the impact of property market developments on the real economy of malaysia

It is conceivable that property booms can reinforce real economic booms since property prices do seem to exert temporary pro- cyclical effects on both consumption and investment.. Recent

Nottingham University Business School, University of Nottingham Malaysia

Jalan Broga, 43500 Semenyih, Selangor Darul Ehsan, Malaysia

E-mail: huihonchung@gmail.com Tel: 603-89248268; Fax: 603-89248019

Abstract

This paper examines the effects of property market developments on the real economy in Malaysia Our findings suggest that in the long-run, domestic demand and GDP are neutral to fluctuations in property prices The reason is that while property booms drive higher gross investments, this is always accompanied by an offsetting decline in private consumption In the short-run however, the neutrality of demand and GDP to property price fluctuations is less certain It is conceivable that property booms can reinforce real economic booms since property prices do seem to exert temporary pro-

cyclical effects on both consumption and investment These findings imply that stimulating property market activities is not an effective way to drive sustained growth in the real economy Nonetheless, there may be room to consider the property market as a policy tool for short-term macroeconomic management

Keywords: Wealth effect, investment channel, cointegration, property market

to remain relevant as the crisis unfolds because they offer important lessons for other developing economies

Recent literatures on the impact of property markets on macroeconomic performance include

Ho and Wong (2008) who assessed the impact of house prices on domestic private demand in Hong Kong and found that housing market booms significantly augment domestic demand Other studies modelled the transmission channels of property market shocks (i.e the investment channel and the wealth effect on consumption) For instance, Ludwig and Slok (2004) and Case, Quigley and Shiller (2005) reported significant positive links between property prices and consumption in the US and a number of OECD economies However, studies on some Asian economies such as Singapore did not confirm such positive links (see among others, Phang, 2004, Edelstein and Lum, 2004) Peng, Cheung

1 See, among others, BIS (2005) and Hunter et al (2003)

and Leung (2001) and Peng, Tam and Yiu (2008) examined both investment and consumption channels in Hong Kong and China respectively The results for Hong Kong suggest that both channels respond positively and significantly to property prices However, in China only the investment channel was positive and significant while the wealth effect on consumption was negative and statistically insignificant The ambiguity of these findings implies that observed relationships between the property sector and macroeconomic performance are far from being conclusive and cannot be generalised across various countries and regions with diverse institutional structures

In this paper, we extend the assessment of the property market-economic performance analysis

to the case of Malaysia Our objectives are to (1) assess the real effects of property price fluctuations

by considering how property prices affect consumption and investment spending, which are the two known transmission channels of property market shocks widely discussed in the extant literature and (2) in the light of the findings from the first objective, to assess the importance of property prices as a driving factor for fluctuations in domestic demand and real GDP over the short-run and long-run Our study covers the period of 1991Q1-2006Q2

This research makes sense for several reasons Despite garnering plentiful attention in other economies, debates on the importance of property markets in economic development have received scant attention in Malaysia Given that the property market and the real economy seem closely intertwined (see Figure 1), there is still very little understanding of the importance of the former in affecting the latter Next, promoting and managing growth in property markets has always been one of the important policy objectives of the government because of claims that such growth would have spillover effects on other sectors of the economy Nonetheless, since there is no empirical evidence to confirm or refute such claims, there is no way for policy-makers to know if promoting property market booms necessarily create the intended effects

Figure 1: Annual growth (%) in real GDP and property prices and in Malaysia

Growth real GDP Growth property prices

Source: Bank Negara Malaysia Monthly Statistical Bulletin (various issues)

A preview of our findings is as follows First, in the long-run, domestic demand and GDP are neutral to fluctuations in property prices We rationalise this outcome by the observation that while property booms drive higher gross investments, there is an offsetting decline in private consumption In the short-run however, the neutrality of demand and GDP to property price fluctuations is less certain

