivity is measured by the impulse responses, while contribution is measured by weighted
impulses, and Cumulative Contribution Ratios (CCR). Several interesting findings emerge. Overall, investment
(followed by net exports) is most sensitive to unanticipated increases in the cash rate. However, the investment
response is sluggish, and takes a few quarters to reach its maximum effect. The sluggish response is likely to
be because firms are unable to change their investment plans in response to an unanticipated cash rate
innovation (which is precisely what an SVAR innovation represents). If firms have priced in an unexpected
increase in their cost of capital (in the form of capital risk) then they are likely to proceed with the project
anyway.
Within the first year, net exports are found to be a key deriver of output reduction, contributing to nearly 15%
of output reduction. However, two to three years after the cash rate increase, net exports increase (due to the
30
Dropping the commodity price term causes a large puzzle. The non-recursive identification results in a small puzzle
(as shown above). Sample period before 1986 also results in the puzzle. However, the period before 1986 is unsuitable
since the AUD was floated at the end of 1983; and the data is likely to result in parameter instability.
P. 20
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
real exchange rate returning to its baseline) thereby reducing the contractionary impact of investment. This
suggests that the exchange rate channel is particularly important for the transmission of Monetary Policy in
Australia. Over a period of five years (due to its sheer size in the economy) consumption contributes around
23% to output deviations, while next exports contribute a similar share. However, overall, investment is found
to contribute most to output deviations (44%). Interestingly, this model does not show evidence of the ‘price
puzzle’ or other puzzles encountered in other SVAR studies. This is likely to be due to the inclusion of
commodity prices, identification scheme, and the sample period.
Model 2 further disaggregates consumption (durables and non-durables), and investment (new dwelling
investment and rest of investment) while retaining the other variables from Model 1 to preserve the accounting
identity. This model finds that within investment, new dwelling investment is the most sensitive to cash rate
shocks. However, the rest of investment contributes more to contractions in output. In contrast, durable
consumption appears to be slightly more sensitive than its non-durable counterpart in the first year. However,
the responses are broadly similar from the second year onwards. This is likely to be because, unlike the US,
Australian consumers do not rely heavily on credit for durable consumption over the sample period, and as such
the transmission of monetary policy to consumer credit is likely to be limited.
Five separate VARs were estimated to test the robustness of the results. Decreased lag length, alternative
identification, inclusion of a trend term, and restricting the sample to the inflation targeting period (sub-sample
analysis) were the variations tested. Broadly speaking, the responses of the GDP components are relatively
consistent across all five estimated VARs. In particular, investment and imports appear to be the most
sensitive. This suggests that the baseline results are relatively robust.
The results from this paper give key insights into the compositional response of GDP to Australian monetary
policy. In particular, the results suggest that non-dwelling investment, and net exports should be the key focus
of the monetary authorities since they largely drive the output response. Moreover, since net exports have a
delayed positive effect on GDP, the monetary authorities need to be aware of the importance of the exchange
rate transmission mechanism in dampening the maximum impact of contractions in investment. These aspects
should be closely monitored to assess the impact of monetary policy.
P. 21
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
8
Bibliography
Angeloni, I., Kashyap, A.K., Mojon, B. & Terlizzese, D. (2003), "The Output Composition Puzzle: A Difference in
the Monetary Transmission Mechanism in the Euro Area and U.S.", Journal of Money Credit, and Banking, Vol.
35, No. 6, pp. 1265-1306.
Beechey, M., Bharucha, N., Cagliarini, A., Gruen, D. & Thompson, C. (2000), “A small model of the Australian
macroeconomy”, Discussion Paper No. 2000-05, Reserve Bank of Australia.
Berkelmans, L. (2005), “Credit and Monetary Policy: An Australian SVAR”, Discussion Paper No. 2005-06,
Reserve Bank of Australia.
Bernanke, B. & Blinder, A. (1992), “The Federal Funds Rate and the Channels of Monetary Transmission”,
American Economic Review, Vol. 82, No. 4, pp. 901-921.
Bernanke, B.S. & Gertler, M. (1995), “Inside the Black Box: The Credit Channel of Monetary Policy
Transmission”, Journal of Economic Perspectives, Vol. 9, No. 4, pp. 27–48.
Brischetto, A. & Voss, G. (1999), “A structural Vector Auto Regression Model of Monetary Policy in Australia”,
Discussion Paper No. 1999 – 11, Reserve Bank of Australia.
Christiano, L. J., Eichenbaum M. & Evans, C. (2001), “Nominal rigidities and the dynamic effects of a shock to
monetary policy”, NBER Working Paper No. W8403, National Bureau of Economic Research.
Christiano, L. J., Eichenbaum, M. & Evans C. (1996), “The Effects of Monetary Policy Shocks: Evidence from the
Flow of Funds”, Review of Economics and Statistics, Vol. 78, No. 1, pp.16-34.
Christiano, L. J., Eichenbaum, M. & Evans, C. (1998), “Monetary Policy Shocks: What Have We Learned and to
What End?”, NBER Working Paper No. 6400, National Bureau of Economic Research.
Cooley, T. & LeRoy, S. (1985), “Atheoretical Macroeconomics: A Critique”, Journal of Monetary Economics, Vol.
16, No. 2, pp.283-308.
Cushman, D. & Zha, T. (1997), “Identifying Monetary Policy in a Small Open Economy Under Flexible Exchange
Rates”, Journal of Monetary Economics, Vol. 39, No. 3, pp.433-448.
De Roos, N. & Russell, B. (1996), “Towards an understanding of Australia’s co-movement with foreign business
cycles”, Discussion Paper No 9607, Reserve Bank of Australia.
Dotsey, M. (1987), “Open market operations in Australia: a US perspective”, Discussion Paper No 8702,
Reserve Bank of Australia.
Dungey, M. & Fry, R. (2003), “International shocks on Australia: The Japanese effect”, Australian Economic
Papers, Vol. 42, No. 2, pp. 499-515.
Dungey, M. & Fry, R. (2007), “The Identification of Fiscal and Monetary Policy in A Structural VAR," Working
Paper No. 2007-29, Centre for Applied Macroeconomic Analysis, Australian National University.
Dungey, M. & Pagan, A. (2000), “A Structural VAR Model of the Australian Economy”, Economic Record, Vol.
76, No. 235, pp. 321–342.
Erceg, C. & Levin, A. (2002), “Optimal monetary policy with durable and non-durable goods”, Working Paper
No. 179, European Central Bank.
Fisher, L.A. & Voss, G.M. (2004), “Consumption, Wealth and Expected Stock Returns in Australia”, Economic
Record, Vol. 80, pp. 359-372.
Friedman, M. & Schwartz, A. (1963), A Monetary History of the United States: 1867-1960, Princeton University
Press.
Fujiwara, I. (2003), “Output composition of monetary policy transmission in Japan”, Discussion Paper No. 0307, Graduate School of Economics and Osaka School of International Public Policy, Osaka University.
Gordon, D. & Leeper, E. (1994), “The Dynamic Impacts of Monetary Policy: An Exercise in Tentative
Identification”, Journal of Political Economy, Vol. 102, No. 6, pp. 1228-1247.
P. 22
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
Grenville, S. (1997), “The Evolution of Monetary Policy: from money targets to inflation targets”, in Proceedings
of a Conference on Monetary Policy and Inflation Targeting, Reserve Bank of Australia, Sydney, pp 125–158.
Gruen, D. & G. Shuetrim (1994), “Internationalisation and the Macroeconomy”, in Lowe, P. &. Dwyer, J. (edn.),
International Integration of the Australian Economy, Reserve Bank of Australia, Sydney, pp. 309–363.
Hall, P. (1992). The Bootstrap and Edgeworth Expansion, Springer, New York.
Hamilton, J. (1994), Times Series Analysis, Princeton University Press, Princeton.