It is conceivable that property booms can reinforce real economic booms since property prices do seem

to exert temporary pro-cyclical effects on both consumption and investment These findings imply that stimulating property market activities is not an effective way to drive sustained growth in the real economy Nonetheless, there may be room to consider the property market as a policy tool for short-term macroeconomic management

2 Overview of Macroeconomic Developments in Malaysia

Before we proceed to our formal analysis, some background macroeconomic developments are presented to provide the proper context Malaysia underwent one major and several more minor property boom-bust cycles over the period 1991-2006 (see Figure 1) The major boom episode occurred in the early 1990s and consisted of ‘twin peaks’ in the growth rate of property prices, with the first peak occurring in 1990-91 and the second in 1994-97 This expansionary phase came abruptly to

an end during the financial crisis in 1997-1998 A sharp recovery followed in 1999-2000 and culminated in another round of modest real estate boom in 2001-06

Closely in line with developments in the property markets, a construction and spending boom had also picked up in the early 1990s as financial institutions accelerated lending activities (Figures 3-5), enabling GDP to grow in the range of 9-10% Notably, the pre-crisis economic expansion was driven mainly by gross investments rather than private consumption, fuelled by demand for more residential, industrial and commercial building space Economic growth was briefly interrupted by the 1997-98 financial crisis, during which construction and gross investment experienced sharp contractions in response to a property market collapse Expansion of the economy resumed in 1999 albeit at more modest rates

An issue which arises from all these statistics is whether and how property markets reinforce real economic activities The next few sections attempt to examine the effects of property markets on real economic activities

Real GDP growth Real construction output growth

Source: Bank Negara Malaysia Monthly Statistical Bulletin (various issues)

Figure 3: Growth (%) in commercial banks lending

-5 0 5 10 15 20 25 30 35 40

1991 Q1 199

1 Q 4 199

2 Q3

1993 Q2

1994 Q1

1994 Q4

1995 Q3 199

6 Q 2 199

7 Q 1 199

7 4

1998 Q3

1999 Q2

2000 Q1

2000 Q4 200

1 Q 3 200

2 Q2 200

3 1

2003 Q4

2004 Q3

2005 Q2

Source: Bank Negara Malaysia Monthly Statistical Bulletin (various issues)

Figure 4: Growth (%) in spending

-80 -60 -40 -20 0 20 40 60

199

1 1

1991 Q4

1992

Q3

199

3 2

1994 Q1199

4 4

1995 Q3 199

6 2

19971

1997

Q4

199

8 3

1999 Q2 200

0 1

2000 Q4

2001

Q3

200

2 2

2003

Q1

200

3 4

2004 Q3 200

5 2

growth in gross fixed capital formation (1987 prices) growth in consumption (1987 prices)

Source: Bank Negara Malaysia Monthly Statistical Bulletin (various issues)

3 Theoretical Frameworks

The property and macroeconomy nexus is a relatively new but important research area (Leung, 2003) Among the salient issues that deserve attention is how property market fluctuations affect the macroeconomy Property market shocks induce fluctuations in real macroeconomic activities via two known channels, namely investment and consumption spending (Zhu, 2003) We elaborate each channel in greater detail in the following illustration

(i) Investment channel

Higher property prices relative to replacement/construction cost of property assets raise the profitability of building construction activities, according to Tobin’s q-ratio theory of investment Hence, developers and non-financial firms would engage in more residential and non-residential building construction This building boom would in turn boost demand and employment in property-related sectors Moreover, construction activities are not the only beneficiaries of a property market boom Higher property prices also provide incentives for firms in other sectors (i.e non-property firms and financial institutions) to increase their investment spending via the liquidity effect In particular, rising property prices tend to strengthen the balance sheet positions of property owners irrespective of the line of business, enabling them to secure external funds more easily and at lower cost to finance new investment projects This effect is what Bernanke et al (1996) referred to as the financial accelerator principle

(ii) Wealth effect channel of consumption

According to the life-cycle hypothesis (Ando and Modigliani, 1963), household consumption spending

is affected by wealth, of which housing is an important constituent Thus, changes in house prices would, via changes in housing wealth, affect consumption expenditure In contrast to the more straightforward investment channel, the existence and magnitude of the wealth effect is harder to rationalise There exist various transmission mechanisms linking property prices with private consumption

First, changes in housing prices would have little effects on the welfare of owner-occupiers Since housing is very much an asset as it is a consumption good, higher house prices also implies a higher cost of consuming housing services An owner-occupier would not be richer in any sense and hence a rise in consumption on other goods would not follow (Edelstein and Lum, 2004) Notably, if the owner-occupier measures the implicit cost of consuming housing in terms of the rental rate for similar types of homes in the same neighbourhood, there could be different short-run and long-run

responses in consumption In the long-run, housing price appreciations would push up rentals As rising house price would be fully reflected in higher housing consumption costs (proxied by rent), owner-occupiers would not feel richer so that increase in consumption would not take place In the short-run however, rental rate movements tend to be stickier than that of house prices due to the existence of rent contracts This being the case, it is possible for housing prices to rise faster than the proxy for cost of housing services, leading to temporary ‘wealth gains’ While a far-sighted owner-occupier in this case would probably not change consumption, seeing that his wealth gains over the long-run would be nil, myopic consumers on the other hand could vary their consumption directly in response to these temporary ‘wealth gains’

Second, higher house prices would benefit existing homeowners if there are ways to withdraw housing equity for consumption This channel, known as collateral enhancement/balance sheet effect,

is operative only if the mortgage markets are sufficiently well developed In less developed mortgage markets where withdrawing housing equity is harder, the collateral enhancement effect would be negligible

Third, the impact of house price changes on consumers is also dependent on whether the changes are temporary or permanent Temporary changes in house prices may produce little effect on consumption compared to permanent changes

Fourth, there may be wealth gains among those trading houses in an environment of rising house prices If trading takes place within the set of existing housing stock (i.e constant housing stock), the number of buyers must be matched by the same number of willing sellers failing which equilibrium would not be achieved Since buyers and sellers are equal in number, losses suffered by buyers of more expensive houses would be exactly offset by the gains reaped by sellers so that the net effect on wealth among those transacting in houses is nil (Edelstein and Lum, 2004)

However, if there is substantial change in housing stock, this argument may not necessarily hold Particularly in developing economies where residential property markets have yet to mature, rapid urbanisation, as seen from the growth rate of the urban population, gives rise to sharp increases in new demand for housing As current urban settlers are unlikely to sell their homes and move out given ample economic opportunities in the urban areas, the number of households wanting to buy would vastly exceed the number of households willing to sell in the secondary market, necessitating large expansions of housing stock to meet the excess demand Ceteris paribus, since there are more buying households (losers) than there are households willing to sell (gainers), housing price increases would cause a net loss among households transacting in houses, yielding a negative link between house price and private consumption

Finally, demographic factors can also explain the negative link from house price to private consumption in the absence of collateral enhancement effects Typically, households who up-grade and buy houses for the first time consist mainly of working adults with young families In contrast, households trading down are constituted mostly by retirees It is a known fact that house price increases would make the former worse off while benefiting the latter Thus, a country with a larger working adult population relative to retirees would have stronger negative wealth effects from rising house prices 2

The interaction of these factors makes it difficult to ascertain the net wealth effect of house price increases on consumption Recent works by Case et al (2005) and Ludwig and Slok (2004) support the claim that housing wealth effect is positively related to consumption in the US and OECD economies However, Phang (2004) and Peng et al (2008) fail to detect such positive links for Singapore and China, respectively

2 Number of households trading up and buying houses for the first time need not equal number of households trading down

4 Empirical Models

We intend to assess the real impacts of property market developments in Malaysia Given our

discussion on theoretical framework in the previous section, our research strategy follows a two-step

procedure:

• In step 1, we model the long-run effects of property prices on consumption and investment

spending, respectively If both the investment and wealth effect channels are operative and

respond positively to property price fluctuations, property price would likely be a driving factor

for aggregate demand and real GDP, a statement which needs validation This leads us to step 2

• In step 2 we test if property price drives domestic demand and real GDP

However, if it turns out that the two channels are not operative, or if each channel responds in a

qualitatively different manner to changes in property prices, the net impact of property prices on

demand and real GDP would be weak or non-existent We would then expect to find that property price

does not drive real GDP in this case Thus, the two strategies tend to reinforce one another

4.1 Modelling consumption and investment channels

4.1.1 Investment channel

The model of investment channel captures two types of impacts of property price changes on

investment spending, namely the Tobin-q effect and the financial accelerator principle To test the

existence of investment channel, a model of investment spending is specified and estimated, with

property price as an explanatory variable The choice of control variables is influenced by the

investment literature particularly Acosta and Loza (2005) and Ang (2007), with the latter bearing more

influence on the model formulation here Hence, the baseline long-run investment function is specified

as follows:

t t

t t

t

t GDP UCC FC HP UNC

where

I= Real gross fixed capital formation

GDP = Real Gross Domestic Product

UCC = Real user cost of capital

FC = Financial constraints

HP = Real property price

UNC = Macroeconomic uncertainty

The investment channel suggests that the sign on β4 should be positive The inclusion of GDP

and user cost of capital is consistent with the neoclassical framework of Jorgensen (1963), which

suggests that investment varies directly with output, but inversely with user cost of capital Hence, β1 is

hypothesised to take a positive sign whereas β2 would take a negative value As noted by Ang (2007),

financial constraints are important to firms in a developing country such as Malaysia We use stock

prices as a proxy for financial constraints Since more robust stock price tends to ease firms’ access to

financing, β3 is hypothesised to take a positive value Macroeconomic uncertainty is also added into the

model because higher uncertainty is reflected in terms of lower investment To the extent that price

instability is one source of uncertainty, we use inflation to proxy uncertainty Hence we expect β5 to be

negative

4.1.2 Wealth effect channel

To test the wealth effects of housing price changes on consumption, a long-run consumption function

is specified in the following:

t t

t t

where

C = Real private consumption

DY = Aggregate real disposable income

SMP = Real stock market price

HMP = Real property price

IR = Real average lending rate of commercial banks

Notably, property and stock prices are proxies for household wealth 3 whereas disposable

income is the proxy for labour income We have also included interest rate as another independent

variable, as what Phang (2004) has done It is important to control for the effects of interest rates since

this may be a common factor driving both house prices and consumption In this conventional model of

consumption behaviour, we expect α1 to be positive because larger disposable incomes encourage more

consumption We expect the sign of α2 to be positive For households investing in the stock market

over the long-term, they make profits not from capital gains but from dividends To the extent that

higher stock prices reflect better corporate performance and dividend payouts, households would be

able to enjoy wealth gains which can be used to finance higher spending Since interest rate represents

cost of credit, α4 should have a negative sign 4 However, the sign on α3 is ambiguous for reasons

discussed in the previous section

4.1.3 Estimating the consumption and investment channels

The investment and wealth effect channels (models (1)-(2)) can be estimated using the Autoregressive

Distributed Lag (ARDL) and bounds testing approach to cointegration introduced by Pesaran, Shin and

Smith (2001) This approach to cointegration is superior to those of Engle and Granger (1987), and

Johansen and Juselius (1990) for two reasons Firstly, the approach particularly suitable for research

involving small samples Second, this approach can be adopted to examine the presence of

cointegration among the underlying variables regardless of whether the underlying variables are I(0),

I(1) or mutually cointegrated The second advantage dispenses with the need for pre-testing the order

of integration since most macroeconomic time series are either I(0) or I(1) The bounds testing

procedure can be applied even when the explanatory variables in models (1)-(2) are endogenous (Tang,