Hoover, K. & Jorda, O. (2001), “Measuring Systematic Monetary Policy”, Working Paper, No. 00-05, University
of California, Davis.
Huh, H. (1999), “How Well Does the Mundell-Fleming Model Fit Australian Data since the Collapse of Bretton
Woods”, Applied Economics, Vol. 31, pp. 397–407.
Kim, S. & Roubini, N. (2000), “Exchange Rate Anomalies in the Industrial Countries: A Solution with a
Structural VAR Approach”, Journal of Monetary Economics, Vol. 45, pp.561-586.
Leeper, E., Sims, C. & Zha, T. (1996), “What Does Monetary Policy Do?”, The Brookings Papers on Economic
Activity, Vol.2, pp. 1-48.
Linde, J. (2003), “Comment on The Output Composition Puzzle: A Difference in the Monetary Transmission
Mechanism in the Euro Area and U.S.", Journal of Money Credit, and Banking, Vol. 35, No. 6, pp.1309-1317.
Lutkepohl, H. & Kratzig, M. (2004), Applied Time Series Econometrics: Themes in Modern Econometrics, ,
Cambridge University Press, Cambridge, UK.
Mojon, B. & Peersman, G. (2003), “A VAR description of the effects of monetary policy in the individual
countries of the euro area”, in Angeloni, I., Kashyap, A. & Mojon, B. (eds.) Monetary policy transmission in the
euro area, Cambridge University Press, UK.
Peersman, G. &Smets, F. (2003), “The monetary transmission mechanism in the euro area: more evidence
from VAR analysis”, in Angeloni, I., Kashyap, A. & Mojon, B. (eds.) Monetary policy transmission in the euro
area, Cambridge University Press, UK.
Raddatz, C. & Rigobon, R. (2003), “Monetary Policy and Sectoral Shocks: Did the Federal Reserve React
Properly to the High-Tech Crisis?”, Working Paper No. 3160, Policy Research, World Bank.
Reserve Bank of Australia (1996), Statement of Monetary Policy, Reserve Bank of Australia, Sydney.
Romer, C. & Romer, D. (1990), '' New Evidence on the Monetary Transmission Mechanism," Brooking Papers on
Economic Activity, Vol. 1, pp. 149-214.
Sims, C. A. (1980), “Macroeconomics and Reality”, Econometrica, Vol. 48, pp. 1-48.
Sims, C. A. (1981), “An autoregressive index model for the U.S. 1948-1975”, in Kmenta, J.& Ramsey, J. B.
(eds.), Large-Scale Macro-Econometric Models, North-Holland, Amsterdam, pp. 283–327.
Sims, C. A. (1992), “Interpreting the macroeconomic time series facts: The effects of Monetary Policy”,
European Economic Review, No. 36, pp 975 – 1000.
Sims, C. A., Stock, J. H. & Watson, M. W. (1990), “Inference in Linear Time Series Models with Some Unit
Roots”, Econometrica, Vol. 58, No. 1, pp 113–144.
Sims, C.A. (1986), “Are Forecasting Models Usable for Policy Analysis?”, Federal Reserve Bank of Minneapolis
Quarterly Review, Vol. 10, No. 1, pp 2–16.
Suzuki, T. (2004), “Is the Lending Channel of Monetary Policy Dominant in Australia?”, Economic Record, Vol.
80, No. 249, pp. 145-156.
Tan, A. & Voss, G. (2000),"Consumption and Wealth," Discussion Papers No. 2000-09, Reserve Bank of
Australia.
P. 23
Output Composition Puzzle / Muheed Jamaldeen (Submission to ACE 2013)
9
Appendix
T abl e 4.
D at a s o u r c es , a n d d e f i n i t i o n s
Variable
US GDP
Commodity prices
Definition
Real US Gross Domestic Product
RBA index of commodity prices – all items
Consumption (C): Households; Final consumption expenditure;
Investment (Invst): All sectors; Gross fixed capital formation;
Government expenditure (G): General government; Final consumption expenditure;
Exports (Ex): Exports of goods and services;
Imports (Im): Imports of goods and services;
New dwelling investment (House): Private Gross fixed capital formation - Dwellings - New;
Rest of investment (Rest-Invest): Invst less House. Derived by the author.