2004) Hence, the presence of endogenous regressors would not invalidate the estimation procedure

The ARDL and bounds testing approach to cointegration starts with tests for the presence of

long-run (cointegrating) relationships in models (1)-(2) To conduct this test, a set of unrestricted error

correction models (UECM) of the following form is estimated:

+ Δ

+ Δ

+ Δ +

=

i

i t h N

i

i t f N

i

i t u

N i

i t g

N i

i t I

t a a I a GDP a UCC a FC a HP

I

i i

i i

i

1 1

1 1

1 0

t t t

t t

t t

++

++

+Δ

+ Δ + Δ +

=

i

i t r N

i

i t h N

i

i t s N

i

i t I N

i

i c

t b b C b DY b SMP b HP b IR

C

i i

i i

i

1 1

1 1

1 0

t t t

t t

C

b1 1 + 2 1+ 3 1+ 4 1 + 5 1+υ2

Equation (3) is set up to test whether cointegration exists between the variables in model (1)

Likewise, equations (4) set up to test whether cointegration exists between the variables in model (2)

In equation (3), the null hypothesis of no cointegration amongst the variables in model is H0: a1=a2=a3=

a4=a5=a6=0 against the alternative hypothesis of H1: a1≠a2≠a3≠a4≠a5≠0 In equation (4), the null

hypothesis of no cointegration amongst the variables in model (4) is H0: b1=b2=b3=b4=b5=0 against the

alternative hypothesis of H1: b1≠b2≠b3≠b4≠b5≠0 Before the bounds test can be conducted, the lag order

(i.e value of N) of each UECM has to be determined To accomplish this task, the approach taken by

Lee (2008) is adopted here i.e a sufficient number of lags (N) in the first differences are added in order

3 According to Zhu (2003) and Phang (2004), changes in asset prices affect financial wealth, which in turn affects

consumption So, asset prices can be use as proxy for wealth In Ludwig and Slok (2004), the authors have used house

price and stock price to proxy for housing and stock market wealth respectively

4 Higher interest rate reduces demand for consumer credit to purchase durable goods

that the disturbance terms in equations (3)-(4) do not have autocorrelation up to lag order of 2,

according to the Breusch-Godfrey Lagrange Multiplier (LM) test The chosen value of N is the lowest

value when the test is unable to reject the null hypothesis of no autocorrelation at 5% level of

significance

For a given level of significance, the critical values in the bounds test consist of a lower and

upper bound The critical value bounds depend on the structure of UECM being used In our case, we

have adopted the ‘unrestricted intercept and no trend’ structure so that the critical value bounds would

be taken from Pesaran et al’s (2001) Table CI(iii) Other studies in the literature which have chosen the

same model structure include Tang (2004), Liang and Cao (2007), Ho and Wong (2003) and Lee

(2008) There is evidence to reject the null of no cointegration if the F-statistic exceeds the upper

bound critical value On the other hand, the null of no cointegration is not rejected if the F-statistic is

smaller than the lower bound critical value Ambiguity arises if the F-statistic lies between the upper

and lower bound, in which case one cannot conclude whether cointegration exists until the order of

integration for the variables are established using unit root tests 5

After the presence of cointegration is found to exist in all relationships, (1)-(2) are then

estimated as ARDL models The ARDL (p, q1, q2,…,qk) model has the following general structure 6:

t t k

i

t i x t

,()

−Φ

q iq i

i t i

L is a lag operator such that Lyt = yt-1 while wt is an sx1 vector of deterministic variables

including dummies, trends and other exogenous variables The estimated ARDL models can be

re-parameterised to obtain the long-run coefficients in the respective cointegrating relationships as well as

their error correction representations (Pesaran and Pesaran, 1997) The magnitude and sign on the

estimated coefficients can subsequently be interpreted, as what had been done in most studies

involving the application of the ARDL and bounds testing procedure (see for instance, Narayan and