Durable consumption (Durc): Clothing and footwear plus, Furnishings and household equipment, plus
Purchase of vehicles. Derived by the author.
Non-Durable consumption (Non-Durc): Households; Final consumption expenditure minus durable
consumption (i.e. C minus Durc). Derived by the author.
The RBA calculates the “Trimmed mean” by ordering all the Consumer Price Index components by their price
change in the quarter and taking the expenditure-weighted average of the middle 70 per cent of these price
changes.
M1 is defined as currency plus bank current deposits of the private non-bank sector.
The “Interbank rate” is a weighted average of the interest rates at which banks have borrowed and lent
exchange settlement funds overnight. Monthly data is convereted to quarterly data by the author.
The ‘Real trade-weighted index’ is the average value of the Australian dollar in relation to currencies of
Australia’s trading partners adjusted for relative price levels, using core consumer price indices from these
countries.
Real GDP components
Inflation
Monetary aggregate
Cash rate
Real trade weighted
index.
T abl e 5.
VAR (4)
VAR (4) with trend
VAR (3)
VAR (3) sub sample
VAR(2)
Units
2000:01 = 100 (seasonally adjusted)
2008:01 = 100, US $
Source
RBA, Table I1
RBA, Table G5
National Income, Expenditure
and Product, ABS Cat. No.
5206.0, Table 2
AU $ millions (seasonally adjusted),
Chain volume measures
Household Final Consumption
Expenditure (HFCE), ABS
Cat. No. 5206.0, Table 8
year ended percentage change
RBA, Table G1
AU $ billions
RBA, Table D3
percent per annum
RBA, Table F1
1995/01 = 100
RBA, Table F15
St abilit y t ests
1.2
1.2
1.4
1.4
2.9
1.2
1.2
1.8
1.4
2.9
1.3
1.3
1.8
1.4
3.2
1.3
1.3
1.9
1.4
3.2
1.5
1.4
1.9
1.5
2.9
1.5
1.4
2.4
1.5
2.9
1.5
1.5
2.5
1.6
3.5
1.7
1.5
2.5
2.2
3.5
1.7
1.8
2.1
2.1
1.7
1.7
1.7
2.1
2.1
1.7
1.8
1.7
1.8
2.1
2.8
2.2
1.9
1.8
2.1
2.8
2.2
1.9
1.3
1.3
2.3
2.8
1.3
1.3
1.3
1.2
1.3
1.3
6.3
6.3
1.2
1.3
1.5
3.6
1.2
1.4
1.4
1.5
3.6
1.2
1.4
1.4
2.1
1.3
1.4
1.1
1.5
2.1
1.3
1.4
1.1
1.5
3.7
1.2
2.4
1.0
2.2
1.2
1.2
2.4
1.1
2.2
1.2
1.7
1.5
1.1
1.3
9.5
1.7
1.5
1.3
1.2
1.3
1.1
1.2
1.2
1.3
1.1
1.2
1.5
3.2
1.2
1.4
1.5
1.1
1.2
1.4
1.6
1.1
1.3
1.6
1.6
1.2
1.3
1.6
1.3
1.2
1.0
6.4
1.3
1.0
1.0
1.2
1.2
1.0
1.1
1.2
1.2
1.0
1.1
1.1
1.1
1.1
1.1
3.5
1.1
1.1
1.1
1.1
1.0
1.1
1.1
1.1
1.1
1.0
1.3
1.0
1.3
1.0
1.6
2.8
3.4
Notes: The above table shows modulus of the eigenvalues of the reverse characteristic polynomial. These values need to be gre ater than one for the VAR to be stable, and imply stationarity (Lutekepohl (2004)). Values
have been rounded to one decimal place for the sake of brevity. All values shown as 1.0 are greater than one.
P. 24