Smyth, 2006, Liang and Cao, 2007 and Ho and Wong, 2003)

4.1.4 Testing the impact of property prices on domestic demand and GDP

After estimating the consumption and investment channels, step 2 of our research involves testing

whether total expenditure and real GDP are driven by fluctuations in property prices

4.1.4.1 The property priceÆdomestic demand link

Following the convention in Ho and Wong (2003, 2008), we define domestic demand or expenditure as

the sum of gross investment and private consumption 7 Given the determinants of consumption and

gross investment spending in equations (1) and (2), a model on determinants of domestic demand

(DEM) can be obtained:

t t

t t

t t

t

Equation (7) is estimated using the ARDL and bounds testing procedure similar to equations (1)

and (2) Particularly, we specify (7) in the following UECM:

5 In this paper, we can dispense with the need to do pre-testing for unit roots, given the advantages of the bounds testing

approach which can be used irrespective of whether the underlying variables are I(1) or I(0) This convention follows Ho

and Wong (2003) and Lee (2008)

6 For more details on ARDL models, interested readers can refer to Pesaran et al (2001)

7 In other words, components of demand which are directly affected by property price

+ Δ

+ Δ

i

i f N

i

i T

N i

i g

N i

i de

t c c DEM c GDP c Tax c FC c HP DEM

i i

i i

i

1 1

1 1

1 0

1 3 1 2 1 1

1 1

+Δ+

Δ+

i

i t u

N

i

i t

c

i i

i

t t t

t t

FC

c4 1+ 5 1+ 6 1+ 7 1+ 8 1+υ1

The null hypothesis of no cointegration amongst the variables in model is H0:

c1=c2=c3=c4=c5=c6=c7=c8=0 against the alternative hypothesis of H1: c1≠c2≠c3≠c4≠c5≠ c6≠c7≠c8≠0 The

bounds test procedure is the same as that used in testing cointegration in equations (3)-(4) If

cointegration is detected in (8), we estimate (7) as an ARDL model and re-parameterise the coefficients

to obtain the long-run coefficients and short-run dynamics

We are interested in the statistical significance of the HP coefficient If property price booms

strongly and significantly increase both consumption and investment, the net impact on total domestic

spending (DEM) would also be significantly positive However, if both channels are not operative,

property prices would have no significant effect on DEM

4.1.4.2 Testing property priceÆreal GDP link

To assess whether property price is a driving factor for real GDP fluctuations, we first test if property

price and GDP are cointegrated when GDP is the dependent variable To test for cointegration, we

employ the ARDL and bounds testing procedure again Particularly, we set a similar UECM just like

what we have done in equation (3), (4) and (8):

1 2 1 1 1

Δ + Δ

+

=

N i

i t h N

i

i t g

t d d GDP d HP d GDP d HP

GDP

i

For equation (9), the null hypothesis of no cointegration amongst the variables in model is H0:

d1=d2=0 against the alternative hypothesis of H1: d1≠d2≠0 If cointegration is detected, we proceed to

estimate the GDP and property price link as an ARDL model

If property prices booms drive up consumption and investment strongly, this would also lead to

an unambiguous and significant increase in real GDP via increases in domestic demand The results of

the test can be used as a basis for conducting test on whether there is long-run and short-run causality

in a Granger sense running from property prices to GDP 8

5 Data Sources and Definitions of Variables

Data for all variables are taken from various issues of Bank Negara Malaysia’s Monthly Statistical

Bulletin Property price is measured by the Malaysian House Price Index (MHPI), published by the

Property and Valuation Services Department under the Ministry of Finance Data are quarterly, and

span 1991Q1-2006Q2 Table 1 summarises the definitions of variables All variables are expressed in

natural logs, except UNC, IR and SPREAD which can take negative values

8 Studies on Granger causality which carried out similar procedures include Liang and Cao (2007) and Lee (2008)

Table 1: Variables and definitions

Variables Definition

1 I a/ Gross fixed capital formation at constant 1987 prices

2 GDP Gross domestic product at constant 1987 prices

X 1-corporate income tax deflated by producer price index (1989=100)

4 HP Malaysian house price index (hereafter, MHPI), deflated by producer price index (1989=100) Quarterly data only became available beginning 1999 Hence, data prior to

1999 were interpolated using cubic-spline method c/

5 FC Kuala Lumpur Composite Index (hereafter, KLCI), deflated by producer price index (1989=100)

6 UNC Producer Price Index inflation

7 C d/ Private consumption at constant 1987 prices

8 DY Disposable income at constant 1987 prices

9 SMP Kuala Lumpur Composite Index (hereafter, KLCI), deflated by producer price index (1989=100)

10 IR Average lending rate of commercial banks, adjusted for producer price index inflation

11 DEM Sum of gross fixed capital formation at constant 1987 prices and private consumption at constant 1987 prices

12 Tax Sum of corporate and individual income taxes deflated by producer price index (1989=100)

a/ This aggregate captures both public and private sector investment in fixed assets

b/ The definition of nominal user cost of capital was taken from Ang (2007) In particular, Ang (2007) assumes rate of depreciation to be 5% whereas price of capital is the gross capital formation deflator In the sample, rate of corporate income tax was 35% from 1991 to 1992, 32% from 1993 to 1995, 30% from 1996 to 1997 and 28% from 1998-2006

c/ The MHPI is the only property price index available for Malaysia It is important to note that the interpolated MHPI data does not obviously contain more information than the original annual data Hence, interpolation merely offers suggestions

as to how the missing quarterly time series may look like One objection to using interpolated data is that the constructed time series seems smooth and devoid of short-term volatility Despite this shortcoming, the interpolated series gives a reasonably good depiction of the actual behaviour of MHPI, because MHPI is in reality a relatively smooth index as well This smoothness is attributed to the characteristics of the housing market such as infrequent trading (Hilbers et al, 2001), lack of short-selling/short-term speculation (Davis and Zhu, 2004), the long-term nature of the market (Wang, 2001) and more importantly, valuation smoothing (Davis and Zhu, 2004), especially because the MHPI was constructed using valuations data (Ting, 2003) Other studies which have interpolated annual real estate data to obtain quarterly data include Chen and Patel (1998), Iacoviello (2002), Chirinko, De Haan and Sterken (2004) and Ludwig and Slok (2004)

d/ Aggregate consumption includes both durables and non-durable consumption We did not use data for consumption of durables and non-durables because no such data exists

6 Main Findings and Discussions

We first report findings in step 1 of our research The results of the bounds test for the baseline models, summarised in Table 2, rejects the null hypothesis of no cointegration at both 5% and 10% significance levels Thus, it can therefore be concluded that the consumption and investment functions are cointegrating equations The consumption and investment functions are thus estimated as ARDL models The underlying ARDL model can be re-parameterised to obtain the long-run cointegrating coefficients and a short-run error correction representation Since quarterly data is used, the maximum order of ARDL is set equal to four The most appropriate lag structure is selected using the Schwarz Bayesian Criterion (SBC) (Pesaran and Shin, 1999) Detailed results of the underlying ARDL model for the baseline investment and consumption functions are not reported, but can be produced upon request9

9 Seasonal dummy variables were included in the initial round of the estimation procedure In the process of estimating the consumption function, the F-test on the joint significance of the seasonal dummies was statistically significant at 5% level, indicating that seasonality effects are present Hence, the seasonal dummies were retained in the final model However, while estimating the investment function, seasonal dummies were not statistically significant The inclusion of these dummies in the investment regression also caused model misspecification as detected by the RESET test As such, the seasonal dummies have been dropped from the final investment function

Price of capital